Wireless Sensor Networks Application Centric Design

Wireless sensor netWorks:
ApplicAtioncentric Design
Edited by Dr. geof V Merret and
Dr. Yen kheng tan (Editor-in-Chief)
Wireless Sensor Networks: Application-Centric Design
Edited by Dr. Geoff V Merret and Dr. Yen Kheng Tan (Editor-in-Chief)
Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia
Copyright 2010 InTech
All chapters are Open Access articles distributed under the Creative Commons
Non Commercial Share Alike Attribution 3.0 license, which permits to copy,
distribute, transmit, and adapt the work in any medium, so long as the original
work is properly cited. After this work has been published by InTech, authors
have the right to republish it, in whole or part, in any publication of which they
are the author, and to make other personal use of the work. Any republication,
referencing or personal use of the work must explicitly identify the original source.
Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted
for the accuracy of information contained in the published articles. The publisher
assumes no responsibility for any damage or injury to persons or property arising out
of the use of any materials, instructions, methods or ideas contained in the book.

Publishing Process Manager Jelena Marusic
Technical Editor Goran Bajac
Cover Designer Martina Sirotic
Image Copyright Ruslan Ivantsova, 2010. Used under license from Shutterstock.com
First published December, 2010
Printed in India
A free online edition of this book is available at www.intechopen.com
Additional hard copies can be obtained from [email protected]
Wireless Sensor Networks: Application-Centric Design, Edited by Dr. Geoff V Merret
and Dr. Yen Kheng Tan (Editor-in-Chief)
p. cm.
ISBN 978-953-307-321-7
free online editions of InTech
Books and Journals can be found at
www.intechopen.com
Part 1
Chapter 1
Chapter 2
Chapter 3
Chapter 4
Chapter 5
Chapter 6
Chapter 7
Chapter 8
Preface IX
Applications and Case Studies 1
Wireless Sensor Networks - An Introduction 3
Qinghua Wang and Ilangko Balasingham
Wireless Sensor Networks for On-field
Agricultural Management Process 17
Luca Bencini, Davide Di Palma, Giovanni Collodi,
Gianfranco Manes and Antonio Manes
Wildlife Assessment using Wireless Sensor Networks 35
Harry Gros-Desormeaux, Philippe Hunel and Nicolas Vidot
Wireless Sensor Network for Disaster Monitoring 51
Dr. Maneesha Vinodini Ramesh
Urban Microclimate and Traffic Monitoring
with Mobile Wireless Sensor Networks 71
Francesco Chiti and Romano Fantacci
Improving Greenhouses Automation and
Data Acquisition with Mobile Robot Controlled
system via Wireless Sensor Network 85
Istvn Matijevics and Simon Jnos
Model Based WSN System Implementations Using PN-WSNA
for Aquarium Environment Control in a House 109
Ting-Shuo Chen and Chung-Hsien Kuo
Wireless Sensor Network for Ambient Assisted Living 127
Juan Zapata and Francisco J. Fernndez-Luque and Ramn Ruiz
Contents
Contents VI
Monitoring of human movements for fall
detection and activities recognition in elderly
care using wireless sensor network: a survey 147
Stefano Abbate, Marco Avvenuti, Paolo Corsini,
Alessio Vecchio and Janet Light
Odor Recognition and Localization
Using Sensor Networks 167
Rabie A. Ramadan
Communication and Networking Technologies 183
Modelling Underwater Wireless Sensor Networks 185
Jess Llor and Manuel P. Malumbres
Prospects and Problems of Optical Diffuse
Wireless Communication for Underwater
Wireless Sensor Networks (UWSNs) 205
Davide Anguita, Davide Brizzolara and Giancarlo Parodi
Estimation of Propagation Characteristics along
Random Rough Surface for Sensor Networks 231
Kazunori Uchida and Junichi Honda
Design of Radio-Frequency Transceivers
for Wireless Sensor Networks 249
Bo Zhao and Huazhong Yang
MAC & Mobility In Wireless Sensor Networks 271
Marwan Al-Jemeli, Vooi Voon Yap and Fawnizu Azmadi Bin Hussin
Hybrid Optical and Wireless Sensor Networks 297
Lianshan Yan, Xiaoyin Li, Zhen Zhang, Jiangtao Liu and Wei Pan
Range-free Area Localization Scheme
for Wireless Sensor Networks 321
Vijay R. Chandrasekhar, Winston K.G. Seah,
Zhi Ang Eu and Arumugam P. Venkatesh
Chapter 9
Chapter 10
Part 2
Chapter 11
Chapter 12
Chapter 13
Chapter 14
Chapter 15
Chapter 16
Chapter 17
Contents VII
Information and Data Processing Technologies 351
Data Fusion Approach for Error
Correction in Wireless Sensor Networks 353
Maen Takruri and Subhash Challa
Target Tracking in Wireless Sensor Networks 373
Jianxun LI and Yan ZHOU
A Gaussian Mixture Model-based Event-Driven Continuous
Boundary Detection in 3D Wireless Sensor Networks 393
Jiehui Chen, Mariam B.Salim and Mitsuji Matsumoto
Monitoring Wireless Sensor Network
Performance by Tracking Node operational Deviation 413
Yaqoob J. Y. Al-raisi and Nazar E. M. Adam
Building Context Aware Network of Wireless
Sensors Using a Scalable Distributed Estimation
Scheme for Real-time Data Manipulation 427
Amir Hossein Basirat and Asad I. Khan
Multimedia Data Processing and
Delivery in Wireless Sensor Networks 449
Javier Molina, Javier M. Mora-Merchan,
Julio Barbancho and Carlos Leon
Imaging in UWB Sensor Networks 469
Ole Hirsch, Rudolf Zetik, and Reiner S. Thom
Part 3
Chapter 18
Chapter 19
Chapter 20
Chapter 21
Chapter 22
Chapter 23
Chapter 24
About this Book
Over the past decade, there has been a prolifc increase in the research, development
and commercialisation of Wireless Sensor Networks (WSNs) and their associated tech-
nologies (see Figure 1). This rise has been a result of a number of contributing factors,
including continued miniaturisation (leading towards an era of truly pervasive and
invisible computing); low-power circuits, devices and computation (for example, the
ultra-low-power sleep states now found in microcontrollers); and efcient short-range
communication (such as ZigBee and Bluetooth). The dramatic rise in WSN activity,
fuelled by the prospect of a new computing paradigm, has resulted in the topic being
researched (and taught) in the electronics and computer science departments of Uni-
versities around the world.
Figure 1. The increase in research into WSNs, shown by the total number of pub-
lished papers (as catalogued on the ISI Web of Knowledge) matching the topic (sensor
network*).
Preface
Preface X
While enabling technologies such as low-power circuitry have permited the concep-
tion and growth of WSNs (for example a microcontrollers ultra-low-power sleep states
enable a vast reduction in the average power consumption obtained through duty cy-
cled operation, a technique which underpins the operation of most implementations),
the principal reason for the explosion of research is, in my opinion, due to the volume of
WSN applications that can be conceived and realised. To name a few, they have found
use in healthcare, defence and security, environmental monitoring, process control,
structural health monitoring, condition monitoring, building automation, multimedia
provision and advertising. However, as a result of the broad array of pertinent appli-
cations, WSN researchers have also realised the application specifcity of the domain;
it is incredibly difcult, if not impossible, to fnd an application-independent solution
to most WSN problems (be it a routing algorithm, MAC protocol, energy harvesting
architecture, or data processing algorithm). Hence, research into WSNs dictates the
adoption of an application-centric design process.
Research into WSN applications not only concerns the technical issues and system in-
tegration strategies of deployment, but also the communication and processing of data,
alongside the analysis, understanding and modelling of the application parameters
that are of interest. As such, this book is not intended to be a comprehensive review
of all WSN applications and deployments to date. Instead, it is a collection of state-of-
the-art research papers discussing current applications and deployment experiences,
but also the communication and data processing technologies that are fundamental in
further developing solutions to applications. Whilst a common foundation is retained
through all chapters, this book contains a broad array of ofen difering interpretations,
confgurations and limitations of WSNs. I believe that these aid to highlight the rich
diversity and sheer scale of this ever-changing research area.
Organisation
The chapters of this book have been categorised into three distinct sections: applica-
tions and case studies (section A), communication and networking technologies (sec-
tion B), and information and data processing technologies (section C). These are de-
scribed below:
Applications and Case Studies: The frst chapter of this section serves as an introduc-
tion

to the book, providing a concise overview of WSNs, discussing their history, plat-
forms and architectures, research challenges, and application. The remainder of the
section discusses current applications and their implementation, from experiences of
monitoring agricultural processes to the diferent methods by which accidental falls of
the elderly can be detected and classifed.
Contents VII
Communication and Networking Technologies: Alongside the predominant theme of
this book, this section contains a collection of state-of-the-art technical papers on ap-
plication-specifc communication and networking problems. These include modelling
wireless propagation through water, routing strategies for hybrid radio frequency/fbre
optic WSNs, through to an overview of radio transceiver design.
Information and Data Processing Technologies: The fnal section of this book provides
an insight into current research on information and data processing, including data
fusion, target tracking, fault tolerance and multimedia provision in WSNs.
The readership of this book is intended to be postgraduate/postdoctoral researchers,
and professional engineers. Some of the chapters may also be of interest to masters
level students that are undertaking modules that are particularly relevant to this feld.
Dr. Geof V Merret and Dr. Yen Kheng Tan (Editor-in-Chief)
Applications and Case Studies
Part 1
Applications and Case Studies
Wireless Sensor Networks - An Introduction
Qinghua Wang and Ilangko Balasingham
0
Wireless Sensor Networks - An Introduction
Qinghua Wang
Dept. of Electronics and Telecommunications
Norwegian University of Science and Technology
Norway
Ilangko Balasingham
Dept. of Electronics and Telecommunications
Norwegian University of Science and Technology
The Interventional Centre, Oslo University Hospital
Institute of Clinical Medicine, University of Oslo
Norway
This chapter provides a detailed introduction to the history and current state of the art with
regard to wireless sensor networks (WSNs).
1. History
The origins of the research on WSNs can be traced back to the Distributed Sensor Networks
(DSN) programat the Defense Advanced Research Projects Agency (DARPA) at around 1980.
By this time, the ARPANET (Advanced Research Projects Agency Network) had been op-
erational for a number of years, with about 200 hosts at universities and research institutes
(Chong & Kumar, 2003). DSNs were assumed to have many spatially distributed low-cost
sensing nodes that collaborated with each other but operated autonomously, with informa-
tion being routed to whichever node was best able to use the information. At that time, this
was actually an ambitious program. There were no personal computers and workstations;
processing was mainly performed on minicomputers and the Ethernet was just becoming
popular (Chong & Kumar, 2003). Technology components for a DSN were identied in a
Distributed Sensor Nets workshop in 1978 (Proceedings of the Distributed Sensor Nets Work-
shop, 1978). These included sensors (acoustic), communication and processing modules, and
distributed software. Researchers at Carnegie Mellon University (CMU) even developed a
communication-oriented operating system called Accent (Rashid & Robertson, 1981), which
allowed exible, transparent accesss to distributedresources requiredfor a fault-tolerant DSN.
A demonstrative application of DSN was a helicopter tracking system (Myers et al., 1984), us-
ing a distributed array of acoustic microphones by means of signal abstractions and matching
techniques, developed at the Massachusetts Institute of Technology (MIT).
Even though early researchers on sensor networks had in mind the vision of a DSN, the tech-
nology was not quite ready. More specically, the sensors were rather large (i.e. shoe box and
This work was carried out during the tenure of an ERCIM Alain Bensoussan Fellowship Pro-
gramme and is part of the MELODY Project, which is funded by the Research Council of Norway under
the contract number 187857/S10.
1
Wireless Sensor Networks: Application-Centric Design 4
up) which limited the number of potential applications. Further, the earliest DSNs were not
tightly associated with wireless connectivity. Recent advances in computing, communication
and microelectromechanical technology have caused a signicant shift in WSN research and
brought it closer to achieving the original vision. The newwave of research in WSNs started in
around 1998 and has been attracting more and more attention and international involvement.
In the new wave of sensor network research, networking techniques and networked informa-
tion processing suitable for highly dynamic ad hoc environments and resource-constrained
sensor nodes have been the focus. Further, the sensor nodes have been much smaller in size
(i.e. pack of cards to dust particle) and much cheaper in price, and thus many new civilian
applications of sensor networks such as environment monitoring, vehicular sensor network
and body sensor network have emerged. Again, DARPA acted as a pioneer in the new wave
of sensor network research by launching an initiative research program called SensIT (Kumar
& Shepherd, 2001) which provided the present sensor networks with new capabilities such
as ad hoc networking, dynamic querying and tasking, reprogramming and multi-tasking. At
the same time, the IEEE noticed the low expense and high capabilities that sensor networks
offer. The organization has dened the IEEE 802.15.4 standard (IEEE 802.15 WPAN Task Group
4, n.d.) for low data rate wireless personal area networks. Based on IEEE 802.15.4, ZigBee
Alliance (ZigBee Alliance, n.d.) has published the ZigBee standard which species a suite of
high level communication protocols which can be used by WSNs. Currently, WSN has been
viewed as one of the most important technologies for the 21st century (21 Ideas for the 21st Cen-
tury, 1999). Countries such as China have involved WSNs in their national strategic research
programmes (Ni, 2008). The commercialization of WSNs are also being accelerated by new
formed companies like Crossbow Technology (Crossbow Technology, n.d.) and Dust Networks
(Dust Networks, Inc., n.d.).
2. Hardware Platform
A WSN consists of spatially distributed sensor nodes. In a WSN, each sensor node is able to
independently perform some processing and sensing tasks. Furthermore, sensor nodes com-
municate with each other in order to forward their sensed information to a central processing
unit or conduct some local coordination such as data fusion. One widely used sensor node
platform is the Mica2 Mote developed by Crossbow Technology (Crossbow Technology, n.d.).
The usual hardware components of a sensor node include a radio transceiver, an embedded
processor, internal and external memories, a power source and one or more sensors.
2.1 Embedded Processor
In a sensor node, the functionality of an embedded processor is to schedule tasks, process
data and control the functionality of other hardware components. The types of embedded
processors that can be used in a sensor node include Microcontroller, Digital Signal Processor
(DSP), Field Programmable Gate Array (FPGA) and Application-Specic Integrated Circuit
(ASIC). Among all these alternatives, the Microcontroller has been the most used embedded
processor for sensor nodes because of its exibility to connect to other devices and its cheap
price. For example, the newest CC2531 development board provided by Chipcon (acquired
by Texas Instruments) uses 8051 microcontroller, and the Mica2 Mote platform provided by
Crossbow uses ATMega128L microcontroller.
2.2 Transceiver
A transceiver is responsible for the wireless communication of a sensor node. The various
choices of wireless transmission media include Radio Frequency (RF), Laser and Infrared. RF
based communication ts to most of WSNapplications. The operational states of a transceiver
are Transmit, Receive, Idle and Sleep. Mica2 Mote uses two kinds of RF radios: RFM TR1000
and Chipcon CC1000. The outdoor transmission range of Mica2 Mote is about 150 meters.
2.3 Memory
Memories in a sensor node include in-chip ash memory and RAM of a microcontroller and
external ash memory. For example, the ATMega128L microcontroller running on Mica2
Mote has 128-Kbyte ash programmemory and 4-Kbyte static RAM. Further, a 4-Mbit Atemel
AT45DB041B serial ash chip can provide external memories for Mica and Mica2 Motes (Hill,
2003).
2.4 Power Source
In a sensor node, power is consumed by sensing, communication and data processing. More
energy is required for data communication than for sensing and data processing. Power can
be stored in batteries or capacitors. Batteries are the main source of power supply for sensor
nodes. For example, Mica2 Mote runs on 2 AA batteries. Due to the limited capacity of bat-
teries, minimizing the energy consumption is always a key concern during WSN operations.
To remove the energy constraint, some preliminary research working on energy-harvesting
techniques for WSNs has also been conducted. Energy-harvesting techniques convert ambi-
ent energy (e.g. solar, wind) to electrical energy and the aim is to revolutionize the power
supply on sensor nodes. A survey about the energy-harvesting sensor nodes is provided by
(Sudevalayam & Kulkarni, 2008).
2.5 Sensors
A sensor is a hardware device that produces a measurable response signal to a change in a
physical condition such as temperature, pressure and humidity. The continual analog signal
sensed by the sensors is digitized by an analog-to-digital converter and sent to the embedded
processor for further processing. Because a sensor node is a micro-electronic device powered
by a limited power source, the attached sensors should also be small in size and consume
extremely low energy. A sensor node can have one or several types of sensors integrated in or
connected to the node.
3. Operating System
The role of any operating system (OS) is to promote the development of reliable application
software by providing a convenient and safe abstraction of hardware resources. OSs for WSN
nodes are typically less complex than general-purpose OSs both because of the special re-
quirements of WSN applications and because of the resource constraints in WSN hardware
platforms.
TinyOS (TinyOS Community Forum, n.d.) is perhaps the rst operating system specically de-
signed for WSNs. It features a component-based architecture which enables rapid innovation
and implementation while minimizing code size as requiredby the severe memory constraints
inherent in WSNs. TinyOSs component library includes network protocols, distributed ser-
vices, sensor drivers, and data acquisition tools - all of which can be further rened for a
up) which limited the number of potential applications. Further, the earliest DSNs were not
tightly associated with wireless connectivity. Recent advances in computing, communication
and microelectromechanical technology have caused a signicant shift in WSN research and
brought it closer to achieving the original vision. The newwave of research in WSNs started in
around 1998 and has been attracting more and more attention and international involvement.
In the new wave of sensor network research, networking techniques and networked informa-
tion processing suitable for highly dynamic ad hoc environments and resource-constrained
sensor nodes have been the focus. Further, the sensor nodes have been much smaller in size
(i.e. pack of cards to dust particle) and much cheaper in price, and thus many new civilian
applications of sensor networks such as environment monitoring, vehicular sensor network
and body sensor network have emerged. Again, DARPA acted as a pioneer in the new wave
of sensor network research by launching an initiative research program called SensIT (Kumar
& Shepherd, 2001) which provided the present sensor networks with new capabilities such
as ad hoc networking, dynamic querying and tasking, reprogramming and multi-tasking. At
the same time, the IEEE noticed the low expense and high capabilities that sensor networks
offer. The organization has dened the IEEE 802.15.4 standard (IEEE 802.15 WPAN Task Group
4, n.d.) for low data rate wireless personal area networks. Based on IEEE 802.15.4, ZigBee
Alliance (ZigBee Alliance, n.d.) has published the ZigBee standard which species a suite of
high level communication protocols which can be used by WSNs. Currently, WSN has been
viewed as one of the most important technologies for the 21st century (21 Ideas for the 21st Cen-
tury, 1999). Countries such as China have involved WSNs in their national strategic research
programmes (Ni, 2008). The commercialization of WSNs are also being accelerated by new
formed companies like Crossbow Technology (Crossbow Technology, n.d.) and Dust Networks
(Dust Networks, Inc., n.d.).
2. Hardware Platform
A WSN consists of spatially distributed sensor nodes. In a WSN, each sensor node is able to
independently perform some processing and sensing tasks. Furthermore, sensor nodes com-
municate with each other in order to forward their sensed information to a central processing
unit or conduct some local coordination such as data fusion. One widely used sensor node
platform is the Mica2 Mote developed by Crossbow Technology (Crossbow Technology, n.d.).
The usual hardware components of a sensor node include a radio transceiver, an embedded
processor, internal and external memories, a power source and one or more sensors.
2.1 Embedded Processor
In a sensor node, the functionality of an embedded processor is to schedule tasks, process
data and control the functionality of other hardware components. The types of embedded
processors that can be used in a sensor node include Microcontroller, Digital Signal Processor
(DSP), Field Programmable Gate Array (FPGA) and Application-Specic Integrated Circuit
(ASIC). Among all these alternatives, the Microcontroller has been the most used embedded
processor for sensor nodes because of its exibility to connect to other devices and its cheap
price. For example, the newest CC2531 development board provided by Chipcon (acquired
by Texas Instruments) uses 8051 microcontroller, and the Mica2 Mote platform provided by
Crossbow uses ATMega128L microcontroller.
2.2 Transceiver
A transceiver is responsible for the wireless communication of a sensor node. The various
choices of wireless transmission media include Radio Frequency (RF), Laser and Infrared. RF
based communication ts to most of WSNapplications. The operational states of a transceiver
are Transmit, Receive, Idle and Sleep. Mica2 Mote uses two kinds of RF radios: RFM TR1000
and Chipcon CC1000. The outdoor transmission range of Mica2 Mote is about 150 meters.
2.3 Memory
Memories in a sensor node include in-chip ash memory and RAM of a microcontroller and
external ash memory. For example, the ATMega128L microcontroller running on Mica2
Mote has 128-Kbyte ash programmemory and 4-Kbyte static RAM. Further, a 4-Mbit Atemel
AT45DB041B serial ash chip can provide external memories for Mica and Mica2 Motes (Hill,
2003).
2.4 Power Source
In a sensor node, power is consumed by sensing, communication and data processing. More
energy is required for data communication than for sensing and data processing. Power can
be stored in batteries or capacitors. Batteries are the main source of power supply for sensor
nodes. For example, Mica2 Mote runs on 2 AA batteries. Due to the limited capacity of bat-
teries, minimizing the energy consumption is always a key concern during WSN operations.
To remove the energy constraint, some preliminary research working on energy-harvesting
techniques for WSNs has also been conducted. Energy-harvesting techniques convert ambi-
ent energy (e.g. solar, wind) to electrical energy and the aim is to revolutionize the power
supply on sensor nodes. A survey about the energy-harvesting sensor nodes is provided by
(Sudevalayam & Kulkarni, 2008).
2.5 Sensors
A sensor is a hardware device that produces a measurable response signal to a change in a
physical condition such as temperature, pressure and humidity. The continual analog signal
sensed by the sensors is digitized by an analog-to-digital converter and sent to the embedded
processor for further processing. Because a sensor node is a micro-electronic device powered
by a limited power source, the attached sensors should also be small in size and consume
extremely low energy. A sensor node can have one or several types of sensors integrated in or
connected to the node.
3. Operating System
The role of any operating system (OS) is to promote the development of reliable application
software by providing a convenient and safe abstraction of hardware resources. OSs for WSN
nodes are typically less complex than general-purpose OSs both because of the special re-
quirements of WSN applications and because of the resource constraints in WSN hardware
platforms.
TinyOS (TinyOS Community Forum, n.d.) is perhaps the rst operating system specically de-
signed for WSNs. It features a component-based architecture which enables rapid innovation
and implementation while minimizing code size as requiredby the severe memory constraints
inherent in WSNs. TinyOSs component library includes network protocols, distributed ser-
vices, sensor drivers, and data acquisition tools - all of which can be further rened for a
custom application. Unlike most other OSs, TinyOS is based on an event-driven program-
ming model instead of multithreading. TinyOS programs are composed into event handlers
and tasks with run-to-completion semantics. When an external event occurs, such as an in-
coming data packet or a sensor reading, TinyOS calls the appropriate event handler to handle
the event. Event handlers can post tasks that are scheduled by the TinyOS kernel at a later
stage. Both the TinyOS system and programs written for TinyOS are written in a special pro-
gramming language called nesC which is an extension of the C programming language. NesC
is designed to detect race conditions between tasks and event handlers. Currently, TinyOS has
been ported to over a dozen platforms and numerous sensor boards. A wide community uses
it in simulation to develop and test various algorithms and protocols. According to the gure
published on TinyOS forum, over 500 research groups and companies are using TinyOS on
the Berkeley/CrossbowMotes. Because TinyOS is open source, numerous groups are actively
contributing code to the development of TinyOS and thus making it even more competitive.
Contiki (Contiki, n.d.) is another open source OS specically designed for WSNs. The Con-
tiki kernel is event-driven, like TinyOS, but the system supports multithreading on a per-
application basis. Furthermore, Contiki includes protothreads that provide a thread-like pro-
gramming abstraction but with a very small memory overhead. Contiki provides IP commu-
nication, both for IPv4 and IPv6. Many key mechanisms and ideas from Contiki have been
widely adopted within the industry. The uIP embedded IP stack, originally released in 2001,
is today used by hundreds of companies in systems such as freighter ships, satellites and oil
drilling equipment. Contikis protothreads, rst released in 2005, have been used in many
different embedded systems, ranging from digital TV decoders to wireless vibration sensors.
Contikis idea of using IP communication in low-power WSNs has lead to an IETF standard
and an international industry alliance - IP for Smart Objects (IPSO) Alliance (IPSO Alliance -
promoting the use of IP for Smart Objects, n.d.).
There are also other OSs that can be used by WSNs. For example, SOS (SOS Embedded Op-
erating System, n.d.) is an event-driven OS for mote-class sensor nodes that adopts a more
dynamic point on the design spectrum. The prime feature of SOS is its support for loadable
modules. A complete system is built from smaller modules, possibly at run-time. To sup-
port the inherent dynamism in its module interface, SOS also focuses on supporting dynamic
memory management. Unfortunately, SOS is no longer under active development due to the
graduation of the core developers. LiteOS (LiteOS, n.d.) is an open source, interactive, UNIX-
like operating system designed for WSNs. With the tools that come from LiteOS, it is possible
to operate one or more WSNs in a Unix-like manner. It is also possible to develop programs
for nodes, and wirelessly distribute such programs to sensor nodes.
4. Networking
4.1 Network Architecture
A WSN is a network consisting of numerous sensor nodes with sensing, wireless communica-
tions and computing capabilities. These sensor nodes are scattered in an unattended environ-
ment (i.e. sensing eld) to sense the physical world. The sensed data can be collected by a few
sink nodes which have accesses to infrastructured networks like the Internet. Finally, an end
user can remotely fetch the sensed data by accessing infrastructured networks. Fig. 1 shows
the operation sketch map of WSNs.
In Fig. 1, two kinds of network topologies are shown. The sensor nodes either form a at
network topology where sensor nodes also act as routers and trasfer data to a sink through
Legend
sensor node
cluster head/
high end sensor node
Sink/
Base Station
User/
Task Manager Node
Internet and
Satellite
(a) Flat sensor network; (b) Clustered sensor network.
Fig. 1. The Operation of WSNs.
multi-hop routing, or a hierarchical network topology where more powerful xed or mobile
relays are used to collect and route the sensor data to a sink.
4.2 Protocol Stack of WSNs
The protocol stack used by the sink, cluster head and sensor nodes are shown in Fig. 2. Ac-
cording to (Akyildiz et al., 2002), the sensor network protocol stack is much like the tradi-
tional protocol stack, with the following layers: application, transport, network, data link,
and physical. The physical layer is responsible for frequency selection, carrier frequency gen-
eration, signal detection, modulation and data encryption. The data link layer is responsible
for the multiplexing of data streams, data frame detection, mediumaccess and error control. It
ensures reliable point-to-point and point-to-multipoint connections in a communication net-
work. The network layer takes care of routing the data supplied by the transport layer. The
network layer design in WSNs must consider the power efciency, data-centric communica-
tion, data aggregation, etc. The transportation layer helps to maintain the data ow and may
be important if WSNs are planned to be accessed through the Internet or other external net-
works. Depending on the sensing tasks, different types of application software can be set up
and used on the application layer.
WSNs must also be aware of the following management planes in order to function efciently:
mobility, power, task, quality of service (QoS) and security management planes. Among
them, the functions of task, mobility and power management planes have been elaborated
in (Akyildiz et al., 2002). The power management plane is responsible for minimizing power
consumption and may turn off functionality in order to preserve energy. The mobility man-
agement plane detects and registers movement of nodes so a data route to the sink is always
maintained. The task management plane balances and schedules the sensing tasks assigned
to the sensing eld and thus only the necessary nodes are assigned with sensing tasks and
the remainder are able to focus on routing and data aggregation. QoS management in WSNs
(Howitt et al., 2006) can be very important if there is a real-time requirement with regard to
custom application. Unlike most other OSs, TinyOS is based on an event-driven program-
ming model instead of multithreading. TinyOS programs are composed into event handlers
and tasks with run-to-completion semantics. When an external event occurs, such as an in-
coming data packet or a sensor reading, TinyOS calls the appropriate event handler to handle
the event. Event handlers can post tasks that are scheduled by the TinyOS kernel at a later
stage. Both the TinyOS system and programs written for TinyOS are written in a special pro-
gramming language called nesC which is an extension of the C programming language. NesC
is designed to detect race conditions between tasks and event handlers. Currently, TinyOS has
been ported to over a dozen platforms and numerous sensor boards. A wide community uses
it in simulation to develop and test various algorithms and protocols. According to the gure
published on TinyOS forum, over 500 research groups and companies are using TinyOS on
the Berkeley/CrossbowMotes. Because TinyOS is open source, numerous groups are actively
contributing code to the development of TinyOS and thus making it even more competitive.
Contiki (Contiki, n.d.) is another open source OS specically designed for WSNs. The Con-
tiki kernel is event-driven, like TinyOS, but the system supports multithreading on a per-
application basis. Furthermore, Contiki includes protothreads that provide a thread-like pro-
gramming abstraction but with a very small memory overhead. Contiki provides IP commu-
nication, both for IPv4 and IPv6. Many key mechanisms and ideas from Contiki have been
widely adopted within the industry. The uIP embedded IP stack, originally released in 2001,
is today used by hundreds of companies in systems such as freighter ships, satellites and oil
drilling equipment. Contikis protothreads, rst released in 2005, have been used in many
different embedded systems, ranging from digital TV decoders to wireless vibration sensors.
Contikis idea of using IP communication in low-power WSNs has lead to an IETF standard
and an international industry alliance - IP for Smart Objects (IPSO) Alliance (IPSO Alliance -
promoting the use of IP for Smart Objects, n.d.).
There are also other OSs that can be used by WSNs. For example, SOS (SOS Embedded Op-
erating System, n.d.) is an event-driven OS for mote-class sensor nodes that adopts a more
dynamic point on the design spectrum. The prime feature of SOS is its support for loadable
modules. A complete system is built from smaller modules, possibly at run-time. To sup-
port the inherent dynamism in its module interface, SOS also focuses on supporting dynamic
memory management. Unfortunately, SOS is no longer under active development due to the
graduation of the core developers. LiteOS (LiteOS, n.d.) is an open source, interactive, UNIX-
like operating system designed for WSNs. With the tools that come from LiteOS, it is possible
to operate one or more WSNs in a Unix-like manner. It is also possible to develop programs
for nodes, and wirelessly distribute such programs to sensor nodes.
4. Networking
4.1 Network Architecture
A WSN is a network consisting of numerous sensor nodes with sensing, wireless communica-
tions and computing capabilities. These sensor nodes are scattered in an unattended environ-
ment (i.e. sensing eld) to sense the physical world. The sensed data can be collected by a few
sink nodes which have accesses to infrastructured networks like the Internet. Finally, an end
user can remotely fetch the sensed data by accessing infrastructured networks. Fig. 1 shows
the operation sketch map of WSNs.
In Fig. 1, two kinds of network topologies are shown. The sensor nodes either form a at
network topology where sensor nodes also act as routers and trasfer data to a sink through
Legend
sensor node
cluster head/
high end sensor node
Sink/
Base Station
User/
Task Manager Node
Internet and
Satellite
(a) Flat sensor network; (b) Clustered sensor network.
Fig. 1. The Operation of WSNs.
multi-hop routing, or a hierarchical network topology where more powerful xed or mobile
relays are used to collect and route the sensor data to a sink.
4.2 Protocol Stack of WSNs
The protocol stack used by the sink, cluster head and sensor nodes are shown in Fig. 2. Ac-
cording to (Akyildiz et al., 2002), the sensor network protocol stack is much like the tradi-
tional protocol stack, with the following layers: application, transport, network, data link,
and physical. The physical layer is responsible for frequency selection, carrier frequency gen-
eration, signal detection, modulation and data encryption. The data link layer is responsible
for the multiplexing of data streams, data frame detection, mediumaccess and error control. It
ensures reliable point-to-point and point-to-multipoint connections in a communication net-
work. The network layer takes care of routing the data supplied by the transport layer. The
network layer design in WSNs must consider the power efciency, data-centric communica-
tion, data aggregation, etc. The transportation layer helps to maintain the data ow and may
be important if WSNs are planned to be accessed through the Internet or other external net-
works. Depending on the sensing tasks, different types of application software can be set up
and used on the application layer.
WSNs must also be aware of the following management planes in order to function efciently:
mobility, power, task, quality of service (QoS) and security management planes. Among
them, the functions of task, mobility and power management planes have been elaborated
in (Akyildiz et al., 2002). The power management plane is responsible for minimizing power
consumption and may turn off functionality in order to preserve energy. The mobility man-
agement plane detects and registers movement of nodes so a data route to the sink is always
maintained. The task management plane balances and schedules the sensing tasks assigned
to the sensing eld and thus only the necessary nodes are assigned with sensing tasks and
the remainder are able to focus on routing and data aggregation. QoS management in WSNs
(Howitt et al., 2006) can be very important if there is a real-time requirement with regard to
Application Layer
Transport Layer
Network Layer
Data Link Layer
Physical Layer
Power Management
Mobility Management
Task Management
QoS Management
Security Management
Fig. 2. The Protocol Stack of WSNs (This gure is an extended version of Figure 3 in (Akyildiz
et al., 2002)).
the data services. QoS management also deals with fault tolerance, error control and perfor-
mance optimization in terms of certain QoS metrics. Security management is the process of
managing, monitoring, and controlling the security related behavior of a network. The pri-
mary function of security management is in controlling access points to critical or sensitive
data. Security management also includes the seamless integration of different security func-
tion modules, including encryption, authentication and intrusion detection. Please refer to the
authors publications (Wang & Zhang, 2008; 2009) for more information about security man-
agement in WSNs. It is obvious that networking protocols developed for WSNs must address
all ve of these management planes.
5. Applications
The original motivation behind the research into WSNs was military application. Examples
of military sensor networks include large-scale acoustic ocean surveillance systems for the
detection of submarines, self-organized and randomly deployed WSNs for battleeld surveil-
lance and attaching microsensors to weapons for stockpile surveillance (Pister, 2000). As the
costs for sensor nodes and communication networks have been reduced, many other poten-
tial applications including those for civilian purposes have emerged. The following are a few
examples.
5.1 Environmental Monitoring
Environmental monitoring (Steere et al., 2000) can be used for animal tracking, forest surveil-
lance, ood detection, and weather forecasting. It is a natural candidate for applying WSNs
(Chong & Kumar, 2003), because the variables to be monitored, e.g. temperature, are usu-
ally distributed over a large region. One example is that researchers from the University of
Southampton have built a glacial environment monitoring system using WSNs in Norway
(Martinez et al., 2005). They collect data from sensor nodes installed within the ice and the
sub-glacial sediment without the use of wires which could disturb the environment. An-
other example is that researchers fromEPFL have performed outdoor WSNdeployments on a
rugged high mountain path located between Switzerland and Italy (Barrenetxea et al., 2008).
Their WSNdeployment is used to provide spatially dense measures to the Swiss authorities in
charge of risk management, and the resulting model will assist in the prevention of avalanches
and accidental deaths.
5.2 Health Monitoring
WSNs can be embedded into a hospital building to track and monitor patients and all medi-
cal resources. Special kinds of sensors which can measure blood pressure, body temperature
and electrocardiograph (ECG) can even be knitted into clothes to provide remote nursing for
the elderly. When the sensors are worn or implanted for healthcare purposes, they form a
special kind of sensor network called a body sensor network (BSN). BSN is a rich interdisci-
plinary area which revolutionizes the healthcare system by allowing inexpensive, continuous
and ambulatory health monitoring with real-time updates of medical records via the Internet.
One of the earliest researches on BSNs was conducted in Imperial College London, where a
specialized BSN sensor node and BSN Development Kit have been developed (BSN Research
in Imperial College London, n.d.).
5.3 Trafc Control
Sensor networks have been used for vehicle trafc monitoring and control for some time.
At many crossroads, there are either overhead or buried sensors to detect vehicles and to
control the trafc lights. Furthermore, video cameras are also frequently used to monitor
road segments with heavy trafc. However, the traditional communication networks used to
connect these sensors are costly, and thus trafc monitoring is usually only available at a few
critical points in a city (Chong &Kumar, 2003). WSNs will completely change the landscape of
trafc monitoring and control by installing cheap sensor nodes in the car, at the parking lots,
along the roadside, etc. Streetline, Inc. (Streetline, Inc., n.d.) is a company which uses sensor
network technology to help drivers nd unoccupied parking places and avoid trafc jams.
The solutions provided by Streetline can signicantly improve the city trafc management
and reduce the emission of carbon dioxide.
5.4 Industrial Sensing
As plant infrastructure ages, equipment failures cause more and more unplanned downtime.
The ARC Advisory Group estimates that 5% of production in North America is lost to un-
planned downtime. Because sensor nodes can be deeply embedded into machines and there
is no infrastructure, WSNs make it economically feasible to monitor the health of machines
and to ensure safe operation. Aging pipelines and tanks have become a major problem in the
oil and gas industry. Monitoring corrosion using manual processes is extremely costly, time
consuming, and unreliable. A network of wireless corrosion sensors can be economically de-
ployed to reliably identify issues before they become catastrophic failures. Rohrback Cosasco
Systems (RCS) (Rohrback Cosasco Systems, n.d.) is the world leader in corrosion monitoring
technology and is applying WSNs in their corrosion monitoring. WSNs have also been sug-
Application Layer
Transport Layer
Network Layer
Data Link Layer
Physical Layer
Power Management
Mobility Management
Task Management
QoS Management
Security Management
Fig. 2. The Protocol Stack of WSNs (This gure is an extended version of Figure 3 in (Akyildiz
et al., 2002)).
the data services. QoS management also deals with fault tolerance, error control and perfor-
mance optimization in terms of certain QoS metrics. Security management is the process of
managing, monitoring, and controlling the security related behavior of a network. The pri-
mary function of security management is in controlling access points to critical or sensitive
data. Security management also includes the seamless integration of different security func-
tion modules, including encryption, authentication and intrusion detection. Please refer to the
authors publications (Wang & Zhang, 2008; 2009) for more information about security man-
agement in WSNs. It is obvious that networking protocols developed for WSNs must address
all ve of these management planes.
5. Applications
The original motivation behind the research into WSNs was military application. Examples
of military sensor networks include large-scale acoustic ocean surveillance systems for the
detection of submarines, self-organized and randomly deployed WSNs for battleeld surveil-
lance and attaching microsensors to weapons for stockpile surveillance (Pister, 2000). As the
costs for sensor nodes and communication networks have been reduced, many other poten-
tial applications including those for civilian purposes have emerged. The following are a few
examples.
5.1 Environmental Monitoring
Environmental monitoring (Steere et al., 2000) can be used for animal tracking, forest surveil-
lance, ood detection, and weather forecasting. It is a natural candidate for applying WSNs
(Chong & Kumar, 2003), because the variables to be monitored, e.g. temperature, are usu-
ally distributed over a large region. One example is that researchers from the University of
Southampton have built a glacial environment monitoring system using WSNs in Norway
(Martinez et al., 2005). They collect data from sensor nodes installed within the ice and the
sub-glacial sediment without the use of wires which could disturb the environment. An-
other example is that researchers fromEPFL have performed outdoor WSNdeployments on a
rugged high mountain path located between Switzerland and Italy (Barrenetxea et al., 2008).
Their WSNdeployment is used to provide spatially dense measures to the Swiss authorities in
charge of risk management, and the resulting model will assist in the prevention of avalanches
and accidental deaths.
5.2 Health Monitoring
WSNs can be embedded into a hospital building to track and monitor patients and all medi-
cal resources. Special kinds of sensors which can measure blood pressure, body temperature
and electrocardiograph (ECG) can even be knitted into clothes to provide remote nursing for
the elderly. When the sensors are worn or implanted for healthcare purposes, they form a
special kind of sensor network called a body sensor network (BSN). BSN is a rich interdisci-
plinary area which revolutionizes the healthcare system by allowing inexpensive, continuous
and ambulatory health monitoring with real-time updates of medical records via the Internet.
One of the earliest researches on BSNs was conducted in Imperial College London, where a
specialized BSN sensor node and BSN Development Kit have been developed (BSN Research
in Imperial College London, n.d.).
5.3 Trafc Control
Sensor networks have been used for vehicle trafc monitoring and control for some time.
At many crossroads, there are either overhead or buried sensors to detect vehicles and to
control the trafc lights. Furthermore, video cameras are also frequently used to monitor
road segments with heavy trafc. However, the traditional communication networks used to
connect these sensors are costly, and thus trafc monitoring is usually only available at a few
critical points in a city (Chong &Kumar, 2003). WSNs will completely change the landscape of
trafc monitoring and control by installing cheap sensor nodes in the car, at the parking lots,
along the roadside, etc. Streetline, Inc. (Streetline, Inc., n.d.) is a company which uses sensor
network technology to help drivers nd unoccupied parking places and avoid trafc jams.
The solutions provided by Streetline can signicantly improve the city trafc management
and reduce the emission of carbon dioxide.
5.4 Industrial Sensing
As plant infrastructure ages, equipment failures cause more and more unplanned downtime.
The ARC Advisory Group estimates that 5% of production in North America is lost to un-
planned downtime. Because sensor nodes can be deeply embedded into machines and there
is no infrastructure, WSNs make it economically feasible to monitor the health of machines
and to ensure safe operation. Aging pipelines and tanks have become a major problem in the
oil and gas industry. Monitoring corrosion using manual processes is extremely costly, time
consuming, and unreliable. A network of wireless corrosion sensors can be economically de-
ployed to reliably identify issues before they become catastrophic failures. Rohrback Cosasco
Systems (RCS) (Rohrback Cosasco Systems, n.d.) is the world leader in corrosion monitoring
technology and is applying WSNs in their corrosion monitoring. WSNs have also been sug-
gested for use in the food industry to prevent the incidents of contaminating the food supply
chain (Connolly & OReilly, 2005).
5.5 Infrastructure Security
WSNs can be used for infrastructure security and counterterrorism applications. Critical
buildings and facilities such as power plants, airports, and military bases have to be protected
from potential invasions. Networks of video, acoustic, and other sensors can be deployed
around these facilities (Chong & Kumar, 2003). An initiative in Shanghai Pudong Interna-
tional Airport has involved the installation of a WSN-aided intrusion prevention system on
its periphery to deter any unexpected intrusions. The Expo 2010 Shanghai China (Expo 2010
Shanghai China, n.d.) has also securedits expo sites with the same intrusion prevention system.
6. Security
While the future of WSNs is very prospective, WSNs will not be successfully deployed if
security, dependability and privacy issues are not addressed adequately. These issues become
more important because WSNs are usually used for very critical applications. Furthermore,
WSNs are very vulnerable and thus attractive to attacks because of their limited prices and
human-unattended deployment (Wang & Zhang, 2009).
6.1 Security Threats in WSNs
A typical WSN consists of hundreds or even thousands of tiny and resource-constrained sen-
sor nodes. These sensor nodes are distributedly deployed in uncontrollable environment for
the collection of security-sensitive information. Individual sensor nodes rely on multi-hop
wireless communication to deliver the sensed data to a remote base station. In a basic WSN
scenario, resource constraint, wireless communication, security-sensitive data, uncontrollable
environment, and even distributed deployment are all vulnerabilities. These vulnerabilities
make WSNs suffer from an amazing number of security threats. WSNs can only be used in
the critical applications after the potential security threats are eliminated.
6.1.1 Physical Layer Threats
Comparing WSNs with traditional networks, there are more threats to WSNs in the physical
layer, due to the non-tamper-resistant WSN nodes and the broadcasting nature of wireless
transmission. Typical types of attacks in the physical layer include physical layer jamming
and the subversion of a node.
6.1.2 Link Layer Threats
The data link layer is responsible for the multiplexing of data streams, data frame detection,
mediumaccess, and error control. The following attacks can happen in the link layer of WSNs:
Link layer jamming; Eavesdropping; Resource exhaustion and trafc analysis.
6.1.3 Network Layer Threats
Threats in the network layer mostly aim at disturbing data-centric and energy efcient multi-
hop routing, which is the main design principle in WSNs. The following threats and attacks
in this layer are identied in (Wang & Zhang, 2009): Spoofed, altered, or replayed routing in-
formation; Sybil attack; Selective forwarding; Sinkhole attack; Wormhole attack and ooding.
6.1.4 Application Layer Threats
Many WSNs applications heavily rely on coordinated services such as localization, time syn-
chronization, and in-network data processing to collaboratively process data (Sabbah et al.,
2006). Unfortunately, these services represent unique vulnerabilities such as: False data lter-
ing; Clock un-synchronization; False data injection.
6.2 Countermeasures
WSN Threats presented above either violate network secrecy and authentication, such as
packet spoong, or violate network availability, such as jamming attack, or violate some other
network functionalities. Generally, countermeasures to the threats in WSNs should fulll the
following security requirements (Wang & Zhang, 2009):
Availability, which ensures that the desired network services are available whenever
required.
Authentication, which ensures that the communication from one node to another node
is genuine.
Condentiality, which provides the privacy of the wireless communication channels.
Integrity, which ensures that the message or the entity under consideration is not al-
tered.
Non-reputation, which prevents malicious nodes to hide or deny their activities.
Freshness, which implies that the data is recent and ensures that no adversary can re-
play old messages.
Survivability, which ensures the acceptable level of network services even in the pres-
ence of node failures and malicious attacks.
Self-security, countermeasures may introduce additional hardware and software infras-
tructures into the network, which must themselves be secure enough to withstand at-
tacks.
Depending on applications, countermeasures should also fulll appropriate performance re-
quirements.
6.2.1 Key Management
When setting up a sensor network, one of the rst security requirements is to establish cryp-
tographic keys for later secure communication. The established keys should be resilient to
attacks and exible to dynamic update. The task that supports the establishment and main-
tenance of key relationships between valid parties according to a security policy is called key
management. Desired features of key management in sensor networks include energy aware-
ness, localized impact of attacks, and scaling to a large number of nodes.
6.2.2 Authentication
As sensor networks are mostly deployedin human-unattended environments for critical sens-
ing measurements, the authentication of the data source as well as the data are critical con-
cerns. Proper authentication mechanisms can provide WSNs with both sensor and user iden-
tication ability, can protect the integrity and freshness of critical data, and can prohibit and
identify impersonating attack. Traditionally, authentication can be provided by public-key
schemes as digital signature and by symmetric-key schemes as message authentication code
(MAC). Besides, key-chain schemes using symmetric keys determined by asymmetric key-
exchange protocols are also popular for broadcast authentication in WSNs.
gested for use in the food industry to prevent the incidents of contaminating the food supply
chain (Connolly & OReilly, 2005).
5.5 Infrastructure Security
WSNs can be used for infrastructure security and counterterrorism applications. Critical
buildings and facilities such as power plants, airports, and military bases have to be protected
from potential invasions. Networks of video, acoustic, and other sensors can be deployed
around these facilities (Chong & Kumar, 2003). An initiative in Shanghai Pudong Interna-
tional Airport has involved the installation of a WSN-aided intrusion prevention system on
its periphery to deter any unexpected intrusions. The Expo 2010 Shanghai China (Expo 2010
Shanghai China, n.d.) has also securedits expo sites with the same intrusion prevention system.
6. Security
While the future of WSNs is very prospective, WSNs will not be successfully deployed if
security, dependability and privacy issues are not addressed adequately. These issues become
more important because WSNs are usually used for very critical applications. Furthermore,
WSNs are very vulnerable and thus attractive to attacks because of their limited prices and
human-unattended deployment (Wang & Zhang, 2009).
6.1 Security Threats in WSNs
A typical WSN consists of hundreds or even thousands of tiny and resource-constrained sen-
sor nodes. These sensor nodes are distributedly deployed in uncontrollable environment for
the collection of security-sensitive information. Individual sensor nodes rely on multi-hop
wireless communication to deliver the sensed data to a remote base station. In a basic WSN
scenario, resource constraint, wireless communication, security-sensitive data, uncontrollable
environment, and even distributed deployment are all vulnerabilities. These vulnerabilities
make WSNs suffer from an amazing number of security threats. WSNs can only be used in
the critical applications after the potential security threats are eliminated.
6.1.1 Physical Layer Threats
Comparing WSNs with traditional networks, there are more threats to WSNs in the physical
layer, due to the non-tamper-resistant WSN nodes and the broadcasting nature of wireless
transmission. Typical types of attacks in the physical layer include physical layer jamming
and the subversion of a node.
6.1.2 Link Layer Threats
The data link layer is responsible for the multiplexing of data streams, data frame detection,
mediumaccess, and error control. The following attacks can happen in the link layer of WSNs:
Link layer jamming; Eavesdropping; Resource exhaustion and trafc analysis.
6.1.3 Network Layer Threats
Threats in the network layer mostly aim at disturbing data-centric and energy efcient multi-
hop routing, which is the main design principle in WSNs. The following threats and attacks
in this layer are identied in (Wang & Zhang, 2009): Spoofed, altered, or replayed routing in-
formation; Sybil attack; Selective forwarding; Sinkhole attack; Wormhole attack and ooding.
6.1.4 Application Layer Threats
Many WSNs applications heavily rely on coordinated services such as localization, time syn-
chronization, and in-network data processing to collaboratively process data (Sabbah et al.,
2006). Unfortunately, these services represent unique vulnerabilities such as: False data lter-
ing; Clock un-synchronization; False data injection.
6.2 Countermeasures
WSN Threats presented above either violate network secrecy and authentication, such as
packet spoong, or violate network availability, such as jamming attack, or violate some other
network functionalities. Generally, countermeasures to the threats in WSNs should fulll the
following security requirements (Wang & Zhang, 2009):
Availability, which ensures that the desired network services are available whenever
required.
Authentication, which ensures that the communication from one node to another node
is genuine.
Condentiality, which provides the privacy of the wireless communication channels.
Integrity, which ensures that the message or the entity under consideration is not al-
tered.
Non-reputation, which prevents malicious nodes to hide or deny their activities.
Freshness, which implies that the data is recent and ensures that no adversary can re-
play old messages.
Survivability, which ensures the acceptable level of network services even in the pres-
ence of node failures and malicious attacks.
Self-security, countermeasures may introduce additional hardware and software infras-
tructures into the network, which must themselves be secure enough to withstand at-
tacks.
Depending on applications, countermeasures should also fulll appropriate performance re-
quirements.
6.2.1 Key Management
When setting up a sensor network, one of the rst security requirements is to establish cryp-
tographic keys for later secure communication. The established keys should be resilient to
attacks and exible to dynamic update. The task that supports the establishment and main-
tenance of key relationships between valid parties according to a security policy is called key
management. Desired features of key management in sensor networks include energy aware-
ness, localized impact of attacks, and scaling to a large number of nodes.
6.2.2 Authentication
As sensor networks are mostly deployedin human-unattended environments for critical sens-
ing measurements, the authentication of the data source as well as the data are critical con-
cerns. Proper authentication mechanisms can provide WSNs with both sensor and user iden-
tication ability, can protect the integrity and freshness of critical data, and can prohibit and
identify impersonating attack. Traditionally, authentication can be provided by public-key
schemes as digital signature and by symmetric-key schemes as message authentication code
(MAC). Besides, key-chain schemes using symmetric keys determined by asymmetric key-
exchange protocols are also popular for broadcast authentication in WSNs.
6.2.3 Intrusion Detection
Security technologies, such as authentication and cryptography, can enhance the security of
sensor networks. Nevertheless, these preventive mechanisms alone cannot deter all possi-
ble attacks (e.g., insider attackers possessing the key). Intrusion detection, which has been
successfully used in Internet, can provide a second line of defense.
6.2.4 Privacy Protection
As WSN applications expand to include increasingly sensitive measurements in both military
tasks and everyday life, privacy protection becomes an increasingly important concern. For
example, few people may enjoy the benets of a body area WSN, if they know that their
personal data such as heart rate, blood pressure, etc., are regularly transmitted without proper
privacy protection. Also, the important data sink in a battleeld surveillance WSN may be
rstly destroyed, if its location can be traced by analyzing the volume of radio activities.
7. Standardization
In the area of WSNs, several standards are currently either ratied or under development. The
major standardization bodies are the Institute of Electrical and Electronics Engineers (IEEE),
the Internet Engineering Task Force (IETF), the International Society for Automation (ISA)
and the HART Communication Foundation, etc. These standardization bodies have different
focuses and they provide global, open standards for interoperable, low-power wireless sensor
devices. Table 1 provides the comparisons of different standards currently available for the
communication protocols of WSNs.
7.1 IEEE 802.15.4
IEEE 802.15.4 is a standard which species the physical layer and MAC layer for low-rate
wireless personal area networks. It is the basis for the ZigBee and WirelessHARTspecication,
each of which further attempts to offer a complete networking solution by developing the
upper layers which are not covered by the standard.
The features of IEEE 802.15.4 include (IEEE 802.15 WPAN Task Group 4, n.d.):
Data rates of 250 kbps, 40 kbps, and 20 kbps.
Two addressing modes; 16-bit short and 64-bit IEEE addressing.
Support for critical latency devices, such as joysticks.
CSMA-CA channel access.
Automatic network establishment by the coordinator.
Fully handshaked protocol for transfer reliability.
Power management to ensure low power consumption.
16 channels in the 2.4GHz ISM band, 10 channels in the 915MHz ISM band and one
channel in the 868MHz band.
7.2 Zigbee
ZigBee is a standard for a suite of high level communication protocols based on the IEEE
802.15.4 standard for low power and low data rate radio communications. Zigbee is initiated
and maintained by the Zigbee Alliance - a large consortium of industry players. The typi-
cal application areas of Zigbee include: Smart energy monitoring; Health care monitoring;
Remote control; Building automation and home automation, etc.
7.3 WirelessHART
WirelessHARTis an open-standard wireless mesh network communications protocol designed
to meet the needs for process automation applications. The protocol utilizes IEEE 802.15.4
compatible DSSS radios and it is operating in the 2.4GHz ISM radio band. On the data link
layer, the protocol uses TDMA technology to arbitrate and coordinate communications be-
tween devices. WirelessHART provides highly secure communications by using AES-128
block ciphers with individual Join and Session Keys and Data-Link level Network Key. Wire-
lessHART supports the standard HART Application Layer and is compatible with existing
HART tools, applications and system integration technology. The other outstanding features
of WirelessHART include reliability and scalability. Typically, the communication reliability
for a well-formed WirelessHART network is greater than 3 and normally greater than 6.
Adding new devices can further improve the network and its communication reliability.
8. Summary
WSNs have been identied as one of the most prospective technologies in this century. This
chapter provides information concerning both its history and current state of the art. In con-
crete terms, the authors provide an overview about the hardware, software and networking
protocol design of this important technology. The authors also discuss the security and on-
going standardization of this technology. Depending on applications, many other techniques
such as localization, synchronization and in-network processing can be important, which are
not discussed in this chapter.
9. References
21 Ideas for the 21st Century (1999). Business Week pp. 78167.
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y. & Cayirci, E. (2002). A survey on sensor
networks, IEEE Communications Magazine 40(8): 102114.
Barrenetxea, G., Ingelrest, F., Schaefer, G. & Vetterli, M. (2008). Wireless sensor networks for
environmental monitoring: The sensorscope experience, Proc. of 20th IEEE Interna-
tional Zurich Seminar on Communications (IZS08).
BSN Research in Imperial College London (n.d.). http://ubimon.doc.ic.ac.uk/bsn/m621.html.
Chong, C.-Y. &Kumar, S. P. (2003). Sensor networks: Evolution, opportunities, and challenges,
Proceedings of the IEEE 91(8): 12471256.
Connolly, M. & OReilly, F. (2005). Sensor networks and the food industry, Proc. of Workshop
on Real-World Wireless Sensor Networks (REALWSN05).
Contiki (n.d.). http://www.sics.se/contiki.
Crossbow Technology (n.d.). http://www.xbow.com.
Dust Networks, Inc. (n.d.). http://www.dustnetworks.com.
Expo 2010 Shanghai China (n.d.). http://www.expo2010.cn.
Hill, J. L. (2003). System Architecture for Wireless Sensor Networks, PhD thesis, Doctor of Philos-
ophy in Computer Science, University of California at Berkeley, USA.
Howitt, I., Manges, W. W., Kuruganti, P. T., Allgood, G., Gutierrez, J. A. & Conrad, J. M.
(2006). Wireless industrial sensor networks: Framework for qos assessment and qos
management, ISA Transactions 45(3): 347359.
IEEE 802.15 WPAN Task Group 4 (n.d.). http://www.ieee802.org/15/pub/TG4.html.
IPSO Alliance - promoting the use of IP for Smart Objects (n.d.). http://www.ipso-alliance.org.
6.2.3 Intrusion Detection
Security technologies, such as authentication and cryptography, can enhance the security of
sensor networks. Nevertheless, these preventive mechanisms alone cannot deter all possi-
ble attacks (e.g., insider attackers possessing the key). Intrusion detection, which has been
successfully used in Internet, can provide a second line of defense.
6.2.4 Privacy Protection
As WSN applications expand to include increasingly sensitive measurements in both military
tasks and everyday life, privacy protection becomes an increasingly important concern. For
example, few people may enjoy the benets of a body area WSN, if they know that their
personal data such as heart rate, blood pressure, etc., are regularly transmitted without proper
privacy protection. Also, the important data sink in a battleeld surveillance WSN may be
rstly destroyed, if its location can be traced by analyzing the volume of radio activities.
7. Standardization
In the area of WSNs, several standards are currently either ratied or under development. The
major standardization bodies are the Institute of Electrical and Electronics Engineers (IEEE),
the Internet Engineering Task Force (IETF), the International Society for Automation (ISA)
and the HART Communication Foundation, etc. These standardization bodies have different
focuses and they provide global, open standards for interoperable, low-power wireless sensor
devices. Table 1 provides the comparisons of different standards currently available for the
communication protocols of WSNs.
7.1 IEEE 802.15.4
IEEE 802.15.4 is a standard which species the physical layer and MAC layer for low-rate
wireless personal area networks. It is the basis for the ZigBee and WirelessHARTspecication,
each of which further attempts to offer a complete networking solution by developing the
upper layers which are not covered by the standard.
The features of IEEE 802.15.4 include (IEEE 802.15 WPAN Task Group 4, n.d.):
Data rates of 250 kbps, 40 kbps, and 20 kbps.
Two addressing modes; 16-bit short and 64-bit IEEE addressing.
Support for critical latency devices, such as joysticks.
CSMA-CA channel access.
Automatic network establishment by the coordinator.
Fully handshaked protocol for transfer reliability.
Power management to ensure low power consumption.
16 channels in the 2.4GHz ISM band, 10 channels in the 915MHz ISM band and one
channel in the 868MHz band.
7.2 Zigbee
ZigBee is a standard for a suite of high level communication protocols based on the IEEE
802.15.4 standard for low power and low data rate radio communications. Zigbee is initiated
and maintained by the Zigbee Alliance - a large consortium of industry players. The typi-
cal application areas of Zigbee include: Smart energy monitoring; Health care monitoring;
Remote control; Building automation and home automation, etc.
7.3 WirelessHART
WirelessHARTis an open-standard wireless mesh network communications protocol designed
to meet the needs for process automation applications. The protocol utilizes IEEE 802.15.4
compatible DSSS radios and it is operating in the 2.4GHz ISM radio band. On the data link
layer, the protocol uses TDMA technology to arbitrate and coordinate communications be-
tween devices. WirelessHART provides highly secure communications by using AES-128
block ciphers with individual Join and Session Keys and Data-Link level Network Key. Wire-
lessHART supports the standard HART Application Layer and is compatible with existing
HART tools, applications and system integration technology. The other outstanding features
of WirelessHART include reliability and scalability. Typically, the communication reliability
for a well-formed WirelessHART network is greater than 3 and normally greater than 6.
Adding new devices can further improve the network and its communication reliability.
8. Summary
WSNs have been identied as one of the most prospective technologies in this century. This
chapter provides information concerning both its history and current state of the art. In con-
crete terms, the authors provide an overview about the hardware, software and networking
protocol design of this important technology. The authors also discuss the security and on-
going standardization of this technology. Depending on applications, many other techniques
such as localization, synchronization and in-network processing can be important, which are
not discussed in this chapter.
9. References
21 Ideas for the 21st Century (1999). Business Week pp. 78167.
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y. & Cayirci, E. (2002). A survey on sensor
networks, IEEE Communications Magazine 40(8): 102114.
Barrenetxea, G., Ingelrest, F., Schaefer, G. & Vetterli, M. (2008). Wireless sensor networks for
environmental monitoring: The sensorscope experience, Proc. of 20th IEEE Interna-
tional Zurich Seminar on Communications (IZS08).
BSN Research in Imperial College London (n.d.). http://ubimon.doc.ic.ac.uk/bsn/m621.html.
Chong, C.-Y. &Kumar, S. P. (2003). Sensor networks: Evolution, opportunities, and challenges,
Proceedings of the IEEE 91(8): 12471256.
Connolly, M. & OReilly, F. (2005). Sensor networks and the food industry, Proc. of Workshop
on Real-World Wireless Sensor Networks (REALWSN05).
Contiki (n.d.). http://www.sics.se/contiki.
Crossbow Technology (n.d.). http://www.xbow.com.
Dust Networks, Inc. (n.d.). http://www.dustnetworks.com.
Expo 2010 Shanghai China (n.d.). http://www.expo2010.cn.
Hill, J. L. (2003). System Architecture for Wireless Sensor Networks, PhD thesis, Doctor of Philos-
ophy in Computer Science, University of California at Berkeley, USA.
Howitt, I., Manges, W. W., Kuruganti, P. T., Allgood, G., Gutierrez, J. A. & Conrad, J. M.
(2006). Wireless industrial sensor networks: Framework for qos assessment and qos
management, ISA Transactions 45(3): 347359.
IEEE 802.15 WPAN Task Group 4 (n.d.). http://www.ieee802.org/15/pub/TG4.html.
IPSO Alliance - promoting the use of IP for Smart Objects (n.d.). http://www.ipso-alliance.org.
Kumar, S. & Shepherd, D. (2001). Sensit: Sensor information technology for the warghter,
Proc. of the 4th International Conference on Information Fusion (FUSION01), pp. 39
(TuC1).
LiteOS (n.d.). http://www.liteos.net.
Martinez, K., Padhy, P., Riddoch, A., Ong, H. L. R. & Hart, J. K. (2005). Glacial environment
monitoring using sensor networks, Proc. of Workshop on Real-World Wireless Sensor
Networks (REALWSN05).
Myers, C., Oppenheim, A., Davis, R. & Dove, W. (1984). Knowledge-based speech analysis
and enhancement, Proc. of the International Conference on Acoustics, Speech and Signal
Processing.
Ni, L. M. (2008). Chinas national researchproject on wireless sensor networks, Proc. of the 2008
IEEE International Conference on Sensor Networks, Ubiquitous, and Trustworthy Comput-
ing (SUTC08), p. 19.
Pister, K. S. J. (2000). Military applications of sensor networks, of Institute for Defense Analyses
Paper P-3531, Defense Science Study Group.
Proceedings of the Distributed Sensor Nets Workshop (1978). Pittsburgh, USA. Department of
Computer Science, Carnegie Mellon University.
Rashid, R. & Robertson, G. (1981). Accent: A communication oriented network operating
system kernel, Proc. of the 8th Symposium on Operating System Principles, pp. 6475.
Rohrback Cosasco Systems (n.d.). http://www.cosasco.com.
Sabbah, E., Majeed, A., Kang, K., Liu, K. & AbuGhazaleh, N. (2006). An application driven
perspective on wireless sensor network security, Proc. of the 2nd ACM Workshop on
QoS and Security for Wireless and Mobile Networks.
SOS Embedded Operating System (n.d.). https://projects.nesl.ucla.edu/public/sos-2x/doc/.
Steere, D., Baptista, A., McNamee, D., Pu, C. & Walpole, J. (2000). Research challenges in en-
vironmental observation and forecasting systems, Proc. of 6th International Conference
on Mobile Computing and Networking (MOBICOMM00), pp. 292299.
Streetline, Inc. (n.d.). http://www.streetlinenetworks.com.
Sudevalayam, S. & Kulkarni, P. (2008). Energy Harvesting Sensor Nodes: Survey and Im-
plications, Technical Report TR-CSE-2008-19, Department of Computer Science and
Engineering, Indian Institute of Technology Bombay.
TinyOS Community Forum (n.d.). http://www.tinyos.net.
Wang, Q. & Zhang, T. (2008). Sec-snmp: Policy-based security management for sensor net-
works, Proc. of the International Conference on Security and Cryptography (SECRYPT08),
in conjunction with ICETE 2008.
Wang, Q. & Zhang, T. (2009). A survey on security in wireless sensor networks, in Y. Zhang
& P. Kitsos (eds), Security in RFID and Sensor Networks, CRC Press, Taylor & Francis
Group, chapter 14, pp. 293320.
ZigBee Alliance (n.d.). http://www.zigbee.org.
T
y
p
e
I
E
E
E
8
0
2
.
1
1
b
/
g
B
l
u
e
t
o
o
t
h
U
W
B
-
I
R
I
E
E
E
8
0
2
.
1
5
.
4
Z
i
g
b
e
e
W
i
r
e
l
e
s
s
H
A
R
T
I
E
E
E
8
0
2
.
1
5
.
6
a
B
a
n
d
I
S
M
2
.
4
G
H
z
I
S
M
2
.
4
G
H
z
3
.
1
G
H
z
-
1
0
.
6
G
H
z
I
S
M
2
.
4
G
H
z
,
9
1
5
M
H
z
,
8
6
8
M
H
z
I
S
M
2
.
4
G
H
z
I
S
M
2
.
4
G
H
z
4
0
0
M
H
z
,
I
S
M
2
.
4
G
H
z
,
3
.
1
-
1
0
.
6
G
H
z
S
p
r
e
a
d
i
n
g
D
S
S
S
F
H
S
S
/
T
D
D
B
a
s
e
b
a
n
d
D
S
S
S
D
S
S
S
D
S
S
S
D
S
S
S
M
o
d
u
l
a
t
i
o
n
B
P
S
K
,
Q
P
S
K
,
C
C
K
,
O
F
D
M
G
F
S
K
I
m
p
u
l
s
e
r
a
d
i
o
,
t
i
m
e
-
d
o
m
a
i
n
O
-
Q
P
S
K
O
-
Q
P
S
K
O
-
Q
P
S
K
G
r
o
u
p
P
P
M
R
a
n
g
e
1
0
0
m
1
0
m
<
5
m
1
0
m
1
0
m
<
2
5
0
m
3
-
1
0
m
R
a
t
e
<
5
4
M
b
p
s
1
M
b
p
s
2
0
K
b
p
s
,
2
5
0
K
b
p
s
,
1
0
M
b
p
s
2
5
0
K
b
p
s
2
5
0
K
b
p
s
2
5
0
K
b
p
s
<
1
0
M
b
p
s
P
o
w
e
r
H
i
g
h
L
o
w
U
l
t
r
a
l
o
w
V
e
r
y
l
o
w
V
e
r
y
l
o
w
B
a
t
t
e
r
y
o
r
l
i
n
e
p
o
w
e
r
U
l
t
r
a
l
o
w
R
o
a
m
i
n
g
Y
e
s
N
o
Y
e
s
Y
e
s
Y
e
s
Y
e
s
Y
e
s
N
o
.
o
f
n
o
d
e
s
3
2
(
p
e
r
a
c
-
c
e
s
s
p
o
i
n
t
)
8
(
p
e
r
p
i
-
c
o
n
e
t
)
1
0
-
1
0
0
0
<
6
5
5
3
6
<
6
5
5
3
6
5
-
6
5
5
3
6
<
2
5
6
P
o
w
e
r
c
o
n
-
s
u
m
p
t
i
o
n
M
e
d
i
u
m
L
o
w
U
l
t
r
a
l
o
w
V
e
r
y
l
o
w
V
e
r
y
l
o
w
V
e
r
y
l
o
w
U
l
t
r
a
l
o
w
C
o
m
p
l
e
x
i
t
y
C
o
m
p
l
e
x
V
e
r
y
c
o
m
-
p
l
e
x
S
i
m
p
l
e
t
r
a
n
s
m
i
t
t
e
r
b
u
t
c
o
m
p
l
e
x
r
e
c
e
i
v
e
r
S
i
m
p
l
e
S
i
m
p
l
e
S
i
m
p
l
e
S
i
m
p
l
e
S
e
c
u
r
i
t
y
W
E
P
,
W
P
A
6
4
b
i
t
o
r
1
2
8
b
i
t
1
2
8
b
i
t
N
U
L
L
,
3
2
b
i
t
,
6
4
b
i
t
o
r
1
2
8
b
i
t
1
2
8
b
i
t
1
2
8
b
i
t
S
c
r
a
m
b
l
e
d
m
a
p
p
i
n
g
c
o
d
e
T
a
r
g
e
t
c
o
s
t
H
i
g
h
M
e
d
i
u
m
V
e
r
y
l
o
w
L
o
w
L
o
w
L
o
w
L
o
w
T
a
b
l
e
1
.
C
o
m
p
a
r
i
s
o
n
s
o
f
d
i
f
f
e
r
e
n
t
s
t
a
n
d
a
r
d
s
a
v
a
i
l
a
b
l
e
f
o
r
t
h
e
c
o
m
m
u
n
i
c
a
t
i
o
n
p
r
o
t
o
c
o
l
s
o
f
W
S
N
s
.
a
A
l
l
v
a
l
u
e
s
i
n
t
h
i
s
c
o
l
u
m
n
a
r
e
n
o
t
o
f
c
i
a
l
.
I
E
E
E
8
0
2
.
1
5
.
6
i
s
a
c
o
m
m
u
n
i
c
a
t
i
o
n
s
t
a
n
d
a
r
d
o
p
t
i
m
i
z
e
d
f
o
r
w
i
r
e
l
e
s
s
b
o
d
y
a
r
e
a
n
e
t
w
o
r
k
s
a
n
d
i
t
i
s
s
t
i
l
l
u
n
d
e
r
a
c
t
i
v
e
d
e
v
e
l
o
p
m
e
n
t
.
Kumar, S. & Shepherd, D. (2001). Sensit: Sensor information technology for the warghter,
Proc. of the 4th International Conference on Information Fusion (FUSION01), pp. 39
(TuC1).
LiteOS (n.d.). http://www.liteos.net.
Martinez, K., Padhy, P., Riddoch, A., Ong, H. L. R. & Hart, J. K. (2005). Glacial environment
monitoring using sensor networks, Proc. of Workshop on Real-World Wireless Sensor
Networks (REALWSN05).
Myers, C., Oppenheim, A., Davis, R. & Dove, W. (1984). Knowledge-based speech analysis
and enhancement, Proc. of the International Conference on Acoustics, Speech and Signal
Processing.
Ni, L. M. (2008). Chinas national researchproject on wireless sensor networks, Proc. of the 2008
IEEE International Conference on Sensor Networks, Ubiquitous, and Trustworthy Comput-
ing (SUTC08), p. 19.
Pister, K. S. J. (2000). Military applications of sensor networks, of Institute for Defense Analyses
Paper P-3531, Defense Science Study Group.
Proceedings of the Distributed Sensor Nets Workshop (1978). Pittsburgh, USA. Department of
Computer Science, Carnegie Mellon University.
Rashid, R. & Robertson, G. (1981). Accent: A communication oriented network operating
system kernel, Proc. of the 8th Symposium on Operating System Principles, pp. 6475.
Rohrback Cosasco Systems (n.d.). http://www.cosasco.com.
Sabbah, E., Majeed, A., Kang, K., Liu, K. & AbuGhazaleh, N. (2006). An application driven
perspective on wireless sensor network security, Proc. of the 2nd ACM Workshop on
QoS and Security for Wireless and Mobile Networks.
SOS Embedded Operating System (n.d.). https://projects.nesl.ucla.edu/public/sos-2x/doc/.
Steere, D., Baptista, A., McNamee, D., Pu, C. & Walpole, J. (2000). Research challenges in en-
vironmental observation and forecasting systems, Proc. of 6th International Conference
on Mobile Computing and Networking (MOBICOMM00), pp. 292299.
Streetline, Inc. (n.d.). http://www.streetlinenetworks.com.
Sudevalayam, S. & Kulkarni, P. (2008). Energy Harvesting Sensor Nodes: Survey and Im-
plications, Technical Report TR-CSE-2008-19, Department of Computer Science and
Engineering, Indian Institute of Technology Bombay.
TinyOS Community Forum (n.d.). http://www.tinyos.net.
Wang, Q. & Zhang, T. (2008). Sec-snmp: Policy-based security management for sensor net-
works, Proc. of the International Conference on Security and Cryptography (SECRYPT08),
in conjunction with ICETE 2008.
Wang, Q. & Zhang, T. (2009). A survey on security in wireless sensor networks, in Y. Zhang
& P. Kitsos (eds), Security in RFID and Sensor Networks, CRC Press, Taylor & Francis
Group, chapter 14, pp. 293320.
ZigBee Alliance (n.d.). http://www.zigbee.org.
T
y
p
e
I
E
E
E
8
0
2
.
1
1
b
/
g
B
l
u
e
t
o
o
t
h
U
W
B
-
I
R
I
E
E
E
8
0
2
.
1
5
.
4
Z
i
g
b
e
e
W
i
r
e
l
e
s
s
H
A
R
T
I
E
E
E
8
0
2
.
1
5
.
6
a
B
a
n
d
I
S
M
2
.
4
G
H
z
I
S
M
2
.
4
G
H
z
3
.
1
G
H
z
-
1
0
.
6
G
H
z
I
S
M
2
.
4
G
H
z
,
9
1
5
M
H
z
,
8
6
8
M
H
z
I
S
M
2
.
4
G
H
z
I
S
M
2
.
4
G
H
z
4
0
0
M
H
z
,
I
S
M
2
.
4
G
H
z
,
3
.
1
-
1
0
.
6
G
H
z
S
p
r
e
a
d
i
n
g
D
S
S
S
F
H
S
S
/
T
D
D
B
a
s
e
b
a
n
d
D
S
S
S
D
S
S
S
D
S
S
S
D
S
S
S
M
o
d
u
l
a
t
i
o
n
B
P
S
K
,
Q
P
S
K
,
C
C
K
,
O
F
D
M
G
F
S
K
I
m
p
u
l
s
e
r
a
d
i
o
,
t
i
m
e
-
d
o
m
a
i
n
O
-
Q
P
S
K
O
-
Q
P
S
K
O
-
Q
P
S
K
G
r
o
u
p
P
P
M
R
a
n
g
e
1
0
0
m
1
0
m
<
5
m
1
0
m
1
0
m
<
2
5
0
m
3
-
1
0
m
R
a
t
e
<
5
4
M
b
p
s
1
M
b
p
s
2
0
K
b
p
s
,
2
5
0
K
b
p
s
,
1
0
M
b
p
s
2
5
0
K
b
p
s
2
5
0
K
b
p
s
2
5
0
K
b
p
s
<
1
0
M
b
p
s
P
o
w
e
r
H
i
g
h
L
o
w
U
l
t
r
a
l
o
w
V
e
r
y
l
o
w
V
e
r
y
l
o
w
B
a
t
t
e
r
y
o
r
l
i
n
e
p
o
w
e
r
U
l
t
r
a
l
o
w
R
o
a
m
i
n
g
Y
e
s
N
o
Y
e
s
Y
e
s
Y
e
s
Y
e
s
Y
e
s
N
o
.
o
f
n
o
d
e
s
3
2
(
p
e
r
a
c
-
c
e
s
s
p
o
i
n
t
)
8
(
p
e
r
p
i
-
c
o
n
e
t
)
1
0
-
1
0
0
0
<
6
5
5
3
6
<
6
5
5
3
6
5
-
6
5
5
3
6
<
2
5
6
P
o
w
e
r
c
o
n
-
s
u
m
p
t
i
o
n
M
e
d
i
u
m
L
o
w
U
l
t
r
a
l
o
w
V
e
r
y
l
o
w
V
e
r
y
l
o
w
V
e
r
y
l
o
w
U
l
t
r
a
l
o
w
C
o
m
p
l
e
x
i
t
y
C
o
m
p
l
e
x
V
e
r
y
c
o
m
-
p
l
e
x
S
i
m
p
l
e
t
r
a
n
s
m
i
t
t
e
r
b
u
t
c
o
m
p
l
e
x
r
e
c
e
i
v
e
r
S
i
m
p
l
e
S
i
m
p
l
e
S
i
m
p
l
e
S
i
m
p
l
e
S
e
c
u
r
i
t
y
W
E
P
,
W
P
A
6
4
b
i
t
o
r
1
2
8
b
i
t
1
2
8
b
i
t
N
U
L
L
,
3
2
b
i
t
,
6
4
b
i
t
o
r
1
2
8
b
i
t
1
2
8
b
i
t
1
2
8
b
i
t
S
c
r
a
m
b
l
e
d
m
a
p
p
i
n
g
c
o
d
e
T
a
r
g
e
t
c
o
s
t
H
i
g
h
M
e
d
i
u
m
V
e
r
y
l
o
w
L
o
w
L
o
w
L
o
w
L
o
w
T
a
b
l
e
1
.
C
o
m
p
a
r
i
s
o
n
s
o
f
d
i
f
f
e
r
e
n
t
s
t
a
n
d
a
r
d
s
a
v
a
i
l
a
b
l
e
f
o
r
t
h
e
c
o
m
m
u
n
i
c
a
t
i
o
n
p
r
o
t
o
c
o
l
s
o
f
W
S
N
s
.
a
A
l
l
v
a
l
u
e
s
i
n
t
h
i
s
c
o
l
u
m
n
a
r
e
n
o
t
o
f
c
i
a
l
.
I
E
E
E
8
0
2
.
1
5
.
6
i
s
a
c
o
m
m
u
n
i
c
a
t
i
o
n
s
t
a
n
d
a
r
d
o
p
t
i
m
i
z
e
d
f
o
r
w
i
r
e
l
e
s
s
b
o
d
y
a
r
e
a
n
e
t
w
o
r
k
s
a
n
d
i
t
i
s
s
t
i
l
l
u
n
d
e
r
a
c
t
i
v
e
d
e
v
e
l
o
p
m
e
n
t
.
Wireless Sensor Networks for On-feld Agricultural Management Process 17
Wireless Sensor Networks for On-feld Agricultural Management Process
Luca Bencini, Davide Di Palma, Giovanni Collodi, Gianfranco Manes and Antonio Manes
0
Wireless Sensor Networks for On-eld
Agricultural Management Process
Luca Bencini, Davide Di Palma, Giovanni Collodi and Gianfranco Manes
Department of Electronics and Telecommunications - University of Florence
Italy
Antonio Manes
Netsens S.r.l.
Italy
1. Introduction
Agriculture is one of the most ancient activities of man in which innovation and technology
are usually accepted with difculty, unless real and immediate solutions are found for specic
problems or for improving production and quality. Nevertheless, a newapproach of gathering
information from the environment could represent an important step towards high quality
and eco-sustainable agriculture.
Nowadays, irrigation, fertilization and pesticides management are often left to the farmers
and agronomists discretion: common criteria used to guarantee safe culture and plant growth
are often giving a greater amount of chemicals and water than necessary. There is no direct
feedback between the decision of treating or irrigating plants and the real effects in the eld.
Plant conditions are usually committed to sporadic and faraway weather stations that cannot
provide accurate and local measurements of the fundamental parameters in each zone of the
eld. Also, agronomic models, based on these monitored data, cannot provide reliable infor-
mation. On the contrary, agriculture needs detailed monitoring in order to obtain real time
feedback between plants, local climate conditions and mans decisions.
The concept of precision agriculture has been around for some time now. Blackmore et al.,
in 1994 (Blackmore, 1994) dened it as a comprehensive system designed to optimize agricul-
tural production by carefully tailoring soil and crop management to correspond to the unique
condition found in each eld while maintaining environmental quality. The early adopters
during that time found precision agriculture to be unprotable and the instances in which it
was implemented were few and far between. Further, the high initial investment in the form
of electronic equipment for sensing and communication meant that only large farms could af-
ford it. The technologies proposed at this point comprised three aspects: Remote Sensing (RS),
Global Positioning System (GPS) and Geographical Information System (GIS). RS coupled with
GPS coordinates produced accurate maps and models of the agricultural elds. The sampling
was typically through electronic sensors such as soil probes and remote optical scanners from
satellites. The collection of such data in the form of electronic computer databases gave birth
to the GIS. Statistical analyses were then conducted on the data and the variability of agri-
cultural land was charted with respect to its properties. The technology, apart from being
2
non-real-time, involved the use of expensive technologies like satellite sensing and was labor
intensive where the maps charting the agricultural elds were mostly manually done.
Over the last seven years, the advancement in sensing and communication technologies has
signicantly brought down the cost of deployment and running of a feasible precision agri-
culture framework. Emerging wireless technologies with low power needs and low data rate
capabilities have been developed which perfectly suit precision agriculture (Wang et al., 2006).
The sensing and communication can now be done on a real-time basis leading to better re-
sponse times. The wireless sensors are cheap enough for wide spread deployment and of-
fer robust communication through redundant propagation paths (Akyildiz & Xudong, 2005).
Thanks to these features, the Wireless Sensor Networks (WSNs) (Akyildiz & Xudong, 2005) have
become the most suitable technology to t an invasive method of monitoring the agricultural
environment.
In this chapter, an end-to-end monitoring WSN technology-based solution is presented, join-
ing hardware optimization with communications protocols design and a suitable interface.
In particular Section 2 discusses the system requirements and illustrates the overall system
characterization in conjunction with related work. Sections 3, 4 and 5 deal respectively with
the system in terms of hardware, protocol and software design. Section 6 describes the actual
experiences, focusing on several case study analyses for highlighting the effectiveness and
accurateness of the developed system. Sections 7 and 8 describe respectively the commercial
system VineSense, born from the experimental solution, and some agronomic results. Fi-
nally, in Section 9 some conclusions are drawn in order to explain the future direction of the
current research study.
2. System Requirement and Architecture
The requirements that adopting a WSN are expected to satisfy in effective agricultural mon-
itoring concern both system level issues (i.e., unattended operation, maximum network life
time, adaptability or even functionality and protocol self-recongurability) and nal user
needs (i.e., communication reliability and robustness, user friendliness, versatile and pow-
erful graphical user interfaces). The most relevant mainly concerns the supply of stand-alone
operations. To this end, the system must be able to run unattended for a long period, as nodes
are expected to be deployed in zones that are difcult to maintain. This calls for optimal en-
ergy management ensuring that the energy spent is directly related to the amount of trafc
handled and not to the overall working time. In fact, energy is nevertheless a limited resource
and the failure of a node may compromise WSN connectivity as the network gets partitioned.
Other issues to be addressed are the capabilities of quickly setting-up an end-to-end com-
munication infrastructure, supporting both synchronous and asynchronous queries, and of
dynamically reconguring it. An additional requirement is robust operative conditions, which
need fault management since a node may fail for several reasons. Other important properties
are scalability and adaptability of the networks topology, in terms of the number of nodes and
their density in unexpected events with a higher degree of responsiveness and recongurabil-
ity. This also implies the development of a plug and play sensor interface and the provisioning
of remote connectivity. Finally, several user-oriented attributes, including fairness, latency,
throughput and enhanced data querying schemes (i.e., time-driven (Al-Karaki & Kamal, 2004)
or query-driven) need to be taken into account even if they could be considered secondary with
respect to our application purposes because the WSNs cost/performance trade-off (Langen-
doen & Halkes, 2004).
A WSN system was developed according to the afore mentioned requirements. The system,
shown in Fig. 1, comprises a self-organizing mesh WSN endowed with sensing capabilities, a
GPRS Gateway, which gathers data and provides a TCP-IP based connection toward a Remote
Server, and a Web Application, which manages information and makes the nal user capable
of monitoring and interacting with the instrumented environment.
NODE 1
NODE 2
NODE 3
NODE 4
GPRS GATEWAY
REMOTE SERVER
TCP-IP
over
GPRS
WEB INTERFACE
Fig. 1. Wireless Sensor Network System
3. Hardware Design
Focusing on an end-to-end system architecture, every constitutive element has to be selected
according to application requirements and scenario issues, especially regarding the hardware
platform. Many details have to be considered, involving the energetic consumption of the sen-
sor readings, the power-on and power-save status management and a good trade-off between
the maximum radio coverage and the transmitted power. After an accurate investigation
of the out-of-the-shelf solutions, 868 MHz Mica2 motes (Mica2 Series, 2002) were adopted ac-
cording to these constraints and to the reference scenarios. The Tiny Operative System(TinyOS)
running on this platform ensures full control of mote communication capabilities to attain op-
timized power management and provides necessary system portability towards future hard-
ware advancements or changes. Nevertheless, Mica2 motes are far from perfection, especially
in the RF section, since the power provided by the transceiver (Chipcon CC1000) is not com-
pletely available for transmission. However, it is lost to imperfect coupling with the antenna,
thus reducing the radio coverage area. An improvement of this section was performed, using
more suitable antennas and coupling circuits and increasing the transmitting power with a
power amplier, thus increasing the output power up to 15 dBm while respecting interna-
tional restrictions and standards. These optimizations allow for greater radio coverage (about
non-real-time, involved the use of expensive technologies like satellite sensing and was labor
intensive where the maps charting the agricultural elds were mostly manually done.
Over the last seven years, the advancement in sensing and communication technologies has
signicantly brought down the cost of deployment and running of a feasible precision agri-
culture framework. Emerging wireless technologies with low power needs and low data rate
capabilities have been developed which perfectly suit precision agriculture (Wang et al., 2006).
The sensing and communication can now be done on a real-time basis leading to better re-
sponse times. The wireless sensors are cheap enough for wide spread deployment and of-
fer robust communication through redundant propagation paths (Akyildiz & Xudong, 2005).
Thanks to these features, the Wireless Sensor Networks (WSNs) (Akyildiz & Xudong, 2005) have
become the most suitable technology to t an invasive method of monitoring the agricultural
environment.
In this chapter, an end-to-end monitoring WSN technology-based solution is presented, join-
ing hardware optimization with communications protocols design and a suitable interface.
In particular Section 2 discusses the system requirements and illustrates the overall system
characterization in conjunction with related work. Sections 3, 4 and 5 deal respectively with
the system in terms of hardware, protocol and software design. Section 6 describes the actual
experiences, focusing on several case study analyses for highlighting the effectiveness and
accurateness of the developed system. Sections 7 and 8 describe respectively the commercial
system VineSense, born from the experimental solution, and some agronomic results. Fi-
nally, in Section 9 some conclusions are drawn in order to explain the future direction of the
current research study.
2. System Requirement and Architecture
The requirements that adopting a WSN are expected to satisfy in effective agricultural mon-
itoring concern both system level issues (i.e., unattended operation, maximum network life
time, adaptability or even functionality and protocol self-recongurability) and nal user
needs (i.e., communication reliability and robustness, user friendliness, versatile and pow-
erful graphical user interfaces). The most relevant mainly concerns the supply of stand-alone
operations. To this end, the system must be able to run unattended for a long period, as nodes
are expected to be deployed in zones that are difcult to maintain. This calls for optimal en-
ergy management ensuring that the energy spent is directly related to the amount of trafc
handled and not to the overall working time. In fact, energy is nevertheless a limited resource
and the failure of a node may compromise WSN connectivity as the network gets partitioned.
Other issues to be addressed are the capabilities of quickly setting-up an end-to-end com-
munication infrastructure, supporting both synchronous and asynchronous queries, and of
dynamically reconguring it. An additional requirement is robust operative conditions, which
need fault management since a node may fail for several reasons. Other important properties
are scalability and adaptability of the networks topology, in terms of the number of nodes and
their density in unexpected events with a higher degree of responsiveness and recongurabil-
ity. This also implies the development of a plug and play sensor interface and the provisioning
of remote connectivity. Finally, several user-oriented attributes, including fairness, latency,
throughput and enhanced data querying schemes (i.e., time-driven (Al-Karaki & Kamal, 2004)
or query-driven) need to be taken into account even if they could be considered secondary with
respect to our application purposes because the WSNs cost/performance trade-off (Langen-
doen & Halkes, 2004).
A WSN system was developed according to the afore mentioned requirements. The system,
shown in Fig. 1, comprises a self-organizing mesh WSN endowed with sensing capabilities, a
GPRS Gateway, which gathers data and provides a TCP-IP based connection toward a Remote
Server, and a Web Application, which manages information and makes the nal user capable
of monitoring and interacting with the instrumented environment.
NODE 1
NODE 2
NODE 3
NODE 4
GPRS GATEWAY
REMOTE SERVER
TCP-IP
over
GPRS
WEB INTERFACE
Fig. 1. Wireless Sensor Network System
3. Hardware Design
Focusing on an end-to-end system architecture, every constitutive element has to be selected
according to application requirements and scenario issues, especially regarding the hardware
platform. Many details have to be considered, involving the energetic consumption of the sen-
sor readings, the power-on and power-save status management and a good trade-off between
the maximum radio coverage and the transmitted power. After an accurate investigation
of the out-of-the-shelf solutions, 868 MHz Mica2 motes (Mica2 Series, 2002) were adopted ac-
cording to these constraints and to the reference scenarios. The Tiny Operative System(TinyOS)
running on this platform ensures full control of mote communication capabilities to attain op-
timized power management and provides necessary system portability towards future hard-
ware advancements or changes. Nevertheless, Mica2 motes are far from perfection, especially
in the RF section, since the power provided by the transceiver (Chipcon CC1000) is not com-
pletely available for transmission. However, it is lost to imperfect coupling with the antenna,
thus reducing the radio coverage area. An improvement of this section was performed, using
more suitable antennas and coupling circuits and increasing the transmitting power with a
power amplier, thus increasing the output power up to 15 dBm while respecting interna-
tional restrictions and standards. These optimizations allow for greater radio coverage (about
200 m) and better power management. In order to manage different kinds of sensors, a compli-
ant sensor board was adopted, allowing up to 16 sensor plugs on the same node;, this makes a
single mote capable of sensing many environmental parameters at a time (Mattoli et al., 2005).
Sensor boards recognize the sensors and send Transducer Electronic Datasheets (TEDS) through
the network up to the server, making it possible for the system to recognize an automatic
sensor. The overall node stack architecture is shown in Fig. 2.
Communication Board
Power Board
Sensor Board
Overall size: 58x32x25 mm
Fig. 2. Node Stack Architecture
The GPRS embedded Gateway, shown in Fig. 3, is a stand-alone communication platform
designed to provide transparent, bi-directional wireless TCP-IP connectivity for remote mon-
itoring. In conjunction with Remote Data Acquisition (RDA) equipment, such as WSN, it acts
when connected with a Master node or when directly connected to sensors and transducers
(i.e., Stand-Alone weather station, Stand-Alone monitoring camera).
Fig. 3. GPRS Gateway
The main hardware components that characterize the gateway are:
a miniaturized GSM/GPRS modem, with embedded TCP/IP stack (Sveda et al.,
2005), (Jain et al., 1990);
a powerful 50 MHz clock microcontroller responsible for coordinating the bidirectional
data exchange between the modem and the master node to handle communication with
the Remote Server;
an additional 128 KB SRAM memory added in order to allow for data buffering, even
if the wide area link is lost;
several A/D channels available for connecting additional analog sensors and a battery
voltage monitor.
Since there is usually no access to a power supply infrastructure, the hardware design has
also been oriented to implement low power operating modalities, using a 12 V rechargeable
battery and a 20 W solar panel.
Data between the Gateway and Protocol Handler are carried out over TCP-IP communica-
tion and encapsulated in a custom protocol; from both local and remote interfaces it is also
possible to access part of the Gateways conguration settings. The low-level rmware im-
plementation of communication protocol also focuses on facing wide area link failures. Since
the gateway is always connected with the Remote Server, preliminary connectivity experi-
ments demonstrated a number of possible inconveniences, most of them involving the Service
Provider Access Point Name (APN) and Gateway GPRS Support Node (GGSN) subsystems. In
order to deal with these drawbacks, custom procedures called Dynamic Session Re-negotiation
(DSR) and Forced Session Re-negotiation (FSR), were implemented both on the gateway and on
the CMS server. This led to a signicant improvement in terms of disconnection periods and
packet loss rates.
The DSR procedure consists in a periodical bi-directional control packet exchange, aimed
at verifying the status of uplink and downlink channels on both sides (gateway and CMS).
This approach makes facing potential deadlocks possible if there is asymmetric socket failure,
which is when one device (acting as client or server) can correctly deliver data packets on the
TCP/IP connection but is unable to receive any. Once this event occurs (it has been observed
during long GPRS client connections, and is probably due to Service Provider Access Point
failures), the DSR procedure makes the client unit to restart the TCP socket connection with
the CMS.
Instead, the FSR procedure is operated on the server side when no data or service packets are
received from a gateway unit and a xed timeout elapses: in this case, the CMS closes the
TCP socket with that unit and waits for a new reconnection. On the other side, the gateway
unit should catch the close event exception and start a recovery procedure, after which a
new connection is re-established. If the close event should not be signaled to the gateway
(for example, the FSR procedure is started during an asymmetric socket failure), the gateway
would anyway enter the DSR recovery procedure.
In any case, once the link is lost, the gateway unit tries to reconnect with the CMS until a
connection is re-established.
4. Protocol Design
The most relevant system requirements, which lead the design of an efcient Medium Access
Control (MAC) and routing protocol for an environmental monitoring WSN, mainly concern
power consumption issues and the possiblity of a quick set-up and end-to-end communication
infrastructure that supports both synchronous and asynchronous queries. The most relevant
challenge is to make a system capable of running unattended for a long period, as nodes are
expected to be deployed in zones that are difcult to maintain. This calls for optimal energy
management since a limited resource and node failure may compromise WSN connectivity.
Therefore, the MAC and the network layer must be perfected ensuring that the energy used
is directly related to the amount of handled trafc and not to the overall working time.
200 m) and better power management. In order to manage different kinds of sensors, a compli-
ant sensor board was adopted, allowing up to 16 sensor plugs on the same node;, this makes a
single mote capable of sensing many environmental parameters at a time (Mattoli et al., 2005).
Sensor boards recognize the sensors and send Transducer Electronic Datasheets (TEDS) through
the network up to the server, making it possible for the system to recognize an automatic
sensor. The overall node stack architecture is shown in Fig. 2.
Communication Board
Power Board
Sensor Board
Overall size: 58x32x25 mm
Fig. 2. Node Stack Architecture
The GPRS embedded Gateway, shown in Fig. 3, is a stand-alone communication platform
designed to provide transparent, bi-directional wireless TCP-IP connectivity for remote mon-
itoring. In conjunction with Remote Data Acquisition (RDA) equipment, such as WSN, it acts
when connected with a Master node or when directly connected to sensors and transducers
(i.e., Stand-Alone weather station, Stand-Alone monitoring camera).
Fig. 3. GPRS Gateway
The main hardware components that characterize the gateway are:
a miniaturized GSM/GPRS modem, with embedded TCP/IP stack (Sveda et al.,
2005), (Jain et al., 1990);
a powerful 50 MHz clock microcontroller responsible for coordinating the bidirectional
data exchange between the modem and the master node to handle communication with
the Remote Server;
an additional 128 KB SRAM memory added in order to allow for data buffering, even
if the wide area link is lost;
several A/D channels available for connecting additional analog sensors and a battery
voltage monitor.
Since there is usually no access to a power supply infrastructure, the hardware design has
also been oriented to implement low power operating modalities, using a 12 V rechargeable
battery and a 20 W solar panel.
Data between the Gateway and Protocol Handler are carried out over TCP-IP communica-
tion and encapsulated in a custom protocol; from both local and remote interfaces it is also
possible to access part of the Gateways conguration settings. The low-level rmware im-
plementation of communication protocol also focuses on facing wide area link failures. Since
the gateway is always connected with the Remote Server, preliminary connectivity experi-
ments demonstrated a number of possible inconveniences, most of them involving the Service
Provider Access Point Name (APN) and Gateway GPRS Support Node (GGSN) subsystems. In
order to deal with these drawbacks, custom procedures called Dynamic Session Re-negotiation
(DSR) and Forced Session Re-negotiation (FSR), were implemented both on the gateway and on
the CMS server. This led to a signicant improvement in terms of disconnection periods and
packet loss rates.
The DSR procedure consists in a periodical bi-directional control packet exchange, aimed
at verifying the status of uplink and downlink channels on both sides (gateway and CMS).
This approach makes facing potential deadlocks possible if there is asymmetric socket failure,
which is when one device (acting as client or server) can correctly deliver data packets on the
TCP/IP connection but is unable to receive any. Once this event occurs (it has been observed
during long GPRS client connections, and is probably due to Service Provider Access Point
failures), the DSR procedure makes the client unit to restart the TCP socket connection with
the CMS.
Instead, the FSR procedure is operated on the server side when no data or service packets are
received from a gateway unit and a xed timeout elapses: in this case, the CMS closes the
TCP socket with that unit and waits for a new reconnection. On the other side, the gateway
unit should catch the close event exception and start a recovery procedure, after which a
new connection is re-established. If the close event should not be signaled to the gateway
(for example, the FSR procedure is started during an asymmetric socket failure), the gateway
would anyway enter the DSR recovery procedure.
In any case, once the link is lost, the gateway unit tries to reconnect with the CMS until a
connection is re-established.
4. Protocol Design
The most relevant system requirements, which lead the design of an efcient Medium Access
Control (MAC) and routing protocol for an environmental monitoring WSN, mainly concern
power consumption issues and the possiblity of a quick set-up and end-to-end communication
infrastructure that supports both synchronous and asynchronous queries. The most relevant
challenge is to make a system capable of running unattended for a long period, as nodes are
expected to be deployed in zones that are difcult to maintain. This calls for optimal energy
management since a limited resource and node failure may compromise WSN connectivity.
Therefore, the MAC and the network layer must be perfected ensuring that the energy used
is directly related to the amount of handled trafc and not to the overall working time.
Other important properties are scalability and adaptability of network topology, in terms of
number of nodes and their density. As a matter of fact, some nodes may either be turned off
may join the network afterward.
Taking these requirements into account, a MAC protocol and a routing protocol were imple-
mented.
4.1 MAC Layer Protocol
Taking the IEEE 802.11 Distributed Coordination Function (DCF) (IEEE St. 802.11, 1999) as
a starting point, several more energy efcient techniques have been proposed in literature to
avoid excessive power waste due to so called idle listening. They are based on periodical
preamble sampling performed at the receiver side in order to leave a low power state and
receive the incoming messages, as in the WiseMAC protocol (El-Hoiydi et al., 2003). Deriving
from the classical contention-based scheme, several protocols (S-MAC (Ye et al., 2002), TMAC
(Dam & Langendoen, 2003) and DMAC (Lu et al., 2004)) have been proposed to address the
overhead idle listening by synchronizing the nodes and implementing a duty cycle within
each slot.
Resorting to the above considerations, a class of MAC protocols was derived, named Syn-
chronous Transmission Asynchronous Reception (STAR) which is particularly suited for a at net-
work topology and benets from both WiseMAC and S-MAC schemes. More specically, due
to the introduction of a duty-cycle, it joins the power saving capability together with the ad-
vantages provided by the offset scheduling, without excessive overhead signaling. According
to the STAR MAC protocol, each node might be either in an idle mode, in which it remains for
a time interval T
l
(listening time), or in an energy saving sleeping state for a T
s
(sleeping time).
The transitions between states are synchronous with a period frame equal to T
f
=T
l
+T
s
par-
titioned in two sub-intervals; as a consequence, a duty-cycle function can also be introduced:
d =
T
l
T
l
+T
s
(1)
To provide the network with full communication capabilities, all the nodes need to be weakly
synchronized, meaning that they are aware at least of the awaking time of all their neighbors.
To this end, as Fig. 4 shows, a node sends a synchronization message (SYNC) frame by frame
to each of its neighbor nodes known to be in the listening mode (Synchronous Transmission),
whereas, during the set-up phase in which each node discovers the network topology, the con-
trol messages are asynchronously broadcasted. On the other hand, its neighbors periodically
awake and enter the listening state independently (Asynchronous Reception). The header of
the synchronization message contains the following elds: a unique node identier, the mes-
sage sequence number and the phase, or the time interval after which the sender claims to be
in the listening status waiting for both synchronization and data messages from its neighbors.
If the node is in the sleeping status, the phase is evaluated according to the following rule:
1
= T
l
(2)
where is the time remaining to the next frame beginning. Conversely, if the mote is in the
listening status, is computed as:
2
= +T
s
(3)
In order to fully characterize the STAR MAC approach, the related energy cost normalized
can be evaluated as it follows:
NODE 1
NODE 3
NODE 2
t
t
t
l
T
s
T
f
T
l
T
s
T
s
T
l
T
f
T
f
T
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC SSYN YN
SYNC YN YN
SYNC YN YN
Fig. 4. STAR MAC Protocol Synchronization Messages Exchange
C = c
rx
dT
f
+ c
sleep
[T
f
(1 d) NT
pkt
] + NC
tx
[mAh] (4)
where c
sleep
and c
rx
represent the sleeping and the receiving costs [mA] and C
tx
is the single
packet transmission costs [mAh], T
f
is the frame interval [s], d is the duty cycle, T
pkt
is the
synchronization packet time length [s] and nally N is the number of neighbors. When the
following inequality is hold:
NT
pkt
T
f
(5)
then:
C c
rx
dT
f
+ c
sleep
T
f
(1 d) + NC
tx
[mAh] (6)
The protocol cost normalized to the synchronization time is nally:
C
T
f
= c
rx
d + c
sleep
(1 d) +
NC
tx
Tf
[mA] (7)
As highlighted in Table 1, it usually happens that c
tx
c
sleep
c
rx
, where c
tx
= C
tx
/T
pkt
and
T
pkt
is the packet transmission time [s] assumed equal to 100 ms as worst case. This means
that the major contribution to the overall cost is represented by the listening period that the
STAR MAC protocol tries to suitably minimize.
c
rx
12 mA
c
sleep
0.01 mA
C
tx
30 mAh
c
tx
0.001 mA
Table 1. Power Consumption Parameters for the Considered Platform.
In Fig. 5(a) the normalized cost versus the number of neighbor nodes is shown for the S-MAC
and STAR MAC schemes. It is worth noticing that the performance of the proposed protocol
is better with respect to the existing approach for a number of neighbor nodes greater than 7.
In Fig. 5(b) the normalized costs of S-MAC and STAR MAC approaches are compared with
Other important properties are scalability and adaptability of network topology, in terms of
number of nodes and their density. As a matter of fact, some nodes may either be turned off
may join the network afterward.
Taking these requirements into account, a MAC protocol and a routing protocol were imple-
mented.
4.1 MAC Layer Protocol
Taking the IEEE 802.11 Distributed Coordination Function (DCF) (IEEE St. 802.11, 1999) as
a starting point, several more energy efcient techniques have been proposed in literature to
avoid excessive power waste due to so called idle listening. They are based on periodical
preamble sampling performed at the receiver side in order to leave a low power state and
receive the incoming messages, as in the WiseMAC protocol (El-Hoiydi et al., 2003). Deriving
from the classical contention-based scheme, several protocols (S-MAC (Ye et al., 2002), TMAC
(Dam & Langendoen, 2003) and DMAC (Lu et al., 2004)) have been proposed to address the
overhead idle listening by synchronizing the nodes and implementing a duty cycle within
each slot.
Resorting to the above considerations, a class of MAC protocols was derived, named Syn-
chronous Transmission Asynchronous Reception (STAR) which is particularly suited for a at net-
work topology and benets from both WiseMAC and S-MAC schemes. More specically, due
to the introduction of a duty-cycle, it joins the power saving capability together with the ad-
vantages provided by the offset scheduling, without excessive overhead signaling. According
to the STAR MAC protocol, each node might be either in an idle mode, in which it remains for
a time interval T
l
(listening time), or in an energy saving sleeping state for a T
s
(sleeping time).
The transitions between states are synchronous with a period frame equal to T
f
=T
l
+T
s
par-
titioned in two sub-intervals; as a consequence, a duty-cycle function can also be introduced:
d =
T
l
T
l
+T
s
(1)
To provide the network with full communication capabilities, all the nodes need to be weakly
synchronized, meaning that they are aware at least of the awaking time of all their neighbors.
To this end, as Fig. 4 shows, a node sends a synchronization message (SYNC) frame by frame
to each of its neighbor nodes known to be in the listening mode (Synchronous Transmission),
whereas, during the set-up phase in which each node discovers the network topology, the con-
trol messages are asynchronously broadcasted. On the other hand, its neighbors periodically
awake and enter the listening state independently (Asynchronous Reception). The header of
the synchronization message contains the following elds: a unique node identier, the mes-
sage sequence number and the phase, or the time interval after which the sender claims to be
in the listening status waiting for both synchronization and data messages from its neighbors.
If the node is in the sleeping status, the phase is evaluated according to the following rule:
1
= T
l
(2)
where is the time remaining to the next frame beginning. Conversely, if the mote is in the
listening status, is computed as:
2
= +T
s
(3)
In order to fully characterize the STAR MAC approach, the related energy cost normalized
can be evaluated as it follows:
NODE 1
NODE 3
NODE 2
t
t
t
l
T
s
T
f
T
l
T
s
T
s
T
l
T
f
T
f
T
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC YN YN
SYNC SSYN YN
SYNC YN YN
SYNC YN YN
Fig. 4. STAR MAC Protocol Synchronization Messages Exchange
C = c
rx
dT
f
+ c
sleep
[T
f
(1 d) NT
pkt
] + NC
tx
[mAh] (4)
where c
sleep
and c
rx
represent the sleeping and the receiving costs [mA] and C
tx
is the single
packet transmission costs [mAh], T
f
is the frame interval [s], d is the duty cycle, T
pkt
is the
synchronization packet time length [s] and nally N is the number of neighbors. When the
following inequality is hold:
NT
pkt
T
f
(5)
then:
C c
rx
dT
f
+ c
sleep
T
f
(1 d) + NC
tx
[mAh] (6)
The protocol cost normalized to the synchronization time is nally:
C
T
f
= c
rx
d + c
sleep
(1 d) +
NC
tx
Tf
[mA] (7)
As highlighted in Table 1, it usually happens that c
tx
c
sleep
c
rx
, where c
tx
= C
tx
/T
pkt
and
T
pkt
is the packet transmission time [s] assumed equal to 100 ms as worst case. This means
that the major contribution to the overall cost is represented by the listening period that the
STAR MAC protocol tries to suitably minimize.
c
rx
12 mA
c
sleep
0.01 mA
C
tx
30 mAh
c
tx
0.001 mA
Table 1. Power Consumption Parameters for the Considered Platform.
In Fig. 5(a) the normalized cost versus the number of neighbor nodes is shown for the S-MAC
and STAR MAC schemes. It is worth noticing that the performance of the proposed protocol
is better with respect to the existing approach for a number of neighbor nodes greater than 7.
In Fig. 5(b) the normalized costs of S-MAC and STAR MAC approaches are compared with
1 2 3 4 5 6 7 8 9 10 11
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Neighbour Nodes
N
o
r
m
a
l
i
z
e
d

C
o
s
t

[
m
A
]
STAR MAC
SMAC
(a) Normalized Cost vs. Neighbor Nodes
1 2 3 4 5 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Duty Cycle [%]
N
o
r
m
a
l
i
z
e
d

C
o
s
t

[
m
A
]
SMAC
STAR MAC
(b) Normalized Cost vs. Duty Cycle Duration
Fig. 5. STAR MAC Performance
respect to the duty cycle duration for a number of neighbor nodes equal to 8. It is possible to
notice that for d <3.5% the proposed protocol provide a signicant gain.
Nevertheless, for densely deployed or high trafc loaded WSN, STAR MAC approach might
suffers the shortcoming of cost increasing due to the large number of unicasted messages. To
limit this effect, an enhanced approach, named STAR+, was introduced, aiming at minimizing
also the packet transmission cost. According to it, only one synchronization packet is multi-
casted to all the neighbor nodes belonging to a subset, i.e., such that they are jointly awake
for a time interval greater than T
l
. This leads to an additional advantage, as the number
of neighbors increases allowing better performance with respect to scalability and a power
saving too. Besides, the synchronization overhead is reduced with a consequent collisions
lowering. Under this hypothesis the normalized cost might be expressed as:
C
T
f
= c
rx
d + c
sleep
(1 d) +
KC
tx
Tf
[mA] (8)
where K is the number of subsets. Since K N, the normalized cost results to be remarkably
lowered, especially if number of nodes and duty-cycle get higher, even if the latter case is
inherently power consuming.
4.2 Network Layer Protocol
In order to evaluate the capability of the proposed MAC scheme in establishing effective end-
to-end communications within a WSN, a routing protocol was introduced and integrated ac-
cording to the cross layer design principle (Shakkottai et al., 2003). In particular, we refer to
a proactive algorithm belonging to the class link-state protocol that enhance the capabilities
of the Link Estimation Parent Selection (LEPS) protocol. It is based on periodically information
needed for building and maintaining the local routing table, depicted in Table 2. However,
our approach resorts both to the signaling introduced by the MAC layer (i.e., synchronization
message) and by the Network layer (i.e., ping message), with the aim of minimizing the over-
head and make the systemmore adaptive in a cross layer fashion. In particular, the parameters
transmitted along a MAC synchronization message, with period T
f
, are the following:
next hop (NH) to reach the gateway, that is, the MAC address of the one hop neighbor;
distance (HC) to the gateway in terms of number of needed hops;
phase (PH) that is the schedule time at which the neighbor enter in listening mode ac-
cording to Equation (2) and Equation (3);
link quality (LQ) estimation as the ratio of correctly received and the expected synchro-
nization messages from a certain neighbor.
Target NH HC PH LQ BL CL
Sink 1 A N
A

A

A
B
A
C
A
B N
B

B

B
B
B
C
B
Sink 2 C N
C

C

C
B
C
C
C
D N
D

D

D
B
D
C
D
Table 2. Routing Table General Structure
On the other hand, the parameters related to long-term phenomena are carried out by the
ping messages, with period T
p
T
f
, in order to avoid unnecessary control trafcs and, thus,
reducing congestion. Particularly, they are:
battery level (BL) (i.e., an estimation of the energy available at that node);
congestion level (CL) in terms of the ratio between the number of packets present in the
local buffer and the maximum number of packets to be stored in.
Once, the routing table has been lled with these parameters, it is possible to derive the proper
metric by means of a weighted summation of them. It is worth mentioning that the routing
table might indicate more than one destination (sink) thanks to the ping messages that keep
trace of the intermediate nodes within the message header.
5. Software and End User Interface Design
The software implementation was developed, considering a node as both a single element
in charge of accomplishing prearranged tasks and as a part of a complex network in which
each component plays a crucial role in the networks maintenance. As far as the former aspect
is concerned, several TinyOS modules were implemented for managing high and low power
states and for realizing a nite state machine, querying sensors at xed intervals and achieving
anti-blocking procedures, in order to avoid software failure or deadlocks and provide a robust
stand alone system. On the other hand, the node has to interact with neighbors and provide
adequate connectivity to carry the messages through the network, regardless of the destina-
tion. Consequently, additional modules were developed according to a cross layer approach
that are in charge of managing STAR MAC and multihop protocols. Furthermore, other mod-
ules are responsible for handling and forwarding messages, coming from other nodes or from
the gateway itself. Messages are not only sensing (i.e., measures, battery level) but also control
and management messages (i.e., synchronization, node reset). As a result, a full interaction
between the nal user and the WSN is guaranteed.
The nal user may check the system status through graphical user interface (GUI) accessible
via web. After the log-in phase, the user can select the proper pilot site. For each site the
deployed WSNtogether with the gateway is schematically represented through an interactive
map. In addition to this, the related sensors display individual or aggregate time diagrams
1 2 3 4 5 6 7 8 9 10 11
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Neighbour Nodes
N
o
r
m
a
l
i
z
e
d

C
o
s
t

[
m
A
]
STAR MAC
SMAC
(a) Normalized Cost vs. Neighbor Nodes
1 2 3 4 5 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Duty Cycle [%]
N
o
r
m
a
l
i
z
e
d

C
o
s
t

[
m
A
]
SMAC
STAR MAC
(b) Normalized Cost vs. Duty Cycle Duration
Fig. 5. STAR MAC Performance
respect to the duty cycle duration for a number of neighbor nodes equal to 8. It is possible to
notice that for d <3.5% the proposed protocol provide a signicant gain.
Nevertheless, for densely deployed or high trafc loaded WSN, STAR MAC approach might
suffers the shortcoming of cost increasing due to the large number of unicasted messages. To
limit this effect, an enhanced approach, named STAR+, was introduced, aiming at minimizing
also the packet transmission cost. According to it, only one synchronization packet is multi-
casted to all the neighbor nodes belonging to a subset, i.e., such that they are jointly awake
for a time interval greater than T
l
. This leads to an additional advantage, as the number
of neighbors increases allowing better performance with respect to scalability and a power
saving too. Besides, the synchronization overhead is reduced with a consequent collisions
lowering. Under this hypothesis the normalized cost might be expressed as:
C
T
f
= c
rx
d + c
sleep
(1 d) +
KC
tx
Tf
[mA] (8)
where K is the number of subsets. Since K N, the normalized cost results to be remarkably
lowered, especially if number of nodes and duty-cycle get higher, even if the latter case is
inherently power consuming.
4.2 Network Layer Protocol
In order to evaluate the capability of the proposed MAC scheme in establishing effective end-
to-end communications within a WSN, a routing protocol was introduced and integrated ac-
cording to the cross layer design principle (Shakkottai et al., 2003). In particular, we refer to
a proactive algorithm belonging to the class link-state protocol that enhance the capabilities
of the Link Estimation Parent Selection (LEPS) protocol. It is based on periodically information
needed for building and maintaining the local routing table, depicted in Table 2. However,
our approach resorts both to the signaling introduced by the MAC layer (i.e., synchronization
message) and by the Network layer (i.e., ping message), with the aim of minimizing the over-
head and make the systemmore adaptive in a cross layer fashion. In particular, the parameters
transmitted along a MAC synchronization message, with period T
f
, are the following:
next hop (NH) to reach the gateway, that is, the MAC address of the one hop neighbor;
distance (HC) to the gateway in terms of number of needed hops;
phase (PH) that is the schedule time at which the neighbor enter in listening mode ac-
cording to Equation (2) and Equation (3);
link quality (LQ) estimation as the ratio of correctly received and the expected synchro-
nization messages from a certain neighbor.
Target NH HC PH LQ BL CL
Sink 1 A N
A

A

A
B
A
C
A
B N
B

B

B
B
B
C
B
Sink 2 C N
C

C

C
B
C
C
C
D N
D

D

D
B
D
C
D
Table 2. Routing Table General Structure
On the other hand, the parameters related to long-term phenomena are carried out by the
ping messages, with period T
p
T
f
, in order to avoid unnecessary control trafcs and, thus,
reducing congestion. Particularly, they are:
battery level (BL) (i.e., an estimation of the energy available at that node);
congestion level (CL) in terms of the ratio between the number of packets present in the
local buffer and the maximum number of packets to be stored in.
Once, the routing table has been lled with these parameters, it is possible to derive the proper
metric by means of a weighted summation of them. It is worth mentioning that the routing
table might indicate more than one destination (sink) thanks to the ping messages that keep
trace of the intermediate nodes within the message header.
5. Software and End User Interface Design
The software implementation was developed, considering a node as both a single element
in charge of accomplishing prearranged tasks and as a part of a complex network in which
each component plays a crucial role in the networks maintenance. As far as the former aspect
is concerned, several TinyOS modules were implemented for managing high and low power
states and for realizing a nite state machine, querying sensors at xed intervals and achieving
anti-blocking procedures, in order to avoid software failure or deadlocks and provide a robust
stand alone system. On the other hand, the node has to interact with neighbors and provide
adequate connectivity to carry the messages through the network, regardless of the destina-
tion. Consequently, additional modules were developed according to a cross layer approach
that are in charge of managing STAR MAC and multihop protocols. Furthermore, other mod-
ules are responsible for handling and forwarding messages, coming from other nodes or from
the gateway itself. Messages are not only sensing (i.e., measures, battery level) but also control
and management messages (i.e., synchronization, node reset). As a result, a full interaction
between the nal user and the WSN is guaranteed.
The nal user may check the system status through graphical user interface (GUI) accessible
via web. After the log-in phase, the user can select the proper pilot site. For each site the
deployed WSNtogether with the gateway is schematically represented through an interactive
map. In addition to this, the related sensors display individual or aggregate time diagrams
for each node with an adjustable time interval (Start/Stop) for the observation. System moni-
toring could be performed both at a high level with a user friendly GUI and at a low level by
means of message logging.
Fig. 6 shows some friendly Flash Player applications that, based on mathematical models, an-
alyze the entire amount of data in a selectable period and provide ready-to-use information.
Fig. 6(a) specically shows the aggregate data models for three macro-parameters, such as
vineyard water management, plant physiological activity and pest management. The appli-
cation, using cross light colors for each parameter, points out normal (green), mild (yellow)
or heavy (red) stress conditions and provides suggestions to the farmer on how to apply pes-
ticides or water in a certain part of the vineyard. Fig. 6(b) shows a graphical representation
of the soil moisture measurement. Soil moisture sensors positioned at different depths in the
vineyard make it possible to verify whether a summer rain runs off on the soil surface or
seeps into the earth and provokes benecial effects on the plants: this can be appreciated with
a rapid look at the soil moisture aggregate report which, shows the moisture sensors at two
depths with the moisture differences colore in green tones. Fig. 6(c) highlights stress condi-
tions on plants, due to dry soil and/or to hot weather thanks to the accurate trunk diametric
growth sensor that can follow each minimal variation in the trunk giving important infor-
mation on plant living activity. Finally, Fig. 6(d) shows a vineyard map: the green spots are
wireless units, distributed in a vineyard of one hectare.
6. Real World Experiences
The WSN system described above was developed and deployed in three pilot sites and in a
greenhouse. Since 2005, an amount of 198 sensors and 50 nodes have continuously sent data
to a remote server. The collected data represents a unique database of information on grape
growth useful for investigating the differences between cultivation procedures, environments
and treatments.
6.1 Pilot Sites Description
The rst pilot site was deployed in November 2005 on a sloped vineyard of the Montepaldi
farm in Chianti Area (Tuscany - Italy). The vineyard is a wide area where 13 nodes (including
the master node) with 24 sensors, running STAR MAC and dynamic routing protocols were
successfully deployed. The deployment took place in two different steps: during the rst one,
6 nodes (nodes 9,10,14,15,16,17) were placed to perform an exhaustive one week test. The
most important result regards the multi-hop routing efciency, estimated as:
MHop
=
M
EU
M
ex
(9)
where
MHop
is the efciency, M
EU
are the messages correctly received by the remote user and
M
ex
are the expected transmitted messages. For the gateway neighbors,
MHop
is very high,
over 90%. However, even nodes far fromthe gateway (i.e., concerning an end-to-end multihop
path) show a message delivery rate (MDR) of over 80%. This means that the implemented
routing protocol does not affect communication reliability. After the second deployment, in
which nodes 11,12,13,18,19,20 were arranged, the increased number of collisions changed the
global efciency, thus decreasing the messages that arrived to the end user, except for nodes
18,19,20, in which an upgraded rmware release was implemented. The related results are
detailed in Table 3.
GoodFood EU Integrated Project
Food Safety and Quality Monitoring with Microsystems
Home Logout
Site 1 - Montepaldi Farm
This site requires Macromedia Flash Player. Download it
GoodFood - EU Integrated Project All rights reserved

Powered by
C.S.I.A.F. Universit degli Studi di Firenze

Page 1 of 1 GoodFood Project
29/09/2009 http://www.unifi.it/midra/goodfood/dashboard.php
(a) Aggregate Data Models for Vineyard Water
Management, Plant Physiological Activity and
Pest Management

Powered by

01/10/2009 http://www.unifi.it/midra/goodfood/showIDW.php
(b) Soil Moisture Aggregation Report: the up-
per map represent soil moisture @ 10 cm in the
soil and the lower map represents soil moisture
@ 35cm in the vineyard after a slipping rain
(c) Trunk Diametric Growth Diagram: daily and
nightly metabolic phases
(d) Distributed Wireless Nodes in a Vineyard
Fig. 6. Flash Player User Interface
This conrms the robustness of the network installed and the reliability of the adopted com-
munications solution, also considering the power consumption issues: batteries were replaced
on March 11
th
2006 in order to face the entire farming season. After that, eleven months passed
before the rst battery replacement occurred on February 11
th
2007, conrming our expecta-
tions and fully matching the user requirements. The overall Montepaldi system has been run-
ning unattended for one year and a half and is going to be a permanent pilot site. So far, nearly
2 million samples from the Montepaldi vineyard have been collected and stored in the server
at the University of Florence Information Services Centre (CSIAF), helping agronomist experts
improve wine quality through deeper insight on physical phenomena (such as weather and
soil) and the relationship with grape growth.
The second pilot site was deployed on a farm in the Chianti Classico with 10 nodes and 50
sensors at about 500 m above sea level on a stony hill area of 2.5 hectares. The environmental
for each node with an adjustable time interval (Start/Stop) for the observation. System moni-
toring could be performed both at a high level with a user friendly GUI and at a low level by
means of message logging.
Fig. 6 shows some friendly Flash Player applications that, based on mathematical models, an-
alyze the entire amount of data in a selectable period and provide ready-to-use information.
Fig. 6(a) specically shows the aggregate data models for three macro-parameters, such as
vineyard water management, plant physiological activity and pest management. The appli-
cation, using cross light colors for each parameter, points out normal (green), mild (yellow)
or heavy (red) stress conditions and provides suggestions to the farmer on how to apply pes-
ticides or water in a certain part of the vineyard. Fig. 6(b) shows a graphical representation
of the soil moisture measurement. Soil moisture sensors positioned at different depths in the
vineyard make it possible to verify whether a summer rain runs off on the soil surface or
seeps into the earth and provokes benecial effects on the plants: this can be appreciated with
a rapid look at the soil moisture aggregate report which, shows the moisture sensors at two
depths with the moisture differences colore in green tones. Fig. 6(c) highlights stress condi-
tions on plants, due to dry soil and/or to hot weather thanks to the accurate trunk diametric
growth sensor that can follow each minimal variation in the trunk giving important infor-
mation on plant living activity. Finally, Fig. 6(d) shows a vineyard map: the green spots are
wireless units, distributed in a vineyard of one hectare.
6. Real World Experiences
The WSN system described above was developed and deployed in three pilot sites and in a
greenhouse. Since 2005, an amount of 198 sensors and 50 nodes have continuously sent data
to a remote server. The collected data represents a unique database of information on grape
growth useful for investigating the differences between cultivation procedures, environments
and treatments.
6.1 Pilot Sites Description
The rst pilot site was deployed in November 2005 on a sloped vineyard of the Montepaldi
farm in Chianti Area (Tuscany - Italy). The vineyard is a wide area where 13 nodes (including
the master node) with 24 sensors, running STAR MAC and dynamic routing protocols were
successfully deployed. The deployment took place in two different steps: during the rst one,
6 nodes (nodes 9,10,14,15,16,17) were placed to perform an exhaustive one week test. The
most important result regards the multi-hop routing efciency, estimated as:
MHop
=
M
EU
M
ex
(9)
where
MHop
is the efciency, M
EU
are the messages correctly received by the remote user and
M
ex
are the expected transmitted messages. For the gateway neighbors,
MHop
is very high,
over 90%. However, even nodes far fromthe gateway (i.e., concerning an end-to-end multihop
path) show a message delivery rate (MDR) of over 80%. This means that the implemented
routing protocol does not affect communication reliability. After the second deployment, in
which nodes 11,12,13,18,19,20 were arranged, the increased number of collisions changed the
global efciency, thus decreasing the messages that arrived to the end user, except for nodes
18,19,20, in which an upgraded rmware release was implemented. The related results are
detailed in Table 3.
GoodFood EU Integrated Project
Food Safety and Quality Monitoring with Microsystems
Home Logout
Site 1 - Montepaldi Farm

Powered by

29/09/2009 http://www.unifi.it/midra/goodfood/dashboard.php
(a) Aggregate Data Models for Vineyard Water
Management, Plant Physiological Activity and
Pest Management

Powered by

01/10/2009 http://www.unifi.it/midra/goodfood/showIDW.php
(b) Soil Moisture Aggregation Report: the up-
per map represent soil moisture @ 10 cm in the
soil and the lower map represents soil moisture
@ 35cm in the vineyard after a slipping rain
(c) Trunk Diametric Growth Diagram: daily and
nightly metabolic phases
(d) Distributed Wireless Nodes in a Vineyard
Fig. 6. Flash Player User Interface
This conrms the robustness of the network installed and the reliability of the adopted com-
munications solution, also considering the power consumption issues: batteries were replaced
on March 11
th
2006 in order to face the entire farming season. After that, eleven months passed
before the rst battery replacement occurred on February 11
th
2007, conrming our expecta-
tions and fully matching the user requirements. The overall Montepaldi system has been run-
ning unattended for one year and a half and is going to be a permanent pilot site. So far, nearly
2 million samples from the Montepaldi vineyard have been collected and stored in the server
at the University of Florence Information Services Centre (CSIAF), helping agronomist experts
improve wine quality through deeper insight on physical phenomena (such as weather and
soil) and the relationship with grape growth.
The second pilot site was deployed on a farm in the Chianti Classico with 10 nodes and 50
sensors at about 500 m above sea level on a stony hill area of 2.5 hectares. The environmental
Location MDR
Node 9 72.2%
Node 10 73.7%
Node 11 88.5%
Node 12 71.4%
Node 13 60.4%
Node 14 57.2%
Node 15 45.6%
Node 16 45.4%
Node 17 92.1%
Node 18 87.5%
Node 19 84.1%
Table 3. Message Delivery Rate for the Montepaldi Farm Pilot Site
variations of the the terroir have been monitored since July 2007, producing one of the most
appreciated wines in the world.
Finally, the third WSN was installed in Southern France in the vineyard of Peach Rouge at
Gruissan. High sensor density was established to guarantee measurement redundancy and to
provide a deeper knowledge of the phenomena variation in an experimental vineyard where
micro-zonation has been applied and where water management experiments have been per-
formed for studying plant reactions and grape quality.
6.2 Greenhouse
An additional deployment at the University of Florence Greenhouse was performed to let the
agronomist experts conduct experiments even in seasons like Fall and Winter, where plants
are quiescent, thus breaking free from the natural growth trend. This habitat also creates the
opportunity to run several experiments on the test plants, in order to evaluate their responses
under different stimuli using in situ sensors.
The greenhouse environmental features are completely different fromthose of the vineyard: as
a matter of fact, the multipath propagation effects become relevant, due to the indoor scenario
and the presence of a metal infrastructure. A highly dense node deployment, in terms of
both nodes and sensors, might imply an increased network trafc load. Nevertheless, the
same node rmware and hardware used in the vineyard are herein adopted; this leads to a
resulting star topology as far as end-to-end communications are concerned.
Furthermore, 6 nodes have been in the greenhouse since June 2005, and 30 sensors have con-
stantly monitored air temperature and humidity, plants soil moisture and temperature, differ-
ential leaf temperature and trunk diametric growth. The sensing period is equal to 10 min-
utes, less than the climate/plant parameter variations, providing redundant data storage. The
WSN message delivery rate is extremely high: the efciency is over 95%, showing that a low
number of messages are lost.
7. VineSense
The fruitful experience of the three pilot sites was gathered by a newItalian company, Netsens,
founded as a spinoff of the University of Florence. Netsens has designed a new monitoring
system called VineSense based on WSN technology and oriented towards market and user
applications.
VineSense exalts the positive characteristics of the experimental system and overcomes the
problems encountered in past experiences, thus achieving an important position in the wire-
less monitoring market.
The rst important outcome of the experimental system, enhanced by VineSense is the idea
of an end-to-end system. Sensors deployed in the eld constantly monitor and send measure-
ments to a remote server through the WSN. Data can be queried and analyzed by nal users
thanks to the professional and user-friendly VineSense web interface. Qualied mathematic
models are applied to monitoring parameters and provide predictions on diseases and plant
growth, increasing agronomists knowledge and reducing costs while paving the road for new
vineyard management.
VineSense improves many aspects of the experimental system, both in electronics and
telecommunications.
The MAC and Routing protocol tested in the previous experimental system showed such im-
portant and signicant results in terms of reliability that the same scheme was also adopted
in the VineSense system and minimal changes were introduced: the routing protocol is lighter
in terms of data exchange, building the route with different parameters, aimed at increasing
the message success rate, such as master node distance and received signal strength.
A more secure data encryption was adopted in data messages to protect customers from ma-
licious snifng or to discourage possible competitors from decrypting network data.
Furthermore, a unique key-lock sequence was also implemented on each wireless node to
prevent stealing, ensuring correct use with only genuine Netsens products and only in com-
bination with its master node, which comes from the factory.
The new wireless nodes are smaller, more economical, more robust and suited for vineyard
operations with machines and tractors. The electronics are more fault-tolerant, easier to in-
stall and more energy efcient: only a 2200 mAh lithium battery for 2-3 years of continuous
running without human intervention. Radio coverage has been improved up to 350 m and
nodes deployment can be easily performed by end users who can rely on a smart installation
system with instantaneous radio coverage recognition. Some users have also experimented
with larger area coverage, measuring a point-to-point communication of about 600 m in the
line of sight.
Hypothetically, a VineSense system could be composed of up to 255 wireless nodes and more
than 2500 sensors, considering a full sensor set per node, but since it is a commercial system
these numbers are much more than necessary to cover farmers needs.
Sensors used in the VineSense system are low-cost, state-of-the-art devices designed by Net-
sens for guaranteeing the best accuracy-reliability-price ratio. The choice of Netsens to de-
velop custom and reliable sensors for the VineSense system is not only strategic from a mar-
keting point of view, since it frees VineSense from any kind of external problems, such as
external supplying, delays, greater costs and compliancy. It is also a consequence of the Sys-
tem Vision, where VineSense is not only a wireless communication system product, but an
entire system with no black holes inside so as to provide the customer with a complete
system with better support.
The VineSense wireless-sensor unit is shown in Fig. 7(a).
Recovery strategies and communication capabilities of the stand-alone GPRS gateway have
been improved: in fact, data received by wireless nodes are both forwarded in real-time to a
remote server and temporarily stored on board in case of abrupt disconnections; moreover,
Location MDR
Node 9 72.2%
Node 10 73.7%
Node 11 88.5%
Node 12 71.4%
Node 13 60.4%
Node 14 57.2%
Node 15 45.6%
Node 16 45.4%
Node 17 92.1%
Node 18 87.5%
Node 19 84.1%
Table 3. Message Delivery Rate for the Montepaldi Farm Pilot Site
variations of the the terroir have been monitored since July 2007, producing one of the most
appreciated wines in the world.
Finally, the third WSN was installed in Southern France in the vineyard of Peach Rouge at
Gruissan. High sensor density was established to guarantee measurement redundancy and to
provide a deeper knowledge of the phenomena variation in an experimental vineyard where
micro-zonation has been applied and where water management experiments have been per-
formed for studying plant reactions and grape quality.
6.2 Greenhouse
An additional deployment at the University of Florence Greenhouse was performed to let the
agronomist experts conduct experiments even in seasons like Fall and Winter, where plants
are quiescent, thus breaking free from the natural growth trend. This habitat also creates the
opportunity to run several experiments on the test plants, in order to evaluate their responses
under different stimuli using in situ sensors.
The greenhouse environmental features are completely different fromthose of the vineyard: as
a matter of fact, the multipath propagation effects become relevant, due to the indoor scenario
and the presence of a metal infrastructure. A highly dense node deployment, in terms of
both nodes and sensors, might imply an increased network trafc load. Nevertheless, the
same node rmware and hardware used in the vineyard are herein adopted; this leads to a
resulting star topology as far as end-to-end communications are concerned.
Furthermore, 6 nodes have been in the greenhouse since June 2005, and 30 sensors have con-
stantly monitored air temperature and humidity, plants soil moisture and temperature, differ-
ential leaf temperature and trunk diametric growth. The sensing period is equal to 10 min-
utes, less than the climate/plant parameter variations, providing redundant data storage. The
WSN message delivery rate is extremely high: the efciency is over 95%, showing that a low
number of messages are lost.
7. VineSense
The fruitful experience of the three pilot sites was gathered by a newItalian company, Netsens,
founded as a spinoff of the University of Florence. Netsens has designed a new monitoring
applications.
VineSense exalts the positive characteristics of the experimental system and overcomes the
problems encountered in past experiences, thus achieving an important position in the wire-
less monitoring market.
The rst important outcome of the experimental system, enhanced by VineSense is the idea
of an end-to-end system. Sensors deployed in the eld constantly monitor and send measure-
ments to a remote server through the WSN. Data can be queried and analyzed by nal users
thanks to the professional and user-friendly VineSense web interface. Qualied mathematic
models are applied to monitoring parameters and provide predictions on diseases and plant
growth, increasing agronomists knowledge and reducing costs while paving the road for new
vineyard management.
VineSense improves many aspects of the experimental system, both in electronics and
telecommunications.
The MAC and Routing protocol tested in the previous experimental system showed such im-
portant and signicant results in terms of reliability that the same scheme was also adopted
in the VineSense system and minimal changes were introduced: the routing protocol is lighter
in terms of data exchange, building the route with different parameters, aimed at increasing
the message success rate, such as master node distance and received signal strength.
A more secure data encryption was adopted in data messages to protect customers from ma-
licious snifng or to discourage possible competitors from decrypting network data.
Furthermore, a unique key-lock sequence was also implemented on each wireless node to
prevent stealing, ensuring correct use with only genuine Netsens products and only in com-
bination with its master node, which comes from the factory.
The new wireless nodes are smaller, more economical, more robust and suited for vineyard
operations with machines and tractors. The electronics are more fault-tolerant, easier to in-
stall and more energy efcient: only a 2200 mAh lithium battery for 2-3 years of continuous
running without human intervention. Radio coverage has been improved up to 350 m and
nodes deployment can be easily performed by end users who can rely on a smart installation
system with instantaneous radio coverage recognition. Some users have also experimented
with larger area coverage, measuring a point-to-point communication of about 600 m in the
line of sight.
Hypothetically, a VineSense system could be composed of up to 255 wireless nodes and more
than 2500 sensors, considering a full sensor set per node, but since it is a commercial system
these numbers are much more than necessary to cover farmers needs.
Sensors used in the VineSense system are low-cost, state-of-the-art devices designed by Net-
sens for guaranteeing the best accuracy-reliability-price ratio. The choice of Netsens to de-
velop custom and reliable sensors for the VineSense system is not only strategic from a mar-
keting point of view, since it frees VineSense from any kind of external problems, such as
external supplying, delays, greater costs and compliancy. It is also a consequence of the Sys-
tem Vision, where VineSense is not only a wireless communication system product, but an
entire system with no black holes inside so as to provide the customer with a complete
system with better support.
The VineSense wireless-sensor unit is shown in Fig. 7(a).
Recovery strategies and communication capabilities of the stand-alone GPRS gateway have
been improved: in fact, data received by wireless nodes are both forwarded in real-time to a
remote server and temporarily stored on board in case of abrupt disconnections; moreover,
automatic reset and restart procedures avoid possible software deadlocks or GPRS network
failures. Finally, a high-gain antenna guarantees good GPRS coverage almost everywhere.
The GPRS gateway rmware has been implemented for remotely managing of the acquisition
settings, relieving users of the necessity of eld maintenance.
The GPRS gateway communication has been greatly improved introducing new different
communication interfaces, such as Ethernet connection (RJ-45), USB data downloading and
the possibility of driving an external Wi-Fi communication system for short-range transmis-
sions.
Since the beginning of 2010, the Always On connection started to be fully used and it
boosted the VineSense system, enlarging its possible eld of application: a complete bidi-
rectional communication was established between the GPRS unit in the eld and the remote
server at Netsens. The previous one way data ow, from the vineyard to the internet, was
gone over by a new software release, able to send instantaneous messages from the VineSense
web interface to the eld: the monitoring systemwas changed into the monitoring and control
system, sending automatic, scheduled or asynchronous commands to the gateway station or
to nodes, i.e. to open or close irrigation systems or simply to download a rmware upgrade.
In Fig. 7(b) the GPRS gateway with weather sensors is shown.
(a) VineSense Wireless Unit (b) VineSense GPRS Gateway
with Weather Sensors
Fig. 7. VineSense Hardware Elements
The web interface is the last part of VineSenses end-to-end: the great amount of data gathered
by the sensors and stored in the database needs a smart analysis tool to become useful and
usable. For this reason different tools are at the disposal of various kinds of users. On one
hand, some innovative tools such as control panels for real time monitoring or 2D chromatic
maps create a quick and easy approach to the interface. On the other hand, professional plots
and data ltering options allow experts or agronomists to study them more closely.
8. Agronomic Results
The use of VineSense in different scenarios with different agronomic aims has brought a large
amount of important results.
When VineSense is adopted to monitor soil moisture positive effects can be obtained for plants
and saving water, thus optimizing irrigation schedules. Some examples of this application
can be found in systems installed in the Egyptian desert where agriculture is successful only
through wise irrigation management. In such a terroir, plants suffer continuous hydric stress
during daylight due to high air temperature, low air humidity and hot sandy soils with a low
water retention capacity. Water is essential for plant survival and growth, an irrigation delay
can be fatal for the seasonal harvest therefore, a reliable monitoring system is necessary. The
adoption of VineSense in this scenario immediately resulted in continuous monitoring of the
irrigating system, providing an early warning whenever pump failure occurred. On the other
hand, the possibility to measure soil moisture at different depths allows agronomists to decide
on the right amount of water to provide plants; depending on different day temperatures and
soil moisture, pipe schedules can be changed in order to reduce water waste and increase
water available for plants.
An example of different pipe schedules is shown in Fig. 8.
Fig. 8. Different Pipe Schedules in Accordance with Soil Moisture Levels
Originally, the irrigation system was opened once a day for 5 hours giving 20 liters per day
(schedule 1); since sandy soils reach saturation very rapidly most of this water was wasted
in deeper soil layers; afterwards irrigation schedules were changed (schedule 2), giving the
same amount of water in two or more times per day; the water remained in upper soil layers
at plant root level, reducing wastes and increasing the amount of available water for plants,
as highlighted by soil moisture at 60 cm (blue plot).
Another important application of the VineSense systemuses the dendrometer to monitor plant
physiology. The trunk diametric sensor is a mechanical sensor with +/ 5 microns of accu-
racy; such an accurate sensor can appreciate stem micro variations occurring during day and
night, due to the xilematic ux inside the plant. Wireless nodes measure plant diameter ev-
ery 15 minutes, an appropriate time interval for following these changes and for creating a
plot showing this trend. In normal weather conditions, common physiologic activity can be
recognized by agronomists the same as a doctor can do reading an electrocardiogram; when
air temperature increases and air humidity falls in combining low soil moisture levels, plants
change their activity in order to face water stress, preserve their grapes and especially them-
automatic reset and restart procedures avoid possible software deadlocks or GPRS network
failures. Finally, a high-gain antenna guarantees good GPRS coverage almost everywhere.
The GPRS gateway rmware has been implemented for remotely managing of the acquisition
settings, relieving users of the necessity of eld maintenance.
The GPRS gateway communication has been greatly improved introducing new different
communication interfaces, such as Ethernet connection (RJ-45), USB data downloading and
the possibility of driving an external Wi-Fi communication system for short-range transmis-
sions.
Since the beginning of 2010, the Always On connection started to be fully used and it
boosted the VineSense system, enlarging its possible eld of application: a complete bidi-
rectional communication was established between the GPRS unit in the eld and the remote
server at Netsens. The previous one way data ow, from the vineyard to the internet, was
gone over by a new software release, able to send instantaneous messages from the VineSense
web interface to the eld: the monitoring systemwas changed into the monitoring and control
system, sending automatic, scheduled or asynchronous commands to the gateway station or
to nodes, i.e. to open or close irrigation systems or simply to download a rmware upgrade.
In Fig. 7(b) the GPRS gateway with weather sensors is shown.
(a) VineSense Wireless Unit (b) VineSense GPRS Gateway
with Weather Sensors
Fig. 7. VineSense Hardware Elements
The web interface is the last part of VineSenses end-to-end: the great amount of data gathered
by the sensors and stored in the database needs a smart analysis tool to become useful and
usable. For this reason different tools are at the disposal of various kinds of users. On one
hand, some innovative tools such as control panels for real time monitoring or 2D chromatic
maps create a quick and easy approach to the interface. On the other hand, professional plots
and data ltering options allow experts or agronomists to study them more closely.
8. Agronomic Results
The use of VineSense in different scenarios with different agronomic aims has brought a large
amount of important results.
When VineSense is adopted to monitor soil moisture positive effects can be obtained for plants
and saving water, thus optimizing irrigation schedules. Some examples of this application
can be found in systems installed in the Egyptian desert where agriculture is successful only
through wise irrigation management. In such a terroir, plants suffer continuous hydric stress
during daylight due to high air temperature, low air humidity and hot sandy soils with a low
water retention capacity. Water is essential for plant survival and growth, an irrigation delay
can be fatal for the seasonal harvest therefore, a reliable monitoring system is necessary. The
adoption of VineSense in this scenario immediately resulted in continuous monitoring of the
irrigating system, providing an early warning whenever pump failure occurred. On the other
hand, the possibility to measure soil moisture at different depths allows agronomists to decide
on the right amount of water to provide plants; depending on different day temperatures and
soil moisture, pipe schedules can be changed in order to reduce water waste and increase
water available for plants.
An example of different pipe schedules is shown in Fig. 8.
Fig. 8. Different Pipe Schedules in Accordance with Soil Moisture Levels
Originally, the irrigation system was opened once a day for 5 hours giving 20 liters per day
(schedule 1); since sandy soils reach saturation very rapidly most of this water was wasted
in deeper soil layers; afterwards irrigation schedules were changed (schedule 2), giving the
same amount of water in two or more times per day; the water remained in upper soil layers
at plant root level, reducing wastes and increasing the amount of available water for plants,
as highlighted by soil moisture at 60 cm (blue plot).
Another important application of the VineSense systemuses the dendrometer to monitor plant
physiology. The trunk diametric sensor is a mechanical sensor with +/ 5 microns of accu-
racy; such an accurate sensor can appreciate stem micro variations occurring during day and
night, due to the xilematic ux inside the plant. Wireless nodes measure plant diameter ev-
ery 15 minutes, an appropriate time interval for following these changes and for creating a
plot showing this trend. In normal weather conditions, common physiologic activity can be
recognized by agronomists the same as a doctor can do reading an electrocardiogram; when
air temperature increases and air humidity falls in combining low soil moisture levels, plants
change their activity in order to face water stress, preserve their grapes and especially them-
selves. This changed behavior can be registered by the dendrometer and plotted in the Vine-
Sense interface, warning agronomists about incoming risks; as a consequence, new irrigation
schedules can be carried into effect.
Fig. 9 shows an example of a plant diametric trend versus air temperature.
Fig. 9. Plant Diametric Trend vs. Air Temperature
The blue plot represents the air temperature in 10 days, from13
th
August until the 22
nd
August
2009 in Italy; the blue line becomes red when the temperature goes over a 35 degree threshold.
During the period in which the temperature is so high, plant stem variations are reduced due
to the lower amount of xilematic ux owing in its vessels, a symptom of water leakage.
WSN in agriculture are also useful for creating new databases with historical data: storing
information highlighting peculiarities and differences of vineyards provides agronomists an
important archive for better understanding variations in plant production capabilities and
grape ripening. Deploying wireless nodes on plants in interesting areas increases the knowl-
edge about a specic vineyard or a specic terroir, thus recording and proving the specicity
of a certain wine. I.E., the quality of important wines such CRU, coming fromonly one specic
vineyard, can be easily related to grape history: data on air temperature and humidity, plant
stress, irrigation and rain occurring during the farming season can assess a quality growing
process, that can be declared to buyers.
Finally, VineSense can be used to reduce environmental impact thanks to a more optimized
management of pesticides in order to reach a sustainable viticulture. Since many of the most
virulent vine diseases can grow in wet leaf conditions, it is very important to monitor leaf
wetness in a continuous and distributed way. Sensors deployed in different parts of vineyards
are a key element for agronomists in monitoring risky conditions: since wetness can change
very rapidly during the night in a vineyard and it is not homogeneous in a eld, a real time
distributed systemis the right solution for identifying risky conditions and deciding when and
where to apply chemical treatments. As a result, chemicals can be used only when they are
strictly necessary and only in small parts of the vineyard where they are really needed, thus
reducing the number of treatments per year and decreasing the amount of active substances
sprayed in the eld and in the environment. In some tests performed in 2009 in Chianti, the
amount of pesticides was reduced by 65% compared to the 2008 season.
Leaf wetness sensors on nodes 2 and 3 measure different wetness conditions as shown in Fig.
10. The upper part of the vineyard is usually wetter (brown plot) than the lower part (blue
plot) and sometimes leaf wetness persists for many hours, increasing the risk of attacks on
plants.
Fig. 10. Different Leaf Wetness Conditions in a Small Vineyard
9. Conclusion
This paper deals with the design, optimization and development of a practical solution for
application to the agro-food chain monitoring and control. The overall system was addressed
in terms of the experienced platform, network issues related both to communication protocols
between nodes and gateway operations up to the suitable remote user interface. Every con-
stitutive element of the system chain was described in detail in order to point out the features
and the remarkable advantages in terms of complexity reduction and usability.
To highlight the effectiveness and accurateness of the developed system, several case studies
were presented. Moreover, the encouraging and unprecedented results achieved by this ap-
proach and supported by several pilot sites into different vineyard in Italy and France were
shown.
The fruitful experience of some pilot sites was gathered by a new Italian company, Netsens,
founded as a spin off of the University of Florence. Netsens has designed a new monitoring
applications. In order to point out the improvements of the new solution respect to the ex-
perimental one, the main features of VineSense were described. Moreover, some important
agronomic results achieved by the use of VineSense in different scenarios were sketched out,
thus emphasizing the positive effects of the WSN technology in the agricultural environment.
Nowadays, the application of the solution described in this paper is under investigation to the
more general eld of environmental monitoring, due to its exibility, scalability, adaptability
and self-recongurability.
selves. This changed behavior can be registered by the dendrometer and plotted in the Vine-
Sense interface, warning agronomists about incoming risks; as a consequence, new irrigation
schedules can be carried into effect.
Fig. 9 shows an example of a plant diametric trend versus air temperature.
Fig. 9. Plant Diametric Trend vs. Air Temperature
The blue plot represents the air temperature in 10 days, from13
th
August until the 22
nd
August
2009 in Italy; the blue line becomes red when the temperature goes over a 35 degree threshold.
During the period in which the temperature is so high, plant stem variations are reduced due
to the lower amount of xilematic ux owing in its vessels, a symptom of water leakage.
WSN in agriculture are also useful for creating new databases with historical data: storing
information highlighting peculiarities and differences of vineyards provides agronomists an
important archive for better understanding variations in plant production capabilities and
grape ripening. Deploying wireless nodes on plants in interesting areas increases the knowl-
edge about a specic vineyard or a specic terroir, thus recording and proving the specicity
of a certain wine. I.E., the quality of important wines such CRU, coming fromonly one specic
vineyard, can be easily related to grape history: data on air temperature and humidity, plant
stress, irrigation and rain occurring during the farming season can assess a quality growing
process, that can be declared to buyers.
Finally, VineSense can be used to reduce environmental impact thanks to a more optimized
management of pesticides in order to reach a sustainable viticulture. Since many of the most
virulent vine diseases can grow in wet leaf conditions, it is very important to monitor leaf
wetness in a continuous and distributed way. Sensors deployed in different parts of vineyards
are a key element for agronomists in monitoring risky conditions: since wetness can change
very rapidly during the night in a vineyard and it is not homogeneous in a eld, a real time
distributed systemis the right solution for identifying risky conditions and deciding when and
where to apply chemical treatments. As a result, chemicals can be used only when they are
strictly necessary and only in small parts of the vineyard where they are really needed, thus
reducing the number of treatments per year and decreasing the amount of active substances
sprayed in the eld and in the environment. In some tests performed in 2009 in Chianti, the
amount of pesticides was reduced by 65% compared to the 2008 season.
Leaf wetness sensors on nodes 2 and 3 measure different wetness conditions as shown in Fig.
10. The upper part of the vineyard is usually wetter (brown plot) than the lower part (blue
plot) and sometimes leaf wetness persists for many hours, increasing the risk of attacks on
plants.
Fig. 10. Different Leaf Wetness Conditions in a Small Vineyard
9. Conclusion
This paper deals with the design, optimization and development of a practical solution for
application to the agro-food chain monitoring and control. The overall system was addressed
in terms of the experienced platform, network issues related both to communication protocols
between nodes and gateway operations up to the suitable remote user interface. Every con-
stitutive element of the system chain was described in detail in order to point out the features
and the remarkable advantages in terms of complexity reduction and usability.
To highlight the effectiveness and accurateness of the developed system, several case studies
were presented. Moreover, the encouraging and unprecedented results achieved by this ap-
proach and supported by several pilot sites into different vineyard in Italy and France were
shown.
The fruitful experience of some pilot sites was gathered by a new Italian company, Netsens,
founded as a spin off of the University of Florence. Netsens has designed a new monitoring
applications. In order to point out the improvements of the new solution respect to the ex-
perimental one, the main features of VineSense were described. Moreover, some important
agronomic results achieved by the use of VineSense in different scenarios were sketched out,
thus emphasizing the positive effects of the WSN technology in the agricultural environment.
Nowadays, the application of the solution described in this paper is under investigation to the
more general eld of environmental monitoring, due to its exibility, scalability, adaptability
and self-recongurability.
10. References
Blackmore, S. (1994). Precision Farming: An Introduction, Outlook on Agriculture Journal, Vol.
23, pp. 275-280.
Wang, N., Zhang, N. & Wang, M. (2006). Wireless sensors in agriculture and food industry -
Recent development and future perspective, Computers and Electronics in Agriculture
Journal, Vol. 50, pp. 114-120.
Akyildiz, I.F. & Xudong, W. (2005). A Survey on Wireless Mesh Networks, IEEE Communica-
tion Magazine, Vol. 43, pp. S23-S30.
Al-Karaki, J. & Kamal, A. (2004). Routing Techniques in Wireless Sensor Networks: a Survey,
IEEE Communication Magazine, Vol. 11, pp. 6-28.
Langendoen, K. & Halkes, G. (2004). Energy-Efcient Medium Access Control, The Embedded
Systems Handbook, pp. 2-30.
Mica2 Series, Avaiable on http://www.xbow.com.
Mattoli, V., Mondini, A., Razeeb, K.M., Oynn, B., Murphy, F., Bellis, S., Collodi, G., Manes,
A., Pennacchia, P., Mazzolai, B., & Dario, P. (2005). Development of a Programmable
Sensor Interface for Wireless Network Nodes for Intelligent Agricultural Applica-
tions, Proceedings of IE 2005, IEEE Computer and Communications Societies, Sydney,
pp. 1-6.
Sveda, M., Benes, P., Vrba, R. & Zezulka, F. (2005). Introduction to Industrial Sensor Net-
working, Handbook of Sensor Networks: Compact Wireless and Wired Sensing Systems,
pp. 10-24.
Jain, J.N. & Agrawala, A.K. (1990). Open Systems Interconnection: Its Architecture and Protocols,
Elsevier.
IEEE Standard 802.11 (1999). Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) Specications, IEEE Computer Society.
El-Hoiydi, A., Decotignie, J., Enz, C. & Le Roux, E. (2003). WiseMAC, an Ultra Low Power
MAC Protocol for the WiseNET Wireless Sensor Network, Proceedings of SENSYS
2003, Association for Computer Machinery, Los Angeles (CA), pp. 244-251.
Ye, W., Heidemann, J. &Estrin, D. (2002). An Energy-Efcient MAC Protocol for Wireless Sen-
sor Networks, Proceedings of INFOCOM 2002, IEEE Computer and Communications
Societies, New York (NY), pp. 1567-1576.
Dam, T. & Langendoen, K. (2003). An Adaptive Energy-Efcient MAC Protocol for Wireless
Sensor Networks, Proceedings of SENSYS 2003, Association for Computer Machinery,
Los Angeles (CA), pp. 171-180.
Lu, G., Krishnamachari, B. & Raghavendra, C. (2004). Adaptive Energy-Efcient and Low-
Latency MAC for Data Gathering in Sensor Networks, Proceedings of WMAN 2004,
Institut fur Medien.Informatik, Ulm (Germany), pp. 2440-2443.
Shakkottai, S., Rappaport, T. &Karlsson, P. (2003). Cross-Layer Design for Wireless Networks,
IEEE Communication Magazine, Vol. 41, pp. 77-80.
Wildlife Assessment using Wireless Sensor Networks
0
Wildlife Assessment using
Wireless Sensor Networks
LAMIA, Universit des Antilles et de la Guyane,
Campus de Schlcher, B.P. 7209, 97275 Schoelcher, French West Indies
France
1. Introduction
The endangered species always drew the attention of the scientic community since their dis-
appearance would cause irreplaceable loss. To help these species to survive, their habitat is
protected by the laws of environmental protection. Sometimes this protection is not enough,
because their natural evolution is the main cause of their disappearance. However, to save
them, it is sometimes possible to transfer them elsewhere that should be similar to their previ-
ous habitat to avoid disturbing the balance of wildlife. To model a habitat, several parameters
must be of interest and are generally dened by experts. This is the case for the number of
singing birds which will be studied in this paper.
Today, advances in sensor technology enable the monitoring of species and their habitat at
a very low cost. Indeed, the increasing sophistication of wireless sensors bids opportunities
that enable new challenges in a lot of areas, including the surveillance one. Progress in their
miniaturization leads to micro-sensors of size of cubic millimeters which, used in large quan-
tity, produce huge amounts of data. This paper promotes the use of sensors for monitoring
bird endangered in their habitat. Actual methods for counting endangered birds use mainly
human labor and because they are not really comprehensive leads to poor estimation. The
use of sensors deployed in critical environments can help the census of these species and even
generate new data on their customs.
Among the challenges that the use of the sensor technology enable, energy efciency is the
most critical for these wireless networks since battery depletion totally disables a sensor. In
addition, designing algorithms for wireless networks stems from the distributed computer
science domain with limited devices. Memory space and computational power are often of a
magnitude less than miles than their desktop counterparts. This paper investigate the problem
and proposes to approximate the number of birds by geometric means derived in a graph
problem.
Our paper is organized as follows. First, Section 2 provides an overview of techniques gen-
erally used to estimate the locations of multiple sources with a unknown sensor network.
Section 4 details our heuristics used to count birds. Section 5 introduces a distributed algo-
rithm for counting birds. Experimentation conrms the effectiveness of our counting systems
in Section 6. Then we conclude in Section 7 and gives an overview of our future work.
3
2. Previous Work
Source localization is an area of interest that has been widely studied in these recent years.
A comprehensive review of incentives techniques and source localization has been written by
Krimand Viberg in (Krim&Viberg, 1996) and it is not difcult to understand that problemhas
been of particular focus for military needs. Indeed, radar and sonars are a direct application
of source localization.
Several acoustic parameters such as bandwidth, distance sensors, reverberation and thus
change the way the location of the sources are handled. In addition, the algorithms of source
localization depends strongly on physics and rely on the sound characteristics of waveform to
calculate location sources. Waveform audio is known to be broadband (30Hz-15kHz) and sen-
sors usually record the sound from near-eld sources. The following presents some algorithms
of interest which satisfy these two properties. Near-elds algorithms like close-formed ones
(Smith. & Abel, 1987) use time delays between sensors location to estimate the source po-
sition. However, though they are computationally less expensive than maximum-likelihood
parametric algorithms (Chen et al., 2001a), they cannot handle efciently multiple sources
(Chen et al., 2001b). Maximum-likelihood (ML) algorithms are inspired by the fact that source
location information is contained in the linear phase shift of the sensor data spectrum ob-
tained through a discrete Fourier Transform applied to the wideband data. However, ML
techniques are dominated by low-cost suboptimal techniques like the well-known MUSIC al-
gorithm (Schmidt, 1986) which leverages spectral calculus on signal and noise subspaces to
nd sources locations.
Unlike these approaches, we do not use the acoustic properties of the song of the bird to nd
its location. Indeed, we assume that our sensors are simple and only detect songs relevant to
the monitored specie. Further, our sensors are wireless and rely on battery power to function.
It is important to notice that our algorithms do not try to pinpoint birds, but rather estimate
the number of songbirds that inhabit a region. In our case, only approximate geometric infor-
mation is sufcient to establish this estimate.
3. Recognizing the birdsong
The recognition process of birdsong is the rst part of our counting systems. Today, it is true
that the performance levels made in the treatment of audio signals are high, but this requires
large memory and processing power of large size which could exclude limited capacity of
devices such as wireless sensors.
Recognition of species based on acoustic analysis has been widely studied in recent years
and usually falls within the scope of the classication eld. This is particularly the case for
recognition of bird songs. Indeed, for a particular song, it is necessary to determine if it belongs
to a specie. For example, the work of Seppo Fagerlund (Fagerlund, 2007) uses support vector
machines to classify the different species of birds based on their songs. Similarly, Jim Cai
et al. (Cai et al., 2007) propose a method recognition based on neural networks to nd the
membership of a song to a bird class. Our recognition process, inspired by the work of Rabiner
(Rabiner & Wilpon, 1979), leverages the same mechanics by means of a clustering algorithm
to classify the song.
Figure 1 gives an overview of our wireless counting system.
Fig. 1. The Counting System
Bird Species Recognition Using Clustering
Our classication method is twofold : a parameterization transformation process of the song
in a certain ngerprint, and clustering process to determine its membership.
The parametrization process uses the songs of the birds to create a series of coefcients that
describe the signal. Although various parameterization methods LPC, LPCC, PLP, dots
exist, we use the MFCC Mel Frequency Cepstral Coefcient because our analysis is limited to
a very limited vocabulary on limited devices. Indeed, Christopher Levy compared in (Lvy
et al., 2006) different parameterization methods on small systems such as mobile phones for
reduced vocabulary and have showed that the parameterization based on MFCC is much
more effective for such systems.
2. Previous Work
Source localization is an area of interest that has been widely studied in these recent years.
A comprehensive review of incentives techniques and source localization has been written by
Krimand Viberg in (Krim&Viberg, 1996) and it is not difcult to understand that problemhas
been of particular focus for military needs. Indeed, radar and sonars are a direct application
of source localization.
Several acoustic parameters such as bandwidth, distance sensors, reverberation and thus
change the way the location of the sources are handled. In addition, the algorithms of source
localization depends strongly on physics and rely on the sound characteristics of waveform to
calculate location sources. Waveform audio is known to be broadband (30Hz-15kHz) and sen-
sors usually record the sound from near-eld sources. The following presents some algorithms
of interest which satisfy these two properties. Near-elds algorithms like close-formed ones
(Smith. & Abel, 1987) use time delays between sensors location to estimate the source po-
sition. However, though they are computationally less expensive than maximum-likelihood
parametric algorithms (Chen et al., 2001a), they cannot handle efciently multiple sources
(Chen et al., 2001b). Maximum-likelihood (ML) algorithms are inspired by the fact that source
location information is contained in the linear phase shift of the sensor data spectrum ob-
tained through a discrete Fourier Transform applied to the wideband data. However, ML
techniques are dominated by low-cost suboptimal techniques like the well-known MUSIC al-
gorithm (Schmidt, 1986) which leverages spectral calculus on signal and noise subspaces to
nd sources locations.
Unlike these approaches, we do not use the acoustic properties of the song of the bird to nd
its location. Indeed, we assume that our sensors are simple and only detect songs relevant to
the monitored specie. Further, our sensors are wireless and rely on battery power to function.
It is important to notice that our algorithms do not try to pinpoint birds, but rather estimate
the number of songbirds that inhabit a region. In our case, only approximate geometric infor-
mation is sufcient to establish this estimate.
3. Recognizing the birdsong
The recognition process of birdsong is the rst part of our counting systems. Today, it is true
that the performance levels made in the treatment of audio signals are high, but this requires
large memory and processing power of large size which could exclude limited capacity of
devices such as wireless sensors.
Recognition of species based on acoustic analysis has been widely studied in recent years
and usually falls within the scope of the classication eld. This is particularly the case for
recognition of bird songs. Indeed, for a particular song, it is necessary to determine if it belongs
to a specie. For example, the work of Seppo Fagerlund (Fagerlund, 2007) uses support vector
machines to classify the different species of birds based on their songs. Similarly, Jim Cai
et al. (Cai et al., 2007) propose a method recognition based on neural networks to nd the
membership of a song to a bird class. Our recognition process, inspired by the work of Rabiner
(Rabiner & Wilpon, 1979), leverages the same mechanics by means of a clustering algorithm
to classify the song.
Figure 1 gives an overview of our wireless counting system.
Fig. 1. The Counting System
Bird Species Recognition Using Clustering
Our classication method is twofold : a parameterization transformation process of the song
in a certain ngerprint, and clustering process to determine its membership.
The parametrization process uses the songs of the birds to create a series of coefcients that
describe the signal. Although various parameterization methods LPC, LPCC, PLP, dots
exist, we use the MFCC Mel Frequency Cepstral Coefcient because our analysis is limited to
a very limited vocabulary on limited devices. Indeed, Christopher Levy compared in (Lvy
et al., 2006) different parameterization methods on small systems such as mobile phones for
reduced vocabulary and have showed that the parameterization based on MFCC is much
more effective for such systems.
Once the ngerprint is obtained from the parameterization process, it is added in a set with
other ngerprints, themselves derived from a database containing a large number of songs of
individuals known as the specie. Subsequently, a clustering algorithm (K-Means or EM) is
used on all the ngerprints to determine their similarity and to create one or more clusters in
which will be the bird cluster. For a given footprint, the problem is then to determine its mem-
bership to the bird cluster. If thats the case, data location + Mote timestamp is stored in the
database for further processing counting algorithms. Our recognition results are compelling
because almost all birds are classied correctly in our case.
4. The Counting Algorithm
This section is devoted to our counting heuristics inspired by the triangulation detection used
by R. E. Bell to count owls in the forest (Bell, 1964). Our method differs essentially fromthe fact
that we do not use semi-directional devices but omni-directional wireless sensors to loosely
locate a birdsong. In our theoretical framework, all motes share the same characteristics build-
ing, which means they have the same (processing power, memory, battery, radius of detection,
etc). Optimizing routes in wireless sensors networks here are out of concern. We only focus
on the manner to detect birds in their habitat viewed as a 2D area. Further, we do not have
any assumptions on the number of birds, on their movements or even their customs.
More formally, let denote M = {m
1
, . . . , m
n
} the set of all the motes which covers the habitat.
Each mote has the same detection radius r. All motes can report information to the base
station B which holds our counting algorithm, assuming that B is always reachable by every
mote. Let F
t
: M {0, 1}, the detection function which returns 1 if a mote m
i
detects a bird,
0 otherwise at time t. The base station stores the detection array D
t
= [F
t
(m
1
), . . . , F
t
(m
n
)]
which reveals the detection state of each mote at time t. Note that the base stores detection
arrays at a sampling rate determined empirically, that is, detection arrays D
i
are stored in a
data set D at the base B. Fig. 2 shows an example of motes placed on a 2D area.
Bird
Mote
Detection radius
Fig. 2. Motes, birds and the hard underlying unity Chart
We propose to count one bird for all the motes which trigger at time t and for which radius
of detection intersect mutually. We call such a set a Maximal Detection Set denoted MDS(N)
with N M where N is the set of the motes which trigger at time t. The grayed area in
gure 2 is a MDS. Lets denote such a subset W = {m M | m
i
, m
j
M, r(m
i
) r(m
j
)}.
Finding the Maximum Detection Set is similar to nd a maximum clique (Bomze et al., 1999).
Lets see why.
A unit disk graph G(V, E) is an intersection graph of disks of unit radius, that is, ij E, the
unit circle of center i intersects the unit circle of center j. The set of each center of these circles
is called the model of the unit disk graph. This class of graph is well studied and is extensively
used in the eld of ad hoc networks (Kuhn et al., 2008). Indeed, UDGs (Unit Disk Graphs) can
represent an ideal viewof an ad hoc networks and provides strong theoretical result due to the
geometric properties of the model. For example, Clark and al. (Clark et al., 1990) show that
nding a maximal clique for an UDG is polynomial given its model. More recently, Raghavan
and Spinrad (Raghavan & Spinrad, 2003) have shown that it is even possible to compute the
maximum clique without the model in polynomial time.
Without loss of generality, let G(V, E) a graph where V is the set of the motes and E, the set
of edges where the edge ij exists if and only if the detection radius of mote i intersects the
detection radius of mote j. Clearly, G is a unit disk graph. Unfortunately, a clique in G only
gives motes which are pairwise adjacent and we are interested in motes which are mutually
adjacent, that is motes which intersect mutually. We propose to alter all triangles (clique of size
3) which do not have a mutual intersection in the graph i.e we remove one edge in the triangle.
As a consequence, all cliques of more than three vertices will have a mutual intersection.
Theorem 4.1. If a graph G(V, E) only has triangles formed from motes whose detection radius inter-
sect mutually, then all motes forming a clique in G have detection radii intersecting mutually.
Proof. By denition, all clique of size three have detection radii which intersect mutually.
Now, assume that all motes clique of size n intersect mutually. Let choose such a clique that
we call S = {m
1
, . . . , m
n
} and lets add a newmote m
n+1
to S. Assume that S +{m
n+1
} forma
clique for which some motes do not intersect mutually. Clearly, m
n+1
form at least two proper
intersections with S, and the detection radius of the mote m
n+1
cannot intersect mutually at
least with two other radii detection. But, by denition, all triangles intersect mutually which
is a contradiction.
Reichling (Reichling, 1988) uses convex programming to nd the common intersection of a set
of disks in O(k) steps where k is the number of constraints of the convex program. Moreover,
all the triangles in a graph can be computed in O(mn) steps where m is the number of edges
and n, the number of vertices. Thus, we can alter all triangles which do not have a common
intersection in O(kmn) steps. Several strategies could be used to alter a triangle. However,
removing the longest edge in a triangle seems to be the most relevant one since the number
of altered triangles would be reduced. Intuitively, a longest edge in a bad triangle is more
likely to be common to another bad triangle. Unfortunately, the underlying unit disk graph
can loose its nature since it might become a quasi-unit disk graph
1
for which the maximum
clique problem is known to be NP-complete (Ceroi, 2002).
Algorithm 1 recursively constructs the maximum set of all motes which triggers at time t and
removes a MDS built from this set. For each MDS removed, the number of birds iterates.
This procedure is run for each detection array and the maximum number found over these
detection arrays is an estimation of the number of singing birds. This algorithm complexity is
bounded by the MDS search which consists in nding a clique in the unit disk graph underly-
ing our network. Breu (Breu, 1996) has given an algorithm which nd a maximum clique in a
unit disk graph with complexity O(n
3.5
log n). However, the alteration of the underlying unit
disk graph leads to a NP-complete algorithm.
1
Model which takes into account non-circular detection area
Once the ngerprint is obtained from the parameterization process, it is added in a set with
other ngerprints, themselves derived from a database containing a large number of songs of
individuals known as the specie. Subsequently, a clustering algorithm (K-Means or EM) is
used on all the ngerprints to determine their similarity and to create one or more clusters in
which will be the bird cluster. For a given footprint, the problem is then to determine its mem-
bership to the bird cluster. If thats the case, data location + Mote timestamp is stored in the
database for further processing counting algorithms. Our recognition results are compelling
because almost all birds are classied correctly in our case.
4. The Counting Algorithm
This section is devoted to our counting heuristics inspired by the triangulation detection used
by R. E. Bell to count owls in the forest (Bell, 1964). Our method differs essentially fromthe fact
that we do not use semi-directional devices but omni-directional wireless sensors to loosely
locate a birdsong. In our theoretical framework, all motes share the same characteristics build-
ing, which means they have the same (processing power, memory, battery, radius of detection,
etc). Optimizing routes in wireless sensors networks here are out of concern. We only focus
on the manner to detect birds in their habitat viewed as a 2D area. Further, we do not have
any assumptions on the number of birds, on their movements or even their customs.
More formally, let denote M = {m
1
, . . . , m
n
} the set of all the motes which covers the habitat.
Each mote has the same detection radius r. All motes can report information to the base
station B which holds our counting algorithm, assuming that B is always reachable by every
mote. Let F
t
: M {0, 1}, the detection function which returns 1 if a mote m
i
detects a bird,
0 otherwise at time t. The base station stores the detection array D
t
= [F
t
(m
1
), . . . , F
t
(m
n
)]
which reveals the detection state of each mote at time t. Note that the base stores detection
arrays at a sampling rate determined empirically, that is, detection arrays D
i
are stored in a
data set D at the base B. Fig. 2 shows an example of motes placed on a 2D area.
Bird
Mote
Detection radius
Fig. 2. Motes, birds and the hard underlying unity Chart
We propose to count one bird for all the motes which trigger at time t and for which radius
of detection intersect mutually. We call such a set a Maximal Detection Set denoted MDS(N)
with N M where N is the set of the motes which trigger at time t. The grayed area in
gure 2 is a MDS. Lets denote such a subset W = {m M | m
i
, m
j
M, r(m
i
) r(m
j
)}.
Finding the Maximum Detection Set is similar to nd a maximum clique (Bomze et al., 1999).
Lets see why.
A unit disk graph G(V, E) is an intersection graph of disks of unit radius, that is, ij E, the
unit circle of center i intersects the unit circle of center j. The set of each center of these circles
is called the model of the unit disk graph. This class of graph is well studied and is extensively
used in the eld of ad hoc networks (Kuhn et al., 2008). Indeed, UDGs (Unit Disk Graphs) can
represent an ideal viewof an ad hoc networks and provides strong theoretical result due to the
geometric properties of the model. For example, Clark and al. (Clark et al., 1990) show that
nding a maximal clique for an UDG is polynomial given its model. More recently, Raghavan
and Spinrad (Raghavan & Spinrad, 2003) have shown that it is even possible to compute the
maximum clique without the model in polynomial time.
Without loss of generality, let G(V, E) a graph where V is the set of the motes and E, the set
of edges where the edge ij exists if and only if the detection radius of mote i intersects the
detection radius of mote j. Clearly, G is a unit disk graph. Unfortunately, a clique in G only
gives motes which are pairwise adjacent and we are interested in motes which are mutually
adjacent, that is motes which intersect mutually. We propose to alter all triangles (clique of size
3) which do not have a mutual intersection in the graph i.e we remove one edge in the triangle.
As a consequence, all cliques of more than three vertices will have a mutual intersection.
Theorem 4.1. If a graph G(V, E) only has triangles formed from motes whose detection radius inter-
sect mutually, then all motes forming a clique in G have detection radii intersecting mutually.
Proof. By denition, all clique of size three have detection radii which intersect mutually.
Now, assume that all motes clique of size n intersect mutually. Let choose such a clique that
we call S = {m
1
, . . . , m
n
} and lets add a newmote m
n+1
to S. Assume that S +{m
n+1
} forma
clique for which some motes do not intersect mutually. Clearly, m
n+1
form at least two proper
intersections with S, and the detection radius of the mote m
n+1
cannot intersect mutually at
least with two other radii detection. But, by denition, all triangles intersect mutually which
is a contradiction.
Reichling (Reichling, 1988) uses convex programming to nd the common intersection of a set
of disks in O(k) steps where k is the number of constraints of the convex program. Moreover,
all the triangles in a graph can be computed in O(mn) steps where m is the number of edges
and n, the number of vertices. Thus, we can alter all triangles which do not have a common
intersection in O(kmn) steps. Several strategies could be used to alter a triangle. However,
removing the longest edge in a triangle seems to be the most relevant one since the number
of altered triangles would be reduced. Intuitively, a longest edge in a bad triangle is more
likely to be common to another bad triangle. Unfortunately, the underlying unit disk graph
can loose its nature since it might become a quasi-unit disk graph
1
for which the maximum
clique problem is known to be NP-complete (Ceroi, 2002).
Algorithm 1 recursively constructs the maximum set of all motes which triggers at time t and
removes a MDS built from this set. For each MDS removed, the number of birds iterates.
This procedure is run for each detection array and the maximum number found over these
detection arrays is an estimation of the number of singing birds. This algorithm complexity is
bounded by the MDS search which consists in nding a clique in the unit disk graph underly-
ing our network. Breu (Breu, 1996) has given an algorithm which nd a maximum clique in a
unit disk graph with complexity O(n
3.5
log n). However, the alteration of the underlying unit
disk graph leads to a NP-complete algorithm.
1
Model which takes into account non-circular detection area
begin
L ;
foreach d D do
NumberOfBirds 0;
Construct the underlying altered unit disk graph G(V, E) from d;
while V = do
Search for a maximum clique in G;
Remove this clique from G;
NumberOfBirds NumberOfBirds +1;
Add NumberOfBirds to L;
return max
lL
l;
end
Algorithm 1: The Counting Heuristic
Rening the Counting Heuristic
In the following, we suggest a little enhancement of our scheme. Indeed, we partition suc-
cessive detection arrays pairwise in order to rene our estimation of the number of birds.
Intuitively, the habitat is divided in such a manner that birds in a part could not have moved
to another one between two instants (for each couple of detection arrays). A threshold is em-
pirically xed for the ight speed of the birds such that no birds can y over that value. This
leads to the decomposition of the environment in several sub-environments. Then, each sub-
environment is processed with algorithm 1. For example, assume that we have 10 birds in an
area. Halve this area and put 5 birds in one part, and 5 in the counterpart. Now, assume that
the 5 birds in the rst part sing together at time t, the other ones sing together at time t + 1
and these parts are too distant such that birds in one part can go in the other part between
the two time steps. In that case, algorithm 1 outputs 5 birds as estimate. Our next algorithm
halves the environment in two parts such that birds in two. As a consequence, we can apply
algorithm 1 on each part independently and take the sum of the estimates found on each part,
which gives 10 birds.
Data: A list of detection arrays D = D
1
, . . . , D
m
Result: An estimation of the number of birds in the habitat
begin
L ;
while |D| > 0 do
Partition detection arrays D
i
and D
i+1
respectively in X = {X
1
, . . . , X
k
} and
Y = {Y
1
, . . . , Y
k
};
Z = ;
for i 1 to k do
Process X
i
and Y
i
with algorithm 1 and put the maximum of the number of birds
counted in Z;
Add
zZ
z to L;
return max
lL
l;
end
Algorithm 2: The Enhanced Counting Algorithm
For sake of clarity, in algorithm 2, the number of detection arrays is even and only two suc-
cessive detection arrays are partitioned. The next section presents another way to count the
singing birds in their habitat. This next version is designed to be partially distributed on the
motes.
5. The Swarm Counting Protocol
Our next counting method can be seen as two levels, a local and a global one. At the local
level, motes cooperates sending information to count the number of singing birds in their
neighborhood. At the global level, motes aggregates data to nd a more accurate estimation
of the number of singing birds in the habitat.
Like the technique previously described, we assume that the motes layout forms a unit disk
graph. First, motes have to estimate locally how many birds had sang. Then, they send this
data to the base station which derives from all the information the estimate for the number
of singing birds. In our scheme, motes all have a set of rules which are the following. They
are all in a passive state until some songs trigger them. When triggered, they switch to an
active state and tell to their neighbors
2
that they detect a bird. Then they listen for their
neighborhood during a specied time. Finally, they deduce the number of singing birds in the
vicinity from their active answering neighbors, and send this number to the base station.
Local Counting
Our local counting is somewhat similar to the one in section 4. It leverages the trilatera-
tion technique to estimate a number of birds in the vicinity. All motes know their neighbors
topology and are in an initial passive state when they are waiting for signals (bird songs).
Whenever a mote is triggered , it sends a signal to its neighbors and listen for whose which
were triggered too. If two or more neighbors have an intersecting detection area, we assume
that only one bird is counted for these motes. In gure 3, the black mote hears a bird song,
asks its neighbors if they heard too and waits for their reply. Remark that the number of birds
counted is the number of neighbors which are independent mutually in each neighborhood,
i.e the cardinal of the maximum independent set
3
in the graph induced by the neighbors.
Global Counting
Now, assume that all motes have counted the birds in their vicinity and have sent their local
count to the base station. Now, all these information have to be aggregated accordingly to
nd an estimate of the number of singing birds at this instant. Because, the neighborhood was
used to derive the local counting, obviously, motes which are neighbors will inuence each
other in the counting process. So, summing up their local count can lead to an over-estimate
of the number of singing birds. Note that is also the case for motes which are at distance 2,
that is neighbors of neighbors in the unit disk graph, since they can share common neighbors.
Therefore, only motes which are more than distant 2 each other will sum up their count. Our
estimation will be the maximum number of birds which could be counted over aggregated
nodes in the underlying graph.
In gure 4, the black nodes are at distant 3. So a global counting of singing birds could be
four. Remark that such a counting has to be done for all set of nodes which are at more than
distance 2 each other. If such a technique seems to lead to a combinatorial explosion of the
2
As previously, neighbors are adjacent nodes in the unit disk graph
3
The largest set of vertices which are not pairwise adjacent
begin
L ;
foreach d D do
NumberOfBirds 0;
Construct the underlying altered unit disk graph G(V, E) from d;
while V = do
Search for a maximum clique in G;
Remove this clique from G;
NumberOfBirds NumberOfBirds +1;
Add NumberOfBirds to L;
return max
lL
l;
end
Algorithm 1: The Counting Heuristic
Rening the Counting Heuristic
In the following, we suggest a little enhancement of our scheme. Indeed, we partition suc-
cessive detection arrays pairwise in order to rene our estimation of the number of birds.
Intuitively, the habitat is divided in such a manner that birds in a part could not have moved
to another one between two instants (for each couple of detection arrays). A threshold is em-
pirically xed for the ight speed of the birds such that no birds can y over that value. This
leads to the decomposition of the environment in several sub-environments. Then, each sub-
environment is processed with algorithm 1. For example, assume that we have 10 birds in an
area. Halve this area and put 5 birds in one part, and 5 in the counterpart. Now, assume that
the 5 birds in the rst part sing together at time t, the other ones sing together at time t + 1
and these parts are too distant such that birds in one part can go in the other part between
the two time steps. In that case, algorithm 1 outputs 5 birds as estimate. Our next algorithm
halves the environment in two parts such that birds in two. As a consequence, we can apply
algorithm 1 on each part independently and take the sum of the estimates found on each part,
which gives 10 birds.
Data: A list of detection arrays D = D
1
, . . . , D
m
Result: An estimation of the number of birds in the habitat
begin
L ;
while |D| > 0 do
Partition detection arrays D
i
and D
i+1
respectively in X = {X
1
, . . . , X
k
} and
Y = {Y
1
, . . . , Y
k
};
Z = ;
for i 1 to k do
Process X
i
and Y
i
with algorithm 1 and put the maximum of the number of birds
counted in Z;
Add
zZ
z to L;
return max
lL
l;
end
Algorithm 2: The Enhanced Counting Algorithm
For sake of clarity, in algorithm 2, the number of detection arrays is even and only two suc-
cessive detection arrays are partitioned. The next section presents another way to count the
singing birds in their habitat. This next version is designed to be partially distributed on the
motes.
5. The Swarm Counting Protocol
Our next counting method can be seen as two levels, a local and a global one. At the local
level, motes cooperates sending information to count the number of singing birds in their
neighborhood. At the global level, motes aggregates data to nd a more accurate estimation
of the number of singing birds in the habitat.
Like the technique previously described, we assume that the motes layout forms a unit disk
graph. First, motes have to estimate locally how many birds had sang. Then, they send this
data to the base station which derives from all the information the estimate for the number
of singing birds. In our scheme, motes all have a set of rules which are the following. They
are all in a passive state until some songs trigger them. When triggered, they switch to an
active state and tell to their neighbors
2
that they detect a bird. Then they listen for their
neighborhood during a specied time. Finally, they deduce the number of singing birds in the
vicinity from their active answering neighbors, and send this number to the base station.
Local Counting
Our local counting is somewhat similar to the one in section 4. It leverages the trilatera-
tion technique to estimate a number of birds in the vicinity. All motes know their neighbors
topology and are in an initial passive state when they are waiting for signals (bird songs).
Whenever a mote is triggered , it sends a signal to its neighbors and listen for whose which
were triggered too. If two or more neighbors have an intersecting detection area, we assume
that only one bird is counted for these motes. In gure 3, the black mote hears a bird song,
asks its neighbors if they heard too and waits for their reply. Remark that the number of birds
counted is the number of neighbors which are independent mutually in each neighborhood,
i.e the cardinal of the maximum independent set
3
in the graph induced by the neighbors.
Global Counting
Now, assume that all motes have counted the birds in their vicinity and have sent their local
count to the base station. Now, all these information have to be aggregated accordingly to
nd an estimate of the number of singing birds at this instant. Because, the neighborhood was
used to derive the local counting, obviously, motes which are neighbors will inuence each
other in the counting process. So, summing up their local count can lead to an over-estimate
of the number of singing birds. Note that is also the case for motes which are at distance 2,
that is neighbors of neighbors in the unit disk graph, since they can share common neighbors.
Therefore, only motes which are more than distant 2 each other will sum up their count. Our
estimation will be the maximum number of birds which could be counted over aggregated
nodes in the underlying graph.
In gure 4, the black nodes are at distant 3. So a global counting of singing birds could be
four. Remark that such a counting has to be done for all set of nodes which are at more than
distance 2 each other. If such a technique seems to lead to a combinatorial explosion of the
2
As previously, neighbors are adjacent nodes in the unit disk graph
3
The largest set of vertices which are not pairwise adjacent
Fig. 3. Motes collaboration at the local level
set of motes which can be aggregated, the underlying graph has some nice properties which
allows to nd the estimate in linear time.
More formally, let N(i) dene the neighbors of a mote i, that is
i V, N(i) = {j V| ij E}
and
A V, N(A) =
iA
N(i)
Let G
2+
(V, E
2+
) dene the graph where
i, j V
2
, ij E
2+
iff j N(N(i))\i
The graph G
2+
is the graph of all motes which are at most 2-distant between them. Let S(.)
denote the mapping which maps a vertex v V to the number of birds counted locally. Let
C be the set of all independent set in graph G
2+
. Our estimation is the sum of birds counted
locally for each motes derived from the maximum weighted independent set of G
2+
, i.e
max
cC

vc
S(v)
Lemma 5.1. G
2+
is a chordal graph.
Proof. Proof Remark that in G
2+
, all vertices are simplicial
4
. Thus, there exists a perfect elimi-
nation ordering on its vertices and de facto G
2+
is chordal.
4
Vertices for which neighbors induce a clique in the graph.
Mote
Detection radius
Neighbor
Fig. 4. Example of underlying unit disk graph in local and global detection
Chordal graphs are graph for which vertices do not induce cycles without chord of size more
or equal to four. They are perfect graphs and well discussed in (Golumbic, 1980). It is also well
known that nding a maximum weighted independent set in chordal graph is linear (Leung,
1984). Thus, our later algorithm nds its estimation of the number of birds in linear time given
G
2+
.
Lets see why and how our algorithm is not so sensible to noise and encompasses non circular
detection area. One of the most interesting features of swarm computing (Blum & Merkle,
2008) is that nodes (swarm entities) create mechanisms which tend to be resilient to disrup-
tion and failure. Similarly, our last counting technique leverages the swarm intelligence since
motes collaborates each other to derive their local count. The more the motes are, better the
estimate is. There are two cases where inconsistencies could appear :
1. Motes can have a different status from what it would be. For example, a mote could
stay in a passive state while it would have heard a bird song. However, neighbor
motes tend to negate this last effect. Conversely, motes could wake up while no birds
have sung. This latter case is somewhat less frequent and is easier to correct since this
mote could be a one-vertex connected component in the underlying graph, fact which
is prone to be an erratic behavior of the mote.
2. Objects can occlude bird songs, that is detection area is no more circular. In that case,
the occluded motes would stay in a passive state. Fortunately, the swarm could correct
this drawback by multiplicity : other closer motes could hear the birds too.
Fig. 3. Motes collaboration at the local level
set of motes which can be aggregated, the underlying graph has some nice properties which
allows to nd the estimate in linear time.
More formally, let N(i) dene the neighbors of a mote i, that is
i V, N(i) = {j V| ij E}
and
A V, N(A) =
iA
N(i)
Let G
2+
(V, E
2+
) dene the graph where
i, j V
2
, ij E
2+
iff j N(N(i))\i
The graph G
2+
is the graph of all motes which are at most 2-distant between them. Let S(.)
denote the mapping which maps a vertex v V to the number of birds counted locally. Let
C be the set of all independent set in graph G
2+
. Our estimation is the sum of birds counted
locally for each motes derived from the maximum weighted independent set of G
2+
, i.e
max
cC

vc
S(v)
Lemma 5.1. G
2+
is a chordal graph.
Proof. Proof Remark that in G
2+
, all vertices are simplicial
4
. Thus, there exists a perfect elimi-
nation ordering on its vertices and de facto G
2+
is chordal.
4
Vertices for which neighbors induce a clique in the graph.
Mote
Detection radius
Neighbor
Fig. 4. Example of underlying unit disk graph in local and global detection
Chordal graphs are graph for which vertices do not induce cycles without chord of size more
or equal to four. They are perfect graphs and well discussed in (Golumbic, 1980). It is also well
known that nding a maximum weighted independent set in chordal graph is linear (Leung,
1984). Thus, our later algorithm nds its estimation of the number of birds in linear time given
G
2+
.
Lets see why and how our algorithm is not so sensible to noise and encompasses non circular
detection area. One of the most interesting features of swarm computing (Blum & Merkle,
2008) is that nodes (swarm entities) create mechanisms which tend to be resilient to disrup-
tion and failure. Similarly, our last counting technique leverages the swarm intelligence since
motes collaborates each other to derive their local count. The more the motes are, better the
estimate is. There are two cases where inconsistencies could appear :
1. Motes can have a different status from what it would be. For example, a mote could
stay in a passive state while it would have heard a bird song. However, neighbor
motes tend to negate this last effect. Conversely, motes could wake up while no birds
have sung. This latter case is somewhat less frequent and is easier to correct since this
mote could be a one-vertex connected component in the underlying graph, fact which
is prone to be an erratic behavior of the mote.
2. Objects can occlude bird songs, that is detection area is no more circular. In that case,
the occluded motes would stay in a passive state. Fortunately, the swarm could correct
this drawback by multiplicity : other closer motes could hear the birds too.
Therefore, note that the layout of the motes is somewhat important and a simple way
to tackle the occlusion problem is to rise the density of the motes on the monitored
environment. It is even possible to only increase the number of motes where occlusion
problems could occur.
The next section is dedicated to experiments which prove our algorithmefciency, even in the
presence of noise.
6. Experiments
6.1 Context
Endangered species receive attention from the scientic community since their disappearance
would lead to irreplaceable losses. To help these species to survive, their habitat is protected
by laws of environmental protection. Sometimes, this protection is not sufcient since their
habitat evolution is the main cause of their vanishing. In order to save them, they must be
transferred elsewhere. Obviously, the new habitat has to be similar to the previous one to
minimally disrupt the equilibrium of the wildlife. To model a habitat, several parameters
have to be xed by an expert. This study precedes the MOM project for which wireless sensor
networks have to be used to monitor an endangered specie. So these simulations are the
rst steps to the deployment of WSNs over the Caravelle location in Martinique (a French
Caribbean island). Indeed, birds called White-breasted Thrasher are a specie which is only
known to be in in the Caravelle. They are considered endangered since specialists think that
only fty of them are still alive there.
6.2 Testbed
Environment
For the need of the simulations, we wrote a tool which aims at generating the data necessary
to run our counting heuristics described previously. Our test environment comprises :
an Intel Core 2 Duo E6750 2.67 GHZ,
4 Go RAM,
Windows Vista 64 bits for Operating System,
and the JDK 1.6 Update 10 (x64) since our tool is written in java.
Parameters
Simulations parameters were calibrated to be the closest to our tested area. The dimension of
our habitat is about 1000m1000m. Birds can y at 2 meters per second, stay at place, take
random directions with uniform probability. They sing with some probability xed empiri-
cally. This latter parameter is xed at 0.2 for each sample record (detection array). Finally,
motes are placed randomly on our area.
6.3 Performance Evaluation
Figure 5 shows our three different algorithms estimation for counting fty living birds in the
habitat. Algo1 stands for the algorithm which only rely on the underlying UDG. Longest is the
algorithm which alters the longest edges of bad triangles. Swarm is the algorithm presented
in section 5. True is the number of birds which really sang. Each test has been driven 50 times
and the mean of the estimations was taken as the nal result. Error deviation is shown for
Motes Algo1 Longest Swarm True
100.00 10.10 10.72 9.98 17.58
200.00 13.86 16.56 12.62 17.38
300.00 15.16 18.64 12.92 18.02
400.00 15.70 20.46 13.84 17.58
500.00 16.76 22.56 15.04 17.36
600.00 17.28 24.12 15.80 17.80
700.00 17.88 25.74 16.08 17.66
800.00 18.38 27.74 17.04 17.64
900.00 18.40 29.50 17.60 17.86
1000.00 18.30 31.22 17.98 17.80
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
100 200 300 400 500 600 700 800 900 1000
E
r
r
o
r

r
a
t
e
Number of motes
Algo1
Longest
Distri
Fig. 5. Number of Birds found by the heuristics and error deviation for 50 Birds in the habitat
each algorithm on the graph near the tables and gives an idea of how the algorithms perform
along the parameters.
Experiments showthat algorithmAlgo1 performs nearly as well as algorithmSwarmwhenever
the number of motes is high. However, Algo1 tends to over-count the birds. The outputted
number of birds depends on the manner the MDS are removed. Lets sketch a brief example on
gure 6. There exists three MDS at rst step in each conguration. Grayed areas represent the
MDS removed on each conguration at each step. Conguration 1 leads to two grayed area
whereas conguration 2 leads to three grayed area. That is how two birds can be counted as
three. To reduce this drawback, we could run several times the counting process which would
remove the MDS randomly and then take the minimumnumber of birds over these countings.
However, such a scheme does not guarantee that we will not over-count and further, will
highly rise the execution time of the whole process.
To validate our schemes, noise is added as a parameter in our simulator. We assume that
20% of the motes malfunction. Table 1 shows the percentage error for Algo1 and Swarm in the
presence of noise.
Therefore, note that the layout of the motes is somewhat important and a simple way
to tackle the occlusion problem is to rise the density of the motes on the monitored
environment. It is even possible to only increase the number of motes where occlusion
problems could occur.
The next section is dedicated to experiments which prove our algorithmefciency, even in the
presence of noise.
6. Experiments
6.1 Context
Endangered species receive attention from the scientic community since their disappearance
would lead to irreplaceable losses. To help these species to survive, their habitat is protected
by laws of environmental protection. Sometimes, this protection is not sufcient since their
habitat evolution is the main cause of their vanishing. In order to save them, they must be
transferred elsewhere. Obviously, the new habitat has to be similar to the previous one to
minimally disrupt the equilibrium of the wildlife. To model a habitat, several parameters
have to be xed by an expert. This study precedes the MOM project for which wireless sensor
networks have to be used to monitor an endangered specie. So these simulations are the
rst steps to the deployment of WSNs over the Caravelle location in Martinique (a French
Caribbean island). Indeed, birds called White-breasted Thrasher are a specie which is only
known to be in in the Caravelle. They are considered endangered since specialists think that
only fty of them are still alive there.
6.2 Testbed
Environment
For the need of the simulations, we wrote a tool which aims at generating the data necessary
to run our counting heuristics described previously. Our test environment comprises :
an Intel Core 2 Duo E6750 2.67 GHZ,
4 Go RAM,
Windows Vista 64 bits for Operating System,
and the JDK 1.6 Update 10 (x64) since our tool is written in java.
Parameters
Simulations parameters were calibrated to be the closest to our tested area. The dimension of
our habitat is about 1000m1000m. Birds can y at 2 meters per second, stay at place, take
random directions with uniform probability. They sing with some probability xed empiri-
cally. This latter parameter is xed at 0.2 for each sample record (detection array). Finally,
motes are placed randomly on our area.
6.3 Performance Evaluation
Figure 5 shows our three different algorithms estimation for counting fty living birds in the
habitat. Algo1 stands for the algorithm which only rely on the underlying UDG. Longest is the
algorithm which alters the longest edges of bad triangles. Swarm is the algorithm presented
in section 5. True is the number of birds which really sang. Each test has been driven 50 times
and the mean of the estimations was taken as the nal result. Error deviation is shown for
Motes Algo1 Longest Swarm True
100.00 10.10 10.72 9.98 17.58
200.00 13.86 16.56 12.62 17.38
300.00 15.16 18.64 12.92 18.02
400.00 15.70 20.46 13.84 17.58
500.00 16.76 22.56 15.04 17.36
600.00 17.28 24.12 15.80 17.80
700.00 17.88 25.74 16.08 17.66
800.00 18.38 27.74 17.04 17.64
900.00 18.40 29.50 17.60 17.86
1000.00 18.30 31.22 17.98 17.80
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
100 200 300 400 500 600 700 800 900 1000
E
r
r
o
r

r
a
t
e
Number of motes
Algo1
Longest
Distri
Fig. 5. Number of Birds found by the heuristics and error deviation for 50 Birds in the habitat
each algorithm on the graph near the tables and gives an idea of how the algorithms perform
along the parameters.
Experiments showthat algorithmAlgo1 performs nearly as well as algorithmSwarmwhenever
the number of motes is high. However, Algo1 tends to over-count the birds. The outputted
number of birds depends on the manner the MDS are removed. Lets sketch a brief example on
gure 6. There exists three MDS at rst step in each conguration. Grayed areas represent the
MDS removed on each conguration at each step. Conguration 1 leads to two grayed area
whereas conguration 2 leads to three grayed area. That is how two birds can be counted as
three. To reduce this drawback, we could run several times the counting process which would
remove the MDS randomly and then take the minimumnumber of birds over these countings.
However, such a scheme does not guarantee that we will not over-count and further, will
highly rise the execution time of the whole process.
To validate our schemes, noise is added as a parameter in our simulator. We assume that
20% of the motes malfunction. Table 1 shows the percentage error for Algo1 and Swarm in the
presence of noise.
Configuration 1
Configuration 2
Fig. 6. Example of bad counting
Table 1. Relative error for the counting algorithms for 50 birds
Motes Algo1 Swarm Algo1 with noise (%) Swarm with noise(%)
100 45,08 46,33 50,51 50,85
200 27,24 33,83 32,69 37,34
300 15,27 25,79 20,59 28,62
400 8,85 19,31 13,33 23,22
500 9,32 19,77 13,75 23,86
600 3,31 12,69 6,86 17,49
700 1,93 4,54 1,36 9,53
800 4,41 3,72 0,93 7,67
900 4,30 2,49 2,04 6,23
1000 7,11 1,49 4,47 1,03
Clearly, the algorithm based on the swarm counting protocol seems more sensitive to noise
than its counterpart. Note that without noise, algorithm Algo1 over-counts the number of
birds. Therefore, in presence of noise, the approximate of the number of birds tends to be more
precise. Conversely, algorithm Swarm already undercounts the number of birds originally. So,
noise degrades even more the approximate of the number of birds which generally leads to a
worse counting.
Finally, we decided to x the number of motes which will be used on the Caravelle habitat
to 1000 and vary the number of birds in our simulator to conrm our estimation. Results are
shown on gure 7.
Algorithm Longest suffers the same drawback seen in gure 5 when the number of motes is
high, so much that estimation are too high. This results from the fact that altering triangles
tends to create much more cliques to remove.
Algorithm Swarm gives slightly better estimations in conguration using a high number of
motes. However, its efciency lowers whenever less motes are used. Indeed, the lesser the
motes you have, the lesser you cover the habitat. Furthermore, our schemes rely on a high
number of motes to better estimate the singing birds except algorithm Longest which could be
used to estimate the songbirds whenever the number of motes are low.
Birds Algo1 Longest Swarm True
10.00 5.94 10.20 5.44 5.34
20.00 9.08 16.08 8.76 8.90
30.00 12.76 20.58 12.20 12.28
40.00 15.90 25.48 15.06 14.76
50.00 18.30 31.22 17.98 17.80
60.00 21.64 34.76 20.62 20.28
70.00 23.92 37.74 22.56 23.08
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
10 20 30 40 50 60 70
E
r
r
o
r

r
a
t
e
Number of motes
Algo1
Longest
Distri
Fig. 7. Number of Birds found by the heuristics and error deviation for 1000 motes in the
habitat
These results suggest to design an hybrid algorithm which will switch along determined
thresholds. However, remind that the data was generated from simulations and these thresh-
olds could be different from what our experimentations outputted. In our case, using the
Swarm algorithm for counting the birds seems to be so far the best solution to apply in the
Caravelle since more motes give accurate estimations.
7. Conclusion
Endangered species are a known problem that drew attention from the community these last
years. Habitat monitoring with wireless sensors networks could lead to several improve-
ments in the way to tackle the problem of the survival of these species. We have proposed a
rst technique to estimate the number of birds using wireless microphone motes scattered in
an habitat. Our method derives from the motes layout a unit disk graph and removes contin-
uously maximum cliques to count the number of birds. A limitation of this technique could be
the maximum clique problem but simulations have shown that estimations are still suitable
if unit disk graphs are used to represent the motes network. We have also proposed a linear
algorithm to estimate the singing birds in the habitat and have shown that it is as much as
Configuration 1
Configuration 2
Fig. 6. Example of bad counting
Table 1. Relative error for the counting algorithms for 50 birds
Motes Algo1 Swarm Algo1 with noise (%) Swarm with noise(%)
100 45,08 46,33 50,51 50,85
200 27,24 33,83 32,69 37,34
300 15,27 25,79 20,59 28,62
400 8,85 19,31 13,33 23,22
500 9,32 19,77 13,75 23,86
600 3,31 12,69 6,86 17,49
700 1,93 4,54 1,36 9,53
800 4,41 3,72 0,93 7,67
900 4,30 2,49 2,04 6,23
1000 7,11 1,49 4,47 1,03
Clearly, the algorithm based on the swarm counting protocol seems more sensitive to noise
than its counterpart. Note that without noise, algorithm Algo1 over-counts the number of
birds. Therefore, in presence of noise, the approximate of the number of birds tends to be more
precise. Conversely, algorithm Swarm already undercounts the number of birds originally. So,
noise degrades even more the approximate of the number of birds which generally leads to a
worse counting.
Finally, we decided to x the number of motes which will be used on the Caravelle habitat
to 1000 and vary the number of birds in our simulator to conrm our estimation. Results are
shown on gure 7.
Algorithm Longest suffers the same drawback seen in gure 5 when the number of motes is
high, so much that estimation are too high. This results from the fact that altering triangles
tends to create much more cliques to remove.
Algorithm Swarm gives slightly better estimations in conguration using a high number of
motes. However, its efciency lowers whenever less motes are used. Indeed, the lesser the
motes you have, the lesser you cover the habitat. Furthermore, our schemes rely on a high
number of motes to better estimate the singing birds except algorithm Longest which could be
used to estimate the songbirds whenever the number of motes are low.
Birds Algo1 Longest Swarm True
10.00 5.94 10.20 5.44 5.34
20.00 9.08 16.08 8.76 8.90
30.00 12.76 20.58 12.20 12.28
40.00 15.90 25.48 15.06 14.76
50.00 18.30 31.22 17.98 17.80
60.00 21.64 34.76 20.62 20.28
70.00 23.92 37.74 22.56 23.08
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
10 20 30 40 50 60 70
E
r
r
o
r

r
a
t
e
Number of motes
Algo1
Longest
Distri
Fig. 7. Number of Birds found by the heuristics and error deviation for 1000 motes in the
habitat
These results suggest to design an hybrid algorithm which will switch along determined
thresholds. However, remind that the data was generated from simulations and these thresh-
olds could be different from what our experimentations outputted. In our case, using the
Swarm algorithm for counting the birds seems to be so far the best solution to apply in the
Caravelle since more motes give accurate estimations.
7. Conclusion
Endangered species are a known problem that drew attention from the community these last
years. Habitat monitoring with wireless sensors networks could lead to several improve-
ments in the way to tackle the problem of the survival of these species. We have proposed a
rst technique to estimate the number of birds using wireless microphone motes scattered in
an habitat. Our method derives from the motes layout a unit disk graph and removes contin-
uously maximum cliques to count the number of birds. A limitation of this technique could be
the maximum clique problem but simulations have shown that estimations are still suitable
if unit disk graphs are used to represent the motes network. We have also proposed a linear
algorithm to estimate the singing birds in the habitat and have shown that it is as much as
efcient (quality) as our rst one. This scheme can be fully distributed on a suitable wireless
sensors network. Such a distributed scheme would deny the need of a powerful base station
since the counting process would totally shift from the base to the motes.
Counting singing birds is a rst step in our habitat monitoring project and surely is not suf-
cient to identify specicities of the monitored specie. One major goal of habitat monitoring
is the reintroduction of the specie in another environment which will share the same charac-
teristics. We intend to work in this way by monitoring several parameters of interests in an
environment to model it and compare it with another ones.
8. References
Bell, R. E. (1964). A Sound Triangulation Method for Counting Barred Owls, The Wilson Bul-
letin, The Wilson Ornithological Society.
Blum, C. & Merkle, D. (eds) (2008). Swarm Intelligence: Introduction and Applications, Natural
Computing Series, Springer.
URL: http://dx.doi.org/10.1007/978-3-540-74089-6
Bomze, I. M., Budinich, M., Pardalos, P. M. & Pelillo, M. (1999). The Maximum Clique Prob-
lem, Research Report CS-99-1, Dipartimento di Informatica, Univerit Ca Foscari di
Venezia.
Breu, H. (1996). Algorithmic aspects of constrained unit disk graphs, Technical Report TR-96-15,
Department of Computer Science, University of British Columbia. Tue, 22 Jul 1997
22:20:10 GMT.
URL: ftp://ftp.cs.ubc.ca/pub/local/techreports/1996/TR-96-15.ps.gz
Cai, J., Ee, D., Pham, B., Roe, P. & Zhang, J. (2007). Sensor network for the monitoring of
ecosystem: Bird species recognition, Intelligent Sensors, Sensor Networks and Informa-
tion, Queensland Univ. of Technol., Brisbane, pp. 293298.
Ceroi, S. (2002). The clique number of unit quasi-disk graphs, Rapport.
URL: http://hal.inria.fr/inria-00072169/en/; http://hal.ccsd.cnrs.fr/docs/00/07/21/69/PDF/RR-
4419.pdf
Chen, J. C., Hudson, R. E. & Yao, K. (2001a). A maximum-likelihood parametric approach
to source localizations, Proceedings of the Acoustics, Speech, and Signal Processing, IEEE
Computer Society, Washington, DC, USA, pp. 30133016.
Chen, J. C., Hudson, R. E. & Yao, K. (2001b). Joint maximum-likelihood source localization
and unknown sensor location estimation for near-eld wideband signals, in F. T. Luk
(ed.), Advanced Signal Processing Algorithms, Architectures, and Implementations XI, Vol.
4474, SPIE, pp. 521532.
URL: http://link.aip.org/link/?PSI/4474/521/1
Clark, B. N., Colbourn, C. J. & Johnson, D. S. (1990). Unit disk graphs, Discrete Math. 86(1-
3): 165177.
Fagerlund, S. (2007). Bird species recognition using support vector machines, EURASIP J.
Appl. Signal Process. 2007(1): 6464.
Golumbic, M. C. (1980). Algorithmic Graph Theory and Perfect Graphs, Academic Press.
Krim, H. & Viberg, M. (1996). Two decades of array signal processing research the para-
metric approach, IEEE Signal Processing Magazine, Vol. 13(3).
Kuhn, F., Wattenhofer, R. & Zollinger, A. (2008). Ad hoc networks beyond unit disk graphs,
Wireless Networks 14(5): 715729.
URL: http://dx.doi.org/10.1007/s11276-007-0045-6
Leung, J. Y. T. (1984). Fast algorithms for generating all maximal independent sets of interval,
circular-arc and chordal graphs, ALGORITHMS: Journal of Algorithms 5.
Lvy, C., Linars, G. & Bonastre, J.-F. (2006). Gmm-based acoustic modeling for embedded
speech recognition, In International Conference on Speech Communication and Technol-
ogy.
Rabiner, L. R. & Wilpon, J. G. (1979). Considerations in applying clustering techniques to
speaker-independent word recognition, The Journal of the Acoustical Society of America
66(3): 663673.
URL: http://link.aip.org/link/?JAS/66/663/1
Raghavan, V. & Spinrad, J. (2003). Robust algorithms for restricted domains, ALGORITHMS:
Journal of Algorithms 48.
Reichling, M. (1988). On the detection of a common intersection of k convex objects in the
plane, Information Processing Letters 29(1): 2529.
Schmidt, R. (1986). Multiple emitter location and signal parameter estimation, IEEE Transac-
tions on Antennas and Propagation 34(3): 276280.
Smith., J. & Abel, J. (1987). Closed-form least-squares source location estimation from range-
difference measurements, IEEE Transactions on Acoustics, Speech and Signal Processing
35(12): 16611669.
efcient (quality) as our rst one. This scheme can be fully distributed on a suitable wireless
sensors network. Such a distributed scheme would deny the need of a powerful base station
since the counting process would totally shift from the base to the motes.
Counting singing birds is a rst step in our habitat monitoring project and surely is not suf-
cient to identify specicities of the monitored specie. One major goal of habitat monitoring
is the reintroduction of the specie in another environment which will share the same charac-
teristics. We intend to work in this way by monitoring several parameters of interests in an
environment to model it and compare it with another ones.
8. References
Bell, R. E. (1964). A Sound Triangulation Method for Counting Barred Owls, The Wilson Bul-
letin, The Wilson Ornithological Society.
Blum, C. & Merkle, D. (eds) (2008). Swarm Intelligence: Introduction and Applications, Natural
Computing Series, Springer.
URL: http://dx.doi.org/10.1007/978-3-540-74089-6
Bomze, I. M., Budinich, M., Pardalos, P. M. & Pelillo, M. (1999). The Maximum Clique Prob-
lem, Research Report CS-99-1, Dipartimento di Informatica, Univerit Ca Foscari di
Venezia.
Breu, H. (1996). Algorithmic aspects of constrained unit disk graphs, Technical Report TR-96-15,
Department of Computer Science, University of British Columbia. Tue, 22 Jul 1997
22:20:10 GMT.
URL: ftp://ftp.cs.ubc.ca/pub/local/techreports/1996/TR-96-15.ps.gz
Cai, J., Ee, D., Pham, B., Roe, P. & Zhang, J. (2007). Sensor network for the monitoring of
ecosystem: Bird species recognition, Intelligent Sensors, Sensor Networks and Informa-
tion, Queensland Univ. of Technol., Brisbane, pp. 293298.
Ceroi, S. (2002). The clique number of unit quasi-disk graphs, Rapport.
URL: http://hal.inria.fr/inria-00072169/en/; http://hal.ccsd.cnrs.fr/docs/00/07/21/69/PDF/RR-
4419.pdf
Chen, J. C., Hudson, R. E. & Yao, K. (2001a). A maximum-likelihood parametric approach
to source localizations, Proceedings of the Acoustics, Speech, and Signal Processing, IEEE
Chen, J. C., Hudson, R. E. & Yao, K. (2001b). Joint maximum-likelihood source localization
and unknown sensor location estimation for near-eld wideband signals, in F. T. Luk
(ed.), Advanced Signal Processing Algorithms, Architectures, and Implementations XI, Vol.
4474, SPIE, pp. 521532.
URL: http://link.aip.org/link/?PSI/4474/521/1
Clark, B. N., Colbourn, C. J. & Johnson, D. S. (1990). Unit disk graphs, Discrete Math. 86(1-
3): 165177.
Fagerlund, S. (2007). Bird species recognition using support vector machines, EURASIP J.
Appl. Signal Process. 2007(1): 6464.
Golumbic, M. C. (1980). Algorithmic Graph Theory and Perfect Graphs, Academic Press.
Krim, H. & Viberg, M. (1996). Two decades of array signal processing research the para-
metric approach, IEEE Signal Processing Magazine, Vol. 13(3).
Kuhn, F., Wattenhofer, R. & Zollinger, A. (2008). Ad hoc networks beyond unit disk graphs,
Wireless Networks 14(5): 715729.
URL: http://dx.doi.org/10.1007/s11276-007-0045-6
Leung, J. Y. T. (1984). Fast algorithms for generating all maximal independent sets of interval,
circular-arc and chordal graphs, ALGORITHMS: Journal of Algorithms 5.
Lvy, C., Linars, G. & Bonastre, J.-F. (2006). Gmm-based acoustic modeling for embedded
speech recognition, In International Conference on Speech Communication and Technol-
ogy.
Rabiner, L. R. & Wilpon, J. G. (1979). Considerations in applying clustering techniques to
speaker-independent word recognition, The Journal of the Acoustical Society of America
66(3): 663673.
URL: http://link.aip.org/link/?JAS/66/663/1
Raghavan, V. & Spinrad, J. (2003). Robust algorithms for restricted domains, ALGORITHMS:
Journal of Algorithms 48.
Reichling, M. (1988). On the detection of a common intersection of k convex objects in the
plane, Information Processing Letters 29(1): 2529.
Schmidt, R. (1986). Multiple emitter location and signal parameter estimation, IEEE Transac-
tions on Antennas and Propagation 34(3): 276280.
Smith., J. & Abel, J. (1987). Closed-form least-squares source location estimation from range-
difference measurements, IEEE Transactions on Acoustics, Speech and Signal Processing
35(12): 16611669.
Wireless Sensor Network for Disaster Monitoring
0
Wireless Sensor Network for Disaster Monitoring
Amrita Center for Wireless Networks and Applications,
Amrita Vishwa Vidyapeetham (Amrita University)
India
1. Introduction
This chapter provides a framework of the methodical steps and considerations required when
designing and deploying a Wireless Sensor Network (WSN) to a given application. A real
example is used to demonstrate WSN deployment in action.
WSN has many possible applications that have not yet been explored. WSN is a fast grow-
ing technology however much written about WSN is still theory. How to deploy WSNs,
although having much theory written still currently lacks a practical guide.
Using our research experience and the practical real life solutions found when deploying a
WSN for the application of Landslide Detection this chapter outlines the steps required when
conducting a real world deployment of a WSN.
In this chapter the application for WSN most focused on is for purpose of detecting natural
disasters. WSN can be useful to disaster management in two ways. Firstly, WSN has enabled
a more convenient early warning system and secondly, WSN provides a system able to learn
about the phenomena of natural disasters.
Natural disasters are increasing world wide due to the global warming and climate change.
The losses due to these disasters are increasing in an alarming rate. Hence, it is would be
benecial to detect the pre-cursors of these disasters, early warn the population, evacuate
them, and save their life. However, these disasters are largely unpredictable and occur within
very short spans of time. Therefore technology has to be developed to capture relevant signals
with minimum monitoring delay. Wireless Sensors are one of the cutting edge technologies
that can quickly respond to rapid changes of data and send the sensed data to a data analysis
center in areas where cabling is inappropriate.
WSNtechnology has the capability of quick capturing, processing, and transmission of critical
data in real-time with high resolution. However, it has its own limitations such as relatively
low amounts of battery power and low memory availability compared to many existing tech-
nologies. It does, though, have the advantage of deploying sensors in hostile environments
with a bare minimum of maintenance. This fullls a very important need for any real time
monitoring, especially in hazardous or remote scenarios.
Our researchers are using WSNs in the landslide scenario for estimating the chance occurrence
of landslides. India faces landslides every year with a large threat to human life causing
annual loss of US $400 million (27). The main goal of this effort is to detect rainfall induced
landslides which occur commonly in India.
Many papers have highlighted the need for a better understanding of landslide phenomena
and attempted to create systems that gather and analyse that data (1), (14) & (31).
4
The capacity of sensors and a WSN to collect and collate and analyse valuable worthwhile
data, in an ordered manner, for studying landslide phenomena or other natural disasters and
has not fully been explored.
Landslide prone-area are usually situated in terrains that are steep, hostile, difcult to access
making monitoring landslides a strenuous activity. The wireless sensor network offers itself
as an effective, reliable, low maintenance solution.
Using WSN for real-time continuous monitoring has been proven possible as shown the ex-
ample of (9) who developed a Drought Forecast and Alert System (DFAS) using a WSN. This
success in conjunction with (4) who developed a durable wireless sensor node able to remotely
monitor soil conditions and (26) who proposed a design for slip surface localization in WSNs
motivated our researchers to the design, develop, and deploy a real-time WSN for landslide
detection. This system is deployed to monitor and detect landslides, in a landslide prone area
of Kerala, India, and is further supported by laboratory setups.
This landslide detection system using a WSN is the rst in India, one of the rst in the world
of its kind. It is also one of the rst landslide eld deployments backed up by a laboratory
setup and modeling software. This system has been operational and collecting data for the
last two years, and has issued landslide warnings in July 2009. The current system can be
replicated in other rainfall induced landslide prone areas around the world.
One particular advance was the design of a Deep Earth Probe (DEP) to support the deploy-
ment of sensors. Previous landslide monitoring procedures have used sensors yet they have
not implemented connecting all the sensors to a single wireless sensor node ((29); (28); (14);
(1)). We have designed a sensor placement strategy that can be adapted for any landslide
prone area and potentially for placing sensors to detect other natural disasters, in other disas-
ter prone areas.
The chapter is arranged as follows: Requirement Analysis consisting of: Analysis of Scenario,
Selection of Geophysical Sensors, Placement of Geophysical Sensors, Spatial Distribution of
the Deep Earth Probe (DEP), Wireless Sensor Network Requirements, Algorithm Require-
ments, Network requirements (data transmission requirements/method), Data Analysis Re-
quirements and Data Visualization Requirements;
Followed by sections on: Wireless Sensor Network Architecture; Wireless Network Design
and Architecture; Wireless Sensor Network Algorithms; Wireless Software Architecture; De-
sign of Interfacing Sensors and Power Management Methods; Field Deployment Methods
and Experiences; Field Selection; Deployment of Deep Earth Probe (DEP); Network Imple-
mentation and Integration; Validation of the Complete System - Landslide Warning Issued;
and lastly, Conclusion and Future Work.
2. Requirement Analysis
This section will describe in detail how to design a real-time Wireless Sensor Network (WSN),
and what are the considerations/requirements that have to be analyzed for designing the
network for any scenario. The different processes that will contribute to a WSN design are:
Analysis of Scenario
Wireless Sensor Networks (WSN) could be useful in a vast and diverse amount of appli-
cations. The chosen target scenario must be understood and investigated thoroughly in
order to choose the most appropriate sensors and network. A comprehensive analysis
of the scenario is one of the rst steps to undertake when considering the design of the
system. The constraints found (from the analysis of the scenario) determine and govern
the overall size and type of network and sensors required.
Understanding the characteristics of a scenario allows logical links to be made about
how to detect the occurrence of land movement. The scenario here is landslides and
is then further specied to become rainfall induced landslides. The importance of
specialization is that landslides would be too generic and there would be too many
other factors to consider.
Each landslide behaves differently. Factors playing strong roles in landslide occurrence
include slope subsurface factors such as: the type of soil and its properties, soil layer
structure, the depth of the soil to bedrock, the presence of quartz or other mineral veins,
and the depth of the water table, among others and slope surface factors such as: the
types of foliage and vegetation, the topographical geography, human alterations to the
landscape, and the amount, intensity, and duration of rainfall.
Landslides are one of the major catastrophic disasters that happen around the world.
Their occurrence can be related to several causes such as geological, morphological and
physical effects, as well as human activities (30). Basically, landslides are the down-
slope movement of soil, rock and organic materials due to the inuence of gravity.
These movements are short-lived and suddenly occurring phenomena that cause ex-
traordinary landscape changes and destruction of life and property. Some slopes are
susceptible to landslides whereas others are more stable. Many factors contribute to the
instability of slopes, but the main controlling factors are the nature of the soil and un-
derlying bedrock, the conguration of the slope, the geometry of the slope, and ground-
water conditions.
In India, (27) the main landslide triggers are intense rainfall and earthquakes. Land-
slides can also be triggered by gradual processes such as weathering, or by external
mechanisms including:
Undercutting of a slope by stream erosion, wave action, glaciers, or human activ-
ity such as road building,
Intense or prolonged rainfall, rapid snowmelt, or sharp uctuations in ground-
water levels,
Shocks or vibrations caused by earthquakes or construction activity,
Loading on upper slopes, or
A combination of these and other factors.
Some of the factors that aggravate the incidence of landslides are environmental degra-
dation on account of the heavy pressure of population, decline in forest cover, change in
agricultural practices, and the development of industry and infrastructure on unstable
hill slopes, among others.
In India, the main landslide triggers are intense rainfall and earthquakes. Under heavy
rainfall conditions, rain inltration on the slope causes instability, a reduction in the
factor of safety, transient pore pressure responses, changes in water table height, a re-
duction in shear strength which holds the soil or rock, an increase in soil weight and
a reduction in the angle of repose. When the rainfall intensity is larger than the slope
saturated hydraulic conductivity, runoff occurs (12).
The key principal parameters that initiate the rainfall induced landslides are:
The capacity of sensors and a WSN to collect and collate and analyse valuable worthwhile
data, in an ordered manner, for studying landslide phenomena or other natural disasters and
has not fully been explored.
Landslide prone-area are usually situated in terrains that are steep, hostile, difcult to access
making monitoring landslides a strenuous activity. The wireless sensor network offers itself
as an effective, reliable, low maintenance solution.
Using WSN for real-time continuous monitoring has been proven possible as shown the ex-
ample of (9) who developed a Drought Forecast and Alert System (DFAS) using a WSN. This
success in conjunction with (4) who developed a durable wireless sensor node able to remotely
monitor soil conditions and (26) who proposed a design for slip surface localization in WSNs
motivated our researchers to the design, develop, and deploy a real-time WSN for landslide
detection. This system is deployed to monitor and detect landslides, in a landslide prone area
of Kerala, India, and is further supported by laboratory setups.
This landslide detection system using a WSN is the rst in India, one of the rst in the world
of its kind. It is also one of the rst landslide eld deployments backed up by a laboratory
setup and modeling software. This system has been operational and collecting data for the
last two years, and has issued landslide warnings in July 2009. The current system can be
replicated in other rainfall induced landslide prone areas around the world.
One particular advance was the design of a Deep Earth Probe (DEP) to support the deploy-
ment of sensors. Previous landslide monitoring procedures have used sensors yet they have
not implemented connecting all the sensors to a single wireless sensor node ((29); (28); (14);
(1)). We have designed a sensor placement strategy that can be adapted for any landslide
prone area and potentially for placing sensors to detect other natural disasters, in other disas-
ter prone areas.
The chapter is arranged as follows: Requirement Analysis consisting of: Analysis of Scenario,
Selection of Geophysical Sensors, Placement of Geophysical Sensors, Spatial Distribution of
the Deep Earth Probe (DEP), Wireless Sensor Network Requirements, Algorithm Require-
ments, Network requirements (data transmission requirements/method), Data Analysis Re-
quirements and Data Visualization Requirements;
Followed by sections on: Wireless Sensor Network Architecture; Wireless Network Design
and Architecture; Wireless Sensor Network Algorithms; Wireless Software Architecture; De-
sign of Interfacing Sensors and Power Management Methods; Field Deployment Methods
and Experiences; Field Selection; Deployment of Deep Earth Probe (DEP); Network Imple-
mentation and Integration; Validation of the Complete System - Landslide Warning Issued;
and lastly, Conclusion and Future Work.
2. Requirement Analysis
This section will describe in detail how to design a real-time Wireless Sensor Network (WSN),
and what are the considerations/requirements that have to be analyzed for designing the
network for any scenario. The different processes that will contribute to a WSN design are:
Analysis of Scenario
Wireless Sensor Networks (WSN) could be useful in a vast and diverse amount of appli-
cations. The chosen target scenario must be understood and investigated thoroughly in
order to choose the most appropriate sensors and network. A comprehensive analysis
of the scenario is one of the rst steps to undertake when considering the design of the
system. The constraints found (from the analysis of the scenario) determine and govern
the overall size and type of network and sensors required.
Understanding the characteristics of a scenario allows logical links to be made about
how to detect the occurrence of land movement. The scenario here is landslides and
is then further specied to become rainfall induced landslides. The importance of
specialization is that landslides would be too generic and there would be too many
other factors to consider.
Each landslide behaves differently. Factors playing strong roles in landslide occurrence
include slope subsurface factors such as: the type of soil and its properties, soil layer
structure, the depth of the soil to bedrock, the presence of quartz or other mineral veins,
and the depth of the water table, among others and slope surface factors such as: the
types of foliage and vegetation, the topographical geography, human alterations to the
landscape, and the amount, intensity, and duration of rainfall.
Landslides are one of the major catastrophic disasters that happen around the world.
Their occurrence can be related to several causes such as geological, morphological and
physical effects, as well as human activities (30). Basically, landslides are the down-
slope movement of soil, rock and organic materials due to the inuence of gravity.
These movements are short-lived and suddenly occurring phenomena that cause ex-
traordinary landscape changes and destruction of life and property. Some slopes are
susceptible to landslides whereas others are more stable. Many factors contribute to the
instability of slopes, but the main controlling factors are the nature of the soil and un-
derlying bedrock, the conguration of the slope, the geometry of the slope, and ground-
water conditions.
In India, (27) the main landslide triggers are intense rainfall and earthquakes. Land-
slides can also be triggered by gradual processes such as weathering, or by external
mechanisms including:
Undercutting of a slope by stream erosion, wave action, glaciers, or human activ-
ity such as road building,
Intense or prolonged rainfall, rapid snowmelt, or sharp uctuations in ground-
water levels,
Shocks or vibrations caused by earthquakes or construction activity,
Loading on upper slopes, or
A combination of these and other factors.
Some of the factors that aggravate the incidence of landslides are environmental degra-
dation on account of the heavy pressure of population, decline in forest cover, change in
agricultural practices, and the development of industry and infrastructure on unstable
hill slopes, among others.
In India, the main landslide triggers are intense rainfall and earthquakes. Under heavy
rainfall conditions, rain inltration on the slope causes instability, a reduction in the
factor of safety, transient pore pressure responses, changes in water table height, a re-
duction in shear strength which holds the soil or rock, an increase in soil weight and
a reduction in the angle of repose. When the rainfall intensity is larger than the slope
saturated hydraulic conductivity, runoff occurs (12).
The key principal parameters that initiate the rainfall induced landslides are:
1. Rainfall: Rainfall is one of the main triggers for the landslide. The increase in the
rainfall rate or its intensity increases the probability of landslide. Hence monitor-
ing rainfall rate is essential for the detection and prediction of landslides. This can
be performed by incorporating a rain gauge with the complete system for moni-
toring landslides.
2. Moisture: The moisture level in the soil will increase as the rainfall increases.
Enormous increase in moisture content is considered to be a primary indication
for landslide initiation. Hence, it is very important to know the soil moisture at
which the soil loses sheer strength and eventually triggers failure.
3. Pore pressure: The pore pressure piezometer is one of the critical sensors needed
for the rainfall induced landslide detection. As rainfall increases rainwater accu-
mulates at the pores of the soil. This exerts a negative pressure and also it causes
the loosening of soil strength. So the groundwater pore pressure must be mea-
sured, as this measurement provides critical information about how much water
is in the ground. As the amount of water in the ground is directly related to the soil
cohesion strength, this parameter is one of the most important for slope stability
and landslide prediction.
4. Tilt: Sliding of soil layers has to be measured for identifying the slope failures.
This can be performed by measuring the angular tilt (angular slide) during the
slope failure.
5. : Vibrations: Vibrations in the earth can be produced during the initiation of a
landslide, as the land mass starts to move, but does not fully slide. These vibra-
tions can be monitored and taken as a precursor to a full landslide.
Selection of Geophysical Sensors
Landslide detection requires measurement of principal parameters discussed in the
above section. The key geophysical sensors such as rain gauge, soil moisture sensors,
pore pressure transducers, strain gauges, tiltmeters, and geophones are identied for
measuring the principal parameters. These sensors are selected based on their rele-
vance in nding the causative geological factors for inducing landslides under heavy
rainfall conditions.
The details of the selected sensors are:
Dielectric moisture sensors: Capacitance-type dielectric moisture sensors are used
to monitor the changes experienced in volumetric water content.
Pore pressure piezometers: Pore pressure piezometers are used to capture the pore
pressure variations, as the rainfall rate varies. Either the vibrating wire piezometer
or the strain gauge type piezometer is used for in this deployment.
Strain gauges: When attached to a DEP (Deep Earth Probe), a strain gauge can
be used to measure the movement of soil layers. Strain gauges of different resis-
tance such as 100, 350, and 1000have been used for deployment, to measure
deections in the DEP of 0.5 mm per meter.
Tiltmeters: Tiltmeters are used for measuring the soil layer movements such as
very slow creep movements or sudden movements. High accuracy tiltmeters are
required for this scenario.
Geophones: The geophone is used for the analysis of vibrations caused during a
landslide. The characteristics of landslides demand the measurement of frequen-
cies up to 250 Hz. The resolution should be within 0.1 Hz and these measurements
need to be collected real-time.
Rain gauges: Maximum rainfall of 5000 mm per year needs to be measured using
the tipping bucket. The tipping bucket type of wireless rain gauge, in which the
tipping event is counted as .001 inch of rainfall, has been deployed.
Temperature sensors: The physical properties of soil and water change with tem-
perature. A resolution of 1/10th degree Celsius, measured every 15 minutes, is
sufcient. Temperature measurements are collected using the rain gauge.
Cost-Effective Considerations Cost-effective solutions have been explored, e.g. using
strain gauges for monitoring slope movement. Investigation into the sensors is a neces-
sary pursuit. Searching for cost effective, yet reliable sensors and accessing their ability
to process that data is an issue. When choosing appropriate sensors for your given ap-
plication it is necessary to access the usefulness of a sensor and its ability to provide the
type of worthwhile data required. Developing the ability of sensors effects the applica-
tions currently available.
Another factor when considering the most appropriate sensors is how cost-effective the
sensor is, for example our team opted to use strain gauges for monitoring slope move-
ment which are signicantly cheaper than tiltmeters. Though in choosing to use strain
gauges it took a much longer to develop the signal conditioning and electronics to inter-
face the strain gauge to the wireless sensor nodes. It also took a longer time to learn how
to accurately interpret the data resulting from the strain gauges since strain gauges cap-
ture more noise (and unwanted signals) than other more expensive sensors. Therefore
signal conditioning was required to extract the relevant signals that determine slope
movements from the strain gauges raw data.
Nested Dielectric Moisture Sensor is another cost effective choice made by the re-
searchers for monitoring the inltration rate.
Placement of Geophysical Sensors
The chosen, above mentioned, sensors or a combination of them can be used for detect-
ing landslides. The terrain and type of landslide will determine the group of sensors
to be used in a particular location for detecting landslides. All the chosen geophysical
sensors are capable of real-time monitoring with bare minimum maintenance. A DEP
(Deep Earth Probe) was devised to deploy these many sensors as a stack, attached to
a vertical pipe, in different locations of the landslide prone site. This generalized de-
sign for the DEP, and the sensor placement procedures at the DEP has been developed
and implemented to simplify future deployments. This design can be adapted for any
landslide prone area and potentially for placing sensors to detect other natural disas-
ters, in other disaster prone areas. Preparation with an eye on the future is an integral
part of the development of a practical system, as this design for a DEP proves. Cur-
rently replication of this particular system is being requested across much of India by
the Government, the design of the DEP will enable each procedure to be much more
organised and simplify deployment.
The ideal depth for the DEP to be deployed would be the same as the depth of the
bedrock in that location.
1. Rainfall: Rainfall is one of the main triggers for the landslide. The increase in the
rainfall rate or its intensity increases the probability of landslide. Hence monitor-
ing rainfall rate is essential for the detection and prediction of landslides. This can
be performed by incorporating a rain gauge with the complete system for moni-
toring landslides.
2. Moisture: The moisture level in the soil will increase as the rainfall increases.
Enormous increase in moisture content is considered to be a primary indication
for landslide initiation. Hence, it is very important to know the soil moisture at
which the soil loses sheer strength and eventually triggers failure.
3. Pore pressure: The pore pressure piezometer is one of the critical sensors needed
for the rainfall induced landslide detection. As rainfall increases rainwater accu-
mulates at the pores of the soil. This exerts a negative pressure and also it causes
the loosening of soil strength. So the groundwater pore pressure must be mea-
sured, as this measurement provides critical information about how much water
is in the ground. As the amount of water in the ground is directly related to the soil
cohesion strength, this parameter is one of the most important for slope stability
and landslide prediction.
4. Tilt: Sliding of soil layers has to be measured for identifying the slope failures.
This can be performed by measuring the angular tilt (angular slide) during the
slope failure.
5. : Vibrations: Vibrations in the earth can be produced during the initiation of a
landslide, as the land mass starts to move, but does not fully slide. These vibra-
tions can be monitored and taken as a precursor to a full landslide.
Selection of Geophysical Sensors
Landslide detection requires measurement of principal parameters discussed in the
above section. The key geophysical sensors such as rain gauge, soil moisture sensors,
pore pressure transducers, strain gauges, tiltmeters, and geophones are identied for
measuring the principal parameters. These sensors are selected based on their rele-
vance in nding the causative geological factors for inducing landslides under heavy
rainfall conditions.
The details of the selected sensors are:
Dielectric moisture sensors: Capacitance-type dielectric moisture sensors are used
to monitor the changes experienced in volumetric water content.
Pore pressure piezometers: Pore pressure piezometers are used to capture the pore
pressure variations, as the rainfall rate varies. Either the vibrating wire piezometer
or the strain gauge type piezometer is used for in this deployment.
Strain gauges: When attached to a DEP (Deep Earth Probe), a strain gauge can
be used to measure the movement of soil layers. Strain gauges of different resis-
tance such as 100, 350, and 1000have been used for deployment, to measure
deections in the DEP of 0.5 mm per meter.
Tiltmeters: Tiltmeters are used for measuring the soil layer movements such as
very slow creep movements or sudden movements. High accuracy tiltmeters are
required for this scenario.
Geophones: The geophone is used for the analysis of vibrations caused during a
landslide. The characteristics of landslides demand the measurement of frequen-
cies up to 250 Hz. The resolution should be within 0.1 Hz and these measurements
need to be collected real-time.
Rain gauges: Maximum rainfall of 5000 mm per year needs to be measured using
the tipping bucket. The tipping bucket type of wireless rain gauge, in which the
tipping event is counted as .001 inch of rainfall, has been deployed.
Temperature sensors: The physical properties of soil and water change with tem-
perature. A resolution of 1/10th degree Celsius, measured every 15 minutes, is
sufcient. Temperature measurements are collected using the rain gauge.
Cost-Effective Considerations Cost-effective solutions have been explored, e.g. using
strain gauges for monitoring slope movement. Investigation into the sensors is a neces-
sary pursuit. Searching for cost effective, yet reliable sensors and accessing their ability
to process that data is an issue. When choosing appropriate sensors for your given ap-
plication it is necessary to access the usefulness of a sensor and its ability to provide the
type of worthwhile data required. Developing the ability of sensors effects the applica-
tions currently available.
Another factor when considering the most appropriate sensors is how cost-effective the
sensor is, for example our team opted to use strain gauges for monitoring slope move-
ment which are signicantly cheaper than tiltmeters. Though in choosing to use strain
gauges it took a much longer to develop the signal conditioning and electronics to inter-
face the strain gauge to the wireless sensor nodes. It also took a longer time to learn how
to accurately interpret the data resulting from the strain gauges since strain gauges cap-
ture more noise (and unwanted signals) than other more expensive sensors. Therefore
signal conditioning was required to extract the relevant signals that determine slope
movements from the strain gauges raw data.
Nested Dielectric Moisture Sensor is another cost effective choice made by the re-
searchers for monitoring the inltration rate.
Placement of Geophysical Sensors
The chosen, above mentioned, sensors or a combination of them can be used for detect-
ing landslides. The terrain and type of landslide will determine the group of sensors
to be used in a particular location for detecting landslides. All the chosen geophysical
sensors are capable of real-time monitoring with bare minimum maintenance. A DEP
(Deep Earth Probe) was devised to deploy these many sensors as a stack, attached to
a vertical pipe, in different locations of the landslide prone site. This generalized de-
sign for the DEP, and the sensor placement procedures at the DEP has been developed
and implemented to simplify future deployments. This design can be adapted for any
landslide prone area and potentially for placing sensors to detect other natural disas-
ters, in other disaster prone areas. Preparation with an eye on the future is an integral
part of the development of a practical system, as this design for a DEP proves. Cur-
rently replication of this particular system is being requested across much of India by
the Government, the design of the DEP will enable each procedure to be much more
organised and simplify deployment.
The ideal depth for the DEP to be deployed would be the same as the depth of the
bedrock in that location.
The DEP design uses a heterogeneous structure with different types of geophysical sen-
sors at different positions. The geological and hydrological properties, at the location of
each of the DEPs, determine the total number of each of the geophysical sensors needed
and their corresponding position on the DEP. These geophysical sensors are deployed
or attached inside or outside of the DEP according to each of their specic deployment
strategies.
All the geological sensors on the DEP are connected to the wireless sensor node via a
data acquisition board as shown in Figure 1. This apparatus, including the DEP with its
sensors, the data acquisition board and the wireless sensor node, is conjunctly termed a
wireless probe (WP).
Fig. 1. Multi Sensor Deep Earth Probe
Spatial Distribution of the DEP (Deep Earth Probe)
Challenges come when wide area monitoring is required. Different approaches can
be used for determining the spatial distribution and deployment of Wireless Probes
(WPs). The different approaches considered are the Random Approach, the Matrix Ap-
proach, the Vulnerability Index Approach, and the Hybrid Approach. In the Random
Approach, WPs can be deployed at all possible locations according to the terrain struc-
ture of a landslide prone mountain. Whereas in the Matrix Approach, the total area of
deployment, A, is sectored into a matrix of NxN size, and one WP is placed in each cell
of the matrix. The cell size of the matrix is selected by the smallest value of the max-
imum range covered by each sensor present with the DEP. In the Vulnerability Index
Approach, WPs are deployed in vulnerable regions that have been identied during the
site investigation, terrain mapping, and soil testing. The Hybrid Approach incorporates
more than one approach stated earlier. After considering these different approaches, a
particular approach suitable for the deployment area has to be selected.
Wireless Sensor Network Requirements
Landslide detection requires wide area monitoring, and real-time, continuous data col-
lection, processing, and aggregation. Wireless Sensor Networks (WSNs) are the key
emerging technology that has the capability to real-time, continuous data collection,
processing, aggregation with minimum maintenance. Any wide area monitoring must
determine the
maximum number of wireless sensor nodes,
maximum number of relay nodes,
maximum frequency of data collection from each node per minute,
maximum data rate required,
maximum power required for sampling, transmitting, processing, and receiving,
maximum tolerance limit of delay,
maximum tolerance limit of data packet loss,
Algorithm Requirements
Wide area monitoring requires efcient algorithm development for data collection, pro-
cessing, and transmission. The different criteria to be analyzed for designing the algo-
rithms are: the total area of deployment, maximum and minimum transmission range,
maximum number of sensor nodes necessary, maximum number of sensor nodes avail-
able, maximumamount of power available (in the battery), the corresponding transmis-
sion range, data storage capability of each node, availability of constant power source,
maximum bandwidth availability, frequency of data collection and transmission spe-
cic to the application scenario, and the data aggregation method suitable for the appli-
cation under consideration.
Analysis of the above requirements contributes to the development of required algo-
rithms for designing the network topology, data collection algorithm, data aggregation
algorithm, data dissemination method, energy optimized network, networks with max-
imum life time, time synchronized network, localization techniques etc.
Network Requirements
The design and development of the complete network architecture requires the knowl-
edge and understanding of relevant technologies such as wireless networks, wired net-
works, cellular networks, satellite networks etc., maximumnumber of nodes, maximum
data rate, available bandwidth, trafc rate, delay, distance between the point of data
initiation and its destination, effect of terrain structure, vegetation index, climate varia-
tion etc., on data transmission, delay, and data packet loss, accessibility/connectivity of
the area, location of DEP (Deep Earth Probe), transmission range, identication of the
communication protocol and radio interface technology, integration of the application
specic algorithms for data collection and aggregation, routing and fault tolerance etc.
These requirements have to be thoroughly analyzed with regard to the conditions of the
deployment area, maximum data transmission distance, trafc rate, and the available
technologies. Choose the best technologies that can be integrated effectively to achieve
minimum data packet loss, delay, minimum power consumption, and fast arrival of
data.
The DEP design uses a heterogeneous structure with different types of geophysical sen-
sors at different positions. The geological and hydrological properties, at the location of
each of the DEPs, determine the total number of each of the geophysical sensors needed
and their corresponding position on the DEP. These geophysical sensors are deployed
or attached inside or outside of the DEP according to each of their specic deployment
strategies.
All the geological sensors on the DEP are connected to the wireless sensor node via a
data acquisition board as shown in Figure 1. This apparatus, including the DEP with its
sensors, the data acquisition board and the wireless sensor node, is conjunctly termed a
wireless probe (WP).
Fig. 1. Multi Sensor Deep Earth Probe
Spatial Distribution of the DEP (Deep Earth Probe)
Challenges come when wide area monitoring is required. Different approaches can
be used for determining the spatial distribution and deployment of Wireless Probes
(WPs). The different approaches considered are the Random Approach, the Matrix Ap-
proach, the Vulnerability Index Approach, and the Hybrid Approach. In the Random
Approach, WPs can be deployed at all possible locations according to the terrain struc-
ture of a landslide prone mountain. Whereas in the Matrix Approach, the total area of
deployment, A, is sectored into a matrix of NxN size, and one WP is placed in each cell
of the matrix. The cell size of the matrix is selected by the smallest value of the max-
imum range covered by each sensor present with the DEP. In the Vulnerability Index
Approach, WPs are deployed in vulnerable regions that have been identied during the
site investigation, terrain mapping, and soil testing. The Hybrid Approach incorporates
more than one approach stated earlier. After considering these different approaches, a
particular approach suitable for the deployment area has to be selected.
Wireless Sensor Network Requirements
Landslide detection requires wide area monitoring, and real-time, continuous data col-
lection, processing, and aggregation. Wireless Sensor Networks (WSNs) are the key
emerging technology that has the capability to real-time, continuous data collection,
processing, aggregation with minimum maintenance. Any wide area monitoring must
determine the
maximum number of wireless sensor nodes,
maximum number of relay nodes,
maximum frequency of data collection from each node per minute,
maximum data rate required,
maximum power required for sampling, transmitting, processing, and receiving,
maximum tolerance limit of delay,
maximum tolerance limit of data packet loss,
Algorithm Requirements
Wide area monitoring requires efcient algorithm development for data collection, pro-
cessing, and transmission. The different criteria to be analyzed for designing the algo-
rithms are: the total area of deployment, maximum and minimum transmission range,
maximum number of sensor nodes necessary, maximum number of sensor nodes avail-
able, maximumamount of power available (in the battery), the corresponding transmis-
sion range, data storage capability of each node, availability of constant power source,
maximum bandwidth availability, frequency of data collection and transmission spe-
cic to the application scenario, and the data aggregation method suitable for the appli-
cation under consideration.
Analysis of the above requirements contributes to the development of required algo-
rithms for designing the network topology, data collection algorithm, data aggregation
algorithm, data dissemination method, energy optimized network, networks with max-
imum life time, time synchronized network, localization techniques etc.
Network Requirements
The design and development of the complete network architecture requires the knowl-
edge and understanding of relevant technologies such as wireless networks, wired net-
works, cellular networks, satellite networks etc., maximumnumber of nodes, maximum
data rate, available bandwidth, trafc rate, delay, distance between the point of data
initiation and its destination, effect of terrain structure, vegetation index, climate varia-
tion etc., on data transmission, delay, and data packet loss, accessibility/connectivity of
the area, location of DEP (Deep Earth Probe), transmission range, identication of the
communication protocol and radio interface technology, integration of the application
specic algorithms for data collection and aggregation, routing and fault tolerance etc.
These requirements have to be thoroughly analyzed with regard to the conditions of the
deployment area, maximum data transmission distance, trafc rate, and the available
technologies. Choose the best technologies that can be integrated effectively to achieve
minimum data packet loss, delay, minimum power consumption, and fast arrival of
data.
Data Analysis Requirements The data received from the deployment area has to be mod-
eled and analyzed according the application scenario requirements. Statistical models
and pattern recognition techniques can be used for further data analysis to determine
the warning levels. Warning levels are the level of indication (from the sensors) that
a landslide maybe becoming possible or about to occur. Along with this data analysis
architecture has to be developed for effective and fast data analysis.
Data Visualization Requirements The development of real-time systems requires the de-
sign and development of: a data dissemination method, a channel or technology that
can be used for data dissemination (within the shortest amount of time), and the data
visualization criteria & methods specic to the application scenario. The method of
data dissemination, and the allowable delay for data dissemination, and the techniques
that should be adopted for data dissemination will depend on the application scenario
under consideration. The architecture for data visualization has to be developed with
the goal of effective and fast streaming of data.
3. Wireless Sensor Network Architecture
This current deployment used a placement strategy using the Hybrid Approach, by incorpo-
rating both the Matrix Approach and the Vulnerability Index Approach. The whole deploy-
ment area was initially sectored using Matrix Approach. In each cell, the deployment location
of the Wireless Probe (WP) is decided after considering the Vulnerability Index Approach.
This has helped to maximize the collection of relevant information from the landslide prone
area.
The wide area monitoring using Wireless Sensor Network (WSN) is achieved using a region-
alized two-layer hierarchical architecture. Since the geological and hydrological properties
of each of the locations, of the landslide prone area, differ with respect to the different re-
gions they belong to they are divided into regions. The data received from each of the sensors
cannot be aggregated together due to the variability in soil geological and hydrological prop-
erties. So the whole landslide prone area is divided into regions possessing soil geological and
hydrological properties unique to their region. In this particular case, the deployment area is
divided into three regions such as crown region, middle region, and toe region of the slope as
shown in Figure 2, and numerous WPs are deployed in these regions.
Fig. 2. Regionalized Wireless Sensor Network Architecture for Landslides
4. Wireless Network Design and Architecture
One of the important requirements for any landslide detection system is the efcient delivery
of data in near real-time. This requires seamless connectivity with minimum delay in the
network. The architecture we have developed for satisfying the above requirements is shown
in the Figure 3.
Fig. 3. Wireless Sensor Network Architecture For Landslide Detection
The wireless sensor network follows a two-layer hierarchy, with lower layer wireless sensor
nodes, sample and collect the heterogeneous data from the DEP (Deep Earth Probe) and the
data packets are transmitted to the upper layer. The upper layer aggregates the data and
forwards it to the sink node (gateway) kept at the deployment site.
The current network has 20 wireless sensor nodes spread on two different hardware platforms.
The rst hardware platform is Crossbow MicaZ. This MicaZ network follows a two-layer hi-
erarchy, with a lower level (wireless probes) and a higher level (cluster head), to reduce the
energy consumption in the total network. The wireless probes (lower level nodes) sample
and collect the heterogeneous data from the DEP (Deep Earth Probe) and the data packets are
transmitted to the higher level. The higher level aggregates the data and forwards it to the
probe gateway (sink node) kept at the deployment site.
The second hardware platform, used, is the newly developed WINSOCwireless sensor nodes.
One purpose of this WINSOC network is to extensively test and validate the WINSOC nodes,
shown in Figure 4, with respect to performance reliability and energy trade-offs between
Data Analysis Requirements The data received from the deployment area has to be mod-
eled and analyzed according the application scenario requirements. Statistical models
and pattern recognition techniques can be used for further data analysis to determine
the warning levels. Warning levels are the level of indication (from the sensors) that
a landslide maybe becoming possible or about to occur. Along with this data analysis
architecture has to be developed for effective and fast data analysis.
Data Visualization Requirements The development of real-time systems requires the de-
sign and development of: a data dissemination method, a channel or technology that
can be used for data dissemination (within the shortest amount of time), and the data
visualization criteria & methods specic to the application scenario. The method of
data dissemination, and the allowable delay for data dissemination, and the techniques
that should be adopted for data dissemination will depend on the application scenario
under consideration. The architecture for data visualization has to be developed with
the goal of effective and fast streaming of data.
3. Wireless Sensor Network Architecture
This current deployment used a placement strategy using the Hybrid Approach, by incorpo-
rating both the Matrix Approach and the Vulnerability Index Approach. The whole deploy-
ment area was initially sectored using Matrix Approach. In each cell, the deployment location
of the Wireless Probe (WP) is decided after considering the Vulnerability Index Approach.
This has helped to maximize the collection of relevant information from the landslide prone
area.
The wide area monitoring using Wireless Sensor Network (WSN) is achieved using a region-
alized two-layer hierarchical architecture. Since the geological and hydrological properties
of each of the locations, of the landslide prone area, differ with respect to the different re-
gions they belong to they are divided into regions. The data received from each of the sensors
cannot be aggregated together due to the variability in soil geological and hydrological prop-
erties. So the whole landslide prone area is divided into regions possessing soil geological and
hydrological properties unique to their region. In this particular case, the deployment area is
divided into three regions such as crown region, middle region, and toe region of the slope as
shown in Figure 2, and numerous WPs are deployed in these regions.
Fig. 2. Regionalized Wireless Sensor Network Architecture for Landslides
4. Wireless Network Design and Architecture
One of the important requirements for any landslide detection system is the efcient delivery
of data in near real-time. This requires seamless connectivity with minimum delay in the
network. The architecture we have developed for satisfying the above requirements is shown
in the Figure 3.
Fig. 3. Wireless Sensor Network Architecture For Landslide Detection
The wireless sensor network follows a two-layer hierarchy, with lower layer wireless sensor
nodes, sample and collect the heterogeneous data from the DEP (Deep Earth Probe) and the
data packets are transmitted to the upper layer. The upper layer aggregates the data and
forwards it to the sink node (gateway) kept at the deployment site.
The current network has 20 wireless sensor nodes spread on two different hardware platforms.
The rst hardware platform is Crossbow MicaZ. This MicaZ network follows a two-layer hi-
erarchy, with a lower level (wireless probes) and a higher level (cluster head), to reduce the
energy consumption in the total network. The wireless probes (lower level nodes) sample
and collect the heterogeneous data from the DEP (Deep Earth Probe) and the data packets are
transmitted to the higher level. The higher level aggregates the data and forwards it to the
probe gateway (sink node) kept at the deployment site.
The second hardware platform, used, is the newly developed WINSOCwireless sensor nodes.
One purpose of this WINSOC network is to extensively test and validate the WINSOC nodes,
shown in Figure 4, with respect to performance reliability and energy trade-offs between
the two hardware platforms in a landslide scenario. WINSOC nodes are endowed with a
WINSOC distributed consensus algorithm. Another purpose of this network is to test and
validate the performance and scalability of the WINSOC distributed consensus algorithm in
a landslide scenario. This network is scalable as it provides the capability to incorporate any
new eld networks to the current network.
Fig. 4. Field Deployment of WINSOC Node With MiniatureAntenna
Data received at the gateway has to be transmitted to the Field Management Center (FMC)
which is approximately 500m away from the gateway. A Wi-Fi network is used between
the gateway and FMC to establish the connection. The FMC incorporates facilities such as
a VSAT (Very Small Aperture Terminal) satellite earth station and a broadband network for
long distant data transmission. The VSAT satellite earth station is used for data transmission
fromthe eld deployment site at Munnar, Kerala, South India to the Data Management Center
(DMC), situated at our university campus 300 km away.
The DMC consists of the database server and an analysis station, which performs data analy-
sis and landslide modeling and simulation on the eld data to determine the landslide prob-
ability. The real-time data and the results of the data analysis are real-time streamed on the
Internet. Alert services such as E-Mail, SMS and MMS are implemented to alert about: the
probability of landslides, status of the network and for monitoring the system components.
Fault tolerance is achieved even during extreme weather conditions. For example, if the VSAT
network becomes unavailable, the WAWN adapts by using the broadband or GPRS connec-
tivity at the FMC for uploading the real-time data directly to a web page with minimumdelay
and thus providing fault tolerance.
The entire system is equipped to remotely monitor the level of battery charges and the level
of solar charging rate, and indicate faulty wireless sensor nodes or geological sensors. A
feedback loop is used that remotely changes, the sampling rate of the geological sensors, with
respect to the real-time climatic variations.
This proposed network architecture is scalable, as any number of nodes and newlandslide de-
ployment elds can be incorporated via a Wi-Fi network to the same FMC. In future, this will
provide the capability to monitor many very large areas and also to incorporate the different
spatio-temporal analysis to provide an even better understanding of landslides.
The Munnar region experiences frequent landslides and has several landslide prone areas
within every 1 sq km, which can be utilized as future extension sites for landslide detection
systems. The different deployment sites can connect to the FMC via a Wi-Fi network.
5. Wireless Sensor Network Algorithms
The wireless sensor network designed and deployed for wide area landslide monitoring
requires efcient data collection, data aggregation, energy management, and fault tolerant
methods.
Regionalized dynamic clustering method is designed and implemented for effective geologi-
cal and hydrological data collection using the wireless sensor network.
Threshold based temporal data collection and data aggregation method (19)is designed and
implemented for effective data aggregation. This algorithm combined with the newly de-
signed state transition algorithm (18) contributes optimum energy consumption by each node
and in increasing the life time of the whole network, avoiding unnecessary collection, pro-
cessing and transmission of redundant data thus achieving increased energy efciency and
the simplication of the data analysis & visualization process.
Fault tolerant methods are designed and integrated in the wireless sensor network for effective
handling of node failure, reduced signal strength, high data packet loss, and low balance
energy per node.
6. Wireless Software Architecture
Real-time monitoring and detection of landslides require seamless connectivity together with
minimum delay for data transmission. The existing Wireless Sensor Network (WSN) sys-
tem for landslide detection incorporates various heterogeneous wireless networks such as the
WSN, Wi-Fi, satellite network, and broadband network. Each of these networks perform at
different frequency range, that contributes to different trafc rate, congestion, data packet loss,
buffering methods, delay, and different data collection, transmission, and processing methods.
Hence to reduce the complexity in dealing with different types of wireless network, generic
software architecture was designed and implemented for achieving all the requirements of
each of the wireless network. This wireless software architecture includes wireless sensor
network software, wireless sensor gateway software, and a middleware for heterogeneous
wireless networks.
7. Design of Interfacing Sensors and Power Management Methods
We designed special purpose interfacing circuits, since the commercially available wireless
sensor nodes do not include implanted geophysical sensors necessary for landslide monitor-
ing, and also the geophysical sensors cannot be connected directly to the data acquisition
board, integrated with the wireless sensor node. The special purpose interfacing circuit act as
an intermediary to remove the variance experienced between the required input voltage for
a data acquisition board and the output voltage received from the geophysical sensors. Thus
design requirements of the interfacing board are described in (17), and the output from the
interfacing board is directly fed into the data acquisition board inputs. Later, the signals were
software adjusted to obtain the original sensor outputs, and hence the sensor data. The details
of the interfacing circuit requirements are shown in Figure 5.
the two hardware platforms in a landslide scenario. WINSOC nodes are endowed with a
WINSOC distributed consensus algorithm. Another purpose of this network is to test and
validate the performance and scalability of the WINSOC distributed consensus algorithm in
a landslide scenario. This network is scalable as it provides the capability to incorporate any
new eld networks to the current network.
Fig. 4. Field Deployment of WINSOC Node With MiniatureAntenna
Data received at the gateway has to be transmitted to the Field Management Center (FMC)
which is approximately 500m away from the gateway. A Wi-Fi network is used between
the gateway and FMC to establish the connection. The FMC incorporates facilities such as
a VSAT (Very Small Aperture Terminal) satellite earth station and a broadband network for
long distant data transmission. The VSAT satellite earth station is used for data transmission
fromthe eld deployment site at Munnar, Kerala, South India to the Data Management Center
(DMC), situated at our university campus 300 km away.
The DMC consists of the database server and an analysis station, which performs data analy-
sis and landslide modeling and simulation on the eld data to determine the landslide prob-
ability. The real-time data and the results of the data analysis are real-time streamed on the
Internet. Alert services such as E-Mail, SMS and MMS are implemented to alert about: the
probability of landslides, status of the network and for monitoring the system components.
Fault tolerance is achieved even during extreme weather conditions. For example, if the VSAT
network becomes unavailable, the WAWN adapts by using the broadband or GPRS connec-
tivity at the FMC for uploading the real-time data directly to a web page with minimumdelay
and thus providing fault tolerance.
The entire system is equipped to remotely monitor the level of battery charges and the level
of solar charging rate, and indicate faulty wireless sensor nodes or geological sensors. A
feedback loop is used that remotely changes, the sampling rate of the geological sensors, with
respect to the real-time climatic variations.
This proposed network architecture is scalable, as any number of nodes and newlandslide de-
ployment elds can be incorporated via a Wi-Fi network to the same FMC. In future, this will
provide the capability to monitor many very large areas and also to incorporate the different
spatio-temporal analysis to provide an even better understanding of landslides.
The Munnar region experiences frequent landslides and has several landslide prone areas
within every 1 sq km, which can be utilized as future extension sites for landslide detection
systems. The different deployment sites can connect to the FMC via a Wi-Fi network.
5. Wireless Sensor Network Algorithms
The wireless sensor network designed and deployed for wide area landslide monitoring
requires efcient data collection, data aggregation, energy management, and fault tolerant
methods.
Regionalized dynamic clustering method is designed and implemented for effective geologi-
cal and hydrological data collection using the wireless sensor network.
Threshold based temporal data collection and data aggregation method (19)is designed and
implemented for effective data aggregation. This algorithm combined with the newly de-
signed state transition algorithm (18) contributes optimum energy consumption by each node
and in increasing the life time of the whole network, avoiding unnecessary collection, pro-
cessing and transmission of redundant data thus achieving increased energy efciency and
the simplication of the data analysis & visualization process.
Fault tolerant methods are designed and integrated in the wireless sensor network for effective
handling of node failure, reduced signal strength, high data packet loss, and low balance
energy per node.
6. Wireless Software Architecture
Real-time monitoring and detection of landslides require seamless connectivity together with
minimum delay for data transmission. The existing Wireless Sensor Network (WSN) sys-
tem for landslide detection incorporates various heterogeneous wireless networks such as the
WSN, Wi-Fi, satellite network, and broadband network. Each of these networks perform at
different frequency range, that contributes to different trafc rate, congestion, data packet loss,
buffering methods, delay, and different data collection, transmission, and processing methods.
Hence to reduce the complexity in dealing with different types of wireless network, generic
software architecture was designed and implemented for achieving all the requirements of
each of the wireless network. This wireless software architecture includes wireless sensor
network software, wireless sensor gateway software, and a middleware for heterogeneous
wireless networks.
7. Design of Interfacing Sensors and Power Management Methods
We designed special purpose interfacing circuits, since the commercially available wireless
sensor nodes do not include implanted geophysical sensors necessary for landslide monitor-
ing, and also the geophysical sensors cannot be connected directly to the data acquisition
board, integrated with the wireless sensor node. The special purpose interfacing circuit act as
an intermediary to remove the variance experienced between the required input voltage for
a data acquisition board and the output voltage received from the geophysical sensors. Thus
design requirements of the interfacing board are described in (17), and the output from the
interfacing board is directly fed into the data acquisition board inputs. Later, the signals were
software adjusted to obtain the original sensor outputs, and hence the sensor data. The details
of the interfacing circuit requirements are shown in Figure 5.
Fig. 5. Interfacing Circuit Requirements (17)
For any Wireless Sensor Network (WSN), power constraints are one of the major problems
faced by wide area deployments, for real-time monitoring and detection. In the current de-
ployment, maximum power is consumed for excitation of geophysical sensors than that of
transmission, processing, or reception by a wireless sensor node. Indigenous power circuits
are developed to provide constant power for the excitation of the geophysical sensors, wireless
sensor nodes, and interfacing circuits, since each of them requires different levels of power.
This power circuit board is designed with high efciency regulator chips to provide multi-
ple outputs from a single power battery input, a non-regulated 6 Volts DC supplied from
rechargeable lead acid batteries. To increase the lifetime of the lead acid batteries, they are
automatically recharged by the solar recharging unit using the charge controller.
Along with hardware power management methods, software methods are also incorporated.
Software power solutions are implemented in the wireless sensor network by integrating
switching on-off of geological sensors and by the dynamic adjustment of frequency of sen-
sor measurements, according to the different state transitions of wireless sensor nodes as de-
scribed in the research paper (18). Efcient use of power and an optimized lifetime has been
achieved by these hardware and software solutions.
8. Field Deployment Methods and Experiences
The Wireless Sensor Network (WSN) for landslide detection system is deployed at Anthoniar
colony, Munnar, Idukki (Dist), Kerala (State), India, shown in Figure 6. The deployment site
has historically experienced several landslides, with the latest one occurring in the year 2005,
which caused a death toll of 8 people.
The WSN for landslide detection system is deployed in an area of 7 acres of mountain. The
whole area consists of approximately 20 wireless sensor nodes, 20 DEPs consisting of approxi-
mately 50 geological sensors. Important research focal points were deciding the DEP locations,
designing and constructing the DEPs, DEP deployment methods, interfacing circuitry, WSN,
Wi-Fi network, satellite network, and power solutions, soil tests, and data analysis.
Extensive eld investigations were conducted for identifying the possible landslide prone ar-
eas for the deployment of the system and also for identifying the possible locations for DEP
deployment. The borehole locations for the DEPs were chosen so as to measure the cumulative
effect of geographically specic parameters that cause landslides.
The eld deployment was performed in two phases. The pilot deployment in January 2008 to
March 2008 and the main deployment from January 2009 to June 2009. The period in between
these phases involved extensive testing and calibration processes.
8.1 Field Selection
eas for the deployment of the system, in the state of Kerala, India. Approximately 15 landslide
prone areas have been visited and studied, that had historically experienced landslides. After
extensive investigation of the 15 sites, ve sites were identied as potential eld deployment
sites for a Wireless Sensor Network (WSN) in landslide monitoring applications. Other sites
were also visited but were not deemed suitable for the eld deployment due to various factors,
including: difculty of access, uncertainty about the landslide risk, lack of communication fa-
cilities, and the size of the potential landslide, among others.
Fig. 6. Deep Earth Probe Deployment Locations at the Anthoniar Colony Site, Munnar, Kerala,
India
From the shortlisted ve landslide prone sites, Anthoniar Colony was selected for deploy-
ment, which is located 700 meters Northwest of Munnar town. A rst slide had occurred
many years earlier. On July 25
th
, 2005, another landslide also occurred in Munnar at the An-
thoniar colony. A torrential rainfall of 460mm in the middle of the monsoon period was the
primary trigger. Two levels of slide area can be observed at the Anthoniar Colony, as shown
in the Figure 6.
There is a high probability for another slide at this location. Some of the factors that indicate a
probability of landslide is the seepage owduring the dry season, long vertical and horizontal
cracks, soil material has large amount of quartz vein, and the soil type is reddish colored sticky
clay. Even now when the rain falls, water will ow down on to the top of the houses that are
at the foot of the hill, indicating water saturation and higher pore pressure at the toe region,
which can indicate a landslide in future.
Fig. 5. Interfacing Circuit Requirements (17)
For any Wireless Sensor Network (WSN), power constraints are one of the major problems
faced by wide area deployments, for real-time monitoring and detection. In the current de-
ployment, maximum power is consumed for excitation of geophysical sensors than that of
transmission, processing, or reception by a wireless sensor node. Indigenous power circuits
are developed to provide constant power for the excitation of the geophysical sensors, wireless
sensor nodes, and interfacing circuits, since each of them requires different levels of power.
This power circuit board is designed with high efciency regulator chips to provide multi-
ple outputs from a single power battery input, a non-regulated 6 Volts DC supplied from
rechargeable lead acid batteries. To increase the lifetime of the lead acid batteries, they are
automatically recharged by the solar recharging unit using the charge controller.
Along with hardware power management methods, software methods are also incorporated.
Software power solutions are implemented in the wireless sensor network by integrating
switching on-off of geological sensors and by the dynamic adjustment of frequency of sen-
sor measurements, according to the different state transitions of wireless sensor nodes as de-
scribed in the research paper (18). Efcient use of power and an optimized lifetime has been
achieved by these hardware and software solutions.
8. Field Deployment Methods and Experiences
The Wireless Sensor Network (WSN) for landslide detection system is deployed at Anthoniar
colony, Munnar, Idukki (Dist), Kerala (State), India, shown in Figure 6. The deployment site
has historically experienced several landslides, with the latest one occurring in the year 2005,
which caused a death toll of 8 people.
The WSN for landslide detection system is deployed in an area of 7 acres of mountain. The
whole area consists of approximately 20 wireless sensor nodes, 20 DEPs consisting of approxi-
mately 50 geological sensors. Important research focal points were deciding the DEP locations,
designing and constructing the DEPs, DEP deployment methods, interfacing circuitry, WSN,
Wi-Fi network, satellite network, and power solutions, soil tests, and data analysis.
eas for the deployment of the system and also for identifying the possible locations for DEP
deployment. The borehole locations for the DEPs were chosen so as to measure the cumulative
effect of geographically specic parameters that cause landslides.
The eld deployment was performed in two phases. The pilot deployment in January 2008 to
March 2008 and the main deployment from January 2009 to June 2009. The period in between
these phases involved extensive testing and calibration processes.
8.1 Field Selection
eas for the deployment of the system, in the state of Kerala, India. Approximately 15 landslide
prone areas have been visited and studied, that had historically experienced landslides. After
extensive investigation of the 15 sites, ve sites were identied as potential eld deployment
sites for a Wireless Sensor Network (WSN) in landslide monitoring applications. Other sites
were also visited but were not deemed suitable for the eld deployment due to various factors,
including: difculty of access, uncertainty about the landslide risk, lack of communication fa-
cilities, and the size of the potential landslide, among others.
Fig. 6. Deep Earth Probe Deployment Locations at the Anthoniar Colony Site, Munnar, Kerala,
India
From the shortlisted ve landslide prone sites, Anthoniar Colony was selected for deploy-
ment, which is located 700 meters Northwest of Munnar town. A rst slide had occurred
many years earlier. On July 25
th
, 2005, another landslide also occurred in Munnar at the An-
thoniar colony. A torrential rainfall of 460mm in the middle of the monsoon period was the
primary trigger. Two levels of slide area can be observed at the Anthoniar Colony, as shown
in the Figure 6.
There is a high probability for another slide at this location. Some of the factors that indicate a
probability of landslide is the seepage owduring the dry season, long vertical and horizontal
cracks, soil material has large amount of quartz vein, and the soil type is reddish colored sticky
clay. Even now when the rain falls, water will ow down on to the top of the houses that are
at the foot of the hill, indicating water saturation and higher pore pressure at the toe region,
which can indicate a landslide in future.
8.2 Deployment of DEP (Deep Earth Probe)
One of the important activities required for deploying the landslide detection system is bore
hole drilling. Bore hole location and its depth, determines the maximumamount of geological
and hydrological properties that can be gathered from the eld for the functioning of the
landslide detection system. Hence, the most important parameters that determine the bore
hole design include the decision of location, depth, and diameter of the planned bore hole,
soil sample extraction methods to be adopted, eld tests involved, and the bore hole drilling
method.
Different types of bore hole drilling are available such as the hand auger method, and the
rotary drilling method. The pilot deployment the hand auger method was used. In the main
deployment the rotary drilling method was used to drill deeper holes as deep as 23 meters as
it consumed less time and labor compared to the hand auger method. Soil sample extraction
and eld permeability tests were performed to collect the relevant soil properties, geologic
and hydrologic properties from each of the locations. Drilling was continued until the bed
rock was observed. If bedrock was too deep, drilling was continued until the observation
of weathered rock. If even weathered rock was too deep, drilling was stopped at a major
soil layer change after the water table. The decision of the bore hole depth was chosen to
be dependent on the location of the hole, vulnerability of the location, sensor deployment
requirement, water table height, and location of weathered rock or bed rock.
The DEP (Deep Earth Probe) design is inuenced by the local geological and hydrological
conditions, the terrain structure, and accessibility of that location. The distribution pattern of
different types of geophysical sensors at different depths of the DEP is unique depending on
the characteristics of the specic location.
The DEPs were designed in a two stage process. Initial DEP designs were made for the pilot
deployment, which consists of two DEPs, and in the main deployment, the spatial granularity
was increased to 20 DEPs and 20 wireless sensor nodes. Multiple DEPs were installed in six
locations (labeled henceforth as either C1, C2, , C6), shown in Figure 6.
In the main deployment DEPs are placed signicantly deeper into the ground than in the
pilot deployment, on average 2 to 5 times deeper, with a maximum depth of up to 23 meters,
penetrating to the weathered rock or bed rock. The geological sensors are attached to the
ABS plastic inclinometer casing according to the geological or hydrological parameter that
will be measured. The maximum number of eight external sensors can be attached to the data
acquisition board of Crossbows wireless sensor nodes. The details of the connected geological
sensors as on June 2009, is detailed in the Figure 7.
External hardware components have been put in two enclosures that are used to protect the
data acquisition and transmission equipment for the wireless sensor node - one electronics
box and one power box and are attached to poles equipped with solar panels and external
antenna. These were designed and then fabricated at the University as shown in Figure 8.
8.3 Network Implementation and Integration
The network consists of a Wireless Sensor Network (WSN), Wi-Fi, a satellite network, a broad-
band network, a GPRS and GSM network. The network integration of all the components
required different software and hardware implementations. The design and development of
a WSN for the landslide scenario involves the consideration of different factors such as terrain
structure, vegetation index, climate variation, accessibility of the area, location of DEP (Deep
Earth Probe), transmission range, identication of the communication protocol and radio in-
terface technology, the application specic algorithms for data collection and aggregation,
Fig. 7. Main Deployment - Details of Geological Sensor Deployment, Location and its Depth
of Deployment
routing and fault tolerance etc. The wireless sensor nodes used for the deployment are 2.4
GHz MicaZ motes from Crossbow. The MDA 320 from Xbow is the data acquisition board
used to interface the sensors with the MicaZ motes. The MicaZ samples and processes the
sensor values from the MDA board that has up to 8 channels of 16-bit analog input, logs it,
and sends it to the communication routines for packetizing, framing, check sum generation,
etc.
Although the manufacturer specied that the MicaZ nodes could transmit up to 100 meters,
in ideal eld conditions (at dirt ground, dry weather) the maximum range was around 50
or 60 meters even when the motes were being placed 2 meters above the ground. Due to
this shorter transmission range, a number of relay nodes are used to maintain communication
between the DEPs. These relay nodes required extensive testing for careful placement such
that the network connectivity can be maintained even in the worst case weather conditions.
External antennas were also used to maintain network connectivity.
The hardware of the original probe gateway is fromCrossbowand is named as Stargate. Later
Stargate became unavailable and out of production thereby forcing the use of a new gateway.
The new gateway is based on an AMD Geode Mini-ITX Motherboard. The base station listens
to the packet transmissions, in the sensor network, and logs the packet transmissions if they
are addressed to the base station. After this, the sensor data is stored either accessed through
the Wi-Fi network or through the Ethernet interface of the probe gateway.
Data received at the probe gateway is transmitted to the Field Management Center (FMC), us-
ing a Wi-Fi network. The Wi-Fi network uses standard, off-the-shelf Wi-Fi components, such
as a compact ash Wi-Fi card, at the gateway, and an Ethernet wireless access point, at the
Field Management Center. The Wi-Fi network allows us to install the gateway at any scal-
able distance from the FMC. The FMC incorporates a VSAT (Very Small Aperture Terminal)
satellite earth station and a broadband/GPRS network for long distant data transmission.
Data received at the FMC is transmitted to the Data Management Center (DMC) using a satel-
lite network. The data received at the DMC is analyzed using an in-house designed data anal-
ysis and visualization software. This software is interfaced with landslide modeling software
and data analysis software developed at Amrita University. Landslide modeling software
provides the factor of safety of the mountain and the probability of landslide occurrence with
respect to the signals received from the deployed sensors. Data analysis software provides
8.2 Deployment of DEP (Deep Earth Probe)
One of the important activities required for deploying the landslide detection system is bore
hole drilling. Bore hole location and its depth, determines the maximumamount of geological
and hydrological properties that can be gathered from the eld for the functioning of the
landslide detection system. Hence, the most important parameters that determine the bore
hole design include the decision of location, depth, and diameter of the planned bore hole,
soil sample extraction methods to be adopted, eld tests involved, and the bore hole drilling
method.
Different types of bore hole drilling are available such as the hand auger method, and the
rotary drilling method. The pilot deployment the hand auger method was used. In the main
deployment the rotary drilling method was used to drill deeper holes as deep as 23 meters as
it consumed less time and labor compared to the hand auger method. Soil sample extraction
and eld permeability tests were performed to collect the relevant soil properties, geologic
and hydrologic properties from each of the locations. Drilling was continued until the bed
rock was observed. If bedrock was too deep, drilling was continued until the observation
of weathered rock. If even weathered rock was too deep, drilling was stopped at a major
soil layer change after the water table. The decision of the bore hole depth was chosen to
be dependent on the location of the hole, vulnerability of the location, sensor deployment
requirement, water table height, and location of weathered rock or bed rock.
The DEP (Deep Earth Probe) design is inuenced by the local geological and hydrological
conditions, the terrain structure, and accessibility of that location. The distribution pattern of
different types of geophysical sensors at different depths of the DEP is unique depending on
the characteristics of the specic location.
The DEPs were designed in a two stage process. Initial DEP designs were made for the pilot
deployment, which consists of two DEPs, and in the main deployment, the spatial granularity
was increased to 20 DEPs and 20 wireless sensor nodes. Multiple DEPs were installed in six
locations (labeled henceforth as either C1, C2, , C6), shown in Figure 6.
In the main deployment DEPs are placed signicantly deeper into the ground than in the
pilot deployment, on average 2 to 5 times deeper, with a maximum depth of up to 23 meters,
penetrating to the weathered rock or bed rock. The geological sensors are attached to the
ABS plastic inclinometer casing according to the geological or hydrological parameter that
will be measured. The maximum number of eight external sensors can be attached to the data
acquisition board of Crossbows wireless sensor nodes. The details of the connected geological
sensors as on June 2009, is detailed in the Figure 7.
External hardware components have been put in two enclosures that are used to protect the
data acquisition and transmission equipment for the wireless sensor node - one electronics
box and one power box and are attached to poles equipped with solar panels and external
antenna. These were designed and then fabricated at the University as shown in Figure 8.
8.3 Network Implementation and Integration
The network consists of a Wireless Sensor Network (WSN), Wi-Fi, a satellite network, a broad-
band network, a GPRS and GSM network. The network integration of all the components
required different software and hardware implementations. The design and development of
a WSN for the landslide scenario involves the consideration of different factors such as terrain
structure, vegetation index, climate variation, accessibility of the area, location of DEP (Deep
Earth Probe), transmission range, identication of the communication protocol and radio in-
terface technology, the application specic algorithms for data collection and aggregation,
Fig. 7. Main Deployment - Details of Geological Sensor Deployment, Location and its Depth
of Deployment
routing and fault tolerance etc. The wireless sensor nodes used for the deployment are 2.4
GHz MicaZ motes from Crossbow. The MDA 320 from Xbow is the data acquisition board
used to interface the sensors with the MicaZ motes. The MicaZ samples and processes the
sensor values from the MDA board that has up to 8 channels of 16-bit analog input, logs it,
and sends it to the communication routines for packetizing, framing, check sum generation,
etc.
Although the manufacturer specied that the MicaZ nodes could transmit up to 100 meters,
in ideal eld conditions (at dirt ground, dry weather) the maximum range was around 50
or 60 meters even when the motes were being placed 2 meters above the ground. Due to
this shorter transmission range, a number of relay nodes are used to maintain communication
between the DEPs. These relay nodes required extensive testing for careful placement such
that the network connectivity can be maintained even in the worst case weather conditions.
External antennas were also used to maintain network connectivity.
The hardware of the original probe gateway is fromCrossbowand is named as Stargate. Later
Stargate became unavailable and out of production thereby forcing the use of a new gateway.
The new gateway is based on an AMD Geode Mini-ITX Motherboard. The base station listens
to the packet transmissions, in the sensor network, and logs the packet transmissions if they
are addressed to the base station. After this, the sensor data is stored either accessed through
the Wi-Fi network or through the Ethernet interface of the probe gateway.
Data received at the probe gateway is transmitted to the Field Management Center (FMC), us-
ing a Wi-Fi network. The Wi-Fi network uses standard, off-the-shelf Wi-Fi components, such
as a compact ash Wi-Fi card, at the gateway, and an Ethernet wireless access point, at the
Field Management Center. The Wi-Fi network allows us to install the gateway at any scal-
able distance from the FMC. The FMC incorporates a VSAT (Very Small Aperture Terminal)
satellite earth station and a broadband/GPRS network for long distant data transmission.
Data received at the FMC is transmitted to the Data Management Center (DMC) using a satel-
lite network. The data received at the DMC is analyzed using an in-house designed data anal-
ysis and visualization software. This software is interfaced with landslide modeling software
and data analysis software developed at Amrita University. Landslide modeling software
provides the factor of safety of the mountain and the probability of landslide occurrence with
respect to the signals received from the deployed sensors. Data analysis software provides
Fig. 8. C3 Sensor Column of Main Deployment
the capability to compare and analyze data from different DEPs, different sensors in the same
DEP, the same sensors in different DEPs, selective comparison, etc.
This data analysis and visualization software is also capable of real streaming the data and
the results of the data analysis, over the Internet. Which makes it possible for the scientists
around the world to analyze the data with very minimal delay and effective warning can be
issued on time.
9. Validation of the Complete System - Landslide Warning Issued
A novel and innovative decision support system for landslide warning has been developed
using a three level warning (Early, Intermediate and Imminent). The decision for each level
depends on the moisture (for an Early warning), pore pressure (for an Intermediate warning),
and movement (for an Imminent warning) sensor data values correlating with the rainfall
intensity. Along with the three level warning system, the results of the landslide modeling
software is compared to avoid false alarms. Landslide modeling software incorporates the
raw sensor data from the eld deployment site, along with data from soil tests, lab setup ,
and other terrain information to determine the Factor of Safety (FS) (term used to quantify
Fig. 9. Snapshot from the real streaming software, for a period of 18 July 2009 (00: 00:17) to 20
July 2009 (08:09:05) for location 5, the middle position of the hill
the slope stability). Dependent on the results reaching a threshold (that is, if FS 1), each
grid point could be pronounced unsafe or safe. This implementation is incorporated into
the data visualization software and the results are real-time streamed to the website.
In July 2009, high rain fall was experienced at our deployment site and multiple landslides
occurred all over the state of Kerala, India. The data analysis showed an increase in pore
pressure and also noticeable soil movements. The pore pressure transducer deployed 14 meter
deep from the surface at location 5 (which is a vulnerable area), showed a gradual increase in
pore pressure. The strain gauges deployed at location 5, at various depths such as 4.25 x,
4.25 y, 10.75 x, 10.5 alpha, 10.5 beta and 15 x show noticeable movements of underneath soil.
More strain gauge soil movement is shown at position 10.75 x, 10.5 alpha and 10.5. Other
sensors at 6 different locations at Anthoniar Colony also showed observable soil movements
and increase in pore pressure.
Our real streaming software currently incorporated to www.winsoc.org website can be used
to view the pattern.
Figure 9 shows the real-time streaming data, for a period of July 18th, 2009 to July 20th, 2009
for location 5. The gure shows an increase in the pore pressure and also soil movements at
the middle position of the hill, which is actually a vulnerable area after the previous landslide
of July, 2005. Additionally, the soil moisture sensor readings at location 1, the toe region of the
hill, were already saturated. The strain gauges at location 1 and location 4 also showed slight
soil movements.All of the above analysis shows the vulnerability of Anthoniar Colony to pos-
sible landslides. In this context, we issued a preliminary warning through television channels,
and the ofcial Kerala State Government authorities were informed. The government author-
ities considered the warning seriously. Higher ofcials made visits to the landslide prone area
and the people were asked to evacuate with the warning given below.
We would like to inform you that in case the torrential rainfall prevails, it would be wiser to alert the
people of this region and advise them to relocate to another area till the region comes back to normalcy
in terms of pore pressure and underneath soil movements.
Fig. 8. C3 Sensor Column of Main Deployment
the capability to compare and analyze data from different DEPs, different sensors in the same
DEP, the same sensors in different DEPs, selective comparison, etc.
This data analysis and visualization software is also capable of real streaming the data and
the results of the data analysis, over the Internet. Which makes it possible for the scientists
around the world to analyze the data with very minimal delay and effective warning can be
issued on time.
9. Validation of the Complete System - Landslide Warning Issued
A novel and innovative decision support system for landslide warning has been developed
using a three level warning (Early, Intermediate and Imminent). The decision for each level
depends on the moisture (for an Early warning), pore pressure (for an Intermediate warning),
and movement (for an Imminent warning) sensor data values correlating with the rainfall
intensity. Along with the three level warning system, the results of the landslide modeling
software is compared to avoid false alarms. Landslide modeling software incorporates the
raw sensor data from the eld deployment site, along with data from soil tests, lab setup ,
and other terrain information to determine the Factor of Safety (FS) (term used to quantify
Fig. 9. Snapshot from the real streaming software, for a period of 18 July 2009 (00: 00:17) to 20
July 2009 (08:09:05) for location 5, the middle position of the hill
the slope stability). Dependent on the results reaching a threshold (that is, if FS 1), each
grid point could be pronounced unsafe or safe. This implementation is incorporated into
the data visualization software and the results are real-time streamed to the website.
In July 2009, high rain fall was experienced at our deployment site and multiple landslides
occurred all over the state of Kerala, India. The data analysis showed an increase in pore
pressure and also noticeable soil movements. The pore pressure transducer deployed 14 meter
deep from the surface at location 5 (which is a vulnerable area), showed a gradual increase in
pore pressure. The strain gauges deployed at location 5, at various depths such as 4.25 x,
4.25 y, 10.75 x, 10.5 alpha, 10.5 beta and 15 x show noticeable movements of underneath soil.
More strain gauge soil movement is shown at position 10.75 x, 10.5 alpha and 10.5. Other
sensors at 6 different locations at Anthoniar Colony also showed observable soil movements
and increase in pore pressure.
Our real streaming software currently incorporated to www.winsoc.org website can be used
to view the pattern.
Figure 9 shows the real-time streaming data, for a period of July 18th, 2009 to July 20th, 2009
for location 5. The gure shows an increase in the pore pressure and also soil movements at
the middle position of the hill, which is actually a vulnerable area after the previous landslide
of July, 2005. Additionally, the soil moisture sensor readings at location 1, the toe region of the
hill, were already saturated. The strain gauges at location 1 and location 4 also showed slight
soil movements.All of the above analysis shows the vulnerability of Anthoniar Colony to pos-
sible landslides. In this context, we issued a preliminary warning through television channels,
and the ofcial Kerala State Government authorities were informed. The government author-
ities considered the warning seriously. Higher ofcials made visits to the landslide prone area
and the people were asked to evacuate with the warning given below.
We would like to inform you that in case the torrential rainfall prevails, it would be wiser to alert the
people of this region and advise them to relocate to another area till the region comes back to normalcy
in terms of pore pressure and underneath soil movements.
As the rainfall reduced, the real-time streaming software showed the pore pressure reducing
and then stabilizing. This situation helped us to validate the complete system. As a result of
the successful warning issuance and system validation, the Indian government now wants to
extent the network to all possible landslide areas.
10. Conclusion
Wireless Sensor Networks (WSNs) are still an emerging technology and much literature avail-
able is still theoretical, therefore practical deployment guides using actual experience are few
if any. Using real practical experience, this overview of operations is one such guide provid-
ing the methodical steps and outlining the basic requirements when designing and deploying
a WSN into any given application.
This chapter discusses the design and deployment of a landslide detection system using a
WSN system at Anthoniar Colony, Munnar, Idukki (Dist), Kerala (State), India, a highly land-
slide prone area The deployment site had historically experienced several landslides, with the
latest one occurring in the year 2005, which caused a death toll of 10 (people).
Our researchers, at Amrita University, designed and deployed a Wireless Sensor Network for
the purpose of landslide detection. The complete functional system consists of 50 geological
sensors and 20 wireless sensor nodes. This network has the capability to provide real-time
data through the Internet and also to issue warnings ahead of time using the innovative three
level warning systemdeveloped as part of this work. The systemincorporates energy efcient
data collection methods, fault tolerant clustering approaches, and threshold based data aggre-
gation techniques. This wireless sensor network system is in place. For two years it has been
gathering vast amounts of data, providing better understanding of landslide scenario and has
been poised to warn of any pertinent landslide disaster in future. The system has proved
its validity by delivering real warning to the local community during heavy rains in the last
monsoon season (July 2009). This system is scalable to other landslide prone areas and also it
can be used for ood, avalanche, and water quality monitoring with minor modications.
This development describes a real experience and makes apparent the signicant advantages
of using Wireless Sensor Networks in Disaster Management. The knowledge gained from this
actual experience is useful in the development of other systems for continuous monitoring
and detection of critical and emergency applications.
Acknowledgments
The author would like to express gratitude for the immense amount of motivation and re-
search solutions provided by Sri. Mata Amritanandamayi Devi, The Chancellor, Amrita Uni-
versity. The authors would also like to acknowledge Dr. P. Venkat Rangan, Dr. H. M. Iyer,
Dr. P. V. Ushakumari, Dr. Bharat Jayaraman, Dr. Nirmala Vasudevan, Mr. Sangeeth Kumar,
Mr. Joshua (Udar) D. Freeman, Mr. Vijayan Selvan, Mr. Kalainger (Kailash) Thangaraju, Mr.
Mukundan T. Raman, Ms Rekha. Prabha, Ms. Thushara. Eranholi, Mr. Manohar. B. Patil, Ms.
Erica (Thapasya) S. Fernandes for their valuable contribution to this work.
11. References
[1] Biavati, G.; Godt, W. & McKenna, J. P. (2006). Drainage effects on the transient, near-
surface hydrologic response of a steep hillslope to rainfall: implications for slope sta-
bility, Edmonds, Washington, USA, Natural Hazards Earth System Science, Vol. 6, pp.
343-355, 2006.
[2] Caine, N. (1980). The rainfall intensity-duration control of shallowlandslides and debris
ows, Geograska Annaler, Vol. 62A, pp. 23-27, 1980.
[3] Christanto, N.; Hadmoko, D.S.; Westen, C. J.; Lavigne, F.; Sartohadi, J. & Setiawan, M.
A. (2009). Characteristic and Behavior of Rainfall Induced Landslides in Java Island,
Indonesia : an Overview, Geophysical Research Abstracts, Vol. 11, 2009.
[4] Garich, E. A. (2007). Wireless, Automated Monitoring For Potential Landslide Hazards,
Masters Thesis, Texas A & M University, 2007.
[5] Hill, C. &Sippel, K. (2002) Modern Deformation Monitoring: AMulti Sensor Approach,
In Proceedings of FIG XXII International Congress, Washington D.C, USA, April 19-26,
2002.
[6] Iverson, R.M. (2000). Landslide triggering by rain inltration, Water Resource Research,
Vol. 36, pp. 1897-1910, 2000.
[7] Kimura, H. & Yamaguchl. Y. (2000). Detection of landslide areas using satellite radar
interferometry, Photogrammetric Engineering & Remote Sensing, Vol. 6 (3), pp. 337-344,
2000.
[8] Kumar, V.S.; Sampath, S.; Vinayak, P. V. S. S. K. &Harikumar, R. (2007). Rainfall intensity
characteristics in at coastal and high altitude stations in Kerala, Journal of Earth Systems
Sciences, Vol. 116 (5), pp. 451-463, 2007.
[9] Kung, H.; Hua, J. & Chen, C. (2006) Drought Forecast Model and Framework Using
Wireless Sensor Networks, Journal of Information Science and Engineering, Vol. 22, pp.
751-769, 2006.
[10] Kunnath, A. T. & Ramesh, M. V. (2010). Integrating geophone network to real-time
wireless sensor network system for landslide detection, In Proceedings of The Third In-
ternational Conference on Sensor Technologies and Applications, SENSORCOMM 2010, IEEE
Xplore, 2010.
[11] Kuriakose, S.; Jetten, V. G.; van Westen, C. J.; Sankar, G. & van Beek, L. P. H. (2008). Pore
Water Pressure as a Trigger of ShallowLandslides in the Western Ghats of Kerala, India:
Some Preliminary Observations from an Experimental Catchment, Physical Geography,
Vol. 29 (4), pp. 374-386, 2008.
[12] LAN, Hengxing.; ZHOU, C.; Lee , C. F.; WANG, S. & Faquan, W. U. (2003). Rainfall-
induced landslide stability analysis in response to transient pore pressure - Acase study
of natural terrain landslide in Hong Kong, Science in China Ser. E Technological Sciences,
Vol. 46, pp. 52-68, 2003.
[13] Liu, H.; Meng, Z. & Cui, S. (2007). A Wireless Sensor Network Prototype for Environ-
mental Monitoring in Greenhouses, IEEE Xplore, 2007.
[14] Marchi, L.; Arattano, M. & Deganutti, A. M. (2002). Ten years of debris-ow monitoring
in the Moscardo Torrent (Italian Alphs), Geomorphology, Elseveir, Vol. 46, pp. 1-17, 2002.
[15] Martinez, C.; Jorge, E. & Olarte Montero, Juan, eds. (2001). Socioeconomic and Envi-
ronmental Impacts of Landslides in the Western Hemisphere, Proceedings of the Third
Panamerican Symposium on Landslides, July 29 to August 3, Cartagena, Colombia, 2001.
[16] Musaloiu-E, R.; Terzis, A.; Szlavecz, K.; Szalay, A.; Cogan, J. &Gray, J. (2006). Life Under
your Feet: A Wireless Soil Ecology Sensor Network, 2006.
[17] Ramesh, M. V. (2009). Real-time Wireless Sensor Network for Landslide Detection, In
Proceedings of The Third International Conference on Sensor Technologies and Applications,
SENSORCOMM 2009, IEEE Digital Library, 2009.
[18] Ramesh, M.V.; Raj, R. T.; Freeman, J. D.; Kumar, S. & Rangan, P. V. (2007). Factors
and Approaches for Energy Optimized Wireless Sensor Network to Detect Rainfall In-
As the rainfall reduced, the real-time streaming software showed the pore pressure reducing
and then stabilizing. This situation helped us to validate the complete system. As a result of
the successful warning issuance and system validation, the Indian government now wants to
extent the network to all possible landslide areas.
10. Conclusion
Wireless Sensor Networks (WSNs) are still an emerging technology and much literature avail-
able is still theoretical, therefore practical deployment guides using actual experience are few
if any. Using real practical experience, this overview of operations is one such guide provid-
ing the methodical steps and outlining the basic requirements when designing and deploying
a WSN into any given application.
This chapter discusses the design and deployment of a landslide detection system using a
WSN system at Anthoniar Colony, Munnar, Idukki (Dist), Kerala (State), India, a highly land-
slide prone area The deployment site had historically experienced several landslides, with the
latest one occurring in the year 2005, which caused a death toll of 10 (people).
Our researchers, at Amrita University, designed and deployed a Wireless Sensor Network for
the purpose of landslide detection. The complete functional system consists of 50 geological
sensors and 20 wireless sensor nodes. This network has the capability to provide real-time
data through the Internet and also to issue warnings ahead of time using the innovative three
level warning systemdeveloped as part of this work. The systemincorporates energy efcient
data collection methods, fault tolerant clustering approaches, and threshold based data aggre-
gation techniques. This wireless sensor network system is in place. For two years it has been
gathering vast amounts of data, providing better understanding of landslide scenario and has
been poised to warn of any pertinent landslide disaster in future. The system has proved
its validity by delivering real warning to the local community during heavy rains in the last
monsoon season (July 2009). This system is scalable to other landslide prone areas and also it
can be used for ood, avalanche, and water quality monitoring with minor modications.
This development describes a real experience and makes apparent the signicant advantages
of using Wireless Sensor Networks in Disaster Management. The knowledge gained from this
actual experience is useful in the development of other systems for continuous monitoring
and detection of critical and emergency applications.
Acknowledgments
The author would like to express gratitude for the immense amount of motivation and re-
search solutions provided by Sri. Mata Amritanandamayi Devi, The Chancellor, Amrita Uni-
versity. The authors would also like to acknowledge Dr. P. Venkat Rangan, Dr. H. M. Iyer,
Dr. P. V. Ushakumari, Dr. Bharat Jayaraman, Dr. Nirmala Vasudevan, Mr. Sangeeth Kumar,
Mr. Joshua (Udar) D. Freeman, Mr. Vijayan Selvan, Mr. Kalainger (Kailash) Thangaraju, Mr.
Mukundan T. Raman, Ms Rekha. Prabha, Ms. Thushara. Eranholi, Mr. Manohar. B. Patil, Ms.
Erica (Thapasya) S. Fernandes for their valuable contribution to this work.
11. References
[1] Biavati, G.; Godt, W. & McKenna, J. P. (2006). Drainage effects on the transient, near-
surface hydrologic response of a steep hillslope to rainfall: implications for slope sta-
bility, Edmonds, Washington, USA, Natural Hazards Earth System Science, Vol. 6, pp.
343-355, 2006.
[2] Caine, N. (1980). The rainfall intensity-duration control of shallowlandslides and debris
ows, Geograska Annaler, Vol. 62A, pp. 23-27, 1980.
[3] Christanto, N.; Hadmoko, D.S.; Westen, C. J.; Lavigne, F.; Sartohadi, J. & Setiawan, M.
A. (2009). Characteristic and Behavior of Rainfall Induced Landslides in Java Island,
Indonesia : an Overview, Geophysical Research Abstracts, Vol. 11, 2009.
[4] Garich, E. A. (2007). Wireless, Automated Monitoring For Potential Landslide Hazards,
Masters Thesis, Texas A & M University, 2007.
[5] Hill, C. &Sippel, K. (2002) Modern Deformation Monitoring: AMulti Sensor Approach,
In Proceedings of FIG XXII International Congress, Washington D.C, USA, April 19-26,
2002.
[6] Iverson, R.M. (2000). Landslide triggering by rain inltration, Water Resource Research,
Vol. 36, pp. 1897-1910, 2000.
[7] Kimura, H. & Yamaguchl. Y. (2000). Detection of landslide areas using satellite radar
interferometry, Photogrammetric Engineering & Remote Sensing, Vol. 6 (3), pp. 337-344,
2000.
[8] Kumar, V.S.; Sampath, S.; Vinayak, P. V. S. S. K. &Harikumar, R. (2007). Rainfall intensity
characteristics in at coastal and high altitude stations in Kerala, Journal of Earth Systems
Sciences, Vol. 116 (5), pp. 451-463, 2007.
[9] Kung, H.; Hua, J. & Chen, C. (2006) Drought Forecast Model and Framework Using
Wireless Sensor Networks, Journal of Information Science and Engineering, Vol. 22, pp.
751-769, 2006.
[10] Kunnath, A. T. & Ramesh, M. V. (2010). Integrating geophone network to real-time
wireless sensor network system for landslide detection, In Proceedings of The Third In-
ternational Conference on Sensor Technologies and Applications, SENSORCOMM 2010, IEEE
Xplore, 2010.
[11] Kuriakose, S.; Jetten, V. G.; van Westen, C. J.; Sankar, G. & van Beek, L. P. H. (2008). Pore
Water Pressure as a Trigger of ShallowLandslides in the Western Ghats of Kerala, India:
Some Preliminary Observations from an Experimental Catchment, Physical Geography,
Vol. 29 (4), pp. 374-386, 2008.
[12] LAN, Hengxing.; ZHOU, C.; Lee , C. F.; WANG, S. & Faquan, W. U. (2003). Rainfall-
induced landslide stability analysis in response to transient pore pressure - Acase study
of natural terrain landslide in Hong Kong, Science in China Ser. E Technological Sciences,
Vol. 46, pp. 52-68, 2003.
[13] Liu, H.; Meng, Z. & Cui, S. (2007). A Wireless Sensor Network Prototype for Environ-
mental Monitoring in Greenhouses, IEEE Xplore, 2007.
[14] Marchi, L.; Arattano, M. & Deganutti, A. M. (2002). Ten years of debris-ow monitoring
in the Moscardo Torrent (Italian Alphs), Geomorphology, Elseveir, Vol. 46, pp. 1-17, 2002.
[15] Martinez, C.; Jorge, E. & Olarte Montero, Juan, eds. (2001). Socioeconomic and Envi-
ronmental Impacts of Landslides in the Western Hemisphere, Proceedings of the Third
Panamerican Symposium on Landslides, July 29 to August 3, Cartagena, Colombia, 2001.
[16] Musaloiu-E, R.; Terzis, A.; Szlavecz, K.; Szalay, A.; Cogan, J. &Gray, J. (2006). Life Under
your Feet: A Wireless Soil Ecology Sensor Network, 2006.
[17] Ramesh, M. V. (2009). Real-time Wireless Sensor Network for Landslide Detection, In
Proceedings of The Third International Conference on Sensor Technologies and Applications,
SENSORCOMM 2009, IEEE Digital Library, 2009.
[18] Ramesh, M.V.; Raj, R. T.; Freeman, J. D.; Kumar, S. & Rangan, P. V. (2007). Factors
and Approaches for Energy Optimized Wireless Sensor Network to Detect Rainfall In-
duced Landslides, In Proceedings of the 2007 International Conference on Wireless Networks
(ICWN07), pp. 435-438, CSREA Press, June, 2007.
[19] Ramesh, M. V. & Ushakumari, P. (2008). Threshold Based Data Aggregation Algorithm
To Detect Rainfall Induced Landslides, In Proceedings of the 2008 International Conference
on Wireless Networks (ICWN08), Vol. 1, pp. 255-261, 2008.
[20] Ramesh, M. V.; Vasudevan, N. & Freeman, J. (2009). Real Time Landslide Monitoring
via Wireless Sensor Network,In EGU Geophysical Research Abstracts, Vol. 11, EGU2009-
14061, EGU General Assembly, 2009.
[21] Ramesh, M. V.; Kumar, S. & Rangan, P. V. (2009). Wireless Sensor Network for Land-
slide Detection, In the Proceedings of the 2009 International Conference on Wireless Networks
(ICWN09), pp. 89-95, 2009.
[22] Ramesh, M. V. &Soman, K. P. (2009). Wireless Sensor Network LocalizationWith Impre-
cise Measurements Using Only a Quadratic Solver, In the Proceedings of the 2009 Interna-
tional Conference on Wireless Networks (ICWN09), pp. 109-112, 2009.
[23] Raj, R.; Ramesh, M. V. & Kumar, S. (2008). Fault Tolerant Clustering Approaches in
Wireless Sensor Network for Landslide Area Monitoring, In Proceedings of the 2008 In-
ternational Conference on Wireless Networks (ICWN08), Vol. 1, pp. 107-113, 2008.
[24] Reid, M. E.; LaHusen, M. E. & Iverson, R.M. (1988). Hydraulic factors triggering a shal-
low landslide failure, Bulletin of the Association of Engineering Geologists, Vol. 25(3), pp.
349-361, 1988.
[25] Takehiko, O. & Yoshitsugu, T. (1998). The Characteristics of Pore Water Pressure Fluctu-
ation on the Slip Surface Based on Drainage Well Works: Study of Tairasawa Landslide
Area, Saitama Prefecture, Landslides, Vol. 35(3), pp. 24-33, 1998.
[26] Terzis, A.; Anandarajah, A.; Moore, K. & Wang, I. J.(2006). Slip Surface Localization in
Wireless Sensor Networks for Landslide Prediction, In Proceedings of Information Process-
ing in Sensor Networks (IPSN06), IEEE, 2006.
[27] Thampi, P. K.; Mathai, J.; Sankar, G. &Sidharthan, S. (1997). Landslides: Causes, Control
and Mitigation, Centre for Earth Science Studies, Trivandrum
[28] Tofani, V.; Dapporto, S.; Vannocci, P. & Casagli, N. (2005). Analysis of inltration, seep-
age processes and slope instability mechanisms during the November 2000 storm event
in Tuscany, Advances in Geosciences, Vol. 2, pp. 301-304, 2005.
[29] Tsaparas, I.; Rahardjo, H.; Toll, D. G. & Leong, E. C. (2002). Controlling parameters for
rainfall-induced landslides, Computers and Geotechnics, Vol. 29, pp. 1-27, 2002.
[30] Turner, A. K. (1996). Landslides: Investigation and Mitigation, United States Transporta-
tion Research Board, National Research Council, 1996.
[31] Van Asch, Th. W. J.; Buma, J. & Van Beek, L. P. H. (1999). A view on some hydrological
triggering systems in landslides, Geomorphology, Vol. 30 (1-2), pp. 25-32, October 1999.
[32] Wang et.al 2003 Wang, G. & Sassa, K. (2003). Pore-pressure generation and movement
of rainfall-induced landslide: Effect of grain size and ne-particle content, Engineering
Geology, Vol. 69, pp. 109-125, 2003.
[33] Wilson, R. C. & Wieczorek, G. F. (1995). Rainfall Thresholds for the Initiation of Debris
Flowa at La Honda, California, Environmental and Engineering Geoscience, Vol. 1 (1), pp.
11-27, 1995.
Urban Microclimate and Traffc Monitoring with Mobile Wireless Sensor Networks 71
Urban Microclimate and Traffc Monitoring with Mobile Wireless Sensor
Networks
0
Urban Microclimate and Trafc Monitoring
with Mobile Wireless Sensor Networks
University of Florence
Italy
1. Introduction
Climate is usually dened as the average of the atmospheric conditions over both an extended
period of time and a large region. Small scale patterns of climate resulting from the combined
inuence of topography, urban buildings structure, watercourses, vegetation, are known as
microclimates, which refers to a specic site or location. The microclimate scale may be at the
level of a settlement (urban or rural), neighborhood, cluster, street or buffer space in between
buildings or within the building itself. Specically, the dispersion and dilution of air pollu-
tants emitted by vehicles is one of the most investigated topics within urban meteorology, for
its fundamental impact on the environment affecting cities of all sizes. This issues concern the
average and peak values of various air pollutants as well as their temporal trends and spatial
variability. The accurate detection of these values might be advantageously exploited by pub-
lic authorities to better plan the public and private transportation by evaluating the impact on
people health, while controlling the greenhouse phenomenon.
As the unpredictable nature of a climate variations requires an incessant and ubiquitous sens-
ing, Wireless Sensor Networks (WSNs) represent a key technology for environmental monitor-
ing, hazard detection and, consequently, for decision making (Martinez et al., 2004). A WSN
is designed to be self-conguring and independent from any pre-existing infrastructure, be-
ing composed of a large number of elementary Sensor Nodes (SNs) that can be large-scale
deployed with small installation and maintenance costs.
Literature contains several examples of frameworks for evaluating the urban air quality with
WSNs, as it is reported in (Santini et al., 2008). In addition, in (Cordova-Lopez et al., 2007) it
is addressed the monitoring of exhaust and environmental pollution through the use of WSN
and GIS technology. As micro-climate monitoring usually requires deploying a large num-
ber of measurement tools, in (Shu-Chiung et al., 2009) it is adopted vehicular wireless sensor
networks (VWSNs) approach to reduce systemcomplexity, while achieving ne-grained mon-
itoring.
Another aspect strictly correlated with microclimate establishment is represented by the eco-
logic footprint of trafc congestion due to inefcient trafc management. As a consequence,
an increasing number of cities are going to develop intelligent transport system(ITS) as an ap-
proach to harmonize roads and vehicles in optimized and green paths. ITSs involves several
technologies as advanced informatics, data communications and transmissions, electronics
and computer control with the aim of real-time trafc reporting and alerting. Such a frame-
work allows remote operation management and self-conguration of trafc ows, as well as
5
specic information delivering to vehicles concerning, for instance, trafc congestion or the
presence of accidents (Pinart et al., 2009)
1
. Thus, the research on data acquisition scheme has
become a key point to enable effective ITSs.
At present, the acquisition of real-time trafc data is by means of installation and use of
wired monitoring equipment in most cities. However several concerns are associated with
this choice: rstly, with the continuous expansion of the city size and the increasing or the
trafc roads, the more the number of wired monitoring equipment increases, the more the
cost grows (scalability). Further, the installation of wired monitoring equipment does not have
the exibility, being difcult to (re)deploy. Finally, as urban trafc congestion has a certain
degree of space-time randomness, then it is inappropriate to install monitoring equipment in
xed locations. On the other hand, large-scale universal installation will cause larger waste.
To solve these problems, a promising approach is currently represented by WSNs applicable
to all types of urban environment (Laisheng et al., 2009), as they have no space constraints,
exible distribution, mobile convenience and quick reaction.
2. Background
2.1 ITS Communications Paradigms Overview
ITS services availability relies on the presence of an infrastructure usually comprising xed
devices interconnected by an underlying network, either wired or wireless. Data exchange
toward or among mobile terminals is inherently wireless, since information should directly
reach the drivers through PDAs or on-board transceivers; in evidence, IEEE 802 committee has
activated 11p Task Group to dene a Wi-Fi extension for Wireless Access in Vehicular Environ-
ments (WAVE) (Jiang & Delgrossi, 2008). Moreover, wireless connections are needed also for
data gathering, according to the Wireless Sensor Network (WSN) paradigm, comprising large
number of devices in charge of sensing and relay informations to the core network (Tubaishat
et al., 2009).
Within the above scenario, several communications paradigms are possible (Youse et al.,
2006). The case in which xed Access Points (APs) allow mobile nodes to join the network is
usually referred as infrastructure-to-vehicle (I2V) communications and can support advanced
applications such as web surng, multimedia streaming, remote vehicle diagnostics, real time
navigation, to name a few; on the other side, vehicle-to-vehicle (V2V) communications represent
the option in which mobile nodes can directly communicate to each other without any need
of infrastructure. Although V2V and I2V communications are both prominent research elds,
this paper is mainly focused on the latter, as it aims to efciently exchange short amounts of
data, collected and aggregated by an in-eld deployed WSN, to nomadic users, while keeping
the complexity of on-board circuitry as low as possible. It is worth noticing that a reliable
I2V scheme is extremely valuable even for V2V communications since, whenever a direct link
among vehicles is not available, message exchange can leverage on the infrastructure instead
of being successively relayed by few low-reliable mobile nodes, as addressed in (Gerla et al.,
2006).
The urban environment is usually composed of a large number of mobile terminals that are
likely to quickly change their reference AP, therefore facing frequent disconnection and re-
connection procedures, so that it may be not viable to deliver the total amount of required
data within a single session. Moreover, the urban channel is affected by long and short term
fading that introduces additional delays for data retransmissions (in the case of TCP trafc)
1
These goals could be summarized in two main elds as trafc ow forecast and trafc congestion control.
or sensibly lowers the data reliability (in the case of UDP trafc); these issues are addressed
in details in (Bychkovsky et al., 2006) and (Ott & Kutscher, 2004), providing a practical case
study involving IEEE 802.11b.
In general, content distribution through overlay networks is more efcient when compared
to traditional solutions using multiple unicasts. In order to achieve higher throughput and
failure resilience, parallel downloading from multiple overlay nodes represents a typical ap-
proach in most recent proposals (Wu & Li, 2007). However, the same content may be unnec-
essarily supplied by multiple nodes, rising the problem of the so called content reconcilia-
tion, which usually is a time and bandwidth consuming operation (Byers, Considine, Mitzen-
macher & Rost, 2002).
2.2 Rateless Codes
Rateless (or digital fountain) codes are a recently introduced class of forward error correction
codes with universally capacity-approaching behavior over erasure channels with arbitrary
erasure statistics. The rst practical rateless codes, called LT codes (Luby, 2002), are based
on a simple encoding process where the source message of length k information symbols is
encoded into a message containing a potentially innite number of encoded symbols. Each
encoded symbol is an independently and randomly created representation of the source mes-
sage and, as soon as the receiver correctly receives any set of k
encoded symbols, where k
is
only slightly larger than k, it is able to recover the source message.
LT encoding process is dened by the LT code degree distribution (d), which is a probability
mass function dened over the interval [1, k]. To create a new encoded symbol, the degree d is
randomly sampled from the degree distribution (d), d information symbols are uniformly
and randomly selected from the information message, and the encoded symbol is obtained
by XOR-ing d selected information symbols. Usually, information and encoded symbols are
equal-length binary data packets and the XOR operation is the bit-wise XOR. Encoded sym-
bols are transmitted over an erasure channel and decoded at the receiver using the iterative
Belief-Propagation (BP) algorithm. BP algorithm for erasure channel-decoding iteratively re-
covers information symbols from the degree one encoded packets, and cancels out the recov-
ered information symbols in all the remaining encoded packets (which may result in a newset
of degree one encoded packets). The iterations of this simple process can lead to the complete
message recovery, otherwise the receiver will have to wait for additional encoded packets in
order to complete the decoding process. The key problem of LT code design is the design of
the degree distribution (d) that will enable source message recovery from any slightly more
than k received encoded symbols using the iterative BP decoding algorithm. This problem is
solved asymptotically in (Luby, 2002), where it is shown that using so called robust soliton
degree distribution, it is possible to recover the source message from any k
encoded symbols,
where k
k asymptotically, with encoding/decoding complexity of the order O(k logk).

Rateless codes are usually applied in multicast scenarios, where the source message is entirely
available to the source node. However, in many practical systems such as wireless ad-hoc
networks, WSN or p2p networks, the message of interest might be distributed over many or
all network nodes. As shown recently - see (Vukobratovic et al., 2010) and references therein -
distributed rateless coding may be performed as efciently as its centralized counterparts, and
may provide a number of benets in distributed network scenarios for applications such as
data gathering, data persistence and distributed data storage.
specic information delivering to vehicles concerning, for instance, trafc congestion or the
presence of accidents (Pinart et al., 2009)
1
. Thus, the research on data acquisition scheme has
become a key point to enable effective ITSs.
At present, the acquisition of real-time trafc data is by means of installation and use of
wired monitoring equipment in most cities. However several concerns are associated with
this choice: rstly, with the continuous expansion of the city size and the increasing or the
trafc roads, the more the number of wired monitoring equipment increases, the more the
cost grows (scalability). Further, the installation of wired monitoring equipment does not have
the exibility, being difcult to (re)deploy. Finally, as urban trafc congestion has a certain
degree of space-time randomness, then it is inappropriate to install monitoring equipment in
xed locations. On the other hand, large-scale universal installation will cause larger waste.
To solve these problems, a promising approach is currently represented by WSNs applicable
to all types of urban environment (Laisheng et al., 2009), as they have no space constraints,
exible distribution, mobile convenience and quick reaction.
2. Background
2.1 ITS Communications Paradigms Overview
ITS services availability relies on the presence of an infrastructure usually comprising xed
devices interconnected by an underlying network, either wired or wireless. Data exchange
toward or among mobile terminals is inherently wireless, since information should directly
reach the drivers through PDAs or on-board transceivers; in evidence, IEEE 802 committee has
activated 11p Task Group to dene a Wi-Fi extension for Wireless Access in Vehicular Environ-
ments (WAVE) (Jiang & Delgrossi, 2008). Moreover, wireless connections are needed also for
data gathering, according to the Wireless Sensor Network (WSN) paradigm, comprising large
number of devices in charge of sensing and relay informations to the core network (Tubaishat
et al., 2009).
Within the above scenario, several communications paradigms are possible (Youse et al.,
2006). The case in which xed Access Points (APs) allow mobile nodes to join the network is
usually referred as infrastructure-to-vehicle (I2V) communications and can support advanced
applications such as web surng, multimedia streaming, remote vehicle diagnostics, real time
navigation, to name a few; on the other side, vehicle-to-vehicle (V2V) communications represent
the option in which mobile nodes can directly communicate to each other without any need
of infrastructure. Although V2V and I2V communications are both prominent research elds,
this paper is mainly focused on the latter, as it aims to efciently exchange short amounts of
data, collected and aggregated by an in-eld deployed WSN, to nomadic users, while keeping
the complexity of on-board circuitry as low as possible. It is worth noticing that a reliable
I2V scheme is extremely valuable even for V2V communications since, whenever a direct link
among vehicles is not available, message exchange can leverage on the infrastructure instead
of being successively relayed by few low-reliable mobile nodes, as addressed in (Gerla et al.,
2006).
The urban environment is usually composed of a large number of mobile terminals that are
likely to quickly change their reference AP, therefore facing frequent disconnection and re-
connection procedures, so that it may be not viable to deliver the total amount of required
data within a single session. Moreover, the urban channel is affected by long and short term
fading that introduces additional delays for data retransmissions (in the case of TCP trafc)
1
These goals could be summarized in two main elds as trafc ow forecast and trafc congestion control.
or sensibly lowers the data reliability (in the case of UDP trafc); these issues are addressed
in details in (Bychkovsky et al., 2006) and (Ott & Kutscher, 2004), providing a practical case
study involving IEEE 802.11b.
In general, content distribution through overlay networks is more efcient when compared
to traditional solutions using multiple unicasts. In order to achieve higher throughput and
failure resilience, parallel downloading from multiple overlay nodes represents a typical ap-
proach in most recent proposals (Wu & Li, 2007). However, the same content may be unnec-
essarily supplied by multiple nodes, rising the problem of the so called content reconcilia-
tion, which usually is a time and bandwidth consuming operation (Byers, Considine, Mitzen-
macher & Rost, 2002).
2.2 Rateless Codes
Rateless (or digital fountain) codes are a recently introduced class of forward error correction
codes with universally capacity-approaching behavior over erasure channels with arbitrary
erasure statistics. The rst practical rateless codes, called LT codes (Luby, 2002), are based
on a simple encoding process where the source message of length k information symbols is
encoded into a message containing a potentially innite number of encoded symbols. Each
encoded symbol is an independently and randomly created representation of the source mes-
sage and, as soon as the receiver correctly receives any set of k
encoded symbols, where k
is
only slightly larger than k, it is able to recover the source message.
LT encoding process is dened by the LT code degree distribution (d), which is a probability
mass function dened over the interval [1, k]. To create a new encoded symbol, the degree d is
randomly sampled from the degree distribution (d), d information symbols are uniformly
and randomly selected from the information message, and the encoded symbol is obtained
by XOR-ing d selected information symbols. Usually, information and encoded symbols are
equal-length binary data packets and the XOR operation is the bit-wise XOR. Encoded sym-
bols are transmitted over an erasure channel and decoded at the receiver using the iterative
Belief-Propagation (BP) algorithm. BP algorithm for erasure channel-decoding iteratively re-
covers information symbols from the degree one encoded packets, and cancels out the recov-
ered information symbols in all the remaining encoded packets (which may result in a newset
of degree one encoded packets). The iterations of this simple process can lead to the complete
message recovery, otherwise the receiver will have to wait for additional encoded packets in
order to complete the decoding process. The key problem of LT code design is the design of
the degree distribution (d) that will enable source message recovery from any slightly more
than k received encoded symbols using the iterative BP decoding algorithm. This problem is
solved asymptotically in (Luby, 2002), where it is shown that using so called robust soliton
degree distribution, it is possible to recover the source message from any k
encoded symbols,
where k
k asymptotically, with encoding/decoding complexity of the order O(k logk).

Rateless codes are usually applied in multicast scenarios, where the source message is entirely
available to the source node. However, in many practical systems such as wireless ad-hoc
networks, WSN or p2p networks, the message of interest might be distributed over many or
all network nodes. As shown recently - see (Vukobratovic et al., 2010) and references therein -
distributed rateless coding may be performed as efciently as its centralized counterparts, and
may provide a number of benets in distributed network scenarios for applications such as
data gathering, data persistence and distributed data storage.
3. Proposed Approach
3.1 System Requirements and Architecture
The reference system model is derived from a real world case study, inspired by the Tus-
cany Region project Metropolitan Mobility Agency Supporting Tools (SSAMM), devoted to
enhance the quality of urban transportation system introducing innovative paradigms. The
addressed urban communications scenario is modeled as a two-level network, as illustrated
in Fig. 1. In particular, the lower level is composed of a large number of Sensor Nodes (SNs),
positioned in such way that suitable and effective sampling of the road trafc is achieved
within the area of interest (Tanner, 1957). Whenever possible, SNs are deployed in correspon-
dence with road infrastructures such as posts, lamps and trafc lights, typically arranged in
a square grid fashion. Their purpose is to collect trafc ow information
2
and relay it to the
higher layer consisting of interconnected network of APs. In addition to this xed SNs, also
mobile sensors are introduced; it could be the case of a public vehicle equipped with gas an-
alyzers for the classical air pollutants NO, NO
2
, O
3
and NO, in order to record air pollution
and meteorological data within different urban zones. In the meanwhile vehicles can deliver
information regarding the interarrival time between adjacent APs, which is useful in estimat-
ing the congestion level. Finally, APs deliver gathered data toward Mobile Collector (MC)
usually referred to as data mule.
As the proposed application scenario is concerned with fast and efcient information retrieval,
these drawbacks could be faced by introducing an appropriate data dissemination algorithm,
enhancing the information persistence throughout the network without an excessive overload
in terms of total packet transmissions. To face MCinherent mobility, a distributed data gathering
protocol has been introduced (Stefanovic et al., 2011) to efciently collect all the sensed data
by visiting only an arbitrary subset of the SNs; this general requirement is extremely impor-
tant in urban scenarios, since path are usually space-time constrained. This has been achieved
resorting to a distributed implementation of rateless codes (Byers, Luby & Mitzenmacher,
2002), a particular class of erasure correction codes that rely on sparse binary coefcient data
combining, being suitable for the envised I2V data dissemination application, in which de-
vices exhibit low computational capabilities. Moreover, it has been introduced an adequate
data dissemination protocol which has been integrated with a MC data gathering scheme speci-
cally designed for a urban wide area monitoring WSN in order to allow reliable and accurate
sensing collection.
3.2 Communications Scheme
Each homogeneous subset of SNs is connected in a star-wise or tree topology to an AP; APs
encode and exchange packets received from SNs, then broadcast the information to MCs.
MCs usually join the network without need of an association with a specic AP by adopting a
passive operation mode and continuously collecting information regarding the surrounding en-
vironment broadcasted by APs
3
. Nevertheless, whenever MCs are involved in disseminating
their own information, they explicitly associate to the best AP and operate both in transmitting
and receiving modes. However, inter-vehicle communications are not hereafter considered.
Finally, we assume that MCs have on-board capabilities to process the downloaded data ac-
cording to suitable applications in order to interpret current trafc information. Specically,
2
Different types of information, e.g., average crossroad waiting time, presence of roadworks or accidents,
could be of interest.
3
This is adopted in order to lower the implementation complexity and cost of the mobile equipment,
minimize the downloading time avoiding access contentions and complex handover procedure.
the collected real-time data provide opportunity for on-board computer to perform optimal
route calculation, delay estimates, and present driver with visual map representation of criti-
cal locations where accidents, high pollutants concentrations or severe road congestions took
place. Although the push mode is possible, data processing is usually implemented in an au-
tomatic (i.e., periodic) manner in order to guarantee as much as possible real-time monitoring
of the trafc-load conditions. In particular, the latter mode of operation makes it possible for
MCs to be informed about dangerous situations (e.g., accidents) in a short time span, hence
allowing for increased safety of people and vehicles.
The communication between SNs (even mounted on board of a MC) and APs, and be-
tween MCs and APs, is assumed to be based on wireless technology. As recently shown,
IEEE 802.11b/g standards demonstrated signicant potential for vehicular applications (By-
chkovsky et al., 2006). Another candidate could be IEEE 802.11p (still in the draft stage),
whose one of the aims is to support efcient data exchange between roadside infrastructure
and vehicles.
Regarding the communications between APs, it is accomplished by leveraging on a pre-
existing infrastructure deployed in a urban area, i.e., connecting APs to wired Metropolitan
Area Network (MAN), such the one adopted by Florence Municipality, called FI-Net, com-
prised of a double ber optics ring with a 22.5 Gbps full-duplex capacity. Full-mesh wireless
interconnections are not considered as the adoption of an IEEE 802.11 unique radio interface
could pose several limitations in terms of coverage or, equivalently, scalability.
However, the communication scenario described above ts in the infrastructure mode of the
IEEE 802.11 standard in which several APs are interconnected using an external distributed
system, forming an Extended Service Set (ESS).
3.3 Distributed Data Gathering, Encoding and Dissemination
The systemapplication, residing in APs, periodically performs the following three procedures,
according to (Stefanovic et al., 2011): (i) data gathering from SNs, (ii) encoding and (iii) dis-
seminating encoded data to MCs. We refer to these three stages as upload, encoding and
download phase, respectively, and the period encompassing all of them as data refreshment
period. According to the IEEE 802.11 standard, the link time in every AP coverage zone is
divided in superframes (IEEE, 2007), and the data refreshment period in each zone is aligned
with superframe boundaries (see Fig. 2).
During the upload phase, every AP polls all SNs in its domain
4
and collects the most recent
measurements. As typically foreseen by most of the IEEE 802.11 standards (IEEE, 2007), su-
perframes are divided into the Contention-Free Period (CFP) and Contention-Based Period
(CBP), where the former is used to avoid MAC collisions and deliver prioritized information
to MCs. The polling phase can be accomplished within the CFP part of a typical frame. CFP
always starts after a beacon with a delivery trafc information map (DTIM) eld sent by AP to
STAs (that is SNs in our case). STAs associated with AP learn when the CFP should begin and
automatically set their NAV to MaxCFPDuration when the CFP is expected to begin. Then AP
individually polls each STA with a CF-poll message waiting for DATA and CF-ACK messages
from it, where messages are separated by a Short Interframe Space (SIFS) period. We assume
4
SNs can be either xed, i.e., infrastructured, or mobile, i.e. on board of MCs. In the latter case it is
necessary that the sojourn time of MC is comparable with superframe. This condition is easily satised
in the case of public transportation means close to a regular or temporary stop where an AP has been
placed.
3. Proposed Approach
3.1 System Requirements and Architecture
The reference system model is derived from a real world case study, inspired by the Tus-
cany Region project Metropolitan Mobility Agency Supporting Tools (SSAMM), devoted to
enhance the quality of urban transportation system introducing innovative paradigms. The
addressed urban communications scenario is modeled as a two-level network, as illustrated
in Fig. 1. In particular, the lower level is composed of a large number of Sensor Nodes (SNs),
positioned in such way that suitable and effective sampling of the road trafc is achieved
within the area of interest (Tanner, 1957). Whenever possible, SNs are deployed in correspon-
dence with road infrastructures such as posts, lamps and trafc lights, typically arranged in
a square grid fashion. Their purpose is to collect trafc ow information
2
and relay it to the
higher layer consisting of interconnected network of APs. In addition to this xed SNs, also
mobile sensors are introduced; it could be the case of a public vehicle equipped with gas an-
alyzers for the classical air pollutants NO, NO
2
, O
3
and NO, in order to record air pollution
and meteorological data within different urban zones. In the meanwhile vehicles can deliver
information regarding the interarrival time between adjacent APs, which is useful in estimat-
ing the congestion level. Finally, APs deliver gathered data toward Mobile Collector (MC)
usually referred to as data mule.
As the proposed application scenario is concerned with fast and efcient information retrieval,
these drawbacks could be faced by introducing an appropriate data dissemination algorithm,
enhancing the information persistence throughout the network without an excessive overload
in terms of total packet transmissions. To face MCinherent mobility, a distributed data gathering
protocol has been introduced (Stefanovic et al., 2011) to efciently collect all the sensed data
by visiting only an arbitrary subset of the SNs; this general requirement is extremely impor-
tant in urban scenarios, since path are usually space-time constrained. This has been achieved
resorting to a distributed implementation of rateless codes (Byers, Luby & Mitzenmacher,
2002), a particular class of erasure correction codes that rely on sparse binary coefcient data
combining, being suitable for the envised I2V data dissemination application, in which de-
vices exhibit low computational capabilities. Moreover, it has been introduced an adequate
data dissemination protocol which has been integrated with a MC data gathering scheme speci-
cally designed for a urban wide area monitoring WSN in order to allow reliable and accurate
sensing collection.
3.2 Communications Scheme
Each homogeneous subset of SNs is connected in a star-wise or tree topology to an AP; APs
encode and exchange packets received from SNs, then broadcast the information to MCs.
MCs usually join the network without need of an association with a specic AP by adopting a
passive operation mode and continuously collecting information regarding the surrounding en-
vironment broadcasted by APs
3
. Nevertheless, whenever MCs are involved in disseminating
their own information, they explicitly associate to the best AP and operate both in transmitting
and receiving modes. However, inter-vehicle communications are not hereafter considered.
Finally, we assume that MCs have on-board capabilities to process the downloaded data ac-
cording to suitable applications in order to interpret current trafc information. Specically,
2
Different types of information, e.g., average crossroad waiting time, presence of roadworks or accidents,
could be of interest.
3
This is adopted in order to lower the implementation complexity and cost of the mobile equipment,
minimize the downloading time avoiding access contentions and complex handover procedure.
the collected real-time data provide opportunity for on-board computer to perform optimal
route calculation, delay estimates, and present driver with visual map representation of criti-
cal locations where accidents, high pollutants concentrations or severe road congestions took
place. Although the push mode is possible, data processing is usually implemented in an au-
tomatic (i.e., periodic) manner in order to guarantee as much as possible real-time monitoring
of the trafc-load conditions. In particular, the latter mode of operation makes it possible for
MCs to be informed about dangerous situations (e.g., accidents) in a short time span, hence
allowing for increased safety of people and vehicles.
The communication between SNs (even mounted on board of a MC) and APs, and be-
tween MCs and APs, is assumed to be based on wireless technology. As recently shown,
IEEE 802.11b/g standards demonstrated signicant potential for vehicular applications (By-
chkovsky et al., 2006). Another candidate could be IEEE 802.11p (still in the draft stage),
whose one of the aims is to support efcient data exchange between roadside infrastructure
and vehicles.
Regarding the communications between APs, it is accomplished by leveraging on a pre-
existing infrastructure deployed in a urban area, i.e., connecting APs to wired Metropolitan
Area Network (MAN), such the one adopted by Florence Municipality, called FI-Net, com-
prised of a double ber optics ring with a 22.5 Gbps full-duplex capacity. Full-mesh wireless
interconnections are not considered as the adoption of an IEEE 802.11 unique radio interface
could pose several limitations in terms of coverage or, equivalently, scalability.
However, the communication scenario described above ts in the infrastructure mode of the
IEEE 802.11 standard in which several APs are interconnected using an external distributed
system, forming an Extended Service Set (ESS).
3.3 Distributed Data Gathering, Encoding and Dissemination
The systemapplication, residing in APs, periodically performs the following three procedures,
according to (Stefanovic et al., 2011): (i) data gathering from SNs, (ii) encoding and (iii) dis-
seminating encoded data to MCs. We refer to these three stages as upload, encoding and
download phase, respectively, and the period encompassing all of them as data refreshment
period. According to the IEEE 802.11 standard, the link time in every AP coverage zone is
divided in superframes (IEEE, 2007), and the data refreshment period in each zone is aligned
with superframe boundaries (see Fig. 2).
During the upload phase, every AP polls all SNs in its domain
4
and collects the most recent
measurements. As typically foreseen by most of the IEEE 802.11 standards (IEEE, 2007), su-
perframes are divided into the Contention-Free Period (CFP) and Contention-Based Period
(CBP), where the former is used to avoid MAC collisions and deliver prioritized information
to MCs. The polling phase can be accomplished within the CFP part of a typical frame. CFP
always starts after a beacon with a delivery trafc information map (DTIM) eld sent by AP to
STAs (that is SNs in our case). STAs associated with AP learn when the CFP should begin and
automatically set their NAV to MaxCFPDuration when the CFP is expected to begin. Then AP
individually polls each STA with a CF-poll message waiting for DATA and CF-ACK messages
from it, where messages are separated by a Short Interframe Space (SIFS) period. We assume
4
SNs can be either xed, i.e., infrastructured, or mobile, i.e. on board of MCs. In the latter case it is
necessary that the sojourn time of MC is comparable with superframe. This condition is easily satised
in the case of public transportation means close to a regular or temporary stop where an AP has been
placed.
Fig. 1. Reference network topology for general I2V applications.
that APs are globally synchronized
5
, so the actual upload takes place in the rst superframe
period following the start of the data refreshment period. Each SN uploads its measurements
within a single data packet of length L bits. Since SNs and APs form an infrastructured net-
work, it has been supposed that nodes have been previously deployed in line-of-sight (LoS)
fashion in order to optimize link quality; however possible packet losses are managed by
means Automatic Repeat reQuest (ARQ) scheme, so that from an application point of view
data delivering could be considered reliable. In particular, to match the constraint of polling
completion within the rst frame, a maximum ARQ retransmission persistency equal to N
A
attempts, where this parameter is selected in order to yield a negligible residual packet error
probability.
On reception, AP stores and uniquely indexes each received data packet, where the indexing
scheme is known to all APs. The total number of stored data packets in APs network per
data refreshment period is k, which is equal to the total number of SNs. These k data packets
represent a single data generation, upon which the rateless coding is performed. The differ-
entiation among data generations can be achieved using appropriate eld in packet header,
allowing MCs to maintain global time-references.
After the upload phase, the system application, distributed over all APs in the network, per-
forms distributed rateless encoding of collected data packets (i.e., rateless coding is used at the
5
It is easily provided by the preexisting MAN communications infrastructure which arrange a sort of
APs network.
Fig. 2. Data refreshment period of the proposed application.
application level). Each AP independently produces k
AP
encoded packets, where the actual
value of k
AP
is chosen such that it is sufcient for successful data recovery by all MCs with
high probability (w.h.p.). Specically,
k
AP

1 +
(max)
1 P
PL
k
where
(max)
is the reception overhead that allows for decoding w.h.p. and which depends
on the properties of the applied rateless codes, while P
PL
is the estimated link-layer packet
loss probability. For each encoded packet, AP draws degree d from the employed degree
distribution and then randomly selects d data packets from the pool of all k data packets
residing in the AP network during the current data refreshment period. In general case, most
of the selected data packets are likely to be stored in other APs, so the AP has to request them
using the known indexing scheme. After reception of the missing data packets, AP creates
encoded packets by simple bitwise XOR of associated data packets.
Finally, in the download phase, each AP disseminates encoded packets by simply broadcasting
them to MCs currently falling in its coverage area. This approach has been adopted in order
to minimize the complexity and the power consumption of MC receiver by always keeping
it in a receiving mode. Depending on the provided service, two kinds of dissemination are
possible: broadcast (BC) and geocast (GC) ones. The BC service covers simultaneous global
distribution of the most important data such as key trafc info to all the associated MCs,
using the broadcast MAC address. The GC service could be used for an additional (e.g., trafc
congestion and air quality) local delivering of uncoded packets containing information on the
actual hot spot (e.g., context aware information for navigation software enahnced services).
Due to its duration and broadcast nature, the BC service is capable of delivering signicantly
larger amount of data per superframe to its users as compared to the GCservice, which is why
the former is preferable for delay-sensitive real-time information delivery. The dissemination
starts in the rst superframe that follows the encoding phase, and lasts until the next data
collection phase (i.e., the next data refreshment period). For the purpose of BC service, the
natural choice is to use CFP part of the superframe, as it guarantees delivery of trafc-info
updates to all subscribed MCs within the service area.
While traveling within the service area, each MC performs a channel sensing at periodic inter-
vals (say ), dynamically selects the best carrier and transparently roams among adjacent APs,
Fig. 1. Reference network topology for general I2V applications.
that APs are globally synchronized
5
, so the actual upload takes place in the rst superframe
period following the start of the data refreshment period. Each SN uploads its measurements
within a single data packet of length L bits. Since SNs and APs form an infrastructured net-
work, it has been supposed that nodes have been previously deployed in line-of-sight (LoS)
fashion in order to optimize link quality; however possible packet losses are managed by
means Automatic Repeat reQuest (ARQ) scheme, so that from an application point of view
data delivering could be considered reliable. In particular, to match the constraint of polling
completion within the rst frame, a maximum ARQ retransmission persistency equal to N
A
attempts, where this parameter is selected in order to yield a negligible residual packet error
probability.
On reception, AP stores and uniquely indexes each received data packet, where the indexing
scheme is known to all APs. The total number of stored data packets in APs network per
data refreshment period is k, which is equal to the total number of SNs. These k data packets
represent a single data generation, upon which the rateless coding is performed. The differ-
entiation among data generations can be achieved using appropriate eld in packet header,
allowing MCs to maintain global time-references.
After the upload phase, the system application, distributed over all APs in the network, per-
forms distributed rateless encoding of collected data packets (i.e., rateless coding is used at the
5
It is easily provided by the preexisting MAN communications infrastructure which arrange a sort of
APs network.
Fig. 2. Data refreshment period of the proposed application.
application level). Each AP independently produces k
AP
encoded packets, where the actual
value of k
AP
is chosen such that it is sufcient for successful data recovery by all MCs with
high probability (w.h.p.). Specically,
k
AP

1 +
(max)
1 P
PL
k
where
(max)
is the reception overhead that allows for decoding w.h.p. and which depends
on the properties of the applied rateless codes, while P
PL
is the estimated link-layer packet
loss probability. For each encoded packet, AP draws degree d from the employed degree
distribution and then randomly selects d data packets from the pool of all k data packets
residing in the AP network during the current data refreshment period. In general case, most
of the selected data packets are likely to be stored in other APs, so the AP has to request them
using the known indexing scheme. After reception of the missing data packets, AP creates
encoded packets by simple bitwise XOR of associated data packets.
Finally, in the download phase, each AP disseminates encoded packets by simply broadcasting
them to MCs currently falling in its coverage area. This approach has been adopted in order
to minimize the complexity and the power consumption of MC receiver by always keeping
it in a receiving mode. Depending on the provided service, two kinds of dissemination are
possible: broadcast (BC) and geocast (GC) ones. The BC service covers simultaneous global
distribution of the most important data such as key trafc info to all the associated MCs,
using the broadcast MAC address. The GC service could be used for an additional (e.g., trafc
congestion and air quality) local delivering of uncoded packets containing information on the
actual hot spot (e.g., context aware information for navigation software enahnced services).
Due to its duration and broadcast nature, the BC service is capable of delivering signicantly
larger amount of data per superframe to its users as compared to the GCservice, which is why
the former is preferable for delay-sensitive real-time information delivery. The dissemination
starts in the rst superframe that follows the encoding phase, and lasts until the next data
collection phase (i.e., the next data refreshment period). For the purpose of BC service, the
natural choice is to use CFP part of the superframe, as it guarantees delivery of trafc-info
updates to all subscribed MCs within the service area.
While traveling within the service area, each MC performs a channel sensing at periodic inter-
vals (say ), dynamically selects the best carrier and transparently roams among adjacent APs,
while downloading encoded packets from APs, until it collects enough for sensor data recov-
ery using the iterative BP algorithm. The number of excessive encoded packets compared to
k sensor packets is measured by the reception overhead
6
; i.e., for successful recovery MC
needs in total k
= (1 +
) k encoded packets, where
usually is a small positive num-

ber. Since each encoded packet is an innovative representation of the original data, any subset
of k
= (1 +
) k taken from the set of all the encoded packets in the network allows for
restoration of the whole original data. This property of rateless codes makes them a perfect
candidate to be used at the application level for content delivery in vehicular networks, since
packet losses caused by the varying link characteristics are compensated simply by reception
of the new packets and there is no need for standard acknowledgment-retransmission mecha-
nisms which can not be supported by a semi-duplex architecture as the one adopted. In other
words, the usage of connection-oriented transport protocols like TCP can be avoided, as UDP-
like transport provides a satisfactory functionality. Moreover, the loosing of packets caused by
channel error or by the receiver deafness during the selection of a different AP does not impact
on BC scheme, as MC continues downloading data without any need for (de/re)association,
session management or content reconciliation.
4. Simulation Results
The simulation setup assumes that the urban area is covered by a regular hexagonal lattice,
where each non-overlapping hexagon represents the coverage area of a single AP and the
hexagon side length is equal to the AP transmission range. MCs move throughout the lattice
using the rectangular grid that models urban road-infrastructure, associating with the nearest
AP. The overlay hexagonal AP lattice is independent and arbitrarily aligned with the under-
lying rectangular road-grid. The MCs move according to the Manhattan mobility model (Bai
et al., 2003), a model commonly used for metropolitan trafc. In brief, Manhattan mobility
model assumes a regular grid consisting of horizontal and vertical (bidirectional) streets; at
each intersection, MC continues in the same direction with probability 0.5 or turns left/right
with probability 0.25 in each case. The MC speed is uniformly chosen from a predened inter-
val and changes on a time-slot basis (time-slot duration is a model parameter), with the speed
in the current time-slot being dependent on the value in the previous time-slot. Besides tem-
poral dependencies, Manhattan mobility model also includes spatial dependencies, since the
velocity of a MC depends on the velocity of other MCs moving in the same road segment and
in the same direction; as we are interested only in I2V communications from the perspective
of a single user (i.e., a single MC), spatial dependencies are omitted in our implementation.
The purpose of the simulations is to estimate the duration of the download phase, as the most
important and the lengthiest phase of the data refreshment period. In each simulation run,
while moving on the road grid, the MC starts receiving the encoded data from the AP in
whose coverage zone it is currently located. The reception of the encoded packets continues
until the MC collects enough to successfully decode all the original data. If during this pro-
cess, MC happens to move to another AP zone, it simply associates to a new local AP (i.e.,
handover takes place) and starts to receive its encoded packets. Also, if the AP has transmit-
ted all of its encoded packets to the MC, but it failed to decode the data (e.g., due to link-layer
packet losses), the MC suspends data reception until it enters the new AP coverage zone. The
6
This takes into account both the decoding overhead as well as the redundancy needed in the presence
of erasure channel.
simulation run ends when the decoding is nished and all the original data packets are re-
trieved. All the presented results are obtained by performing 1000 simulation runs for each
set of parameters.
System Parameter Value
AP transmission range 400 m
N
AP
(no. of APs in the system) 40
N
s
(no. of sensor nodes per AP) 50
k (no. of data packets) 2000
L (data packet length) 250 byte
k L (total amount of original data) 4 Mbit 0.48 Mbyte
c, (rateless code parameters) 0.03, 0.5
k
AP
(no. of encoded packets per AP) 3600
R (bit-rate) 6, 11, 12, 24 Mbit/s
T
SF
(superframe duration) 100 ms
HO
(handover time) 0.5 s
P
PL
(packet-loss probability) 0.3
road-segment length 150 m
velocity 4 - 17 m/s
acceleration 0.6 m/s
2
mobility model time-slot duration 2 s
Table 1. Simulation Parameters
Table 1 summarizes the values for the communication and mobility model parameters used
in simulations. The number of APs is chosen such that it provides a coverage area which is
approximately equal to a medium-sized city area. The data packet length is estimated in such
way that is sufcient to accommodate single sensor readings and additional headers (i.e., IEEE
802.11 MAC and LLC, network and transport layer). The values for bit-rate and superframe
duration are selected as suggested in (Bohm & Jonsson, 2008) and (Eriksson et al., 2008),
pessimistic assumption on packet-loss rate and estimate of the mean MC handover time were
taken from (Bychkovsky et al., 2006), the average road segment length (i.e., average distance
between two intersections) from (Peponis et al., 2007). The number of encoded packets per
AP, k
AP
is chosen such that a MC could decode all original data with probability of 0.99,
when downloading from a single AP and considering employed rateless code properties and
assumed link-layer packet-loss rate. In other words, k
AP
>
1 +
(max)
k L/(1 P
PL
).
Fig. 3 presents the probability P
SD
that the MC successfully decodes the sensor data as a
function of time, for the BC service and T
(BC)
SF
= 0.1 T
SF
. The value for T
SF
is selected such
that it leaves enough room for the GC service and other usual best-effort services. As it can
be observed from the gure, for higher bit-rates (i.e., R > 6 Mbit/s), the MCN is able to
while downloading encoded packets from APs, until it collects enough for sensor data recov-
ery using the iterative BP algorithm. The number of excessive encoded packets compared to
k sensor packets is measured by the reception overhead
6
; i.e., for successful recovery MC
needs in total k
= (1 +
) k encoded packets, where
usually is a small positive num-

ber. Since each encoded packet is an innovative representation of the original data, any subset
of k
= (1 +
) k taken from the set of all the encoded packets in the network allows for
restoration of the whole original data. This property of rateless codes makes them a perfect
candidate to be used at the application level for content delivery in vehicular networks, since
packet losses caused by the varying link characteristics are compensated simply by reception
of the new packets and there is no need for standard acknowledgment-retransmission mecha-
nisms which can not be supported by a semi-duplex architecture as the one adopted. In other
words, the usage of connection-oriented transport protocols like TCP can be avoided, as UDP-
like transport provides a satisfactory functionality. Moreover, the loosing of packets caused by
channel error or by the receiver deafness during the selection of a different AP does not impact
on BC scheme, as MC continues downloading data without any need for (de/re)association,
session management or content reconciliation.
4. Simulation Results
The simulation setup assumes that the urban area is covered by a regular hexagonal lattice,
where each non-overlapping hexagon represents the coverage area of a single AP and the
hexagon side length is equal to the AP transmission range. MCs move throughout the lattice
using the rectangular grid that models urban road-infrastructure, associating with the nearest
AP. The overlay hexagonal AP lattice is independent and arbitrarily aligned with the under-
lying rectangular road-grid. The MCs move according to the Manhattan mobility model (Bai
et al., 2003), a model commonly used for metropolitan trafc. In brief, Manhattan mobility
model assumes a regular grid consisting of horizontal and vertical (bidirectional) streets; at
each intersection, MC continues in the same direction with probability 0.5 or turns left/right
with probability 0.25 in each case. The MC speed is uniformly chosen from a predened inter-
val and changes on a time-slot basis (time-slot duration is a model parameter), with the speed
in the current time-slot being dependent on the value in the previous time-slot. Besides tem-
poral dependencies, Manhattan mobility model also includes spatial dependencies, since the
velocity of a MC depends on the velocity of other MCs moving in the same road segment and
in the same direction; as we are interested only in I2V communications from the perspective
of a single user (i.e., a single MC), spatial dependencies are omitted in our implementation.
The purpose of the simulations is to estimate the duration of the download phase, as the most
important and the lengthiest phase of the data refreshment period. In each simulation run,
while moving on the road grid, the MC starts receiving the encoded data from the AP in
whose coverage zone it is currently located. The reception of the encoded packets continues
until the MC collects enough to successfully decode all the original data. If during this pro-
cess, MC happens to move to another AP zone, it simply associates to a new local AP (i.e.,
handover takes place) and starts to receive its encoded packets. Also, if the AP has transmit-
ted all of its encoded packets to the MC, but it failed to decode the data (e.g., due to link-layer
packet losses), the MC suspends data reception until it enters the new AP coverage zone. The
6
This takes into account both the decoding overhead as well as the redundancy needed in the presence
of erasure channel.
simulation run ends when the decoding is nished and all the original data packets are re-
trieved. All the presented results are obtained by performing 1000 simulation runs for each
set of parameters.
System Parameter Value
AP transmission range 400 m
N
AP
(no. of APs in the system) 40
N
s
(no. of sensor nodes per AP) 50
k (no. of data packets) 2000
L (data packet length) 250 byte
k L (total amount of original data) 4 Mbit 0.48 Mbyte
c, (rateless code parameters) 0.03, 0.5
k
AP
(no. of encoded packets per AP) 3600
R (bit-rate) 6, 11, 12, 24 Mbit/s
T
SF
(superframe duration) 100 ms
HO
(handover time) 0.5 s
P
PL
(packet-loss probability) 0.3
road-segment length 150 m
velocity 4 - 17 m/s
acceleration 0.6 m/s
2
mobility model time-slot duration 2 s
Table 1. Simulation Parameters
Table 1 summarizes the values for the communication and mobility model parameters used
in simulations. The number of APs is chosen such that it provides a coverage area which is
approximately equal to a medium-sized city area. The data packet length is estimated in such
way that is sufcient to accommodate single sensor readings and additional headers (i.e., IEEE
802.11 MAC and LLC, network and transport layer). The values for bit-rate and superframe
duration are selected as suggested in (Bohm & Jonsson, 2008) and (Eriksson et al., 2008),
pessimistic assumption on packet-loss rate and estimate of the mean MC handover time were
taken from (Bychkovsky et al., 2006), the average road segment length (i.e., average distance
between two intersections) from (Peponis et al., 2007). The number of encoded packets per
AP, k
AP
is chosen such that a MC could decode all original data with probability of 0.99,
when downloading from a single AP and considering employed rateless code properties and
assumed link-layer packet-loss rate. In other words, k
AP
>
1 +
(max)
k L/(1 P
PL
).
Fig. 3 presents the probability P
SD
that the MC successfully decodes the sensor data as a
function of time, for the BC service and T
(BC)
SF
= 0.1 T
SF
. The value for T
SF
is selected such
that it leaves enough room for the GC service and other usual best-effort services. As it can
be observed from the gure, for higher bit-rates (i.e., R > 6 Mbit/s), the MCN is able to
successfully decode w.h.p. all the data in the time span of several seconds. The positive effect
of rateless coding is inherent in the fact that, even in the worst case, the data refreshment
period is below15s, a value that still allows for real-time information updates and which could
be decreased further by assigning a larger superframe fraction to the BC service. As opposed
to rateless encoded data delivery, the uncoded data delivery would result in retransmission
feedback implosion for BC service, overwhelming the sender (i.e., AP) with unwanted trafc.
The probability of successful decoding for GC service is presented in Fig. 4, where the fraction
of the superframe assigned to a single user is assumed to be T
(GC)
SF
/N
MN
= 0.01 T
SF
; the val-
ues for T
(UC)
SF
and N
MN
are taken from the realistic analysis given in (Bohm & Jonsson, 2008).
Fig. 4 demonstrates that for the standard GC service, the data refreshment period is of the
order of minutes rather than seconds, which limits its usage for the applications that tolerate
larger update periods. However, this period would be signicantly longer if rateless coding
was not used, since the link layer retransmissions would make the data delivery process con-
siderably less efcient. Finally, it can be observed that, for the GC service, the differences in
transmission bit-rate have a signicant impact on the download delay, which makes higher
bit-rates desirable.
Fig. 5 presents the duration of the time interval T
0.99
for which a MN, using the GC service,
decodes all the original data with probability P
SD
= 0.99, as a function of the number of
users N
MN
and for the xed T
(GC)
SF
= 0.8 T
SF
. The gure shows a linear increase in T
0.99
as
the rate decreases or N
MN
increases, verifying that the content reconciliation phase is indeed
unnecessary, since the change of the AP does not introduce additional delays apart from the
handover time. In other words, after a handover, MC seamlessly advances both with the
receiving and decoding processes.
Finally, Fig. 6 shows the cumulative distribution function F
T
of the number of transmitted
packets using the GC service from an AP to any MC within a single AP domain. The
T
(GC)
SF
/N
MN
ratio of the UC service is set to 0.005 T
SF
or 0.01 T
SF
. As it can be observed,
the number of transmitted packets to a MC reaches the threshold value k
AP
equal to
3600 for the selected parameter values (Table I), in all cases but for R = 6 Mbit/s and
T
(GC)
SF
/N
MN
= 0.005 T
SF
. This means that number of encoded packets per AP (i.e., k
AP
) is
properly dimensioned to allow a single user to collect enough of encoded packets to decode
all the data w.h.p. while moving through a single AP coverage zone.
To summarize the benets provided by the proposed I2V data dissemination based on the
rateless codes over traditional methods, it is worth noticing rst of all that, by their design,
rateless codes are tuned to the changing wireless link conditions and have a close-to-the-
minimal reception overhead. Furthermore, each rateless coded packet is an equally important
representation of the original data, which makes lengthy TCP-like reliability mechanisms un-
necessary. These factors inuence the time allocations within the superframe, allowing larger
number of mobile nodes to be serviced during designated service-time portion of the super-
frame, or alternatively, service-time portion shortening, providing larger time allocations for
best-effort trafc. Finally, while roaming through the network, mobile users can simply con-
tinue with data download from the new local AP after a handover, avoiding the redundant
content reconciliation phase.
0 2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time [s]
P
S
D
6 Mbit/s
11 Mbit/s
12 Mbit/s
24 Mbit/s
Fig. 3. Probability of successful decoding P
SD
for BC service, T
(BC)
SF
= 0.1 T
SF
.
20 40 60 80 100 120 140
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time [s]
P
S
D
6 Mbit/s
11 Mbit/s
12 Mbit/s
24 Mbit/s
SD
for GC service, T
(GC)
SF
/N
MN
= 0.01 T
SF
.
successfully decode w.h.p. all the data in the time span of several seconds. The positive effect
of rateless coding is inherent in the fact that, even in the worst case, the data refreshment
period is below15s, a value that still allows for real-time information updates and which could
be decreased further by assigning a larger superframe fraction to the BC service. As opposed
to rateless encoded data delivery, the uncoded data delivery would result in retransmission
feedback implosion for BC service, overwhelming the sender (i.e., AP) with unwanted trafc.
The probability of successful decoding for GC service is presented in Fig. 4, where the fraction
of the superframe assigned to a single user is assumed to be T
(GC)
SF
/N
MN
= 0.01 T
SF
; the val-
ues for T
(UC)
SF
and N
MN
are taken from the realistic analysis given in (Bohm & Jonsson, 2008).
Fig. 4 demonstrates that for the standard GC service, the data refreshment period is of the
order of minutes rather than seconds, which limits its usage for the applications that tolerate
larger update periods. However, this period would be signicantly longer if rateless coding
was not used, since the link layer retransmissions would make the data delivery process con-
siderably less efcient. Finally, it can be observed that, for the GC service, the differences in
transmission bit-rate have a signicant impact on the download delay, which makes higher
bit-rates desirable.
Fig. 5 presents the duration of the time interval T
0.99
for which a MN, using the GC service,
decodes all the original data with probability P
SD
= 0.99, as a function of the number of
users N
MN
and for the xed T
(GC)
SF
= 0.8 T
SF
. The gure shows a linear increase in T
0.99
as
the rate decreases or N
MN
increases, verifying that the content reconciliation phase is indeed
unnecessary, since the change of the AP does not introduce additional delays apart from the
handover time. In other words, after a handover, MC seamlessly advances both with the
receiving and decoding processes.
Finally, Fig. 6 shows the cumulative distribution function F
T
of the number of transmitted
packets using the GC service from an AP to any MC within a single AP domain. The
T
(GC)
SF
/N
MN
ratio of the UC service is set to 0.005 T
SF
or 0.01 T
SF
. As it can be observed,
the number of transmitted packets to a MC reaches the threshold value k
AP
equal to
3600 for the selected parameter values (Table I), in all cases but for R = 6 Mbit/s and
T
(GC)
SF
/N
MN
= 0.005 T
SF
. This means that number of encoded packets per AP (i.e., k
AP
) is
properly dimensioned to allow a single user to collect enough of encoded packets to decode
all the data w.h.p. while moving through a single AP coverage zone.
To summarize the benets provided by the proposed I2V data dissemination based on the
rateless codes over traditional methods, it is worth noticing rst of all that, by their design,
rateless codes are tuned to the changing wireless link conditions and have a close-to-the-
minimal reception overhead. Furthermore, each rateless coded packet is an equally important
representation of the original data, which makes lengthy TCP-like reliability mechanisms un-
necessary. These factors inuence the time allocations within the superframe, allowing larger
number of mobile nodes to be serviced during designated service-time portion of the super-
frame, or alternatively, service-time portion shortening, providing larger time allocations for
best-effort trafc. Finally, while roaming through the network, mobile users can simply con-
tinue with data download from the new local AP after a handover, avoiding the redundant
content reconciliation phase.
0 2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time [s]
P
S
D
6 Mbit/s
11 Mbit/s
12 Mbit/s
24 Mbit/s
SD
for BC service, T
(BC)
SF
= 0.1 T
SF
.
20 40 60 80 100 120 140
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time [s]
P
S
D
6 Mbit/s
11 Mbit/s
12 Mbit/s
24 Mbit/s
SD
for GC service, T
(GC)
SF
/N
MN
= 0.01 T
SF
.
20 40 60 80 100 120 140 160
0
50
100
150
200
250
N
MN
T
0
.
9
9

[
s
]
6 Mbit/s
11 Mbit/s
12 Mbit/s
24 Mbit/s
Fig. 5. Duration of time-interval T
0.99
for which MC decodes all data with P
SD
= 0.99 for GC
service.
0 500 1000 1500 2000 2500 3000 3600
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
no. of transmitted packets
F
T
6 Mbit/s, T
SF
(UC)
/N
MN
=0.005 T
SF
11 Mbit/s, T
SF
(UC)
/N
MN
=0.005 T
SF
6 Mbit/s, T
SF
(UC)
/N
MN
=0.01 T
SF
11 Mbit/s, T
SF
(UC)
/N
MN
=0.01 T
SF
Fig. 6. Cumulative distribution function F
T
of number of transmitted packets to MN in single
AP cell for GC service.
Acknowledgment
This work was supported in part by by the Italian National Project Wireless multiplatfOrm
mimo active access netwoRks for QoS-demanding muLtimedia Delivery (WORLD), under
grant number 2007R989S, as well as by the Tuscany Region projects Metropolitan Mobility
Agency Supporting Tools (SSAMM) and Microparticulate Monitoring via Wireless Sensor
Networks (MAPPS) The authors would like also to thank the partners of EU FP7-REGPOT-
2007-3 - AgroSense project for their fruitful discussion and comments.
5. References
Bai, F., Sadagopan, N. & Helmy, A. (2003). IMPORTANT: A framework to systematically an-
alyze the Impact of Mobility on Performance of RouTing protocols for Adhoc NeT-
works, Proc. of IEEE INFOCOM 2003, San Francisco, CA, USA.
Bohm, A. & Jonsson, M. (2008). Supporting real-time data trafc in safety-critical vehicle-to-
infrastructure communication, Proc. of IEEE LCN 2008, Montreal, QC, Canada.
Bychkovsky, V., Hull, B., Miu, A., Balakrishnan, H. & Madden, S. (2006). A Measurement
Study of Vehicular Internet Access Using in Situ Wi-Fi Networks, Proc. of ACM Mobi-
Com 2006, Los Angeles, CA, USA.
Byers, J., Considine, J., Mitzenmacher, M. &Rost, S. (2002). Informed Content Delivery Across
Adaptive Overlay Networks, Proc. of ACM SIGCOMM 2002, Pittsburg, PA, USA.
Byers, J., Luby, M. &Mitzenmacher, M. (2002). ADigital Fountain Approach to Asynchronous
Reliable Multicast, IEEE Journal on Selected Areas in Communications 20(8): 15281540.
Cordova-Lopez, L. E., Mason, A., Cullen, J. D., Shaw, A. & Al-Shammaa, A. (2007). Online
Vehicle and Atmospheric Pollution Monitoring using GIS and Wireless Sensor Net-
works, Proc. of ACM Intl Conference on Embedded Networked Sensor Systems (SenSys),
pp. 87
U101.
Eriksson, J., Balakrishnan, H. & Madden, S. (2008). Cabernet: Vehicular Content Delivery
Using WiFi, Proc. of ACM MobiCom 2008, San Francisco, CA, USA.
Gerla, M., Zhou, B., Lee, Y. Z., Soldo, F., Lee, U. & Mara, G. (2006). Vehicular Grid Commu-
nications: The Role of the Internet Infrastructure, Proc. of WICON 06, Boston, MA,
USA.
IEEE (2007). Ieee 802.11-2007 wireless lan medium access control and physical layers speci-
cations.
Jiang, D. &Delgrossi, L. (2008). IEEE 802.11p: Towards an International Standard for Wireless
Access in Vehicular Environments, Proc. of IEEE VTC2008-Spring, Singapore.
Laisheng, X., Xiaohong, P., Zhengxia, W., Bing, X. & Pengzhi, H. (2009). Research on trafc
monitoring network and its trafc ow forecast and congestion control model based
on wireless sensor networks, Measuring Technology and Mechatronics Automation, 2009.
ICMTMA 09. International Conference on, Vol. 1, pp. 142 147.
Luby, M. (2002). LT Codes, Proc. of IEEE FOCS 2002, Vancouver, BC, Canada.
Martinez, K., Hart, J. & Ong, R. (2004). Environmental Sensor Networks, IEEE Computer Jour-
nal 37: 5056.
Ott, J. & Kutscher, D. (2004). Drive-thru Internet: IEEE 802.11b for Automobile Users, Proc. of
IEEE Infocom 2004, Hong Kong.
Peponis, J., Allen, D., Haynie, D., Scoppa, M. & Zhang, Z. (2007). MEASURING THE CON-
FIGURATION OF STREET NETWORKS: the Spatial proles of 118 urban areas in
20 40 60 80 100 120 140 160
0
50
100
150
200
250
N
MN
T
0
.
9
9

[
s
]
6 Mbit/s
11 Mbit/s
12 Mbit/s
24 Mbit/s
Fig. 5. Duration of time-interval T
0.99
for which MC decodes all data with P
SD
= 0.99 for GC
service.
0 500 1000 1500 2000 2500 3000 3600
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
no. of transmitted packets
F
T
6 Mbit/s, T
SF
(UC)
/N
MN
=0.005 T
SF
11 Mbit/s, T
SF
(UC)
/N
MN
=0.005 T
SF
6 Mbit/s, T
SF
(UC)
/N
MN
=0.01 T
SF
11 Mbit/s, T
SF
(UC)
/N
MN
=0.01 T
SF
Fig. 6. Cumulative distribution function F
T
of number of transmitted packets to MN in single
AP cell for GC service.
Acknowledgment
This work was supported in part by by the Italian National Project Wireless multiplatfOrm
mimo active access netwoRks for QoS-demanding muLtimedia Delivery (WORLD), under
grant number 2007R989S, as well as by the Tuscany Region projects Metropolitan Mobility
Agency Supporting Tools (SSAMM) and Microparticulate Monitoring via Wireless Sensor
Networks (MAPPS) The authors would like also to thank the partners of EU FP7-REGPOT-
2007-3 - AgroSense project for their fruitful discussion and comments.
5. References
Bai, F., Sadagopan, N. & Helmy, A. (2003). IMPORTANT: A framework to systematically an-
alyze the Impact of Mobility on Performance of RouTing protocols for Adhoc NeT-
works, Proc. of IEEE INFOCOM 2003, San Francisco, CA, USA.
Bohm, A. & Jonsson, M. (2008). Supporting real-time data trafc in safety-critical vehicle-to-
infrastructure communication, Proc. of IEEE LCN 2008, Montreal, QC, Canada.
Bychkovsky, V., Hull, B., Miu, A., Balakrishnan, H. & Madden, S. (2006). A Measurement
Study of Vehicular Internet Access Using in Situ Wi-Fi Networks, Proc. of ACM Mobi-
Com 2006, Los Angeles, CA, USA.
Byers, J., Considine, J., Mitzenmacher, M. &Rost, S. (2002). Informed Content Delivery Across
Adaptive Overlay Networks, Proc. of ACM SIGCOMM 2002, Pittsburg, PA, USA.
Byers, J., Luby, M. &Mitzenmacher, M. (2002). ADigital Fountain Approach to Asynchronous
Reliable Multicast, IEEE Journal on Selected Areas in Communications 20(8): 15281540.
Cordova-Lopez, L. E., Mason, A., Cullen, J. D., Shaw, A. & Al-Shammaa, A. (2007). Online
Vehicle and Atmospheric Pollution Monitoring using GIS and Wireless Sensor Net-
works, Proc. of ACM Intl Conference on Embedded Networked Sensor Systems (SenSys),
pp. 87
U101.
Eriksson, J., Balakrishnan, H. & Madden, S. (2008). Cabernet: Vehicular Content Delivery
Using WiFi, Proc. of ACM MobiCom 2008, San Francisco, CA, USA.
Gerla, M., Zhou, B., Lee, Y. Z., Soldo, F., Lee, U. & Mara, G. (2006). Vehicular Grid Commu-
nications: The Role of the Internet Infrastructure, Proc. of WICON 06, Boston, MA,
USA.
IEEE (2007). Ieee 802.11-2007 wireless lan medium access control and physical layers speci-
cations.
Jiang, D. &Delgrossi, L. (2008). IEEE 802.11p: Towards an International Standard for Wireless
Access in Vehicular Environments, Proc. of IEEE VTC2008-Spring, Singapore.
Laisheng, X., Xiaohong, P., Zhengxia, W., Bing, X. & Pengzhi, H. (2009). Research on trafc
monitoring network and its trafc ow forecast and congestion control model based
on wireless sensor networks, Measuring Technology and Mechatronics Automation, 2009.
ICMTMA 09. International Conference on, Vol. 1, pp. 142 147.
Luby, M. (2002). LT Codes, Proc. of IEEE FOCS 2002, Vancouver, BC, Canada.
Martinez, K., Hart, J. & Ong, R. (2004). Environmental Sensor Networks, IEEE Computer Jour-
nal 37: 5056.
Ott, J. & Kutscher, D. (2004). Drive-thru Internet: IEEE 802.11b for Automobile Users, Proc. of
IEEE Infocom 2004, Hong Kong.
Peponis, J., Allen, D., Haynie, D., Scoppa, M. & Zhang, Z. (2007). MEASURING THE CON-
FIGURATION OF STREET NETWORKS: the Spatial proles of 118 urban areas in
the 12 most populated metropilitan regions in the US, Proc. of 6th International Space
Syntax Symposium, Istanbul, Turkey.
Pinart, C., Calvo, J. C., Nicholson, L. & Villaverde, J. A. (2009). ECall-compliant early crash
notication service for portable and nomadic devices, Proc. of IEEE VTC2009-Spring,
Barcelona, Spain.
Santini, S., Ostermaier, B. & Vitaletti, A. (2008). First Experiences using Wireless Sensor Net-
work for Noise Pollution Monitoring, Proc. of 3rd ACM Workshop on Real-World Wire-
less Sensor Networks (REALWSN08), Glasgow, United Kingdom.
Shu-Chiung, H., You-Chiun, W., Chiuan-Yu, H. & Yu-Chee, T. (2009). A Vehicular Wireless
Sensor Network for CO2 Monitoring, Proc. of IEEE Sensors, pp. 1498 1501.
Stefanovic, C., Crnojevic, V., Vukobratovic, D., Niccolai, L., Chiti, F. & Fantacci, R. (2011).
Urban Infrastructure-to-Vehicle Trafc Data Dissemination Using Rateless Codes, to
appear in IEEE Journal on Selected Areas on Communications special issue on Vehicular
Communications and Networks .
Tanner, J. C. (1957). The Sampling of Road Trafc, Journal of the Royal Statistical Society. Series
C (Applied Statistics) 6(3): 161170.
Tubaishat, M., Zhuang, P., Qi, Q. & Shang, Y. (2009). Wireless sensor networks in intelligent
transportation systems, Wireless Communications and Mobile Computing 2009 9(3): 287
302.
Vukobratovic, D., Stefanovic, C., Crnojevic, V., Chiti, F. & Fantacci, R. (2010). Rateless Packet
Approach for Data Gathering in Wireless Sensor Networks, to appear in IEEE Journal
on Selected Areas in Communications special issue on Simple Wireless Sensor Networking
Solutions) 28(7).
Wu, C. & Li, B. (2007). Outburst: Efcient Overlay Content Distribution with Rateless Codes,
Proc. of IFIP Networking 2007, Atlanta, GA, USA.
Youse, S., Mousavi, M. S. & Fathy, M. (2006). Vehicular Ad Hoc Networks (VANETs): Chal-
lenges and Perspectives, Proc. of ITST 2006, Chengdu, China.
Improving Greenhouses Automation and Data Acquisition
with Mobile Robot Controlled system via Wireless Sensor Network 85
Improving Greenhouses Automation and Data Acquisition with Mobile
Robot Controlled system via Wireless Sensor Network
Istvn Matijevics and Simon Jnos
X

Improving Greenhouses Automation and
Data Acquisition with Mobile Robot Controlled
system via Wireless Sensor Network

Istvn Matijevics* and Simon Jnos**
*University of Szeged, Institute of Informatics
Hungary
**Subotica Tech, Department of Informatics
Serbia

1. Introduction
The function of a greenhouse is to create the optimal growing conditions for the full lifecycle
of the plants. Using autonomous measuring stations helps to monitor all the necessary
parameters for creating the optimal environment in the greenhouse. The robot equipped
with sensors is capable of driving to the end and back along crop rows inside the
greenhouse. This chaper deals with the implementation of mobile measuring station in
greenhouse environment. It introduces a wireless sensor network that was used for the
purpose of measuring and controlling the greenhouse application. Continuous
advancements in wireless technology and miniaturization have made the deployment of
sensor networks to monitor various aspects of the environment increasingly flexible.
Climate monitoring is vitally important to the operation in greenhouses and the quality of
the collected information has a great influence on the precision and accuracy of control
results. Currently, the agro-alimentary market field incorporates diverse data acquisition
techniques. Normally, the type of acquisition system is chosen to be optimal for the control
algorithm to be used. For traditional climate monitoring and control systems, all sensors are
distributed through the greenhouse and connected to the device performing the control
tasks. These equipments use time-based data sampling techniques as a consequence of using
time-based controllers. Typical applications of WSNs include monitoring, tracking, and
controlling. Some of the specific applications are habitat monitoring, object tracking, etc. In a
typical application, a WSN is scattered in a region where it is meant to collect data through
its sensor node. The WSN-based controller has allowed a considerable decrease in the
number of changes in the control action and made possible a study of the compromise
between quantity of transmission and control performance. In modern greenhouses, several
measurement points are required to trace down the local climate parameters in different
parts of the big greenhouse to make the greenhouse automation system work properly.
Cabling would make the measurement system expensive and vulnerable. Moreover, the
cabled measurement points are difficult to relocate once they are installed. Thus, a wireless
6

sensor network (WSN) consisting of small-size wireless sensor nodes equipped with radio
and one or several sensors, is an attractive and cost-efficient option to build the required
measurement system. In this work, we developed a wireless sensor node for greenhouse
monitoring by integrating a sensor platform provided SunSPOT by Sun Microsystems with
few sensors capable to measure four climate variables. Continuous advancements in
wireless technology and miniaturization have made the deployment of sensor networks to
monitor various aspects of the environment increasingly flexible.

2. Mobile platform
Mobile robotics is a young field of research. Its roots include many engineering and science
disciplines, from mechanical, electrical and electronics engineering to computer, cognitive
and social sciences. The Board Of Education is a complete, low-cost development platform
equipped with the needed sensors for humidity, temperature, light, etc. As shown in
Figure 1, the Boe-Bot is a great tool with which to get started with robotics.

Fig. 1. Assembled Boe-Bot

The SunSPOT WSN module makes it possible for the Boe-Bot robots BASIC Stamp 2
microcontroller brain to communicate wirelessly with a web based user interface running on
a nearby PC. The BASIC Stamp microcontroller runs a small PBASIC program that controls
the Boe-Bot robots servos and optionally monitors sensors while it communicates wirelessly
with the web server.

3. Control scheme for mobile robots
A mobile robot needs locomotion mechanisms that enable it to move throughout its known
or unknown environment. But there are a large variety of possible ways to move, and so the
selection of a robots approach to locomotion is an important aspect of mobile robot design.

Figure 2, presents the control scheme for mobile robot systems. In the laboratory, there are
research robots that can walk, jump, run, slide, skate, swim, fly, and, of course, roll. Any of
these activities has its own control algorithm (Gy. Mester, 2009).

Fig. 2. Reference control scheme for mobile robot systems

Locomotion is the complement of manipulation. In manipulation, the robot arm is fixed but
moves objects in the workspace by imparting force to them. In locomotion, the environment
is fixed and the robot moves by imparting force to the environment. In both cases, the
scientific basis is the study of actuators that generate interaction forces, and mechanisms
that implement desired kinematical and dynamic properties. The wheel has been by far the
most popular mechanism in mobile robotics and in man-made vehicles in general. It can
achieve very good efficiencies, and does so with a relatively simple mechanical
implementation. On Figure 3, the kinematics of the mobile robot is depicted. In addition,
balance is not usually a research problem in wheeled robot designs, because wheeled robots
are almost always designed so that all wheels are in ground contact at all times (Gy. Mester,
2009).

sensor network (WSN) consisting of small-size wireless sensor nodes equipped with radio
and one or several sensors, is an attractive and cost-efficient option to build the required
measurement system. In this work, we developed a wireless sensor node for greenhouse
monitoring by integrating a sensor platform provided SunSPOT by Sun Microsystems with
few sensors capable to measure four climate variables. Continuous advancements in
wireless technology and miniaturization have made the deployment of sensor networks to
monitor various aspects of the environment increasingly flexible.

2. Mobile platform
Mobile robotics is a young field of research. Its roots include many engineering and science
disciplines, from mechanical, electrical and electronics engineering to computer, cognitive
and social sciences. The Board Of Education is a complete, low-cost development platform
equipped with the needed sensors for humidity, temperature, light, etc. As shown in
Figure 1, the Boe-Bot is a great tool with which to get started with robotics.

Fig. 1. Assembled Boe-Bot

The SunSPOT WSN module makes it possible for the Boe-Bot robots BASIC Stamp 2
microcontroller brain to communicate wirelessly with a web based user interface running on
a nearby PC. The BASIC Stamp microcontroller runs a small PBASIC program that controls
the Boe-Bot robots servos and optionally monitors sensors while it communicates wirelessly
with the web server.

3. Control scheme for mobile robots
A mobile robot needs locomotion mechanisms that enable it to move throughout its known
or unknown environment. But there are a large variety of possible ways to move, and so the
selection of a robots approach to locomotion is an important aspect of mobile robot design.

Figure 2, presents the control scheme for mobile robot systems. In the laboratory, there are
research robots that can walk, jump, run, slide, skate, swim, fly, and, of course, roll. Any of
these activities has its own control algorithm (Gy. Mester, 2009).

Fig. 2. Reference control scheme for mobile robot systems

Locomotion is the complement of manipulation. In manipulation, the robot arm is fixed but
moves objects in the workspace by imparting force to them. In locomotion, the environment
is fixed and the robot moves by imparting force to the environment. In both cases, the
scientific basis is the study of actuators that generate interaction forces, and mechanisms
that implement desired kinematical and dynamic properties. The wheel has been by far the
most popular mechanism in mobile robotics and in man-made vehicles in general. It can
achieve very good efficiencies, and does so with a relatively simple mechanical
implementation. On Figure 3, the kinematics of the mobile robot is depicted. In addition,
balance is not usually a research problem in wheeled robot designs, because wheeled robots
are almost always designed so that all wheels are in ground contact at all times (Gy. Mester,
2009).

Fig. 3. Robot kinematics and its frames of interests

Thus, three wheels are sufficient to guarantee stable balance, although, as we shall see
below, two-wheeled robots can also be stable (R. Siegwart, 2004). When more than three
wheels are used, a suspension system is required to allow all wheels to maintain ground
contact when the robot encounters uneven terrain. Motion control might not be an easy task
for this kind of systems. However, it has been studied by various research groups, and some
adequate solutions for motion control of a mobile robot system are available (Gy. Mester,
2009).

4. Using Potential Fields method for navigation
A potential field consists of two imaginary fields (attractive potential and repulsive potential)
and used to avoid a collision with unexpected obstacle while moving in a predetermined
path. The Attractive Potential forces the robot to move through a predetermined path and
the Repulsive Field, assumed to be generated by obstacles, forces the robot to move a
different way to avoid the collision (O. Khatib, 1986). The Artificial Potential Field approach
is a local path planner method that was introduced by Khatib. This method defines obstacles
as repelling force sources, and goals as attracting force sources. The path is then influenced
by the composition of the two forces, which produces a robot motion that moves away from
obstacles while moving towards the target goal. The approach is mathematically simple and
is able to produce real-time acceptable results for collision avoidance even in dynamic
environments. The most known limitation of this approach is the local minima, which refers
to locations that trap the robot and prevent it from reaching the target goal location. This
main problem has been addressed by many different techniques that try to solve or at least
minimize its impact (O. Khatib, 1985).

4.1 Attractive Potential Field
The attractive potential field corresponds to the component responsible for the potentials
that attract the robot towards the target goal position. At all locations in the environment the
action vector will point to the target goal.

Fig. 4. Attractive potential field action vectors pointing to the goal and goal representation
(M. Goodrich, 2002)

Usually, the action vector is found by applying a scalar potential field function to the robot's
position and then calculating the gradient of that function.

] , [ ] , [
y
U
x
U
y x
c
c
c
c
= V V = V (1)
After defining: (M. Goodrich, 2002)

- ] , [
G G
y x as the position of the goal;
- r as the radius of the goal;
- ] , [
R R
y x as the position of the robot;
- s as the size of the goal's area of influence;
- o as the strength of the attractive field ( ) 0 > o

We can compute x V and y V using the following steps:

1. Find the distance d between the goal and the robot:

2 2
) ( ) (
G R R G
y y x x d + = (2)

2. Find the angle u between the robot and the goal:

|
|
.
|
\
|
=

R G
R G
x x
y y
1
tan u (3)


Fig. 3. Robot kinematics and its frames of interests

Thus, three wheels are sufficient to guarantee stable balance, although, as we shall see
below, two-wheeled robots can also be stable (R. Siegwart, 2004). When more than three
wheels are used, a suspension system is required to allow all wheels to maintain ground
contact when the robot encounters uneven terrain. Motion control might not be an easy task
for this kind of systems. However, it has been studied by various research groups, and some
adequate solutions for motion control of a mobile robot system are available (Gy. Mester,
2009).

4. Using Potential Fields method for navigation
A potential field consists of two imaginary fields (attractive potential and repulsive potential)
and used to avoid a collision with unexpected obstacle while moving in a predetermined
path. The Attractive Potential forces the robot to move through a predetermined path and
the Repulsive Field, assumed to be generated by obstacles, forces the robot to move a
different way to avoid the collision (O. Khatib, 1986). The Artificial Potential Field approach
is a local path planner method that was introduced by Khatib. This method defines obstacles
as repelling force sources, and goals as attracting force sources. The path is then influenced
by the composition of the two forces, which produces a robot motion that moves away from
obstacles while moving towards the target goal. The approach is mathematically simple and
is able to produce real-time acceptable results for collision avoidance even in dynamic
environments. The most known limitation of this approach is the local minima, which refers
to locations that trap the robot and prevent it from reaching the target goal location. This
main problem has been addressed by many different techniques that try to solve or at least
minimize its impact (O. Khatib, 1985).

4.1 Attractive Potential Field
The attractive potential field corresponds to the component responsible for the potentials
that attract the robot towards the target goal position. At all locations in the environment the
action vector will point to the target goal.

Fig. 4. Attractive potential field action vectors pointing to the goal and goal representation
(M. Goodrich, 2002)

Usually, the action vector is found by applying a scalar potential field function to the robot's
position and then calculating the gradient of that function.

] , [ ] , [
y
U
x
U
y x
c
c
c
c
= V V = V (1)

- ] , [
G G
y x as the position of the goal;
- r as the radius of the goal;
- ] , [
R R
- s as the size of the goal's area of influence;
- o as the strength of the attractive field ( ) 0 > o


1. Find the distance d between the goal and the robot:

2 2
) ( ) (
G R R G
y y x x d + = (2)

2. Find the angle u between the robot and the goal:

|
|
.
|
\
|
=

R G
R G
x x
y y
1
tan u (3)


3. Set x V and y V according to the rules:

If r d < then 0 = V = V y x

If r s d r + s s then
= V
= V
) sin( ) (
) cos( ) (
u o
u o
r d y
r d x
(4)

If r s d + > then
= V
= V
) sin(
) cos(
u o
u o
s y
s x

The last step presents three simple rules that characterize three different behaviors for the
robot according to its relative position towards the goal:
- In the first rule of step 3,
r d <
means that the robot is in the goal area. In this
case, no forces act and
x V
and
y V
are set to zero.

- In the second rule,
r s d r + s s
means that the robot is inside the area of
inuence of the goal. The action vector is set using
o
,
d
and
s
.

- In the third and last rule,
r s d + >
means that the robot is outside the goal
area and also outside its area of influence. The action vector is set to with
s
and
o
thus reaching higher values.

4.2 Repulsive Potential Field
The repulsive potential field is the component that is responsible for forcing the robot to
stay away from the obstacles it encounters on its path. All repulsive action vectors point
away from the obstacle surface driving the robot away from the obstacle.

Fig. 5. Repulsive potential field action vectors pointing away from the obstacle and obstacle
representation (M. Goodrich, 2002)

Similarly to the Attractive Potential, we calculate the repulsive action vector.


- ] , [
O O
y x as the position of the obstacle;
- r as the radius of the obstacle;
- ] , [
R R
- s as the size of the obstacle's area of influence;
- | as the strength of the repulsive field ( ) 0 > |


1. Find the distance d between the obstacle and the robot:

2 2
) ( ) (
O R R O
y y x x d + = (5)

2. Find the angle u between the robot and the obstacle:

|
|
.
|
\
|
=

R O
R O
x x
y y
1
tan u (6)


If r d < then
= V
= V
)) (sin(
)) (cos(
u
u
sign y
sign x

+ = V
+ = V
) sin( ) (
) cos( ) (
u |
u |
d r s y
d r s x
(7)

If r s d + > then 0 = V = V y x

Similar to the attractive potential rules, these rules are also simple and characterize three
different behaviors for the robot according to its position relative to the obstacle. It is
important to notice that all action vectors need to point away from the obstacle, hence the
need to use negative values (M. Goodrich, 2002).

- In the first rule of step 3, the robot is within the radius of the obstacle, so the
action vector needs to be infinite, expressing the need to escape from the robot.


If r d < then 0 = V = V y x

= V
= V
) sin( ) (
) cos( ) (
u o
u o
r d y
r d x
(4)

If r s d + > then
= V
= V
) sin(
) cos(
u o
u o
s y
s x

The last step presents three simple rules that characterize three different behaviors for the
robot according to its relative position towards the goal:
- In the first rule of step 3,
r d <
means that the robot is in the goal area. In this
case, no forces act and
x V
and
y V
are set to zero.

- In the second rule,
r s d r + s s
means that the robot is inside the area of
inuence of the goal. The action vector is set using
o
,
d
and
s
.

- In the third and last rule,
r s d + >
means that the robot is outside the goal
area and also outside its area of influence. The action vector is set to with
s
and
o
thus reaching higher values.

4.2 Repulsive Potential Field
The repulsive potential field is the component that is responsible for forcing the robot to
stay away from the obstacles it encounters on its path. All repulsive action vectors point
away from the obstacle surface driving the robot away from the obstacle.

Fig. 5. Repulsive potential field action vectors pointing away from the obstacle and obstacle
representation (M. Goodrich, 2002)

Similarly to the Attractive Potential, we calculate the repulsive action vector.


- ] , [
O O
y x as the position of the obstacle;
- r as the radius of the obstacle;
- ] , [
R R
- s as the size of the obstacle's area of influence;
- | as the strength of the repulsive field ( ) 0 > |


1. Find the distance d between the obstacle and the robot:

2 2
) ( ) (
O R R O
y y x x d + = (5)

2. Find the angle u between the robot and the obstacle:

|
|
.
|
\
|
=

R O
R O
x x
y y
1
tan u (6)


If r d < then
= V
= V
)) (sin(
)) (cos(
u
u
sign y
sign x

+ = V
+ = V
) sin( ) (
) cos( ) (
u |
u |
d r s y
d r s x
(7)

If r s d + > then 0 = V = V y x

Similar to the attractive potential rules, these rules are also simple and characterize three
different behaviors for the robot according to its position relative to the obstacle. It is
important to notice that all action vectors need to point away from the obstacle, hence the
need to use negative values (M. Goodrich, 2002).

- In the first rule of step 3, the robot is within the radius of the obstacle, so the
action vector needs to be infinite, expressing the need to escape from the robot.

In the second rule, where the robot is outside the obstacle's radius but inside its
area of influence, the action vector is set to a high value in order to express the
need to escape the current location.
In the third rule, where the robot is outside the area of influence of the obstacle,
the action vector is set to zero, meaning that no repulsive forces are acting on the
robot (M. Goodrich, 2002).

Since the repulsive force only acts when the robot is inside the area of influence of the
obstacle, the value of s must be carefully chosen. A small value for s can cause trajectory
problems by causing abrupt changes on the path and some constraints on the speed of the
robot. A large value for s may cause also problems on the robot's movement since it can
constrain movement in small places where the robot could pass.

Fig. 6. Potential Fields simulation

The repulsive force has the objective of repelling the robot only if it is close to an obstacle
and its velocity points towards that obstacle (M. Goodrich, 2002).

4. WSN and Event-Based System for Greenhouse Climate Control
A wireless sensor network (WSN) is a computer network consisting of spatially distributed
autonomous devices using sensors to cooperatively monitor physical or environmental
conditions, such as temperature, sound, vibration, pressure, motion or pollutants, at
different locations (Sun Microsystems, 2002). The development of wireless sensor networks
was originally motivated by military applications such as battlefield surveillance. Figure 7,
presents the sensor node architecture. However, wireless sensor networks are now used in
many civilian application areas, including environment and habitat monitoring, healthcare
applications, home automation, and traffic control.

Fig. 7. Sensor Node Architecture

In addition to one or more sensors, each node in a sensor network is typically equipped with
a radio transceiver or other wireless communications device, a small microcontroller, and an
energy source, usually a battery. Figure 8, shows the typical wireless sensor network.

Fig. 8. Typical wireless sensor network (WSN)

The size a single sensor node can vary from shoebox-sized nodes down to devices the size of
grain of dust. The cost of sensor nodes is similarly variable, ranging from hundreds of
dollars to a few cents, depending on the size of the sensor network and the complexity
required of individual sensor nodes (Sun Microsystems, 2005). Size and cost constraints on
sensor nodes result in corresponding constraints on resources such as energy, memory,
computational speed and bandwidth. In computer science, wireless sensor networks are an
active research area with numerous workshops and conferences arranged each year (S.
Scaglia, 2008). As commented above, this paper is devoted to analyzing diurnal and
nocturnal temperature control with natural ventilation and heating systems, and humidity
control as a secondary control objective. Under diurnal conditions, the controlled variable is
the inside temperature and the control signal is the vent opening. The use of natural
ventilation produces an exchange between the inside and outside air, usually provoking a
decrease in the inside temperature of the greenhouse. The controller must calculate the

In the second rule, where the robot is outside the obstacle's radius but inside its
area of influence, the action vector is set to a high value in order to express the
need to escape the current location.
In the third rule, where the robot is outside the area of influence of the obstacle,
the action vector is set to zero, meaning that no repulsive forces are acting on the
robot (M. Goodrich, 2002).

Since the repulsive force only acts when the robot is inside the area of influence of the
obstacle, the value of s must be carefully chosen. A small value for s can cause trajectory
problems by causing abrupt changes on the path and some constraints on the speed of the
robot. A large value for s may cause also problems on the robot's movement since it can
constrain movement in small places where the robot could pass.

Fig. 6. Potential Fields simulation

The repulsive force has the objective of repelling the robot only if it is close to an obstacle
and its velocity points towards that obstacle (M. Goodrich, 2002).

4. WSN and Event-Based System for Greenhouse Climate Control
A wireless sensor network (WSN) is a computer network consisting of spatially distributed
autonomous devices using sensors to cooperatively monitor physical or environmental
conditions, such as temperature, sound, vibration, pressure, motion or pollutants, at
different locations (Sun Microsystems, 2002). The development of wireless sensor networks
was originally motivated by military applications such as battlefield surveillance. Figure 7,
presents the sensor node architecture. However, wireless sensor networks are now used in
many civilian application areas, including environment and habitat monitoring, healthcare
applications, home automation, and traffic control.

Fig. 7. Sensor Node Architecture

In addition to one or more sensors, each node in a sensor network is typically equipped with
a radio transceiver or other wireless communications device, a small microcontroller, and an
energy source, usually a battery. Figure 8, shows the typical wireless sensor network.

Fig. 8. Typical wireless sensor network (WSN)

The size a single sensor node can vary from shoebox-sized nodes down to devices the size of
grain of dust. The cost of sensor nodes is similarly variable, ranging from hundreds of
dollars to a few cents, depending on the size of the sensor network and the complexity
required of individual sensor nodes (Sun Microsystems, 2005). Size and cost constraints on
sensor nodes result in corresponding constraints on resources such as energy, memory,
computational speed and bandwidth. In computer science, wireless sensor networks are an
active research area with numerous workshops and conferences arranged each year (S.
Scaglia, 2008). As commented above, this paper is devoted to analyzing diurnal and
nocturnal temperature control with natural ventilation and heating systems, and humidity
control as a secondary control objective. Under diurnal conditions, the controlled variable is
the inside temperature and the control signal is the vent opening. The use of natural
ventilation produces an exchange between the inside and outside air, usually provoking a
decrease in the inside temperature of the greenhouse. The controller must calculate the

necessary vent opening to reach the desired setpoint. The commonest controller used is a
gain scheduling PI scheme where the controller parameters are changed based on some
disturbances: outside temperature and wind speed. In the case of nocturnal temperature
control, forced-air heaters are used to increase the inside temperature and an on/off control
with dead/zone was selected as heating controller.

Fig. 9. Humidity control

techniques (S. Scaglia, 2008). Normally, the type of acquisition system is chosen to be
optimal for the control algorithm to be used. For traditional climate monitoring and control
systems, all sensors are distributed through the greenhouse and connected to the device
performing the control tasks. These equipments use time-based data sampling techniques as
a consequence of using time-based controllers. Figure 10, presents the temperature
controller.

Fig. 10. Temperature controller

Nowadays, commercial systems present more flexibility in the implementation of control
algorithms and sampling techniques, especially WSN, where each node of the network can
be programmed with a different sampling algorithm or local control algorithm with the
main goal of optimizing the overall performance.

4.1 Description of the Sun SPOT WSN module
One of the most poplar technologies in the WSN area is Sun SPOT (Small Programmable
Object Technology). It contains 32-bit ARM9 CPU, 512K memory, 2 Mb flash storage and
wireless networking is based on ChipCon CC2420 following the 802.15.4 standard with
integrated antenna and operates in the 2.4GHz to 2.4835GHz ISM unlicensed bands. The IC
contains a 2.4GHz RF transmitter/receiver with digital direct sequence spread spectrum
(DSSS) baseband modem with MAC support.

Fig. 11. Sun SPOT processor board

The sensor board integrates multiple sensors, monitoring LED and interactive switches into
one board. All the facilities of this board are programmable in Java. The Sun SPOT SDK
comes with two important tools for managing the software on your SPOTs: SPOTManager
and SPOTWorld. The SPOTManager is a tool for managing the Sun SPOT SDK software.
You can use it to download from the Internet both new and old versions of the Sun SPOT
SDK. You can use it to make one or another SDK the active SDK on your host workstation,
and you can use it to download system software to your Sun SPOTs (Sun Microsystems,
2005).


necessary vent opening to reach the desired setpoint. The commonest controller used is a
gain scheduling PI scheme where the controller parameters are changed based on some
disturbances: outside temperature and wind speed. In the case of nocturnal temperature
control, forced-air heaters are used to increase the inside temperature and an on/off control
with dead/zone was selected as heating controller.

Fig. 9. Humidity control

techniques (S. Scaglia, 2008). Normally, the type of acquisition system is chosen to be
optimal for the control algorithm to be used. For traditional climate monitoring and control
systems, all sensors are distributed through the greenhouse and connected to the device
performing the control tasks. These equipments use time-based data sampling techniques as
a consequence of using time-based controllers. Figure 10, presents the temperature
controller.

Fig. 10. Temperature controller

Nowadays, commercial systems present more flexibility in the implementation of control
algorithms and sampling techniques, especially WSN, where each node of the network can
be programmed with a different sampling algorithm or local control algorithm with the
main goal of optimizing the overall performance.

4.1 Description of the Sun SPOT WSN module
One of the most poplar technologies in the WSN area is Sun SPOT (Small Programmable
Object Technology). It contains 32-bit ARM9 CPU, 512K memory, 2 Mb flash storage and
wireless networking is based on ChipCon CC2420 following the 802.15.4 standard with
integrated antenna and operates in the 2.4GHz to 2.4835GHz ISM unlicensed bands. The IC
contains a 2.4GHz RF transmitter/receiver with digital direct sequence spread spectrum
(DSSS) baseband modem with MAC support.

Fig. 11. Sun SPOT processor board

The sensor board integrates multiple sensors, monitoring LED and interactive switches into
one board. All the facilities of this board are programmable in Java. The Sun SPOT SDK
comes with two important tools for managing the software on your SPOTs: SPOTManager
and SPOTWorld. The SPOTManager is a tool for managing the Sun SPOT SDK software.
You can use it to download from the Internet both new and old versions of the Sun SPOT
SDK. You can use it to make one or another SDK the active SDK on your host workstation,
and you can use it to download system software to your Sun SPOTs (Sun Microsystems,
2005).


Fig. 12. Sun SPOT sensor board

The facilities of the sensor board are:

One 2G/6G 3-axis accelerometer
One temperature sensor
One light sensor
Two 8-bit tri-color LEDs
6 analog inputs
Two momentary switches
5 general purpose I/O pins

The internal battery is a 3.7V rechargeable lithium-ion prismatic cell. The battery has
internal protection circuit to guard against over discharge, under voltage and overcharge
conditions. The battery can be charged from either the USB type mini-B device connector or
from an external source with a 5V power supply.

4.2 Wireless radio
The wireless network communications uses an integrated radio transceiver, the TI CC2420
(formerly ChipCon). The CC2420 is IEEE 802.15.4 compliant device and operates in the
2.4GHz to 2.4835GHz ISM unlicensed bands. Regulations for these bands are covered by
FCC CFR47 part 15 (USA), ETSI EN 300 328 and EN 300 440 class 2 device (Europe) and
ARIB STD-T66 (Japan).

The IC contains a 2.4GHz RF transmitter/receiver with digital direct sequence spread
spectrum (DSSS) baseband modem with MAC support. Other features include separate TX
and RX 128 byte FIFOs, AES encryption (currently not supported), received signal strength
indication (RSSI) with 100dB sensitivity and transmit output power setting from -24dBm to
0dBm. Effective bit rate is 250kbps and chip rate is 2000kChips/s. Receive sensitivity is -

90dBm. The digital control and data communications with the CC2420 use PIO port bits and
the SPI channel. The CC2420 is a slave SPI bidirectional device addressed when RF_CS
(PCS2) is asserted active low. PIO ports reset the CC2420 (RF_RST), power it down
(RF_PWDOWN), or check the status of the receive FIFO (FIFO and FIFOP), clear channel
assessment (CCA) and start of frame (SFD). There are 33 configuration and status registers,
15 command registers and two 8-bit registers for the separate transmit and receive FIFOs.
The first byte sent to the CC2420 is the address made up of 6-bit address, RAM/Register
select (Bit 7) and Read/Write select (Bit 6). Following bytes are data read from or written to
the CC2420.

Fig. 13. Typical application circuit for CC2420

The CC2420 is housed in a 48pin quad leadless package (QLP or QFN) that is 7mm square. It
is powered with +3.3V Vcc supply. The CC2420 has an internal 1.8V low drop out regulator
for powering the internal RF and analog circuitry. It consumes 20mA during receive
operation and 18mA for 0dBm transmit. The frequency generation uses an accurate 16MHz
crystal with 10ppm accuracy, 10ppm stability and 1ppm aging. The entire RF section is
enclosed in an upper and lower RF shield and has modular FCC approval.

4.3 I/O pin Manipulation of the SunSPOT module
The SunSPOTs sensor board has an Atmega88 and operates the 8 tricolor LEDs, the
accelerometer configuration, and the following pins on the I/O header: I/O pins D0 through
D3 can be set as either an output or input.

Fig. 12. Sun SPOT sensor board

The facilities of the sensor board are:

One 2G/6G 3-axis accelerometer
One temperature sensor
One light sensor
Two 8-bit tri-color LEDs
6 analog inputs
Two momentary switches
5 general purpose I/O pins

The internal battery is a 3.7V rechargeable lithium-ion prismatic cell. The battery has
internal protection circuit to guard against over discharge, under voltage and overcharge
conditions. The battery can be charged from either the USB type mini-B device connector or
from an external source with a 5V power supply.

4.2 Wireless radio
The wireless network communications uses an integrated radio transceiver, the TI CC2420
(formerly ChipCon). The CC2420 is IEEE 802.15.4 compliant device and operates in the
2.4GHz to 2.4835GHz ISM unlicensed bands. Regulations for these bands are covered by
FCC CFR47 part 15 (USA), ETSI EN 300 328 and EN 300 440 class 2 device (Europe) and
ARIB STD-T66 (Japan).

The IC contains a 2.4GHz RF transmitter/receiver with digital direct sequence spread
spectrum (DSSS) baseband modem with MAC support. Other features include separate TX
and RX 128 byte FIFOs, AES encryption (currently not supported), received signal strength
indication (RSSI) with 100dB sensitivity and transmit output power setting from -24dBm to
0dBm. Effective bit rate is 250kbps and chip rate is 2000kChips/s. Receive sensitivity is -

90dBm. The digital control and data communications with the CC2420 use PIO port bits and
the SPI channel. The CC2420 is a slave SPI bidirectional device addressed when RF_CS
(PCS2) is asserted active low. PIO ports reset the CC2420 (RF_RST), power it down
(RF_PWDOWN), or check the status of the receive FIFO (FIFO and FIFOP), clear channel
assessment (CCA) and start of frame (SFD). There are 33 configuration and status registers,
15 command registers and two 8-bit registers for the separate transmit and receive FIFOs.
The first byte sent to the CC2420 is the address made up of 6-bit address, RAM/Register
select (Bit 7) and Read/Write select (Bit 6). Following bytes are data read from or written to
the CC2420.

Fig. 13. Typical application circuit for CC2420

The CC2420 is housed in a 48pin quad leadless package (QLP or QFN) that is 7mm square. It
is powered with +3.3V Vcc supply. The CC2420 has an internal 1.8V low drop out regulator
for powering the internal RF and analog circuitry. It consumes 20mA during receive
operation and 18mA for 0dBm transmit. The frequency generation uses an accurate 16MHz
crystal with 10ppm accuracy, 10ppm stability and 1ppm aging. The entire RF section is
enclosed in an upper and lower RF shield and has modular FCC approval.

4.3 I/O pin Manipulation of the SunSPOT module
The SunSPOTs sensor board has an Atmega88 and operates the 8 tricolor LEDs, the
accelerometer configuration, and the following pins on the I/O header: I/O pins D0 through
D3 can be set as either an output or input.

Fig. 14. The SunSPOTs sensor board component location

The high current driver pins, H0 to H3, can only be used as an output. If configured as an
output, the pin may be set hi, low, or toggled to its opposite state.

Developing environment for Sun SPOTs

The Java programming language is a general-purpose concurrent class-based object-
oriented programming language, specifically designed to have as few implementation
dependencies as possible. It allows application developers to write a program once and then
be able to run it everywhere on the Internet. Java is a programming language originally
developed by James Gosling at Sun Microsystems and released in 1995 as a core component
of Sun Microsystems' Java platform. The language derives much of its syntax from C and
C++ but has a simpler object model and fewer low-level facilities. Java applications are
typically compiled to bytecode that can run on any Java virtual machine (JVM) regardless of
computer architecture. The original and reference implementation Java compilers, virtual
machines, and class libraries were developed by Sun from 1995. As of May 2007, in
compliance with the specifications of the Java Community Process, Sun made available most
of their Java technologies as free software under the GNU General Public License. One
characteristic of Java is portability, which means that computer programs written in the Java
language must run similarly on any supported hardware/operating-system platform. One
should be able to write a program once, compile it once, and run it anywhere. This is
achieved by compiling the Java language code, not to machine code but to Java bytecode
instructions analogous to machine code but intended to be interpreted by a virtual machine
(VM) written specifically for the host hardware. End-users commonly use a Java Runtime
Environment (JRE) installed on their own machine for standalone Java applications, or in a
Web browser for Java applets. Standardized libraries provide a generic way to access host
specific features such as graphics, threading and networking. In some JVM versions,
bytecode can be compiled to native code, either before or during program execution,

resulting in faster execution (J. Gosling, 2005). The most popular developing environment
for Java is Netbeans IDE.

Fig. 15. NetBeans IDE 6.5 Sun SPOT SDK

A major benefit of using bytecode is porting. However, the overhead of interpretation
means that interpreted programs almost always run more slowly than programs compiled
to native executables would, and Java suffered a reputation for poor performance. This gap
has been narrowed by a number of optimization techniques introduced in the more recent
JVM implementations. One such technique, known as just-in-time (JIT) compilation,
translates Java bytecode into native code the first time that code is executed, then caches it.
This results in a program that starts and executes faster than pure interpreted code can, at
the cost of introducing occasional compilation overhead during execution. More
sophisticated VMs also use dynamic recompilation, in which the VM analyzes the behavior
of the running program and selectively recompiles and optimizes parts of the program.
Dynamic recompilation can achieve optimizations superior to static compilation because the
dynamic compiler can base optimizations on knowledge about the runtime environment
and the set of loaded classes, and can identify hot spots - parts of the program, often inner
loops, that take up the most execution time. JIT compilation and dynamic recompilation
allow Java programs to approach the speed of native code without losing portability.

4.4 The Sun SPOT emulator
Solarium includes an emulator capable of running a Sun SPOT application on your desktop
computer. This allows for testing a program before deploying it to a real SPOT, or if a real
SPOT is not available. Instead of a physical sensor board, Solarium displays a virtual SPOT
with a control panel where you can set any of the potential sensor inputs (e.g. light level,
temperature, digital pin inputs, analog input voltages, and accelerometer values). Your
application can control the LEDs' color that is displayed in the virtual SPOT image, just like
it would a real SPOT. You can click with the mouse on the push button switches in the
virtual SPOT image to press and release the switches. Receiving and sending via the radio is
also supported. Each virtual SPOT is assigned its own address and can broadcast or unicast

Fig. 14. The SunSPOTs sensor board component location

The high current driver pins, H0 to H3, can only be used as an output. If configured as an
output, the pin may be set hi, low, or toggled to its opposite state.

Developing environment for Sun SPOTs

The Java programming language is a general-purpose concurrent class-based object-
oriented programming language, specifically designed to have as few implementation
dependencies as possible. It allows application developers to write a program once and then
be able to run it everywhere on the Internet. Java is a programming language originally
developed by James Gosling at Sun Microsystems and released in 1995 as a core component
of Sun Microsystems' Java platform. The language derives much of its syntax from C and
C++ but has a simpler object model and fewer low-level facilities. Java applications are
typically compiled to bytecode that can run on any Java virtual machine (JVM) regardless of
computer architecture. The original and reference implementation Java compilers, virtual
machines, and class libraries were developed by Sun from 1995. As of May 2007, in
compliance with the specifications of the Java Community Process, Sun made available most
of their Java technologies as free software under the GNU General Public License. One
characteristic of Java is portability, which means that computer programs written in the Java
language must run similarly on any supported hardware/operating-system platform. One
should be able to write a program once, compile it once, and run it anywhere. This is
achieved by compiling the Java language code, not to machine code but to Java bytecode
instructions analogous to machine code but intended to be interpreted by a virtual machine
(VM) written specifically for the host hardware. End-users commonly use a Java Runtime
Environment (JRE) installed on their own machine for standalone Java applications, or in a
Web browser for Java applets. Standardized libraries provide a generic way to access host
specific features such as graphics, threading and networking. In some JVM versions,
bytecode can be compiled to native code, either before or during program execution,

resulting in faster execution (J. Gosling, 2005). The most popular developing environment
for Java is Netbeans IDE.

Fig. 15. NetBeans IDE 6.5 Sun SPOT SDK

A major benefit of using bytecode is porting. However, the overhead of interpretation
means that interpreted programs almost always run more slowly than programs compiled
to native executables would, and Java suffered a reputation for poor performance. This gap
has been narrowed by a number of optimization techniques introduced in the more recent
JVM implementations. One such technique, known as just-in-time (JIT) compilation,
translates Java bytecode into native code the first time that code is executed, then caches it.
This results in a program that starts and executes faster than pure interpreted code can, at
the cost of introducing occasional compilation overhead during execution. More
sophisticated VMs also use dynamic recompilation, in which the VM analyzes the behavior
of the running program and selectively recompiles and optimizes parts of the program.
Dynamic recompilation can achieve optimizations superior to static compilation because the
dynamic compiler can base optimizations on knowledge about the runtime environment
and the set of loaded classes, and can identify hot spots - parts of the program, often inner
loops, that take up the most execution time. JIT compilation and dynamic recompilation
allow Java programs to approach the speed of native code without losing portability.

4.4 The Sun SPOT emulator
Solarium includes an emulator capable of running a Sun SPOT application on your desktop
computer. This allows for testing a program before deploying it to a real SPOT, or if a real
SPOT is not available. Instead of a physical sensor board, Solarium displays a virtual SPOT
with a control panel where you can set any of the potential sensor inputs (e.g. light level,
temperature, digital pin inputs, analog input voltages, and accelerometer values). Your
application can control the LEDs' color that is displayed in the virtual SPOT image, just like
it would a real SPOT. You can click with the mouse on the push button switches in the
virtual SPOT image to press and release the switches. Receiving and sending via the radio is
also supported. Each virtual SPOT is assigned its own address and can broadcast or unicast

to the other virtual SPOTs. If a shared base station is available a virtual SPOT can also
interact over the radio with real SPOTs.

Fig. 16. The Sun SPOT Emulator

Virtual SPOTs can communicate with each other by opening radio connections, both
broadcast and point-to-point. Instead of using an actual radio these connections take place
over regular and multicast sockets. When a base station SPOT is connected to the host
computer and a shared base station is running, virtual SPOTs can also use it to communicate
with real SPOTs using the base stations radio. The advantage of using a shared base station
is that multiple host applications can then all access the radio. One disadvantage is that
communication from a host application to a target SPOT takes two radio hops, in contrast to
the one hop needed with a dedicated base station. Another disadvantage is that run-time
manipulation of the base station SPOTs radio channel, pan id or output power is not
currently possible. Each virtual SPOT has its own Squawk VM running in a separate process
on the host computer. Each Squawk VM contains a complete host-side radio stack as part of
the SPOT library, which allows the SPOT application to communicate with other SPOT
applications running on the host computer, such as other virtual SPOTs, using sockets or
real SPOTs via radio if a shared base station is running. The current Solarium
implementation is primarily an emulator since it actually runs a SPOT application in a
Squawk VM, just like the VM on a real SPOT. Likewise radio interaction between virtual
SPOTs is emulated with data sent via packets and streams from one (virtual) SPOT to
another. Only the SPOT's interaction with the environment is simulated using a simple
model where the user needs to explicitly set the current sensor values. Future versions may
incorporate more simulation of SPOT properties like battery level or radio range.

5. Solution
Building and programming a robot is a combination of mechanics, electronics, and problem
solving. What you're about to learn while doing the activities and projects in this text will be
relevant to "real world" applications that use robotic control, the only difference being the size
and sophistication. Robotics has come a long way, especially for mobile robots. In the past,
mobile robots were controlled by heavy, large, and expensive computer systems that could not

be carried and had to be linked via cable or wireless devices. As shown in Figure 8, the mobile
measuring station is navigating inside the greenhouse. Today, however, we can build small
mobile robots with numerous actuators and sensors that are controlled by inexpensive, small,
and light embedded computer systems that are carried on-board the robot.

Fig. 17. Greenhouse top view with the mobile measuring station

The mechanical principles, example program listings, and circuits you will use are very similar
to, and sometimes the same as, industrial applications developed by engineers. In this project
we have used SunSPOT-s to achieve remote control over a Boe-Bot. For this project we have
used 2 SunSPOT-s from the kit (free range and base station module) as depicted on Figure 18.
SunSPOT's wireless protocol is Zigbee based protocol (I. Matijevics, 2008).

Fig. 18. Connection of the system

The Hardware basically centers around Sun SPOT and DC Motors controlled by Basic
Stamp. The Sun SPOT base station will send data to Sun SPOT on the mobile measuring
station which will drive the Basic Stamp controller to DC IO pins (J. Simon, 2009). The
microcontroller will drive the Motors which will run the measuring station. Figure 19,
shows the testing phase of the mobile measuring station.

to the other virtual SPOTs. If a shared base station is available a virtual SPOT can also
interact over the radio with real SPOTs.

Fig. 16. The Sun SPOT Emulator

Virtual SPOTs can communicate with each other by opening radio connections, both
broadcast and point-to-point. Instead of using an actual radio these connections take place
over regular and multicast sockets. When a base station SPOT is connected to the host
computer and a shared base station is running, virtual SPOTs can also use it to communicate
with real SPOTs using the base stations radio. The advantage of using a shared base station
is that multiple host applications can then all access the radio. One disadvantage is that
communication from a host application to a target SPOT takes two radio hops, in contrast to
the one hop needed with a dedicated base station. Another disadvantage is that run-time
manipulation of the base station SPOTs radio channel, pan id or output power is not
currently possible. Each virtual SPOT has its own Squawk VM running in a separate process
on the host computer. Each Squawk VM contains a complete host-side radio stack as part of
the SPOT library, which allows the SPOT application to communicate with other SPOT
applications running on the host computer, such as other virtual SPOTs, using sockets or
real SPOTs via radio if a shared base station is running. The current Solarium
implementation is primarily an emulator since it actually runs a SPOT application in a
Squawk VM, just like the VM on a real SPOT. Likewise radio interaction between virtual
SPOTs is emulated with data sent via packets and streams from one (virtual) SPOT to
another. Only the SPOT's interaction with the environment is simulated using a simple
model where the user needs to explicitly set the current sensor values. Future versions may
incorporate more simulation of SPOT properties like battery level or radio range.

5. Solution
Building and programming a robot is a combination of mechanics, electronics, and problem
solving. What you're about to learn while doing the activities and projects in this text will be
relevant to "real world" applications that use robotic control, the only difference being the size
and sophistication. Robotics has come a long way, especially for mobile robots. In the past,
mobile robots were controlled by heavy, large, and expensive computer systems that could not

be carried and had to be linked via cable or wireless devices. As shown in Figure 8, the mobile
measuring station is navigating inside the greenhouse. Today, however, we can build small
mobile robots with numerous actuators and sensors that are controlled by inexpensive, small,
and light embedded computer systems that are carried on-board the robot.

Fig. 17. Greenhouse top view with the mobile measuring station

The mechanical principles, example program listings, and circuits you will use are very similar
to, and sometimes the same as, industrial applications developed by engineers. In this project
we have used SunSPOT-s to achieve remote control over a Boe-Bot. For this project we have
used 2 SunSPOT-s from the kit (free range and base station module) as depicted on Figure 18.
SunSPOT's wireless protocol is Zigbee based protocol (I. Matijevics, 2008).

Fig. 18. Connection of the system

The Hardware basically centers around Sun SPOT and DC Motors controlled by Basic
Stamp. The Sun SPOT base station will send data to Sun SPOT on the mobile measuring
station which will drive the Basic Stamp controller to DC IO pins (J. Simon, 2009). The
microcontroller will drive the Motors which will run the measuring station. Figure 19,
shows the testing phase of the mobile measuring station.

Fig. 19. Boe-bot with SunSPOT mounted

6. Experimental results
The applications for WSNs are many and varied. They are used in commercial and
industrial applications to monitor data that would be difficult or expensive to monitor using
wired sensors. They could be deployed in wilderness areas, where they would remain for
many years (monitoring some environmental variable) without the need to
recharge/replace their power supplies. They could form a perimeter about a property and
monitor the progression of intruders (passing information from one node to the next). There
are a many uses for WSNs (I. Matijevics, 2009).

Fig. 20. Crops in greenhouse

Typical applications of WSNs include monitoring, tracking, and controlling. Some of the
specific applications are habitat monitoring, object tracking, nuclear reactor controlling, fire
detection, traffic monitoring, etc. In a typical application, a WSN is scattered in a region
where it is meant to collect data through its sensor node. Figure 23, shows the complete
control system of the greenhouse. The WSN-based controller has allowed a considerable
decrease in the number of changes in the control action and made possible a study of the
compromise between quantity of transmission and control performance.

Fig. 21. Capsicum

Motion control of mobile robots is a very important research field today, because mobile
robots are a very interesting subject both in scientific research and practical applications. In
this paper the object of the remote control is the Boe-Bot. The vehicle has two driving wheels
and the angular velocities of the two wheels are independently controlled (A. Pawlowski,
2009). When the vehicle is moving towards the target and the sensors detect an obstacle, an
avoiding strategy is necessary. The host system connects to the mobile robot with the
SunSPOT module. A remote control program has been implemented as shown on Figure 22.


Fig. 19. Boe-bot with SunSPOT mounted

6. Experimental results
The applications for WSNs are many and varied. They are used in commercial and
industrial applications to monitor data that would be difficult or expensive to monitor using
wired sensors. They could be deployed in wilderness areas, where they would remain for
many years (monitoring some environmental variable) without the need to
recharge/replace their power supplies. They could form a perimeter about a property and
monitor the progression of intruders (passing information from one node to the next). There
are a many uses for WSNs (I. Matijevics, 2009).

Fig. 20. Crops in greenhouse

Typical applications of WSNs include monitoring, tracking, and controlling. Some of the
specific applications are habitat monitoring, object tracking, nuclear reactor controlling, fire
detection, traffic monitoring, etc. In a typical application, a WSN is scattered in a region
where it is meant to collect data through its sensor node. Figure 23, shows the complete
control system of the greenhouse. The WSN-based controller has allowed a considerable
decrease in the number of changes in the control action and made possible a study of the
compromise between quantity of transmission and control performance.

Fig. 21. Capsicum

Motion control of mobile robots is a very important research field today, because mobile
robots are a very interesting subject both in scientific research and practical applications. In
this paper the object of the remote control is the Boe-Bot. The vehicle has two driving wheels
and the angular velocities of the two wheels are independently controlled (A. Pawlowski,
2009). When the vehicle is moving towards the target and the sensors detect an obstacle, an
avoiding strategy is necessary. The host system connects to the mobile robot with the
SunSPOT module. A remote control program has been implemented as shown on Figure 22.


Fig. 22. Control surface of the system

The limit of the level crossing sampling has presented a great influence on the event based
control performance where, for the greenhouse climate control problem, the system has
provided promising results.

Fig. 23. Greenhouse control system

The code snippet below gives an example for testing the communication devices in
broadcast mode as we can see on Figure 24. It is written in Java and runs on SunSPOT
modules. Each SPOT is assigned its own address and can broadcast or unicast to the other
SPOTs. This code is implemented for testing purposes only.

Fig. 24. Sending broadcast packets via wsn from base station

protected void startApp() throws MIDletStateChangeException {
System.out.println("Broadcast Counter MIDlet");
//showColor(color);
//switches[0].addISwitchListener(this);
try {
tx =
(RadiogramConnection)Connector.open("radiogram://broadcast:123");
xdg = (Radiogram)tx.newDatagram(20); //transmitting the
radiogram
RadiogramConnection rx =
(RadiogramConnection)Connector.open("radiogram://:123");
Radiogram rdg = (Radiogram)rx.newDatagram(20);
//outs[0].setHigh();
while (true) {
try {
rx.receive(rdg);
int cmd = rdg.readInt();
//int newCount = rdg.readInt();
//int newColor = rdg.readInt();
/*if (cmd == CHANGE_COLOR) {
System.out.println("Received packet from " +
rdg.getAddress());
//showColor(newColor);
} else {
//showCount(newCount, newColor);
}*/
switch (cmd){
case 0: outs[demo.H0].setLow();
outs[demo.H1].setLow(); leds[0].setRGB(200, 0, 0); leds[0].setOn();
leds[1].setOff();leds[2].setOff();leds[3].setOff(); break;
case 4: outs[demo.H0].setHigh();
outs[demo.H1].setHigh(); leds[2].setRGB(200, 0, 0); leds[2].setOn();
//setting up the diagnostic leds
default: leds[4].setRGB(200, 0, 0);
leds[4].setOn(); break;
}
} catch (IOException ex) {
System.out.println("Error receiving packet: " + ex);
ex.printStackTrace(); // Error detection
}
}
System.out.println("Error opening connections: " + ex);
}
}

Fig. 22. Control surface of the system

The limit of the level crossing sampling has presented a great influence on the event based
control performance where, for the greenhouse climate control problem, the system has
provided promising results.

Fig. 23. Greenhouse control system

The code snippet below gives an example for testing the communication devices in
broadcast mode as we can see on Figure 24. It is written in Java and runs on SunSPOT
modules. Each SPOT is assigned its own address and can broadcast or unicast to the other
SPOTs. This code is implemented for testing purposes only.

Fig. 24. Sending broadcast packets via wsn from base station

protected void startApp() throws MIDletStateChangeException {
System.out.println("Broadcast Counter MIDlet");
//showColor(color);
try {
tx =
(RadiogramConnection)Connector.open("radiogram://broadcast:123");
xdg = (Radiogram)tx.newDatagram(20); //transmitting the
radiogram
RadiogramConnection rx =
(RadiogramConnection)Connector.open("radiogram://:123");
Radiogram rdg = (Radiogram)rx.newDatagram(20);
//outs[0].setHigh();
while (true) {
try {
rx.receive(rdg);
int cmd = rdg.readInt();
//int newCount = rdg.readInt();
//int newColor = rdg.readInt();
/*if (cmd == CHANGE_COLOR) {
System.out.println("Received packet from " +
rdg.getAddress());
//showColor(newColor);
} else {
//showCount(newCount, newColor);
}*/
switch (cmd){
//setting up the diagnostic leds
default: leds[4].setRGB(200, 0, 0);
leds[4].setOn(); break;
}
System.out.println("Error receiving packet: " + ex);
}
}
System.out.println("Error opening connections: " + ex);
}
}

The Sun SPOT is a Java programmable embedded device designed for flexibility. The basic
unit includes accelerometer, temperature and light sensors, radio transmitter, eight
multicolored LEDs, 2 push-button control switches, 5 digital I/O pins, 6 analog inputs, 4
digital outputs, and a rechargeable battery. Java implementation and programming the Sun
SPOT is surprisingly easy. Experimental testing has demonstrated the validity of our
approach.

7. Comparison of the fruit production
Tomatoes are a warm season vegetable crop. They grow best under conditions of high light
and warm temperatures. Low light in a fall or winter greenhouse, when it is less than 15% of
summer light levels, greatly reduces fruit yield when heating costs are highest. For this
reason, it is difficult to recommend that a greenhouse operator should grow and harvest
fruit from December 15 to February 15. Based on few years of experience, tomato production
is most successful in the spring. Excellent light, moderate heating costs and good prices
annually demonstrate this is the best time for greenhouse tomato production. Tomato plants
grow best when the night temperature is maintained at 16 - 18 C. Temperatures below 16
C will prevent normal pollination and fruit development. In warm or hot outdoor
conditions, tomato greenhouses must be ventilated to keep temperatures below 35 C. High
temperatures not only effect the leaves and fruit, but increased soil temperatures also reduce
root growth. Table 1, gives an overview of effectiveness of the control system.

Tested plants Average weight
of fruit (with
WSN control)
Average weight
of fruit (without
WSN control)
Average
number of fruit
per plant (with
WSN control)
Average
number of
fruit per plant
(without WSN
control)
Tomato 210 g 180 g 17 11
Capsicum 135 g 110 g 15 12
Cucumber 70 g 60 g 13 10
Table 1. The average total weight and number of fruit harvested

Success in greenhouse plants depends completely on fruit yield. Yields of 20 25 % gain per
plant are very good for annual costs.

8. Conclusion
The system and its implementation have been successful, however there are still possibilities
for further development. The first cycle of plant development has just passed, and it has
provided numerous valuable data. For the next cycle will better conditions will be provided,
with a more experienced staff. With further developments the application of professional
industrial electronics will also have to be taken into consideration, which would
significantly decrease possible problems. One of the research areas inside the Greenhouse
program is wirelessly controlled mobile measuring station. Traditionally research into
autonomous robotics has been performed on robotics platforms that cost tens of thousands
of dollars. As an alternative, several research groups have developed low-cost robots that

are controlled by a SunSPOT node running Java VM. One of the goals of this project is to
develop algorithms for coordination and navigation inside the greenhouse.

9. Acknowledgement
This research was partially supported by the TMOP 4.2.1/B-09/1/KONV-2010-0005
program of the Hungarian National Development Agency.

10. References
A. Pawlowski, J. Luis Guzman, F. Rodrguez, M. Berenguel, J. Snchez and S. Dormido,
(2009) Simulation of Greenhouse Climate Monitoring and Control with Wireless
Sensor Network and Event-Based Control Proceedings of the Conference
Gy. Mester, (2009) Wireless Sensor-based Control of Mobile Robot Motion, Proceeding of
the IEEE SISY 2009, pp 81-84, Subotica, Serbia
Gy. Mester, (2009) Intelligent Wheeled Mobile Robot Navigation, Proceedings of the
Conference Eurpai Kihvsok V, pp. 1-5, SZTE, Szeged, Hungary
I. Matijevics, J. Simon, (2008) Advantages of Remote Greenhouse Laboratory for Distant
Monitoring, Proceedings of the Conference ICoSTAF 2008, pp 1-5, Szeged,
Hungary
I. Matijevics, J. Simon, (2009) Comparison of various wireless sensor networks and their
implementation, Proceedings of the Conference SIP 2009, pp 1-3, Pcs, Hungary
J. Gosling, (2005) The Java Language Specification Third Edition
J. Simon, G. Martinovi, (2009) Web Based Distant Monitoring and Control for Greenhouse
Systems Using the Sun SPOT Modules, Proceedings of the Conference SISY 2009,
pp. 1-5, Subotica, Serbia
J. Simon, I. Matijevics, (2009) Distant Monitoring And Control For Greenhouse Systems
Via Internet, Zbornik radova konferencije Yuinfo 2009, pp. 1-3, Kopaonik, Srbija
J. Vasu, L. Shahram, (2008) Comprehensive Study of Routing Management in Wireless
Sensor Networks- Part-1
L. Gonda, , C. Cugnasca, (2006) A proposal of greenhouse control using wireless sensor
networks In Proceedings of 4thWorld Congress Conference on Computers in
Agriculture and Natural Resources, Orlando, Florida, USA
M. J. Matari, (2007) The Robotics Primer. The MIT Press, 1st edition.
M. A. Goodrich, (2002) Potential fields tutorial.
O. Khatib, (1986) "Real-Time Obstacle Avoidance for Manipulators and Mobile Robots", Int.
J. of Robotic Research, Vol.5, No.1, p.60
O. Khatib, (1985) The Potential Field Approach and Operational Space Formulation in
Robot Control Proc. Fourth Yale Workshop on Applications of Adaptive Systems
Theory, Yale University, New Haven, Connecticut, pp. 208-214.
P. Kucsera, (2006) Sensors For Mobile Robot Systems , Academic and Applied Research in
Military Science, Volume 5, Issue 4, p.645-658,Hungary
P. Kucsera, (2006) Industrial Component-based Sample Mobile Robot System , Acta
Polytechnica Hungarica, Volume 4 Issue Number 4, Hungary
Roland Siegwart and Illah R., (2004) Introduction to Autonomous Mobile Robots,
Nourbakhsh

The Sun SPOT is a Java programmable embedded device designed for flexibility. The basic
unit includes accelerometer, temperature and light sensors, radio transmitter, eight
multicolored LEDs, 2 push-button control switches, 5 digital I/O pins, 6 analog inputs, 4
digital outputs, and a rechargeable battery. Java implementation and programming the Sun
SPOT is surprisingly easy. Experimental testing has demonstrated the validity of our
approach.

7. Comparison of the fruit production
Tomatoes are a warm season vegetable crop. They grow best under conditions of high light
and warm temperatures. Low light in a fall or winter greenhouse, when it is less than 15% of
summer light levels, greatly reduces fruit yield when heating costs are highest. For this
reason, it is difficult to recommend that a greenhouse operator should grow and harvest
fruit from December 15 to February 15. Based on few years of experience, tomato production
is most successful in the spring. Excellent light, moderate heating costs and good prices
annually demonstrate this is the best time for greenhouse tomato production. Tomato plants
grow best when the night temperature is maintained at 16 - 18 C. Temperatures below 16
C will prevent normal pollination and fruit development. In warm or hot outdoor
conditions, tomato greenhouses must be ventilated to keep temperatures below 35 C. High
temperatures not only effect the leaves and fruit, but increased soil temperatures also reduce
root growth. Table 1, gives an overview of effectiveness of the control system.

Tested plants Average weight
of fruit (with
WSN control)
Average weight
of fruit (without
WSN control)
Average
number of fruit
per plant (with
WSN control)
Average
number of
fruit per plant
(without WSN
control)
Tomato 210 g 180 g 17 11
Capsicum 135 g 110 g 15 12
Cucumber 70 g 60 g 13 10
Table 1. The average total weight and number of fruit harvested

Success in greenhouse plants depends completely on fruit yield. Yields of 20 25 % gain per
plant are very good for annual costs.

8. Conclusion
The system and its implementation have been successful, however there are still possibilities
for further development. The first cycle of plant development has just passed, and it has
provided numerous valuable data. For the next cycle will better conditions will be provided,
with a more experienced staff. With further developments the application of professional
industrial electronics will also have to be taken into consideration, which would
significantly decrease possible problems. One of the research areas inside the Greenhouse
program is wirelessly controlled mobile measuring station. Traditionally research into
autonomous robotics has been performed on robotics platforms that cost tens of thousands
of dollars. As an alternative, several research groups have developed low-cost robots that

are controlled by a SunSPOT node running Java VM. One of the goals of this project is to
develop algorithms for coordination and navigation inside the greenhouse.

9. Acknowledgement
This research was partially supported by the TMOP 4.2.1/B-09/1/KONV-2010-0005
program of the Hungarian National Development Agency.

10. References
A. Pawlowski, J. Luis Guzman, F. Rodrguez, M. Berenguel, J. Snchez and S. Dormido,
(2009) Simulation of Greenhouse Climate Monitoring and Control with Wireless
Sensor Network and Event-Based Control Proceedings of the Conference
Gy. Mester, (2009) Wireless Sensor-based Control of Mobile Robot Motion, Proceeding of
the IEEE SISY 2009, pp 81-84, Subotica, Serbia
Gy. Mester, (2009) Intelligent Wheeled Mobile Robot Navigation, Proceedings of the
Conference Eurpai Kihvsok V, pp. 1-5, SZTE, Szeged, Hungary
I. Matijevics, J. Simon, (2008) Advantages of Remote Greenhouse Laboratory for Distant
Monitoring, Proceedings of the Conference ICoSTAF 2008, pp 1-5, Szeged,
Hungary
I. Matijevics, J. Simon, (2009) Comparison of various wireless sensor networks and their
implementation, Proceedings of the Conference SIP 2009, pp 1-3, Pcs, Hungary
J. Gosling, (2005) The Java Language Specification Third Edition
J. Simon, G. Martinovi, (2009) Web Based Distant Monitoring and Control for Greenhouse
Systems Using the Sun SPOT Modules, Proceedings of the Conference SISY 2009,
pp. 1-5, Subotica, Serbia
J. Simon, I. Matijevics, (2009) Distant Monitoring And Control For Greenhouse Systems
Via Internet, Zbornik radova konferencije Yuinfo 2009, pp. 1-3, Kopaonik, Srbija
J. Vasu, L. Shahram, (2008) Comprehensive Study of Routing Management in Wireless
Sensor Networks- Part-1
L. Gonda, , C. Cugnasca, (2006) A proposal of greenhouse control using wireless sensor
networks In Proceedings of 4thWorld Congress Conference on Computers in
Agriculture and Natural Resources, Orlando, Florida, USA
M. J. Matari, (2007) The Robotics Primer. The MIT Press, 1st edition.
M. A. Goodrich, (2002) Potential fields tutorial.
O. Khatib, (1986) "Real-Time Obstacle Avoidance for Manipulators and Mobile Robots", Int.
J. of Robotic Research, Vol.5, No.1, p.60
O. Khatib, (1985) The Potential Field Approach and Operational Space Formulation in
Robot Control Proc. Fourth Yale Workshop on Applications of Adaptive Systems
Theory, Yale University, New Haven, Connecticut, pp. 208-214.
P. Kucsera, (2006) Sensors For Mobile Robot Systems , Academic and Applied Research in
Military Science, Volume 5, Issue 4, p.645-658,Hungary
P. Kucsera, (2006) Industrial Component-based Sample Mobile Robot System , Acta
Polytechnica Hungarica, Volume 4 Issue Number 4, Hungary
Roland Siegwart and Illah R., (2004) Introduction to Autonomous Mobile Robots,
Nourbakhsh

Sun Microsystems Inc. , (2007) Sun Small Programmable Object Technology (Sun SPOT)
Owners Manual Release 3.0
Sun Microsystems Inc. , (2005) Sun Spot Developers Guide
Sun Microsystems Inc. , (2005) Demo Sensor Board Library
S. Scaglia, (2008) The Embedded Internet
X. Feng, T. Yu-Chu, S. Yanjun, S. Youxian, (2007) Wireless Sensor/Actuator Network
Design for Mobile Control Applications. Sensors Proceedings of the Conference

Model Based WSN System Implementations Using
PN-WSNA for Aquarium Environment Control in a House 109
Model Based WSN System Implementations Using PN-WSNA for
Aquarium Environment Control in a House
X

Model Based WSN System
Implementations Using PN-WSNA for
Aquarium Environment Control in a House

Department of Electrical Engineering
National Taiwan University of Science and Technology
Taiwan

1. Introduction
Ubiquitous computing architectures are implemented for cognitive sensor networks.
Wireless sensor networks cooperating with cognitive science and artificial intelligence are
used to develop cognitive sensor networks (Shenai et al., 2008). Therefore, a cognitive sensor
network is generally represented as a closed loop control system (Ruiz et al., 2008), where
the feedback data is collected from remote sensor nodes. At the same time, the control
approaches are desired to deal with regulations of desired operation scenarios and sensor
feedbacks.

Wireless sensor networks (WSN) (Romer et al., 2004; Akyildiz et al., 2008) are developed
using autonomous sensor nodes (Dalola et al., 2009) to collect remote sensor data for
decision systems with low power consumptions and failure tolerable mechanisms. In
general, the WSN system can be applied to the factory automation (Zhuang et al., 2008),
intelligent diagnosis (Zhuang et al., 2008), intelligent monitoring and control systems
(Sridhar et al., 2007), smart home (Suh et al., 2008), etc. Practically, the challenging issues for
developing WSN systems are large amounts of coding and program maintenance efforts for
various sensor oriented applications as well as interdisciplinary integrations of domain
engineers and WSN engineers.

For the first issue, diverse control and decision scenarios in a WSN system are developed for
different sensor nodes. The coding and maintenance for large scale WSN systems would be
huge challenges. The second issue is the problems of interdisciplinary integrations. Coding
for sensor nodes is a challenge for domain engineers who are not familiar with
programming. According to aforementioned challenging issues, model based system
implementation approaches are proposed to eliminate the efforts for programming native
codes with cross compilers.

7

In order to perform model based implementation approaches, several discrete event
dynamic system (DEDS) modeling approaches are surveyed, such as finite state machine
(FSM) (Avnur et al., 1990), unified modeling language (UML) (Manasseh et al., 2010) and
Petri net (PN) (Murata et al., 1989; Kuo et al., 2009). The Petri net (Murata et al., 1989) was
proposed by C.A. Petri. A PN model may model the system using events and conditions.
Events are represented as transitions; conditions are represented as places. Arcs are used to
describe pre- and post-conditions between places and transitions. In general, an
autonomous sensor node can be also described as conditions, events, and their relationships.
Sensor events are generated in terms of the changes of sensor conditions.

Although the PN is suitable for modeling a WSN, the ordinary PN is not applicable due to
the lack of interfaces and intercommunications. Therefore, the Petri net based wireless
sensor node architecture (PN-WSNA) (Kuo et al., 2009) is selected in this book chapter to
model an aquarium environment control in a house. Interface functions are desired for
collecting sensor data and controlling actuators. Intercommunication functions provide
wireless data exchanges among different autonomous sensor nodes. In our approach, the
PN-WSNA system is composed of a PN-WSNA kernel program and a PN-WSNA
management program. The PN-WSNA kernel program is developed as an inference engine
which is implemented inside the sensor node. The PN-WSNA kernel program is responsible
of receiving and interpreting PN-WSNA models, collecting sensor data from analog and
digital channels, intercommunication between sensor nodes, PN model inference, decision
making, and controlling actuators.

In this book chapter, the Petri net based wireless sensor node architecture (PN-WSNA) (Kuo
et al., 2009) is used to construct an aquarium environment control system in a house. This
aquarium environment control system demonstrates the modelling and implementation
procedures for two PN-WSNA sensor node systems, where one sensor node is deployed for
aquarium environment control and the other one is desired for entrance counting system.
The entrance counting system counts the people in a house. The aquarium environment
control system acquires the data from temperature sensor and dissolved oxygen sensor as
well as the people number collected from the entrance counting system. Meanwhile, the
light, heater and pump are also activated using the sensor node. As a consequence, the
aquarium environment control system is capable of autonomously controlling the
temperature and dissolved oxygen concentration in a desired condition. In addition, the
light can also be controlled in terms of the presence of people in a house.

2. PN-WSNA Definitions
In this book chapter, the PN-WSNA is developed by inheriting the definitions of the
ordinary PN. In order to deal with real-time sensor data acquisitions, intercommunications
and actuator controls, additional interface and intercommunication places are defined based
upon the ordinary PN for practical cognitive sensor network applications. In addition to the
interface and communication places, periodic executions of the system are also defined
using timed transitions. The sensor data is further categorized as high enable and low enable
situations. Therefore, the PN-WSNA structure is a eleven-tuple, PN-WSNA = (P, P
s
, P
ci
, P
co
,
P
a
, T
0
, T
t
, T
H
, T
L
, I, O) structure; where P is a finite set of normal places; P
s
is a finite set of

sensor places; P
ci
is a finite set of receiver places; P
co
is a finite set of transmitter places; P
a
is
a finite set of actuation places; T
0
is a finite set of immediate transitions; T
t
is a finite set of
timed transitions; T
H
is a finite set of high-enable transitions; T
L
is a finite set of low-enable
transitions; I is the input function; and O is the output functions. The PN-WSNA graphical
definitions are shown in Fig. 1. The PN-WSNA definitions are further elaborated as follows.

Normal
Place
Sensor
Place
Receiver
Place
Transmitter
Place
Actuation
Place
S Ci Co
A
Immediate
Transition
Timed
Transition
High-Enable
Transition
Low-Enable
Transition
H
L
Token
Arc

Fig. 1. PN-WSNA graphical definitions.

I. Place: P = {p1,p2,p3,...,pn}: P is a finite set of places, n1, and it is denoted a circle. Places
of the PN-WSNA are refined as normal places, sensor places, receiver places, transmitter
places, and actuation places. Brief introduction is defined as follows. Detailed definitions
may refer to (Kuo et al., 2009).
a. Normal places: the definition of a normal place is the same as the place defined in the
ordinary PN. Tokens in normal places may represent the corresponding status,
condition, command, etc.
b. Sensor places: A sensor place is desired for data collections. For a PN-WSNA, the
sensor place may collect sensor signals in terms of analog value (0 3 V), binary digits
(0 and 1), or serial communication packets (0 255) manners, and the sensor data of a
place (pi) is denoted as (pi). A sensor interface is required to be corresponded to an
analog-digital-converter (ADC) address, a generalized input-output (GIO) address or
the universal asynchronous receiver /transmitter (UART). Because the PN-WSNA
does not define the color token (Kuo et al., 2003), the sensor status is eventually
represented as high or low status. Hence, a threshold value is defined for the
sensor place to divide the analog value into high or low status, and the threshold
value of a place (pi) is denoted as (pi).
c. Receiver and transmitter places: With the PN-WSNA, a receiver place and a transmitter
place can be combined as a communication pair, and they appear in different sensor
node models. Hence, the transmitter place is a sink place (Kuo et al., 2009); and the
receiver place is a source place (Kuo et al., 2009).
d. Actuation places: The actuation place play similar roles to transmitter places; however,
the token in an actuation place are converted as actuation signals to control peripheral
devices. As a consequence, an actuation place is a sink place, any token in an
actuation place may directly control peripheral devices and then the actuation place
releases this token.

In order to perform model based implementation approaches, several discrete event
dynamic system (DEDS) modeling approaches are surveyed, such as finite state machine
(FSM) (Avnur et al., 1990), unified modeling language (UML) (Manasseh et al., 2010) and
Petri net (PN) (Murata et al., 1989; Kuo et al., 2009). The Petri net (Murata et al., 1989) was
proposed by C.A. Petri. A PN model may model the system using events and conditions.
Events are represented as transitions; conditions are represented as places. Arcs are used to
describe pre- and post-conditions between places and transitions. In general, an
autonomous sensor node can be also described as conditions, events, and their relationships.
Sensor events are generated in terms of the changes of sensor conditions.

Although the PN is suitable for modeling a WSN, the ordinary PN is not applicable due to
the lack of interfaces and intercommunications. Therefore, the Petri net based wireless
sensor node architecture (PN-WSNA) (Kuo et al., 2009) is selected in this book chapter to
model an aquarium environment control in a house. Interface functions are desired for
collecting sensor data and controlling actuators. Intercommunication functions provide
wireless data exchanges among different autonomous sensor nodes. In our approach, the
PN-WSNA system is composed of a PN-WSNA kernel program and a PN-WSNA
management program. The PN-WSNA kernel program is developed as an inference engine
which is implemented inside the sensor node. The PN-WSNA kernel program is responsible
of receiving and interpreting PN-WSNA models, collecting sensor data from analog and
digital channels, intercommunication between sensor nodes, PN model inference, decision
making, and controlling actuators.

In this book chapter, the Petri net based wireless sensor node architecture (PN-WSNA) (Kuo
et al., 2009) is used to construct an aquarium environment control system in a house. This
aquarium environment control system demonstrates the modelling and implementation
procedures for two PN-WSNA sensor node systems, where one sensor node is deployed for
aquarium environment control and the other one is desired for entrance counting system.
The entrance counting system counts the people in a house. The aquarium environment
control system acquires the data from temperature sensor and dissolved oxygen sensor as
well as the people number collected from the entrance counting system. Meanwhile, the
light, heater and pump are also activated using the sensor node. As a consequence, the
aquarium environment control system is capable of autonomously controlling the
temperature and dissolved oxygen concentration in a desired condition. In addition, the
light can also be controlled in terms of the presence of people in a house.

2. PN-WSNA Definitions
In this book chapter, the PN-WSNA is developed by inheriting the definitions of the
ordinary PN. In order to deal with real-time sensor data acquisitions, intercommunications
and actuator controls, additional interface and intercommunication places are defined based
upon the ordinary PN for practical cognitive sensor network applications. In addition to the
interface and communication places, periodic executions of the system are also defined
using timed transitions. The sensor data is further categorized as high enable and low enable
situations. Therefore, the PN-WSNA structure is a eleven-tuple, PN-WSNA = (P, P
s
, P
ci
, P
co
,
P
a
, T
0
, T
t
, T
H
, T
L
, I, O) structure; where P is a finite set of normal places; P
s
is a finite set of

sensor places; P
ci
is a finite set of receiver places; P
co
is a finite set of transmitter places; P
a
is
a finite set of actuation places; T
0
is a finite set of immediate transitions; T
t
is a finite set of
timed transitions; T
H
is a finite set of high-enable transitions; T
L
is a finite set of low-enable
transitions; I is the input function; and O is the output functions. The PN-WSNA graphical
definitions are shown in Fig. 1. The PN-WSNA definitions are further elaborated as follows.

Normal
Place
Sensor
Place
Receiver
Place
Transmitter
Place
Actuation
Place
S Ci Co
A
Immediate
Transition
Timed
Transition
High-Enable
Transition
Low-Enable
Transition
H
L
Token
Arc

Fig. 1. PN-WSNA graphical definitions.

I. Place: P = {p1,p2,p3,...,pn}: P is a finite set of places, n1, and it is denoted a circle. Places
of the PN-WSNA are refined as normal places, sensor places, receiver places, transmitter
places, and actuation places. Brief introduction is defined as follows. Detailed definitions
may refer to (Kuo et al., 2009).
a. Normal places: the definition of a normal place is the same as the place defined in the
ordinary PN. Tokens in normal places may represent the corresponding status,
condition, command, etc.
b. Sensor places: A sensor place is desired for data collections. For a PN-WSNA, the
sensor place may collect sensor signals in terms of analog value (0 3 V), binary digits
(0 and 1), or serial communication packets (0 255) manners, and the sensor data of a
place (pi) is denoted as (pi). A sensor interface is required to be corresponded to an
analog-digital-converter (ADC) address, a generalized input-output (GIO) address or
the universal asynchronous receiver /transmitter (UART). Because the PN-WSNA
does not define the color token (Kuo et al., 2003), the sensor status is eventually
represented as high or low status. Hence, a threshold value is defined for the
sensor place to divide the analog value into high or low status, and the threshold
value of a place (pi) is denoted as (pi).
c. Receiver and transmitter places: With the PN-WSNA, a receiver place and a transmitter
place can be combined as a communication pair, and they appear in different sensor
node models. Hence, the transmitter place is a sink place (Kuo et al., 2009); and the
receiver place is a source place (Kuo et al., 2009).
d. Actuation places: The actuation place play similar roles to transmitter places; however,
the token in an actuation place are converted as actuation signals to control peripheral
devices. As a consequence, an actuation place is a sink place, any token in an
actuation place may directly control peripheral devices and then the actuation place
releases this token.

II. Transitions: T = {t1,t2,t3,...,tm}: T is a finite set of transitions, m1, and it is denoted a bar.
Transitions of the PN-WSNA are further refined as immediate transitions, timed
transitions, high-enable transitions, and low-enable transitions. Detailed definitions are
illustrated as below:
a. Immediate transitions: the definition of an immediate transition is the same as the
transition defined in the ordinary PN, and it can be used to model events and decisions.
b. Timed transitions: the definition of a timed transition is similar to the transition
defined in the ordinary PN; however, tokens in the input places of a timed transition
do not deliver to its output places directly. Instead, a fired transition keeps these
tokens until a predefined elapsed time is expired. Therefore, an elapsed time factor is
further defined for the timed transition.
c. High-enable and low-enable transitions: high-enable and low-enable transitions are
defined for sensor places. Basically, high-enable and low-enable transitions serve as
output transitions of a sensor place. They must be appeared in a pair configuration;
hence conflicts of these transitions are happened. The firing of conflict high-enable
and low-enable transitions depends on the sensor data and threshold value defined in
the input sensor place. A high-enable transition is fired when the sensor data is
greater than or equal to the threshold value defined in the input sensor place; and a
low-enable transition is fired when the sensor data is less than the threshold value
defined in the input sensor place.

III. In a PN-WSNA model, the places and transitions follow the rules of PT, and PT .

IV. Token, marking and initial marking: tokens are quantitative representations of bag set in
places. The marking is denoted as , which represents the token distributions in all
places of a PN-WSNA model. is a q 1 column vector, the j-th element of indicates
the number of tokens in place j. Note that q is a nonnegative integer, and it is equal to the
number of places in a PN-WSNA model. The initial marking (
0
) is defined for the
marking of system startup.

VI. Input, output functions, enabling and firing: input and output functions are defined via
directed arcs graphically, and they are represented as I(pi,tj)Ni,j and O(pr,ts) Nr,s,
respectively. Ni,j and Nr,s are nonnegative integers, and they defines the pre- and post-
conditions of the PN-WSNA models. In this study, directed and inhibited functions are
further defined. A transition (tj) is said to be enabled when (1) satisfies.

, ,
1 1
( ( , ) ) 0 ( ( , ) ) 0
k k
i j i j i j i j
i i Directed Inhibited
I p t N and I p t N

(1)
where i = 1 to k, and k equals the number of input places of tj; pi
input places of tj
with directed arcs.
At the same time, an enabled transition is not necessarily to be fired because of the conflict
situations. The conflict exists when the number of enabled transitions for a place is greater
than unity. With a conflict situation, only one of the enabled transitions can be fired. For
the PN-WSNA, conflict transitions are resolved in terms of the following approaches.

a. Immediate and timed transitions: Random selections of an enabled and conflict
transitions are desired for immediate and timed transitions because of identical token
and transition characteristics.
b. High-enable and low-enable transitions: to resolve the conflict situations of a pair of
high-enable and low-enable transitions, the sensor data, (pi), and threshold value,
(pi), are evaluated for the same input place (pi). A high-enable transition is fired if (2)
satisfies.
(pi) (pi) (2)

Meanwhile, a low-enable transition is fired if (3) satisfies.

(pi) < (pi) (3)

3. PN-WSNA Based Aquarium Environment Control System
3.1 System Descriptions
To verify the proposed PN-WSNA approaches, an aquarium environment control system is
implemented. Fig. 2 shows the facilities used in this experiment. An aquarium is the major
environment for this study. A dissolved oxygen meter (with type: DO-5510 from Lutron Co.
Ltd.) is used for measuring the dissolved oxygen concentration and the temperature as well
in the water. In addition, two infrared human motion detection sensors are used for
detecting the entry and exit of visitors. In case of insufficient dissolved oxygen concentration
in the water, a pump is activated for increasing the dissolved oxygen. The pump stops when
the dissolved oxygen concentration satisfies the setting conditions. On the other hand, in
case of low temperature in the water, a heater is also activated for increasing the
temperature in the water. Similarly, the heater stops when the water temperature satisfies
the setting conditions. It is noted that hysteresis ranges are desired for the activation and
termination conditions with respect to their threshold values.

Aquarium System
Heater Pump Dissolved Oxygen & Temperature Instrument
Door
Infra-ray Human
Body Sensor A
Infra-ray Human
Body Sensor B

Fig. 2. PN-WSNA model construction architecture.

II. Transitions: T = {t1,t2,t3,...,tm}: T is a finite set of transitions, m1, and it is denoted a bar.
Transitions of the PN-WSNA are further refined as immediate transitions, timed
transitions, high-enable transitions, and low-enable transitions. Detailed definitions are
illustrated as below:
a. Immediate transitions: the definition of an immediate transition is the same as the
transition defined in the ordinary PN, and it can be used to model events and decisions.
b. Timed transitions: the definition of a timed transition is similar to the transition
defined in the ordinary PN; however, tokens in the input places of a timed transition
do not deliver to its output places directly. Instead, a fired transition keeps these
tokens until a predefined elapsed time is expired. Therefore, an elapsed time factor is
further defined for the timed transition.
c. High-enable and low-enable transitions: high-enable and low-enable transitions are
defined for sensor places. Basically, high-enable and low-enable transitions serve as
output transitions of a sensor place. They must be appeared in a pair configuration;
hence conflicts of these transitions are happened. The firing of conflict high-enable
and low-enable transitions depends on the sensor data and threshold value defined in
the input sensor place. A high-enable transition is fired when the sensor data is
greater than or equal to the threshold value defined in the input sensor place; and a
low-enable transition is fired when the sensor data is less than the threshold value
defined in the input sensor place.

III. In a PN-WSNA model, the places and transitions follow the rules of PT, and PT .

IV. Token, marking and initial marking: tokens are quantitative representations of bag set in
places. The marking is denoted as , which represents the token distributions in all
places of a PN-WSNA model. is a q 1 column vector, the j-th element of indicates
the number of tokens in place j. Note that q is a nonnegative integer, and it is equal to the
number of places in a PN-WSNA model. The initial marking (
0
) is defined for the
marking of system startup.

VI. Input, output functions, enabling and firing: input and output functions are defined via
directed arcs graphically, and they are represented as I(pi,tj)Ni,j and O(pr,ts) Nr,s,
respectively. Ni,j and Nr,s are nonnegative integers, and they defines the pre- and post-
conditions of the PN-WSNA models. In this study, directed and inhibited functions are
further defined. A transition (tj) is said to be enabled when (1) satisfies.

, ,
1 1
( ( , ) ) 0 ( ( , ) ) 0
k k
i j i j i j i j
i i Directed Inhibited
I p t N and I p t N

(1)
where i = 1 to k, and k equals the number of input places of tj; pi
input places of tj
with directed arcs.
At the same time, an enabled transition is not necessarily to be fired because of the conflict
situations. The conflict exists when the number of enabled transitions for a place is greater
than unity. With a conflict situation, only one of the enabled transitions can be fired. For
the PN-WSNA, conflict transitions are resolved in terms of the following approaches.

a. Immediate and timed transitions: Random selections of an enabled and conflict
transitions are desired for immediate and timed transitions because of identical token
and transition characteristics.
b. High-enable and low-enable transitions: to resolve the conflict situations of a pair of
high-enable and low-enable transitions, the sensor data, (pi), and threshold value,
(pi), are evaluated for the same input place (pi). A high-enable transition is fired if (2)
satisfies.
(pi) (pi) (2)

Meanwhile, a low-enable transition is fired if (3) satisfies.

(pi) < (pi) (3)

3. PN-WSNA Based Aquarium Environment Control System
3.1 System Descriptions
To verify the proposed PN-WSNA approaches, an aquarium environment control system is
implemented. Fig. 2 shows the facilities used in this experiment. An aquarium is the major
environment for this study. A dissolved oxygen meter (with type: DO-5510 from Lutron Co.
Ltd.) is used for measuring the dissolved oxygen concentration and the temperature as well
in the water. In addition, two infrared human motion detection sensors are used for
detecting the entry and exit of visitors. In case of insufficient dissolved oxygen concentration
in the water, a pump is activated for increasing the dissolved oxygen. The pump stops when
the dissolved oxygen concentration satisfies the setting conditions. On the other hand, in
case of low temperature in the water, a heater is also activated for increasing the
temperature in the water. Similarly, the heater stops when the water temperature satisfies
the setting conditions. It is noted that hysteresis ranges are desired for the activation and
termination conditions with respect to their threshold values.

Aquarium System
Heater Pump Dissolved Oxygen & Temperature Instrument
Door
Infra-ray Human
Body Sensor A
Infra-ray Human
Body Sensor B


Two sensor nodes are cooperated to control the proposed aquarium system. PN-WSNA
models are implanted inside two sensor nodes to autonomously control aquarium
environment system. Fig. 3 shows the architecture this system, respectively. The functions in
our system are elaborated as follows.

1. Dissolved oxygen control: The concentration of dissolved oxygen is important for
aquarium environment. In general, adequate concentrations of dissolved oxygen
depend on the species of fishes. In our system, concentrations of dissolved oxygen are
desired for our experiment with 4.5 mg/l. The value of concentration of dissolved
oxygen is collected from dissolved oxygen sensor. This sensor can transmit the packets
of concentration of dissolved oxygen and temperature via RS-232. In here, an AVR
micro-controller is used to collect the data, and then converted it into analog signals
(DAC) to meet the interface requirements of the PN-WSNA (ADC). The signal would be
transmitted in to Mote-1 via ADC port. If the sensor value is less than 4.5 mg/l, the
pump will switch on for increasing concentration of dissolved oxygen. On contrary, if
the sensor value is greater than 5.0 mg/l (0.5 mg/l hysteresis range), pump will switch
off. The control scenario is shown in Fig. 4.
2. Temperature control: The working process of temperature is similar to concentration of
dissolved oxygen. The lower threshold of temperature is 25C; and the upper threshold
of temperature is 27C. The corresponding action is used to turn on /off the heater. The
control scenario is also shown in Fig. 4.
3. Light control: Light control in our system depends on the number of visitors in the
house. Counting the number of visitors is realized by comparing the rising edges of two
Infra-ray human detection sensors. The control scenario is also shown in Fig. 4.

1
RF
Mote
PN-WSNA Kernel
2
RF
Mote
PN-WSNA Kernel
: UART
: ADC and DI
: DAC and DO
Mote Interfaces
Power
Circuit
Pump
Light
AC
110V
IR
sensor
T O
H
e
a
t
e
r
Note:
O: Dissolve oxygen sensor
T : Temperature sensor
PN-WSNA-1
PN-WSNA-2A
PN-WSNA-2B
PN-WSNA-2C
PN-WSNA-2

Fig. 3. System architecture of the proposed aquarium system.

Dissolved oxygen
control
Temperature
control
Light
control
Dissolved oxygen
concentration
D.O.
< 4.5
mg/l
D.O.
> 5.0
mg/l
Temperature
Temp.
< 25C
Temp.
> 27 C
Visitors
Visitors in
the house
No
visitor
in the
house
Pump
On
Pump
Off
Heater
On
Heater
Off
Light
On
Light
Off

Fig. 4. System operation control scenario.

3.2 PN-WSNA Integrated Development Environment
In order to construct the PN-WSNA models, an integrated development environment (IDE)
for constructing the PN-WSNA model is developed. The WSN developer may construct
their domain-based PN-WSNA models by using the IDE, and then simulate the PN-WSNA
models using the IDE to verify their models before these models are deployed. The PN-
WSNA system is composed of the PN-WSNA kernel program and a PN-WSNA
management server. Fig. 5 shows the PN-WSNA system implementation architecture. The
management server is composed of an IDE which provides a graphical user interface for the
domain engineers to construct or modify their PN-WSNA models. At the same time, the
databases are also constructed for recording the PN-WSNA models and route tables of
sensor nodes. The route tables are used to explore a specific route path for delivering PN-
WSNA models to a remote mote in terms of wireless media.

On the other hand, the kernel program is implanted inside a sensor node. In addition, radio
frequency (RF) interface with Zigbee protocol (IEEE 802.15.4) as well as physically
connected interfaces of UART, ADC, DCA, digital input (DI) and digital output (DO) are
also available for cognitive sensing and controls. It is noted that the PN-WSNA IDE is
realized using the Microsoft visual C++ and the nesC (Avvenuti et al., 2007) program is used
to implement the kernel program.

The PN-WSNA IDE is a plug-and-play model construction environment. Fig. 6 shows the
proposed IDE, and the toolbar icons are used for model constructions, editing, revisions,
manipulations, run-time simulations, model drawing auxiliaries as well as model deliveries.
Detailed descriptions of the tool bars and algorithms for implementing the PN-WSNA
inference engine are referred to (Kuo et al., 2009). Finally, the interface functions are coded
within the kernel program. Fig. 7 shows the I/O, ADC, DAC and communication interfaces
of PN-WSNA motes. The mote is capable of collecting sensor status and actuating the
actuators using the sensor interface. At the same time, PN-WSNA motes may also

Two sensor nodes are cooperated to control the proposed aquarium system. PN-WSNA
models are implanted inside two sensor nodes to autonomously control aquarium
environment system. Fig. 3 shows the architecture this system, respectively. The functions in
our system are elaborated as follows.

1. Dissolved oxygen control: The concentration of dissolved oxygen is important for
aquarium environment. In general, adequate concentrations of dissolved oxygen
depend on the species of fishes. In our system, concentrations of dissolved oxygen are
desired for our experiment with 4.5 mg/l. The value of concentration of dissolved
oxygen is collected from dissolved oxygen sensor. This sensor can transmit the packets
of concentration of dissolved oxygen and temperature via RS-232. In here, an AVR
micro-controller is used to collect the data, and then converted it into analog signals
(DAC) to meet the interface requirements of the PN-WSNA (ADC). The signal would be
transmitted in to Mote-1 via ADC port. If the sensor value is less than 4.5 mg/l, the
pump will switch on for increasing concentration of dissolved oxygen. On contrary, if
the sensor value is greater than 5.0 mg/l (0.5 mg/l hysteresis range), pump will switch
off. The control scenario is shown in Fig. 4.
2. Temperature control: The working process of temperature is similar to concentration of
dissolved oxygen. The lower threshold of temperature is 25C; and the upper threshold
of temperature is 27C. The corresponding action is used to turn on /off the heater. The
control scenario is also shown in Fig. 4.
3. Light control: Light control in our system depends on the number of visitors in the
house. Counting the number of visitors is realized by comparing the rising edges of two
Infra-ray human detection sensors. The control scenario is also shown in Fig. 4.

1
RF
Mote
PN-WSNA Kernel
2
RF
Mote
PN-WSNA Kernel
: UART
: ADC and DI
: DAC and DO
Mote Interfaces
Power
Circuit
Pump
Light
AC
110V
IR
sensor
T O
H
e
a
t
e
r
Note:
O: Dissolve oxygen sensor
T : Temperature sensor
PN-WSNA-1
PN-WSNA-2A
PN-WSNA-2B
PN-WSNA-2C
PN-WSNA-2

Fig. 3. System architecture of the proposed aquarium system.

Dissolved oxygen
control
Temperature
control
Light
control
Dissolved oxygen
concentration
D.O.
< 4.5
mg/l
D.O.
> 5.0
mg/l
Temperature
Temp.
< 25C
Temp.
> 27 C
Visitors
Visitors in
the house
No
visitor
in the
house
Pump
On
Pump
Off
Heater
On
Heater
Off
Light
On
Light
Off

Fig. 4. System operation control scenario.

3.2 PN-WSNA Integrated Development Environment
In order to construct the PN-WSNA models, an integrated development environment (IDE)
for constructing the PN-WSNA model is developed. The WSN developer may construct
their domain-based PN-WSNA models by using the IDE, and then simulate the PN-WSNA
models using the IDE to verify their models before these models are deployed. The PN-
WSNA system is composed of the PN-WSNA kernel program and a PN-WSNA
management server. Fig. 5 shows the PN-WSNA system implementation architecture. The
management server is composed of an IDE which provides a graphical user interface for the
domain engineers to construct or modify their PN-WSNA models. At the same time, the
databases are also constructed for recording the PN-WSNA models and route tables of
sensor nodes. The route tables are used to explore a specific route path for delivering PN-
WSNA models to a remote mote in terms of wireless media.

On the other hand, the kernel program is implanted inside a sensor node. In addition, radio
frequency (RF) interface with Zigbee protocol (IEEE 802.15.4) as well as physically
connected interfaces of UART, ADC, DCA, digital input (DI) and digital output (DO) are
also available for cognitive sensing and controls. It is noted that the PN-WSNA IDE is
realized using the Microsoft visual C++ and the nesC (Avvenuti et al., 2007) program is used
to implement the kernel program.

The PN-WSNA IDE is a plug-and-play model construction environment. Fig. 6 shows the
proposed IDE, and the toolbar icons are used for model constructions, editing, revisions,
manipulations, run-time simulations, model drawing auxiliaries as well as model deliveries.
Detailed descriptions of the tool bars and algorithms for implementing the PN-WSNA
inference engine are referred to (Kuo et al., 2009). Finally, the interface functions are coded
within the kernel program. Fig. 7 shows the I/O, ADC, DAC and communication interfaces
of PN-WSNA motes. The mote is capable of collecting sensor status and actuating the
actuators using the sensor interface. At the same time, PN-WSNA motes may also

communicate with each other to deliver tokens among different PN-WSNA motes so that
distributed decisions can be achieved.

PN-WSNA Kernel
Management Server
DB for PN-
WSNA Models
DB for ad-hoc
Route Table
Integrated Development
Environment (IDE)


Fig. 6. PN-WSNA IDE workspace.

3.3 PN-WSNA Models
In this subsection, the PN-WSNA models for the proposed aquarium system are presented.
These PN-WSNA models are implemented using two PN-WSNA motes. Two motes are
communicated via the Zigbee for the delivery of tokens in the corresponding
communication places. The proposed overall PN-WSNA architecture was shown in Fig. 3,
where PN-WSNA-1 is desired for the entrance counting system using a PN-WSNA mote;

and PN-WSNA-2 is desired for the temperature, dissolved oxygen concentration, and light
control system using another PN-WSNA mote.
1
RF
Mote
PN-WSNA Kernel
2
RF
Mote
PN-WSNA Kernel
: UART
: ADC and DI
: DAC and DO
PN-WSNA Mote Interfaces
PN-WSNA-2
PN-WSNA-1

Fig. 7. Interfaces of PN-WSNA motes and their communications.

The first PN-WSNA model (PN-WSNA-1) is an entrance visitor counting and light control
system. Fig. 8 shows this model. It is can be classified into three parts, including rising edge
detections of two infrared human motion detection sensors, event sequence determinations
and light control command generations. For the rising edge detection of an infrared human
motion detection sensor, two conditions are considered for pulses generated from each
infrared human motion detection sensor including the signals from low-to-high and high-to-
low TTL voltage level. The second part is desired for recognizing activated event sequences
of two infra-ray sensors (A and B). When infra-ray sensor A is activated first, it means a
visitor entering the house. Contrarily, if sensor B is activated first, it means a visitor exiting
the house. The last one part is to determine the total number of visitors. If the number of
visitors is greater than or equal to one, a command with turning on the light is generated;
otherwise, a command with turning off the light is generated. Because the light is installed
a far away mote (PN-WSNA-2), two communication places are desired to transmit the
tokens for these light control commands in PN-WSNA1.

For the rising edge detections of two infrared human body sensors model, two similar sub-
models are shown first in the left-hand side of the Fig. 8. P001 and P010 indicate the
availability of each sensor. T001 and T011 are timed transitions for periodic sampling of the
sensors. P003 and P012 indicate the ready signals of sensors A and B, respectively. When the
sensor places are ready, the sensor data will be attached within the corresponding places.
T002 and T012 are low-enable transitions; T003 and T013 are high-enable transition, and
they are used to percept sensor data as high or low status. P004 and P013 indicate the
conclusions of low-enable transitions of T002 and T012; P005 and P014 indicate the
conclusions of high-enable transitions of T003 and T013. T005, T007, T015 and T017 have the
higher priority compared with T004, T006, T014 and T016. It is noted that, the places P006,
P007, P015 and P016 are safe (i.e., boundedness with unity token) to keep the current high/

communicate with each other to deliver tokens among different PN-WSNA motes so that
distributed decisions can be achieved.

PN-WSNA Kernel
Management Server
DB for PN-
WSNA Models
DB for ad-hoc
Route Table
Integrated Development
Environment (IDE)


Fig. 6. PN-WSNA IDE workspace.

3.3 PN-WSNA Models
In this subsection, the PN-WSNA models for the proposed aquarium system are presented.
These PN-WSNA models are implemented using two PN-WSNA motes. Two motes are
communicated via the Zigbee for the delivery of tokens in the corresponding
communication places. The proposed overall PN-WSNA architecture was shown in Fig. 3,
where PN-WSNA-1 is desired for the entrance counting system using a PN-WSNA mote;

and PN-WSNA-2 is desired for the temperature, dissolved oxygen concentration, and light
control system using another PN-WSNA mote.
1
RF
Mote
PN-WSNA Kernel
2
RF
Mote
PN-WSNA Kernel
: UART
: ADC and DI
: DAC and DO
PN-WSNA Mote Interfaces
PN-WSNA-2
PN-WSNA-1

Fig. 7. Interfaces of PN-WSNA motes and their communications.

The first PN-WSNA model (PN-WSNA-1) is an entrance visitor counting and light control
system. Fig. 8 shows this model. It is can be classified into three parts, including rising edge
detections of two infrared human motion detection sensors, event sequence determinations
and light control command generations. For the rising edge detection of an infrared human
motion detection sensor, two conditions are considered for pulses generated from each
infrared human motion detection sensor including the signals from low-to-high and high-to-
low TTL voltage level. The second part is desired for recognizing activated event sequences
of two infra-ray sensors (A and B). When infra-ray sensor A is activated first, it means a
visitor entering the house. Contrarily, if sensor B is activated first, it means a visitor exiting
the house. The last one part is to determine the total number of visitors. If the number of
visitors is greater than or equal to one, a command with turning on the light is generated;
otherwise, a command with turning off the light is generated. Because the light is installed
a far away mote (PN-WSNA-2), two communication places are desired to transmit the
tokens for these light control commands in PN-WSNA1.

For the rising edge detections of two infrared human body sensors model, two similar sub-
models are shown first in the left-hand side of the Fig. 8. P001 and P010 indicate the
availability of each sensor. T001 and T011 are timed transitions for periodic sampling of the
sensors. P003 and P012 indicate the ready signals of sensors A and B, respectively. When the
sensor places are ready, the sensor data will be attached within the corresponding places.
T002 and T012 are low-enable transitions; T003 and T013 are high-enable transition, and
they are used to percept sensor data as high or low status. P004 and P013 indicate the
conclusions of low-enable transitions of T002 and T012; P005 and P014 indicate the
conclusions of high-enable transitions of T003 and T013. T005, T007, T015 and T017 have the
higher priority compared with T004, T006, T014 and T016. It is noted that, the places P006,
P007, P015 and P016 are safe (i.e., boundedness with unity token) to keep the current high/

low level status of the sensor. Once the high and low status are both detected (i.e., both P006
and P007 have tokens), the system detects a high/ low level change event. At this moment,
the T009, T010, T019 and T020 are used to detect the rising or falling edge event of the sensor
in terms of its current level (sensor place P009; the same sensor as P003). For example, if a
level change event is detected and its current level is high, then the rising edge from a low-
level to high-level voltage would be concluded. It is noted that, only the rising edges (T010
and T020) are used in this project.

Fig. 8. PN-WSNA model for entrance visitor counting and light control system.

The second part covers the places of P019 - P021 and the transitions of T021 - T023. P019 and
P020 represent the rising edges of sensors A and B, respectively. The token arriving
sequences determine the entering and exiting events of visitors. If a token arrives at P019
and there is not any token in P020, then the system detects a visor passing through sensor A
first. In this situation, the token would enable and fire T021, and then the token enters P021.
Once, a token arrives at P020 the T023 will be enabled and fired. As a consequence, a token
will be released to P024. On the other hand, a leaving visitors can also be defined similarly.
For the case of leaving visitors, token(s) will enter P023.

The remaining part is the light control system. In this sub-model, the tokens in P024 indicate
the total number of visitors in a home. The token number would be decreased if a visitor

leaves the home (P023). Therefore, the third part of this model realizes such a scenario. If
there is no visitor in the home, inhibit arcs from P023 and P024 with respect to T024 would
not be inhibited. In this situation, the token in P022 would periodically enable and fire T024
and then the token enters to the transmitter place P025 for delivering tokens to another mote.
For another situation, if any visitor is in the home, token(s) would be appeared in P024.
Inhibit arc from P024 for T024 would inhibit the activation of T024. For this situation, the
token would enable and fire T025, and then the token enters to the transmitter place P026 for
delivering tokens to another mote. Other situation is the case of leaving visitors. In this
situation, token(s) would be appeared in P023 and P024 as well. Because of using inhibited
arcs, the token will enable and fire T026 only.

The second PN-WSNA model (PN-WSNA-2A) is desired for temperature control in our
system, as shown in Fig. 9. For conventional temperature control systems, the threshold is
setting for turn on and off to modulate the temperature. In order to reduce the switch turn
on and off frequency, a hysteresis temperature is desired. In our PN-WSNA, high-enable
transition and low enable transition are created. The activation thresholds are defined using
specific values. Because of the conflict between high-enable transition and low enable
transition, the only one transition would be enabled by comparing the sensor data. In order
to create the hysteresis voltage range, two threshold values for upper and lower bounds are
desired in this model.

Fig. 9. PN-WSNA model for temperature control.

Two sensory places with the same sensor device and signal as well as two high-enable
transitions and low-enable transitions are used in our approach. In this model, P005
indicates the availability of aquarium system. P002 and P004 indicate the ready signals of
the temperature sensor. P006 and P007 is the actuation place for turning on and off of the

low level status of the sensor. Once the high and low status are both detected (i.e., both P006
and P007 have tokens), the system detects a high/ low level change event. At this moment,
the T009, T010, T019 and T020 are used to detect the rising or falling edge event of the sensor
in terms of its current level (sensor place P009; the same sensor as P003). For example, if a
level change event is detected and its current level is high, then the rising edge from a low-
level to high-level voltage would be concluded. It is noted that, only the rising edges (T010
and T020) are used in this project.

Fig. 8. PN-WSNA model for entrance visitor counting and light control system.

The second part covers the places of P019 - P021 and the transitions of T021 - T023. P019 and
P020 represent the rising edges of sensors A and B, respectively. The token arriving
sequences determine the entering and exiting events of visitors. If a token arrives at P019
and there is not any token in P020, then the system detects a visor passing through sensor A
first. In this situation, the token would enable and fire T021, and then the token enters P021.
Once, a token arrives at P020 the T023 will be enabled and fired. As a consequence, a token
will be released to P024. On the other hand, a leaving visitors can also be defined similarly.
For the case of leaving visitors, token(s) will enter P023.

The remaining part is the light control system. In this sub-model, the tokens in P024 indicate
the total number of visitors in a home. The token number would be decreased if a visitor

leaves the home (P023). Therefore, the third part of this model realizes such a scenario. If
there is no visitor in the home, inhibit arcs from P023 and P024 with respect to T024 would
not be inhibited. In this situation, the token in P022 would periodically enable and fire T024
and then the token enters to the transmitter place P025 for delivering tokens to another mote.
For another situation, if any visitor is in the home, token(s) would be appeared in P024.
Inhibit arc from P024 for T024 would inhibit the activation of T024. For this situation, the
token would enable and fire T025, and then the token enters to the transmitter place P026 for
delivering tokens to another mote. Other situation is the case of leaving visitors. In this
situation, token(s) would be appeared in P023 and P024 as well. Because of using inhibited
arcs, the token will enable and fire T026 only.

The second PN-WSNA model (PN-WSNA-2A) is desired for temperature control in our
system, as shown in Fig. 9. For conventional temperature control systems, the threshold is
setting for turn on and off to modulate the temperature. In order to reduce the switch turn
on and off frequency, a hysteresis temperature is desired. In our PN-WSNA, high-enable
transition and low enable transition are created. The activation thresholds are defined using
specific values. Because of the conflict between high-enable transition and low enable
transition, the only one transition would be enabled by comparing the sensor data. In order
to create the hysteresis voltage range, two threshold values for upper and lower bounds are
desired in this model.

Fig. 9. PN-WSNA model for temperature control.

Two sensory places with the same sensor device and signal as well as two high-enable
transitions and low-enable transitions are used in our approach. In this model, P005
indicates the availability of aquarium system. P002 and P004 indicate the ready signals of
the temperature sensor. P006 and P007 is the actuation place for turning on and off of the

heater. Three tokens are initially assigned to these places for the initial marking. A timed
transition (T003) is desired for the periodically sampling and control of this model. If the
firing time of T003 is expired, a token in P005 will initiate a decision process. T001 and T004
are low-enable transition; T002 and T005 are high-enable transition, and they are used to
percept sensor data as high and low status. Specially, the two different thresholds are
defined in the sensory place P001 and P003 for hysteresis ranges. The threshold in sensory
place P001 is the lower one. Hence, if data in sensory place P001 is below the threshold, the
transition T001 is enabled and then the corresponding actuation place will turn off the
heater. The activation of then sensory place P003 is similar with sensory place P001.
Therefore, the two thresholds for control the heater is done by aforementioned process. On
the other hand, the PN-WSNA model for dissolved oxygen (PN-WSNA-2B) is similar to the
temperature model, and it just needs to adjust the threshold of sensory places as specific
values. Then, this model can turn on and off of the pump in terms of actuation places.

The last PN-WSNA model (PN-WSNA-2C) is desired for communication between two motes
so that the lighting device at a remote mote can be controlled. P001 in PN-WSNA-2C is
corresponded P026 in PN-WSNA-1; and P002 in PN-WSNA-2C is corresponded P025 in PN-
WSNA-1. Once the receiver place P001 receives a token and then the token would enable fire
T001. This token will be released to actuation place P003. The corresponding action of P003 is
turning on the light. The activation process of P004 is similar to P003 and the corresponding
action of P004 is turning off the light. Fig. 10 shows the model of PN-WSNA-2C.

Fig. 10. PN-WSNA model for mote communication and control the light.

4. Experiments and Discussions
In this section, the experiment is proposed and discussed. At first, the initial marking of PN-
WSNA-1 was shown in Fig. 8. Tokens are initially paaeared in the sensory places (P003,
P009, P011 and P018) and the normal places (P001, P010 and P022). In here, an experimental
example is proposed to describe the control procedures of light control system. A time chart
of sensor signals with respect to two infrared human motion detection sensors of this
experiment is shown in Fig. 11.

t
1
t
2
t
3
t
4
t
A
B

Fig. 11. Time chart of sensor signal

Initial signals of sensor A and B are both in low states before the time t
1
. For the sensor A,
when time transition T001 is fired, the token enters P002. After token is appeared in P002,
the sensory place P003 would enable and fire low-enable transition T002, and then the token
enters P004. After that, T005 will fire and then the token transmits to P006. For the setting of
safeness, the number of token in P006 is one maximally. The procedures of sensor B are the
same as sensor A. The inference of PN-WSNA model for this part is shown in Fig. 12 (a) and
Fig. 12 (b).

At the moment of t
1
, a high-level signal is generated from sensor A, and then high-enable
transition T003 would be fired. The token would consequently enter P007, as shown in Fig.
12 (c). At that time, T008 is fired, and then the token enters P008, as shown in Fig. 12 (d).
Because the signal is in a high state, signal from sensor A would enable and fire high-enable
transition T010, and then the token enters P019, as shown in Fig. 12 (e). As a consequence,
the token in P019 represents a low-to-high rising edge signal is detected, as shown in Fig. 12
(f). After an edge signal is detected from sensor A, token would enter to P021 for waiting
edge signal which comes from sensor B, as shown in Fig. 13.

At the moment of t
2
, a high-level signal is generated from sensor B. Similar procedures will
be executed, and consequently deliver a token to P020. In this situation, the sequence of
sensory A and B are that A comes before B. Therefore, T023 is fired, and then the token
enters P024 for presenting a visitor entering in the house. After that time, T025 is enabled
and fired, and then the token enters the transmitter place P026 to send the command with
turning on the light to mote-1. The inferences of PN-WSAN model is shown in Fig. 14.

W At the moment of t
1
, a visitor leaves the house, and then the corresponding signal of
sensory B is activated first. In this situation, T022 is fired and then generates a token for
presenting a visitor leaving the house. Then the token in P023 and P024 would be released
accordingly. The inferences of PN-WSAN model is shown in Fig. 15.


heater. Three tokens are initially assigned to these places for the initial marking. A timed
transition (T003) is desired for the periodically sampling and control of this model. If the
firing time of T003 is expired, a token in P005 will initiate a decision process. T001 and T004
are low-enable transition; T002 and T005 are high-enable transition, and they are used to
percept sensor data as high and low status. Specially, the two different thresholds are
defined in the sensory place P001 and P003 for hysteresis ranges. The threshold in sensory
place P001 is the lower one. Hence, if data in sensory place P001 is below the threshold, the
transition T001 is enabled and then the corresponding actuation place will turn off the
heater. The activation of then sensory place P003 is similar with sensory place P001.
Therefore, the two thresholds for control the heater is done by aforementioned process. On
the other hand, the PN-WSNA model for dissolved oxygen (PN-WSNA-2B) is similar to the
temperature model, and it just needs to adjust the threshold of sensory places as specific
values. Then, this model can turn on and off of the pump in terms of actuation places.

The last PN-WSNA model (PN-WSNA-2C) is desired for communication between two motes
so that the lighting device at a remote mote can be controlled. P001 in PN-WSNA-2C is
corresponded P026 in PN-WSNA-1; and P002 in PN-WSNA-2C is corresponded P025 in PN-
WSNA-1. Once the receiver place P001 receives a token and then the token would enable fire
T001. This token will be released to actuation place P003. The corresponding action of P003 is
turning on the light. The activation process of P004 is similar to P003 and the corresponding
action of P004 is turning off the light. Fig. 10 shows the model of PN-WSNA-2C.

Fig. 10. PN-WSNA model for mote communication and control the light.

4. Experiments and Discussions
In this section, the experiment is proposed and discussed. At first, the initial marking of PN-
WSNA-1 was shown in Fig. 8. Tokens are initially paaeared in the sensory places (P003,
P009, P011 and P018) and the normal places (P001, P010 and P022). In here, an experimental
example is proposed to describe the control procedures of light control system. A time chart
of sensor signals with respect to two infrared human motion detection sensors of this
experiment is shown in Fig. 11.

t
1
t
2
t
3
t
4
t
A
B

Fig. 11. Time chart of sensor signal

Initial signals of sensor A and B are both in low states before the time t
1
. For the sensor A,
when time transition T001 is fired, the token enters P002. After token is appeared in P002,
the sensory place P003 would enable and fire low-enable transition T002, and then the token
enters P004. After that, T005 will fire and then the token transmits to P006. For the setting of
safeness, the number of token in P006 is one maximally. The procedures of sensor B are the
same as sensor A. The inference of PN-WSNA model for this part is shown in Fig. 12 (a) and
Fig. 12 (b).

At the moment of t
1
, a high-level signal is generated from sensor A, and then high-enable
transition T003 would be fired. The token would consequently enter P007, as shown in Fig.
12 (c). At that time, T008 is fired, and then the token enters P008, as shown in Fig. 12 (d).
Because the signal is in a high state, signal from sensor A would enable and fire high-enable
transition T010, and then the token enters P019, as shown in Fig. 12 (e). As a consequence,
the token in P019 represents a low-to-high rising edge signal is detected, as shown in Fig. 12
(f). After an edge signal is detected from sensor A, token would enter to P021 for waiting
edge signal which comes from sensor B, as shown in Fig. 13.

At the moment of t
2
, a high-level signal is generated from sensor B. Similar procedures will
be executed, and consequently deliver a token to P020. In this situation, the sequence of
sensory A and B are that A comes before B. Therefore, T023 is fired, and then the token
enters P024 for presenting a visitor entering in the house. After that time, T025 is enabled
and fired, and then the token enters the transmitter place P026 to send the command with
turning on the light to mote-1. The inferences of PN-WSAN model is shown in Fig. 14.

W At the moment of t
1
, a visitor leaves the house, and then the corresponding signal of
sensory B is activated first. In this situation, T022 is fired and then generates a token for
presenting a visitor leaving the house. Then the token in P023 and P024 would be released
accordingly. The inferences of PN-WSAN model is shown in Fig. 15.


(a) Both signal of sensors are in low-status (b) At t
1
signal of sensor A goes to high

(c) Token transmits to P007 (d) Token transmits to P008

(e) Concluding T010 the be fired (f) Token transmits to P0019
Fig. 12. Decision procedures of sensor As signal (from low-to-high).

Fig. 13. Concluding rising edge status of sensor A.

5. Conclusions and Future Works
In this book chapter, the PN-WSNA is used to construct an aquarium environment control
system in a house. The major advantages of using PN-WSNA are to use a model based WSN
realization approach so that the coding efforts from domain engineers can be significantly
reduced. In addition, the control scenarios can be verified in terms of the PN-WSNA
simulations before the sensor algorithm are deployed. This book chapter use an aquarium
environment control system to demonstrate the modelling and implementation procedures
for two PN-WSNA sensor node systems, where one sensor node is deployed for aquarium
environment control and the other one is desired for entrance counting system. Two PN-
WSAN motes are communicated using the communication places of the PN-WSNA. The
aquarium environment control system acquires the data from temperature sensor and
dissolved oxygen sensor as well as the people number collected from the entrance counting
system. Meanwhile, the light, heater and pump are also activated using the sensor node. In
the future, the PN-WSNA will be used to construct more complicated WSN system to
demonstrate the powerful modelling and control capability of the PN-WSNA.

6. Acknowledgement
This work was supported by the National Science Council, Taiwan, R.O.C., under Grants
NSC 98-2218-E-011-017.


(a) Both signal of sensors are in low-status (b) At t
1
signal of sensor A goes to high

(c) Token transmits to P007 (d) Token transmits to P008

(e) Concluding T010 the be fired (f) Token transmits to P0019
Fig. 12. Decision procedures of sensor As signal (from low-to-high).

Fig. 13. Concluding rising edge status of sensor A.

5. Conclusions and Future Works
In this book chapter, the PN-WSNA is used to construct an aquarium environment control
system in a house. The major advantages of using PN-WSNA are to use a model based WSN
realization approach so that the coding efforts from domain engineers can be significantly
reduced. In addition, the control scenarios can be verified in terms of the PN-WSNA
simulations before the sensor algorithm are deployed. This book chapter use an aquarium
environment control system to demonstrate the modelling and implementation procedures
for two PN-WSNA sensor node systems, where one sensor node is deployed for aquarium
environment control and the other one is desired for entrance counting system. Two PN-
WSAN motes are communicated using the communication places of the PN-WSNA. The
aquarium environment control system acquires the data from temperature sensor and
dissolved oxygen sensor as well as the people number collected from the entrance counting
system. Meanwhile, the light, heater and pump are also activated using the sensor node. In
the future, the PN-WSNA will be used to construct more complicated WSN system to
demonstrate the powerful modelling and control capability of the PN-WSNA.

6. Acknowledgement
This work was supported by the National Science Council, Taiwan, R.O.C., under Grants
NSC 98-2218-E-011-017.


(a) Rising edge detected from sensor B (P020) (b) A->B sequence determined

(c) Token enters to P024 (d) T025 enable fire

(e) Token enters to P026
Fig. 14. Concluding status of A->B sequence determinations.

(a) Marking of B rising edge detected (b) Marking of A rising edge detected

(c) B->A sequence is concluded (d) Token enters to P023

(e) Release a token (f) Return to initial marking
Fig. 15. Concluding status of B->A sequence determinations.

(a) Rising edge detected from sensor B (P020) (b) A->B sequence determined

(c) Token enters to P024 (d) T025 enable fire

(e) Token enters to P026
Fig. 14. Concluding status of A->B sequence determinations.

(a) Marking of B rising edge detected (b) Marking of A rising edge detected

(c) B->A sequence is concluded (d) Token enters to P023

(e) Release a token (f) Return to initial marking
Fig. 15. Concluding status of B->A sequence determinations.

7. References
Akyildiz I.F.; Melodia T.; & Chowdhury K.R. (2008). Wireless Multimedia Sensor Networks:
Applications and Testbeds, Proceedings of the IEEE, Vol. 96, No. 10, pp. 1588 1605.
Avnur A. (1990). Finite State Machines for Real-time Software Engineering, Computing &
Control Engineering Journal, Vol. 1, No. 6, pp. 275 278.
Avvenuti M.; Corsini P.; Masci P.; & Vecchio A. (2007). An Application Adaptation Layer for
Wireless Sensor Networks, Pervasive and Mobile Computing, Vol. 3, No. 4, pp. 413
438.
Dalola S.; Ferrari V.; Guizzetti M.; Marioli D.; Sardini E.; Serpelloni M.; & Taroni A. (2009).
Autonomous Sensor System with Power Harvesting for Telemetric Temperature
Measurements of Pipes, IEEE Transactions on Instrumentation and Measurement, Vol.
58, No. 5, pp. 1471 1478.
Kuo C.H.; Wang C.H.; & Huang K.W. (2003). Behavior Modeling and Control of 300 mm
Fab iIntrabays Using Distributed Agent Oriented Petri Net, IEEE Transaction on
Systems, Man and Cybernetics, Part A, Vol. 33, No. 5, pp. 641 648.
Kuo C.H.; & Siao J.W. (2009). Petri Net Based Reconfigurable Wireless Sensor Networks for
Intelligent Monitoring Systems, International Conference on Computational Science and
Engineering, Vol. 2, pp. 897 902.
Manasseh C.; & Sengupta R. (2010). Middleware to Enhance Mobile Communications for
Road Safety and Traffic Mobility Applications, IET Intelligent Transport Systems,
Vol. 4, No. 1, pp. 24 36.
Murata T. (1989). Petri Nets: Properties, Analysis and Applications, Proceedings of the IEEE,
Vol. 77, No. 4, pp. 541 580.
Romer K.; & Mattern F. (2004). The Design Space of Wireless Sensor Networks, IEEE Wireless
Communications, Vol. 11, No. 6, pp. 54 61.
Ruiz A.F.; Rocon E.; Raya R.; & Pons J.L. (2008). Coupled Control of Human-exoskeleton
Systems: an Adaptative Process, IEEE Conference on Human System Interactions, pp.
242 246.
Shenai K.; & Mukhopadhyay S. (2008). Cognitive Sensor Networks, International Conference
on Microelectronics, pp. 315 320.
Sridhar P.; Madni A.M.; & Jamshidi M. (2007). Hierarchical Aggregation and Intelligent
Monitoring and Control in Fault-tolerant Wireless Sensor Networks, IEEE Systems
Journal, Vol. 1, No. 1, pp. 38 54.
Suh C; & Ko Y.B. (2008). Design and Implementation of Intelligent Home Hontrol Systems
Based on Active Sensor Networks, IEEE Transactions on Consumer Electronics, Vol.
54, No. 3, pp. 1177 1184.
Zhuang L.Q.; Liu W.; Zhang J.B.; Zhang D.H.; & Kamajaya I. (2008). Distributed Asset
Tracking Using Wireless Sensor Network, IEEE International Conference on Emerging
Technologies and Factory Automation, pp. 1165 1168.
Zhuang X.; Yang Y.; & Ding W. (2008). The Wireless Sensor Network Node Design for
Electrical Equipment On-line Monitoring, IEEE International Conference on Industrial
Technology, pp. 1 4.

Wireless Sensor Network for Ambient Assisted Living
0
Wireless Sensor Network
for Ambient Assisted Living
Universidad Politcnica de Cartagena
Spain
1. Introduction
There is a great demand, in both the public and the private sector, to take the actions needed to
expand uses for electronic devices, assistive and monitoring software, and home health com-
munication technologies to provide assisted living and health care to those in need. By means
of using Information and Communication Technologies (ICT) to assist and monitor elderly,
disabled, and chronically ill individuals in the home can improve quality of life, improve
health outcomes, and help control health care.
Assisted living technologies are for people needing assistance with Activities of Daily Living
(ADLs) but wishing to live for as long as possible independently. Assisted living exists to
bridge the gap between independent living and nursing homes. Residents in assisted living
centers are not able to live by themselves but do not require constant care either. Assisted liv-
ing facilities offer help with ADLs such as eating, bathing, dressing, laundry, housekeeping,
and assistance with medications. Many facilities also have centers for medical care; how-
ever, the care offered may not be as intensive or available to residents as the care offered at
a nursing home. Assisted living is not an alternative to a nursing home, but an intermedi-
ate level of long-term care appropriate for many seniors. Increasing health care costs and an
aging population are placing signicant strains upon the health care system. Small pilot stud-
ies have shown that meeting seniors needs for independence and autonomy, coupled with
expanded use of home health technologies, mitigate against the circumstances above, and
provide improved health outcomes. Difculty with reimbursement policies, governmental
approval processes, and absence of efcient deployment strategies has hampered adopting
such technologies.
Most efforts to-date of applying ICT to health care in the home have focused on telemedicine,
in which ICT is used to connect the patient and his biometric information to a health care
provider either in real-time or into storage for analysis at a later time (Biemer & Hampe,
2005; Lubrin et al., 2006; Ross, 2004; Rowan & Mynatt, 2005). While such functionality will be
included in Information Technology for Assisted Living at Home (ITALH), the main objective
in tele-assistence is to provide smart monitors and sensors that will alert the user and/or their
care provider (be that a doctor, nurse, family member, neighbor, friend, etc.) of events such
as accidents or acute illness and to diagnostic events that could indicate a deterioration in his
health condition. This differs from efforts (Sixsmith & Johnson, 2004a), in which IR sensors
array are used to detect such events as accidental falls, in that the sensors themselves detect
8
the event, not a central system. This reduces wireless bandwidth and greatly improves the
privacy of the system by not streaming data constantly.
In particular, the chapter is focused in the introduction of an ubiquitous wireless network in-
frastructure to support an assisted living at home system, called DIA (Dispositivo Inteligente
de Alerta, in spanish) which is being developed by Universidad Politcnica de Cartagena, Uni-
versidad de Murcia and Ambiental Intelligence & Interaction S.L.L. (Ami2) company. Speci-
cally, the system is constructed based on a wireless communication network in order to trans-
fer data and events of elderly. A typical scenario consists of a private home which is instru-
mented based on WSN. In this context, the concept of a individual assisted by monitoring
via radio-frequency is evident. The wireless infrastructure is a heterogeneous and ubiqui-
tous, being present everywhere at once, wireless network that connects sensor devices within
the home to a central Home Health System Gateway and/or a mobile Gateway. The sensor
nodes themselves have embedded processing capability and are required to transmit only oc-
casional information about their own status and messages notifying the central system when
they detect a signicant event. The central system, in a smart sense, connect this network to
the outside world via secure Internet and telephone service so that intelligent alerts can be
sent out, and authorized caregivers can have access to the system to check up on the user.
Privacy and security are fundamental concerns in these systems. The chapter is organized
as follows. Section 2 provides an overview of the state of the art in terms of sensor network
technology. Section 3 explains the application scenario and where and how the node sensors
was deployed and Section 4 discusses data processing issues. The chapter concludes with a
brief summary and some nal remarks.
2. Technology Overview
2.1 Sensor Network Technology Overview
Asensor network is an infrastructure comprised of sensing (measuring), computing, and com-
munication elements that gives an administrator the ability to instrument, observe, and react
to events and phenomena in a specied environment (Sohraby et al., 2007). Typical applica-
tions include, but are not limited to, data collection, monitoring, surveillance, and medical
telemetry. In addition to sensing, one is often also interested in control and activation.
There are four basic components in a sensor network: (1) a set of distributed or localized
sensors; (2) a communication network (usually, but not always, wireless-based); (3) a central
point of information clustering (usually called base station or sink); and (4) a set of comput-
ing resources at the central point (or beyond, e.g., personal computer board or other device
like PDA) to handle data correlation, event trending, status querying, and data mining. In
this context, the sensing and computation nodes are considered part of the sensor network; in
fact, some of the basic computation may be done in the network itself. The computation and
communication infrastructure associated with sensor networks is often specic to this envi-
ronment and rooted in the device and application-based nature of these networks. Figure 1
shows a generic protocol stack model that can be utilized to describe the WSN. Issues here re-
late to the following: (1) Physical layer treats about connectivity and coverage. (2) Data Link
Layer is the protocol layer which transfers data between adjacent network nodes in a wide
area network or between nodes on the same local area network segment. (3) Network Layer is
responsible for end-to-end (source to destination) packet delivery including routing through
intermediate hosts, whereas the data link layer is responsible for node-to-node (hop-to-hop)
frame delivery on the same link. (4) Transport Layer is a group of methods and protocols
within a layered architecture of network components within which it is responsible for encap-
sulating application data blocks into data units. (5) Upper (application, presentation, session)
layer treats about processing application, in what could be an environment with highly corre-
lated and time-dependent arrivals.
Management planes are needed, so that sensor nodes can work together in a power efcient
way, route data in a wireless (mobile or not) sensor network, and share resources between
sensor nodes. Without them, each sensor node will just work individually. From the whole
sensor network point of view, it is more efcient if sensor nodes can collaborate with each
other, so the lifetime of the sensor networks can be prolonged (Kahn et al., 1999).
C
o
m
m
u
n
i
c
a
t
i
o
n
p
r
o
t
o
c
o
l
s
Physical layer
M
a
n
a
g
e
m
e
n
t
p
r
o
t
o
c
o
l
s
Mobilitymanagement plane
Taskmanagement plane
Power management plane
Data Linklayer
Networklayer
Transport layer
Upper layers
Fig. 1. Generic protocol stack for sensor networks
Sensors in a WSNhave a variety of purposes, functions, and capabilities. Sensor networking is
a multidisciplinary area that involves, among others, radio and networking, signal processing,
articial intelligence, database management, systems architectures for operator-friendly in-
frastructure administration, resource optimization, power management algorithms, and plat-
form technology (hardware and software, such as operating systems)
The technology for sensing and control includes electric and magnetic eld sensors; radio-
wave frequency sensors; optical-, electrooptic-, and infrared sensors; radars; lasers; loca-
tion/navigation sensors; seismic and pressure-wave sensors; environmental parameter sen-
sors (e.g., wind, humidity, heat); and biochemical national security oriented sensors. Todays
sensors can be described as smart inexpensive devices equipped with multiple onboard sens-
ing elements; they are low-cost low-power untethered multifunctional nodes that are logically
homed to a central sink node. Sensor devices, or wireless nodes (WNs), are also (sometimes)
called motes. Therefore, a WSN consists of densely distributed nodes that support sensing,
signal processing, embedded computing, and connectivity; sensors are logically linked by self-
organizing means. WNs typically transmit information to collecting (monitoring) stations that
aggregate some or all of the information. WSNs have unique characteristics, such as, but not
limited to, power constraints and limited battery life for the WNs, redundant data acquisition,
low duty cycle, and, many-to-one ows. Power efciency in WSNs is generally accomplished
in three way: Low-duty-cycle operation, Local/in-network processing to reduce data volume
the event, not a central system. This reduces wireless bandwidth and greatly improves the
privacy of the system by not streaming data constantly.
In particular, the chapter is focused in the introduction of an ubiquitous wireless network in-
frastructure to support an assisted living at home system, called DIA (Dispositivo Inteligente
de Alerta, in spanish) which is being developed by Universidad Politcnica de Cartagena, Uni-
versidad de Murcia and Ambiental Intelligence & Interaction S.L.L. (Ami2) company. Speci-
cally, the system is constructed based on a wireless communication network in order to trans-
fer data and events of elderly. A typical scenario consists of a private home which is instru-
mented based on WSN. In this context, the concept of a individual assisted by monitoring
via radio-frequency is evident. The wireless infrastructure is a heterogeneous and ubiqui-
tous, being present everywhere at once, wireless network that connects sensor devices within
the home to a central Home Health System Gateway and/or a mobile Gateway. The sensor
nodes themselves have embedded processing capability and are required to transmit only oc-
casional information about their own status and messages notifying the central system when
they detect a signicant event. The central system, in a smart sense, connect this network to
the outside world via secure Internet and telephone service so that intelligent alerts can be
sent out, and authorized caregivers can have access to the system to check up on the user.
Privacy and security are fundamental concerns in these systems. The chapter is organized
as follows. Section 2 provides an overview of the state of the art in terms of sensor network
technology. Section 3 explains the application scenario and where and how the node sensors
was deployed and Section 4 discusses data processing issues. The chapter concludes with a
brief summary and some nal remarks.
2. Technology Overview
2.1 Sensor Network Technology Overview
Asensor network is an infrastructure comprised of sensing (measuring), computing, and com-
munication elements that gives an administrator the ability to instrument, observe, and react
to events and phenomena in a specied environment (Sohraby et al., 2007). Typical applica-
tions include, but are not limited to, data collection, monitoring, surveillance, and medical
telemetry. In addition to sensing, one is often also interested in control and activation.
There are four basic components in a sensor network: (1) a set of distributed or localized
sensors; (2) a communication network (usually, but not always, wireless-based); (3) a central
point of information clustering (usually called base station or sink); and (4) a set of comput-
ing resources at the central point (or beyond, e.g., personal computer board or other device
like PDA) to handle data correlation, event trending, status querying, and data mining. In
this context, the sensing and computation nodes are considered part of the sensor network; in
fact, some of the basic computation may be done in the network itself. The computation and
communication infrastructure associated with sensor networks is often specic to this envi-
ronment and rooted in the device and application-based nature of these networks. Figure 1
shows a generic protocol stack model that can be utilized to describe the WSN. Issues here re-
late to the following: (1) Physical layer treats about connectivity and coverage. (2) Data Link
Layer is the protocol layer which transfers data between adjacent network nodes in a wide
area network or between nodes on the same local area network segment. (3) Network Layer is
responsible for end-to-end (source to destination) packet delivery including routing through
intermediate hosts, whereas the data link layer is responsible for node-to-node (hop-to-hop)
frame delivery on the same link. (4) Transport Layer is a group of methods and protocols
within a layered architecture of network components within which it is responsible for encap-
sulating application data blocks into data units. (5) Upper (application, presentation, session)
layer treats about processing application, in what could be an environment with highly corre-
lated and time-dependent arrivals.
Management planes are needed, so that sensor nodes can work together in a power efcient
way, route data in a wireless (mobile or not) sensor network, and share resources between
sensor nodes. Without them, each sensor node will just work individually. From the whole
sensor network point of view, it is more efcient if sensor nodes can collaborate with each
other, so the lifetime of the sensor networks can be prolonged (Kahn et al., 1999).
C
o
m
m
u
n
i
c
a
t
i
o
n
p
r
o
t
o
c
o
l
s
Physical layer
M
a
n
a
g
e
m
e
n
t
p
r
o
t
o
c
o
l
s
Mobilitymanagement plane
Taskmanagement plane
Power management plane
Data Linklayer
Networklayer
Transport layer
Upper layers
Fig. 1. Generic protocol stack for sensor networks
Sensors in a WSNhave a variety of purposes, functions, and capabilities. Sensor networking is
a multidisciplinary area that involves, among others, radio and networking, signal processing,
articial intelligence, database management, systems architectures for operator-friendly in-
frastructure administration, resource optimization, power management algorithms, and plat-
form technology (hardware and software, such as operating systems)
The technology for sensing and control includes electric and magnetic eld sensors; radio-
wave frequency sensors; optical-, electrooptic-, and infrared sensors; radars; lasers; loca-
tion/navigation sensors; seismic and pressure-wave sensors; environmental parameter sen-
sors (e.g., wind, humidity, heat); and biochemical national security oriented sensors. Todays
sensors can be described as smart inexpensive devices equipped with multiple onboard sens-
ing elements; they are low-cost low-power untethered multifunctional nodes that are logically
homed to a central sink node. Sensor devices, or wireless nodes (WNs), are also (sometimes)
called motes. Therefore, a WSN consists of densely distributed nodes that support sensing,
signal processing, embedded computing, and connectivity; sensors are logically linked by self-
organizing means. WNs typically transmit information to collecting (monitoring) stations that
aggregate some or all of the information. WSNs have unique characteristics, such as, but not
limited to, power constraints and limited battery life for the WNs, redundant data acquisition,
low duty cycle, and, many-to-one ows. Power efciency in WSNs is generally accomplished
in three way: Low-duty-cycle operation, Local/in-network processing to reduce data volume
(and hence transmission time), and multihop. Multihop networking reduces the requirement
for long-range transmission since signal path loss is an inverse exponent with range or dis-
tance. Each node in the sensor network can act as a repeater, thereby reducing the link range
coverage required and, in turn, the transmission power.
For a number of years, vendors have made use of proprietary technology for collecting perfor-
mance data from devices. In the early 2000s, sensor device suppliers were researching ways
of introducing standardization, rst designers ruled out Wi-Fi (wireless delity, IEEE 802.11b)
standards for sensors as being too complex and supporting more bandwidth than is actually
needed for typical sensors. Infrared systems require line of sight, which is not always achiev-
able; Bluetooth (IEEE 802.15.1) technology was at rst considered a possibility, but it was soon
deemed too complex and expensive. This opened the door for a new standard IEEE 802.15.4
along with ZigBee (more specically, ZigBee comprises the software layers above the newly
adopted IEEE 802.15.4 standard and supports a plethora of applications). IEEE 802.15.4 op-
erates in the 2.4 GHz industrial, scientic, and medical (ISM) radio band and supports data
transmission at rates up to 250 kbit s
1
at ranges from 10 to 70 m. ZigBee/IEEE 802.15.4 is
designed to complement wireless technologies such as Bluetooth, Wi-Fi, and ultra-wideband
(UWB).
2.1.1 Requeriments for Wireless Sensor Network in Ambient Assisted Living Environments
Wireless sensor network integration in Ambient Assisted Living frameworks is usually not
described in the literature, and normally is treated as a black box. Following, a list of the
functional requirements for WSN in Ambient Assisted Living environments which need to be
addressed in order to have a reliable systembased on WSN(Martin et al., 2009) is enumerated.
1. Ambient Assisted Living Systems: Type of Networks. The design objectives are related to
the development of an Ambient Assisted Living System capable of offering its services
at home and nursing houses. The AAL systems could be composed by:
ABody Sensor Network (BSN), which will include all the devices that a person must
wear (accelerometers, gyroscopes, spirometers, oxymeters, etc) or use to allow the
services to work. Depending on the elderly prole and the services to be con-
gured, the BSN may include continuous monitoring sensors and other health
sensors. To congure the BSN, it is always important to bear in mind the usability
restrictions imposed by the users acceptance of personal devices in home environ-
ments. Basically, BSN is a mobile subset of the Wireless Sensor Network.
A Wireless Sensor Network, which will include home infrastructure sensors (ambi-
ent, presence, pressure, home automation sensors, etc.), actuators and appliances
capable of notifying their status. The Wireless Sensor Network station base will
be able to communicate with the BSN by using ad hoc networking capabilities. It
will include local intelligent features to dispatch events and orders depending on
the situation. These processing capabilities will be part of a home gateway which
will connect the home ambient via station base with the Core Care Network.
ACore Care Network, serving as a bridge of communication between the home sen-
sorial infrastructure and third parties and service providers (caregivers). Services
may be enabled through the Core Care Network. It can also authorize the connec-
tion of external service providers, centralize systemmonitoring and guarantee the
security of personal data.
2. Ambient Assisted Living Systems: General design requirements for WSN. The previous sce-
nario imposes some functional requirements to the Wireless Sensor Network nally
composed by infrastructure and personal sensing nodes. Next there is a brief list of the
most important features to consider.
Interoperability. Wireless Sensor Networks in real deployments need to be ready to
manage heterogeneous sensors, which need to share a common communication
scheme.
Network self-conguration and maintenance. It is desirable that the WSN demands
as little attention from a human operator as possible.
Easy and robust deployment. When designing WSN functionalities, it is important
to consider the deployment requirements to make the network fully operational.
Multihop routing. A WSN for AAL usually consists of several sensor nodes that
send their measurements to a sink node, which collects all the information and
typically sends it to a PC, where all the data are stored and elaborated. In a home,
the sink node may not provide coverage over the whole area. As a consequence,
it is necessary to implement routing algorithms that transmit the information to-
wards the sink through other nodes.
Positioning service. This kind of service is required for many operational and ser-
vice purposes. For instance, to process the information related to the place where
the user is, avoiding the storage and computation of information that is not rele-
vant in a specic moment.
Energy saving strategies. As the devices that are integrated in the network have
limited computational and radio communication capabilities, collaborative algo-
rithms with energy-aware communication are required to achieve multi-modal
collaboration and energy conservation.
Scalability of sensors and actuators. AALs services may impose different type of
sensing and actuation requirements. For example, in a scenario considering ser-
vices for COPD patients, devices sensing the quality of air may be needed. For
that reason, WSN for ACSs need to be ready to include new sensors and actu-
ators, which may be connected to existent network nodes or congure as nodes
themselves. Methodologies and software architectures making easier to scale the
network sensing capabilities are needed.
Security. As wireless networks are based on a standard and data are sent over a
broadcast channel, it is possible to make packet snifng and data spoong attacks.
IEEE 802.15.4 MAC layer offers some facilities which can be used by upper layers
to achieve a good level of security.
2.2 Sensor Node Technology Overview
Figure 2 shows the general architecture soft and hardware of a sensor node. The terms sensor
node, wireless node (WN), Smart Dust, mote, and COTS (commercial off-the-shelf) mote are
used somewhat interchangeably in the industry; the most general terms used here are sen-
sor node and WN. WSNs that combine physical sensing of parameters such as temperature,
light, or others events with computation and networking capabilities are expected to become
ubiquitous in the next future.
(and hence transmission time), and multihop. Multihop networking reduces the requirement
for long-range transmission since signal path loss is an inverse exponent with range or dis-
tance. Each node in the sensor network can act as a repeater, thereby reducing the link range
coverage required and, in turn, the transmission power.
For a number of years, vendors have made use of proprietary technology for collecting perfor-
mance data from devices. In the early 2000s, sensor device suppliers were researching ways
of introducing standardization, rst designers ruled out Wi-Fi (wireless delity, IEEE 802.11b)
standards for sensors as being too complex and supporting more bandwidth than is actually
needed for typical sensors. Infrared systems require line of sight, which is not always achiev-
able; Bluetooth (IEEE 802.15.1) technology was at rst considered a possibility, but it was soon
deemed too complex and expensive. This opened the door for a new standard IEEE 802.15.4
along with ZigBee (more specically, ZigBee comprises the software layers above the newly
adopted IEEE 802.15.4 standard and supports a plethora of applications). IEEE 802.15.4 op-
erates in the 2.4 GHz industrial, scientic, and medical (ISM) radio band and supports data
transmission at rates up to 250 kbit s
1
at ranges from 10 to 70 m. ZigBee/IEEE 802.15.4 is
designed to complement wireless technologies such as Bluetooth, Wi-Fi, and ultra-wideband
(UWB).
2.1.1 Requeriments for Wireless Sensor Network in Ambient Assisted Living Environments
Wireless sensor network integration in Ambient Assisted Living frameworks is usually not
described in the literature, and normally is treated as a black box. Following, a list of the
functional requirements for WSN in Ambient Assisted Living environments which need to be
addressed in order to have a reliable systembased on WSN(Martin et al., 2009) is enumerated.
1. Ambient Assisted Living Systems: Type of Networks. The design objectives are related to
the development of an Ambient Assisted Living System capable of offering its services
at home and nursing houses. The AAL systems could be composed by:
ABody Sensor Network (BSN), which will include all the devices that a person must
wear (accelerometers, gyroscopes, spirometers, oxymeters, etc) or use to allow the
services to work. Depending on the elderly prole and the services to be con-
gured, the BSN may include continuous monitoring sensors and other health
sensors. To congure the BSN, it is always important to bear in mind the usability
restrictions imposed by the users acceptance of personal devices in home environ-
ments. Basically, BSN is a mobile subset of the Wireless Sensor Network.
A Wireless Sensor Network, which will include home infrastructure sensors (ambi-
ent, presence, pressure, home automation sensors, etc.), actuators and appliances
capable of notifying their status. The Wireless Sensor Network station base will
be able to communicate with the BSN by using ad hoc networking capabilities. It
will include local intelligent features to dispatch events and orders depending on
the situation. These processing capabilities will be part of a home gateway which
will connect the home ambient via station base with the Core Care Network.
ACore Care Network, serving as a bridge of communication between the home sen-
sorial infrastructure and third parties and service providers (caregivers). Services
may be enabled through the Core Care Network. It can also authorize the connec-
tion of external service providers, centralize systemmonitoring and guarantee the
security of personal data.
2. Ambient Assisted Living Systems: General design requirements for WSN. The previous sce-
nario imposes some functional requirements to the Wireless Sensor Network nally
composed by infrastructure and personal sensing nodes. Next there is a brief list of the
most important features to consider.
Interoperability. Wireless Sensor Networks in real deployments need to be ready to
manage heterogeneous sensors, which need to share a common communication
scheme.
Network self-conguration and maintenance. It is desirable that the WSN demands
as little attention from a human operator as possible.
Easy and robust deployment. When designing WSN functionalities, it is important
to consider the deployment requirements to make the network fully operational.
Multihop routing. A WSN for AAL usually consists of several sensor nodes that
send their measurements to a sink node, which collects all the information and
typically sends it to a PC, where all the data are stored and elaborated. In a home,
the sink node may not provide coverage over the whole area. As a consequence,
it is necessary to implement routing algorithms that transmit the information to-
wards the sink through other nodes.
Positioning service. This kind of service is required for many operational and ser-
vice purposes. For instance, to process the information related to the place where
the user is, avoiding the storage and computation of information that is not rele-
vant in a specic moment.
Energy saving strategies. As the devices that are integrated in the network have
limited computational and radio communication capabilities, collaborative algo-
rithms with energy-aware communication are required to achieve multi-modal
collaboration and energy conservation.
Scalability of sensors and actuators. AALs services may impose different type of
sensing and actuation requirements. For example, in a scenario considering ser-
vices for COPD patients, devices sensing the quality of air may be needed. For
that reason, WSN for ACSs need to be ready to include new sensors and actu-
ators, which may be connected to existent network nodes or congure as nodes
themselves. Methodologies and software architectures making easier to scale the
network sensing capabilities are needed.
Security. As wireless networks are based on a standard and data are sent over a
broadcast channel, it is possible to make packet snifng and data spoong attacks.
IEEE 802.15.4 MAC layer offers some facilities which can be used by upper layers
to achieve a good level of security.
2.2 Sensor Node Technology Overview
Figure 2 shows the general architecture soft and hardware of a sensor node. The terms sensor
node, wireless node (WN), Smart Dust, mote, and COTS (commercial off-the-shelf) mote are
used somewhat interchangeably in the industry; the most general terms used here are sen-
sor node and WN. WSNs that combine physical sensing of parameters such as temperature,
light, or others events with computation and networking capabilities are expected to become
ubiquitous in the next future.
Driver
Sensor ADC
Sensing unit #1 Sensing unit #2 Processing unit
Processor
Transceiver
Antenna
Storage
ADC Sensor
Power generator Actuator
Power Unit
Sensor
system
Actuator
Applications
Mini-
Communication:
Networking/
topology
Communication:
Radio
Antenna
Storage Memory
Operating System(OS)
Processor
Driver
Driver
Driver
Fig. 2. Hardware and software components of WNs
Many of these examples share some basic characteristics. In most of them, there is a clear
difference between sources of data the actual nodes that sense data and sinks nodes where the
data should be delivered to. The interaction patterns between sources and sinks show some
typical patterns. The most relevant ones are:
Event detection. Sensor nodes should report to the sink(s) once they have detected the
occurrence of a specied event. The simplest events can be detected locally by a single
sensor node in isolation (e.g., a temperature threshold is exceeded); more complicated
types of events require the collaboration of nearby or even remote sensors to decide
whether a (composite) event has occurred (e.g., a temperature gradient becomes too
steep). If several different events can occur, event classication might be an additional
issue.
Periodic measurements. Sensors can be tasked with periodically reporting measured val-
ues. Often, these reports can be triggered by a detected event; the reporting period is
application dependent.
Function approximation and edge detection. The way a physical value like temperature
changes from one place to another can be regarded as a function of location. A WSN
can be used to approximate this unknown function (to extract its spatial characteristics),
using a limited number of samples taken at each individual sensor node. This approx-
imate mapping should be made available at the sink. How and when to update this
mapping depends on the applications needs, as do the approximation accuracy and the
inherent trade-off against energy consumption. Similarly, a relevant problem can be
to nd areas or points of the same given value. An example is to nd the isothermal
points in a forest re application to detect the border of the actual re. This can be gen-
eralized to nding edges in such functions or to sending messages along the boundaries
of patterns in both space and/or time.
Tracking. The source of an event can be mobile (e.g., an intruder in surveillance sce-
narios). The WSN can be used to report updates on the event sources position to the
sink(s), potentially with estimates about speed and direction as well. To do so, typically
sensor nodes have to cooperate before updates can be reported to the sink.
Embedded sensing refers to the synergistic incorporation of sensors in structures or environ-
ments; embedded sensing enables spatially and temporally dense monitoring of the system
under consideration (e.g., a home). In biological systems, the sensors themselves must not
affect the system or organism adversely. The technology for sensing and control includes
electric and magnetic eld sensors; radio-wave frequency sensors; optical-, electrooptic-, and
infrared sensors; radars; lasers; location and navigation sensors; seismic and pressure-wave
sensors; environmental parameter sensors (e.g., wind, humidity, heat); and biochemical na-
tional security oriented sensors.
Small, low-cost, robust, reliable, and sensitive sensors are needed to enable the realization of
practical and economical sensor networks. Although a large number measurements are of in-
terest for WSN applications, commercially available sensors exist for many of these measure-
ments. Sensor nodes come in a variety of hardware congurations: from nodes connected to a
LAN and attached to permanent power sources, to nodes communicating via wireless multi-
hop RF radio powered by small batteries. The trend is toward very large scale integration
(VLSI), integrated optoelectronics, and nanotechnology; in particular, work is under way in
earnest in the biochemical arena.
2.2.1 Hardware and software architecture of WNs
Normally, the hardware components of a WN include the sensing and actuation unit (single
element or array), the processing unit, the communication unit, the power unit, and other
application-dependent units. Sensors, particularly Smart Dust and COTS motes, have four
basic hardware subsystems:
1. Sensor transducer(s). The interface between the environment and the WN is the sen-
sor. Basic environmental sensors include, but are not limited to, acceleration, humidity,
light, magnetic ux, temperature, pressure, and sound.
2. Computational logic and storage. These are used to handle onboard data processing and
manipulation, transient and short-term storage, encryption, digital modulation, and
digital transmission.
3. Communication. WNs must have the ability to communicate either in C1-WSN arrange-
ments (mesh-based systems with multi-hop radio connectivity among or between WNs,
utilizing dynamic routing in both the wireless and wireline portions of the network),
and/or in C2-WSN arrangements (point-to-point or multipoint-to-point systems gen-
erally with single-hop radio connectivity to WNs, utilizing static routing over the wire-
less network with only one route from the WNs to the companion terrestrial or wireline
forwarding node).
Driver
Sensor ADC
Sensing unit #1 Sensing unit #2 Processing unit
Processor
Transceiver
Antenna
Storage
ADC Sensor
Power generator Actuator
Power Unit
Sensor
system
Actuator
Applications
Mini-
Communication:
Networking/
topology
Communication:
Radio
Antenna
Storage Memory
Operating System(OS)
Processor
Driver
Driver
Driver
Fig. 2. Hardware and software components of WNs
Many of these examples share some basic characteristics. In most of them, there is a clear
difference between sources of data the actual nodes that sense data and sinks nodes where the
data should be delivered to. The interaction patterns between sources and sinks show some
typical patterns. The most relevant ones are:
Event detection. Sensor nodes should report to the sink(s) once they have detected the
occurrence of a specied event. The simplest events can be detected locally by a single
sensor node in isolation (e.g., a temperature threshold is exceeded); more complicated
types of events require the collaboration of nearby or even remote sensors to decide
whether a (composite) event has occurred (e.g., a temperature gradient becomes too
steep). If several different events can occur, event classication might be an additional
issue.
Periodic measurements. Sensors can be tasked with periodically reporting measured val-
ues. Often, these reports can be triggered by a detected event; the reporting period is
application dependent.
Function approximation and edge detection. The way a physical value like temperature
changes from one place to another can be regarded as a function of location. A WSN
can be used to approximate this unknown function (to extract its spatial characteristics),
using a limited number of samples taken at each individual sensor node. This approx-
imate mapping should be made available at the sink. How and when to update this
mapping depends on the applications needs, as do the approximation accuracy and the
inherent trade-off against energy consumption. Similarly, a relevant problem can be
to nd areas or points of the same given value. An example is to nd the isothermal
points in a forest re application to detect the border of the actual re. This can be gen-
eralized to nding edges in such functions or to sending messages along the boundaries
of patterns in both space and/or time.
Tracking. The source of an event can be mobile (e.g., an intruder in surveillance sce-
narios). The WSN can be used to report updates on the event sources position to the
sink(s), potentially with estimates about speed and direction as well. To do so, typically
sensor nodes have to cooperate before updates can be reported to the sink.
Embedded sensing refers to the synergistic incorporation of sensors in structures or environ-
ments; embedded sensing enables spatially and temporally dense monitoring of the system
under consideration (e.g., a home). In biological systems, the sensors themselves must not
affect the system or organism adversely. The technology for sensing and control includes
electric and magnetic eld sensors; radio-wave frequency sensors; optical-, electrooptic-, and
infrared sensors; radars; lasers; location and navigation sensors; seismic and pressure-wave
sensors; environmental parameter sensors (e.g., wind, humidity, heat); and biochemical na-
tional security oriented sensors.
Small, low-cost, robust, reliable, and sensitive sensors are needed to enable the realization of
practical and economical sensor networks. Although a large number measurements are of in-
terest for WSN applications, commercially available sensors exist for many of these measure-
ments. Sensor nodes come in a variety of hardware congurations: from nodes connected to a
LAN and attached to permanent power sources, to nodes communicating via wireless multi-
hop RF radio powered by small batteries. The trend is toward very large scale integration
(VLSI), integrated optoelectronics, and nanotechnology; in particular, work is under way in
earnest in the biochemical arena.
2.2.1 Hardware and software architecture of WNs
Normally, the hardware components of a WN include the sensing and actuation unit (single
element or array), the processing unit, the communication unit, the power unit, and other
application-dependent units. Sensors, particularly Smart Dust and COTS motes, have four
basic hardware subsystems:
1. Sensor transducer(s). The interface between the environment and the WN is the sen-
sor. Basic environmental sensors include, but are not limited to, acceleration, humidity,
light, magnetic ux, temperature, pressure, and sound.
2. Computational logic and storage. These are used to handle onboard data processing and
manipulation, transient and short-term storage, encryption, digital modulation, and
digital transmission.
3. Communication. WNs must have the ability to communicate either in C1-WSN arrange-
ments (mesh-based systems with multi-hop radio connectivity among or between WNs,
utilizing dynamic routing in both the wireless and wireline portions of the network),
and/or in C2-WSN arrangements (point-to-point or multipoint-to-point systems gen-
erally with single-hop radio connectivity to WNs, utilizing static routing over the wire-
less network with only one route from the WNs to the companion terrestrial or wireline
forwarding node).
4. Power supply. An appropriate energy infrastructure or supply is necessary to support
operation from a few hours to months or years (depending on the application).
Sensors typically have ve basic software subsystems:
1. Operating system (OS) microcode (also called middleware). This is the board common mi-
crocode that is used by all high-level node-resident software modules to support vari-
ous functions. As is generally the case, the purpose of an operating system is to shield
the software from the machine-level functionality of the microprocessor. It is desirable
to have open-source operating systems designed specically for WSNs; these OSs typ-
ically utilize an architecture that enables rapid implementation while minimizing code
size. TinyOS is one such example of a commonly used OS.
2. Sensor drivers. These are the software modules that manage basic functions of the sen-
sor transceivers; sensors may possibly be of the modular/plug-in type, and depending
on the type and sophistication, the appropriate conguration and settings must be up-
loaded into the sensor (drivers shield the application software from the machine-level
functionality of the sensor or other peripheral).
3. Communication processors. This code manages the communication functions, including
routing, packet buffering and forwarding, topology maintenance, medium access con-
trol (e.g., contention mechanisms, direct-sequence spread-spectrum mechanisms), en-
cryption, and FEC, to list a few.
4. Communication drivers (encoding and the physical layer). These software modules man-
age the minutia of the radio channel transmission link, including clocking and synchro-
nization, signal encoding, bit recovery, bit counting, signal levels, and modulation.
5. Data processing mini-apps. These are numerical, data-processing, signal value storage
and manipulations, or other basic applications that are supported at the node level for
in-network processing.
2.3 A Survey of Sensor Nodes for Wireless Sensor Networks
A signicant and prime research of sensor network is a project named Smart-Dust developed
by University of California at Berkeley, USA(Pister, 2008) in the late 90s. The main objective of
the project was to develop a compact size node that includes sensor, capability to compute the
sensor data onboard, lowcost, minimal power consumption and having bidirectional wireless
communication capability. It was later sold commercially by Crossbow Inc. and Moteiv. The
rst mote was the WeC, which appeared in 1998 and was followed in the next year by the
Ren mote and the next year by the Ren2 mote and Dot mote. Based on the eld trials of
these platforms, a second-generation platform called MICA was developed; it appeared in
2001. A third generation of motes were named MICA2 which was appeared in 2002. MICAz
was appeared in 2002 and it was the rst generation with a 2.4 GHz frequency of radio. Iris
Mote is built upon the IEEE 802.15.4 standard. It is regarded as the successor of MICAz, one
of the most commonly used mote systems in the world. The most recent signicant step in
commercial mote development was the Telos in 2004 y Telosb in 2005.
Another interesting research testbed was the Spec platform (JLHLabs, 2008), which integrated
the functionality of Mica onto a single 5 mm
2
chip. Spec was built with a micro-radio, an
analog-to-digital converter, and a temperature sensor on a single chip, which lead to a 30-fold
reduction in total power consumption. This single-chip integration also opened the path to
lowcost sensor nodes. The integrated RAMand cache memory architecture greatly simplied
the design of the mote family. However, the tiny footprint also requires a specialized operating
system, which was developed by UC Berkeley, called TinyOS (TinyOS, 2009). TinyOS features
component based architecture and event driven model that are suitable for programming with
small embedded devices, such as motes. The combination of Motes and TinyOS is gradually
becoming a popular experimental platform for many research efforts in the eld of WSNs.
The Medusa MK-2 (UCLA, 2009) sensor node was carried out by the Center for Embedded
Networked Sensing (CENS) at UCLA in 2002 to target both high and low-end processing ap-
plications. It integrates two microcontrollers, the rst one; ATmega128 was dedicated to less
computationally demanding tasks, including radio base band processing and sensor sam-
pling. The second one, AT91FR4081, was a more powerful microcontroller (40 MHz, 1 MB
ash, 136 kB RAM) that was designed to handle more sophisticated, but less frequent signal
processing tasks (e.g., the Kalman lter). The combination of these two microcontrollers pro-
vided more exibility in WSN development and deployment, especially for applications that
require both high computation capabilities and long lifetime.
In 2002 the Berkeley Wireless Research Center (BWRC) developed Systemon Chip (SoC) based
sensor node, named PicoNode II (Cho et al., 2005). It was built using two ASIC chips that
implemented the entire node functionality. In the following year, the same team developed a
rst radio transmitter (that used power less than 400 W), PicoBeacon was purely powered by
solar and vibrational energy sources.
Another ASIC based approach was taken by the AMPS group from MIT. Following their
rst testbed, AMPS-I (MIT, 2008), the team then tried to build a highly integrated sensor
node comprised of a digital and an analog/RF ASIC, AMPS-II. The interesting feature of
AMPS-II was the nodes capability to operate in several modes. It can operate either as low-
end stand-alone guarding node, a fully functional node for middle-end sensor networks or
as a companion component in a more powerful high-end sensor system. Thus, it favored a
network with heterogeneous sensor nodes for a more efcient utilization of resources.
The Free2move wireless sensor node (Bilstrup & Wiberg, 2004) is based on a transceiver oper-
ating in the 2.4 GHz ISM band. The node was initially thought of as an active RFID tag for
monitoring temperature in goods. However, it has been shown that it is also possible to use
it as a wireless sensor network node. The node is equipped with an extremely low power mi-
crocontroller (Microchip PIC16F87), for executing communication protocols and sensor func-
tionality. The memory and processing resources are very limited to keep the price and energy
consumption as low as possible. The node is also equipped with a temperature sensor.
3. Application Scenario
A rst prototype scenario has been developed in which a user will have a home assistence
system that is able to monitor his or her activity in order to detect incidents and uncommon
activities (Fernndez-Luque et al., 2009) and (Bota-Blaya et al., 2009). The prototype house or
scenario has a bedroom, a hall, a corridor, a toilet, a kitchen, and a living room. Movement
infrared sensors are installed in each location. Moreover, in the bedroom there is a pressure
sensor in bed; in the hall, a magnetic sensor to detect the opening and closening of the en-
trance door, and in the sofa of living room another pressure sensor. All sensor boards have a
complementary temperature sensor. The data is gathered from sensors mounted in the home.
The sensor events are transmitted by the wireless sensor network to the base station by means
ZigBee technology. A gateway is also included in the system to allow continuous monitor-
ing. The gateway receives the events from the sensors through base station and decides what
the appropriate action to take will be. Options could include querying the user to check on
their status, storing (or forwarding) data on the event for future analysis by a assistential care
4. Power supply. An appropriate energy infrastructure or supply is necessary to support
operation from a few hours to months or years (depending on the application).
Sensors typically have ve basic software subsystems:
1. Operating system (OS) microcode (also called middleware). This is the board common mi-
crocode that is used by all high-level node-resident software modules to support vari-
ous functions. As is generally the case, the purpose of an operating system is to shield
the software from the machine-level functionality of the microprocessor. It is desirable
to have open-source operating systems designed specically for WSNs; these OSs typ-
ically utilize an architecture that enables rapid implementation while minimizing code
size. TinyOS is one such example of a commonly used OS.
2. Sensor drivers. These are the software modules that manage basic functions of the sen-
sor transceivers; sensors may possibly be of the modular/plug-in type, and depending
on the type and sophistication, the appropriate conguration and settings must be up-
loaded into the sensor (drivers shield the application software from the machine-level
functionality of the sensor or other peripheral).
3. Communication processors. This code manages the communication functions, including
routing, packet buffering and forwarding, topology maintenance, medium access con-
trol (e.g., contention mechanisms, direct-sequence spread-spectrum mechanisms), en-
cryption, and FEC, to list a few.
4. Communication drivers (encoding and the physical layer). These software modules man-
age the minutia of the radio channel transmission link, including clocking and synchro-
nization, signal encoding, bit recovery, bit counting, signal levels, and modulation.
5. Data processing mini-apps. These are numerical, data-processing, signal value storage
and manipulations, or other basic applications that are supported at the node level for
in-network processing.
2.3 A Survey of Sensor Nodes for Wireless Sensor Networks
A signicant and prime research of sensor network is a project named Smart-Dust developed
by University of California at Berkeley, USA(Pister, 2008) in the late 90s. The main objective of
the project was to develop a compact size node that includes sensor, capability to compute the
sensor data onboard, lowcost, minimal power consumption and having bidirectional wireless
communication capability. It was later sold commercially by Crossbow Inc. and Moteiv. The
rst mote was the WeC, which appeared in 1998 and was followed in the next year by the
Ren mote and the next year by the Ren2 mote and Dot mote. Based on the eld trials of
these platforms, a second-generation platform called MICA was developed; it appeared in
2001. A third generation of motes were named MICA2 which was appeared in 2002. MICAz
was appeared in 2002 and it was the rst generation with a 2.4 GHz frequency of radio. Iris
Mote is built upon the IEEE 802.15.4 standard. It is regarded as the successor of MICAz, one
of the most commonly used mote systems in the world. The most recent signicant step in
commercial mote development was the Telos in 2004 y Telosb in 2005.
Another interesting research testbed was the Spec platform (JLHLabs, 2008), which integrated
the functionality of Mica onto a single 5 mm
2
chip. Spec was built with a micro-radio, an
analog-to-digital converter, and a temperature sensor on a single chip, which lead to a 30-fold
reduction in total power consumption. This single-chip integration also opened the path to
lowcost sensor nodes. The integrated RAMand cache memory architecture greatly simplied
the design of the mote family. However, the tiny footprint also requires a specialized operating
system, which was developed by UC Berkeley, called TinyOS (TinyOS, 2009). TinyOS features
component based architecture and event driven model that are suitable for programming with
small embedded devices, such as motes. The combination of Motes and TinyOS is gradually
becoming a popular experimental platform for many research efforts in the eld of WSNs.
The Medusa MK-2 (UCLA, 2009) sensor node was carried out by the Center for Embedded
Networked Sensing (CENS) at UCLA in 2002 to target both high and low-end processing ap-
plications. It integrates two microcontrollers, the rst one; ATmega128 was dedicated to less
computationally demanding tasks, including radio base band processing and sensor sam-
pling. The second one, AT91FR4081, was a more powerful microcontroller (40 MHz, 1 MB
ash, 136 kB RAM) that was designed to handle more sophisticated, but less frequent signal
processing tasks (e.g., the Kalman lter). The combination of these two microcontrollers pro-
vided more exibility in WSN development and deployment, especially for applications that
require both high computation capabilities and long lifetime.
In 2002 the Berkeley Wireless Research Center (BWRC) developed Systemon Chip (SoC) based
sensor node, named PicoNode II (Cho et al., 2005). It was built using two ASIC chips that
implemented the entire node functionality. In the following year, the same team developed a
rst radio transmitter (that used power less than 400 W), PicoBeacon was purely powered by
solar and vibrational energy sources.
Another ASIC based approach was taken by the AMPS group from MIT. Following their
rst testbed, AMPS-I (MIT, 2008), the team then tried to build a highly integrated sensor
node comprised of a digital and an analog/RF ASIC, AMPS-II. The interesting feature of
AMPS-II was the nodes capability to operate in several modes. It can operate either as low-
end stand-alone guarding node, a fully functional node for middle-end sensor networks or
as a companion component in a more powerful high-end sensor system. Thus, it favored a
network with heterogeneous sensor nodes for a more efcient utilization of resources.
The Free2move wireless sensor node (Bilstrup & Wiberg, 2004) is based on a transceiver oper-
ating in the 2.4 GHz ISM band. The node was initially thought of as an active RFID tag for
monitoring temperature in goods. However, it has been shown that it is also possible to use
it as a wireless sensor network node. The node is equipped with an extremely low power mi-
crocontroller (Microchip PIC16F87), for executing communication protocols and sensor func-
tionality. The memory and processing resources are very limited to keep the price and energy
consumption as low as possible. The node is also equipped with a temperature sensor.
3. Application Scenario
A rst prototype scenario has been developed in which a user will have a home assistence
system that is able to monitor his or her activity in order to detect incidents and uncommon
activities (Fernndez-Luque et al., 2009) and (Bota-Blaya et al., 2009). The prototype house or
scenario has a bedroom, a hall, a corridor, a toilet, a kitchen, and a living room. Movement
infrared sensors are installed in each location. Moreover, in the bedroom there is a pressure
sensor in bed; in the hall, a magnetic sensor to detect the opening and closening of the en-
trance door, and in the sofa of living room another pressure sensor. All sensor boards have a
complementary temperature sensor. The data is gathered from sensors mounted in the home.
The sensor events are transmitted by the wireless sensor network to the base station by means
ZigBee technology. A gateway is also included in the system to allow continuous monitor-
ing. The gateway receives the events from the sensors through base station and decides what
the appropriate action to take will be. Options could include querying the user to check on
their status, storing (or forwarding) data on the event for future analysis by a assistential care
Fig. 3. Schematic overview of the system installed at prototype home
provider, placing a telephone call to a care provider, relative or health care service, or other
options. Figure 3 shows a schematic overview of the system.
The main idea consists in monitoring the person living alone in his home without interacting
with him. To start, it is needed to knowif he is at home in order to activate the ubiquitous cus-
todial care system. It is easy to know by the context if a resident is at home knowing that the
entrance door was opened and movement in the hall was detected. By means of distributed
sensors installed in each room at home we can know the activities and the elderly location.
The sensor boards developed by us allow to distinguish if movement event is induced by a
pet or person thanks to a dual passive infrared sensor. On the other hand, as the pressure
sensors are located in the bed and the favorite sofa in the living room, we can know more of
where he is even if he is not in movement. All this sensorial assembly will be ruled by an
articial intelligent software which will allow to learn of elderly diary activities. If the system
detects a suspicious event, i.e., movement in any room at 12 a.m and pressure in the bed, then
the system give an alert to the caregiver.
3.1 Assembly of Distributed Sensors
The terms sensor node, wireless node (WN), Smart Dust, mote, and COTS (commercial off-
the-shelf) mote are used somewhat interchangeably in the industry; the most general terms
used here are sensor node and WN. WSNs that combine physical sensing of parameters such
Radio Transceiver
M
u
l
t
i
s
e
n
s
o
r
U
A
R
T
B
U
S
I
/
O
I
n
t
e
r
f
a
c
e
8 Kb
1
2
8
K
B
F
l
a
s
h
M
e
m
o
r
y
Microcontroller
ADC
CPU
4 Kb
MOTEBOARD
Fig. 4. Sensor node scheme
as temperature, light, or others events with computation and networking capabilities are ex-
pected to become ubiquitous in the next future. The basic functionality of a WN generally
depends on the application and type of sensor device. Sensors are either passive or active
devices. Passive sensors in single-element form include, among others, seismic-, acoustic-,
strain-, humidity-, and temperature-measuring devices. Passive sensors in array forminclude
optical- (visible, infrared 1 mm, infrared 10 mm) and biochemical-measuring devices. Arrays
are geometrically regular clusters of WNs (i.e., following some topographical grid arrange-
ment). Passive sensors tend to be low-energy devices. Active sensors include radar and sonar;
these tend to be high-energy systems.
Activity monitoring can be benecial for elderly people who live alone at home. By means of
using electronic technologies to assist and monitor elderly, disabled, and chronically ill indi-
viduals in the home can improve quality of life, improve health outcomes, and help control
asistential care. This is done with mote devices developed ad hoc for this purpose which are
based on Iris mote from Crossbow (Horton & Suh, 2005). The mote board developed uses
a single channel 2.4 GHz radio to provide bi-directional communications at 40 kbit s
1
, and
an Atmel Atmega 1281 microcontroller running at 8 MHz controls the signal sampling and
data transmission. The wireless sensor node is powered by a pair conventional AA batteries
and a DC boost converter provides a stable voltage source. Figure 4 and 5 shows a schematic
overview of sensor node architecture and station base architecture.
This mote board was designed in order to provide basic environmental sensing, and expan-
sion for other sensing functionality. Actually, wearable sensors are also being included which
could measure and analyze the users health as biomedicals signals (ECG, heart rate, etc) and
activity such falls. Among other things because we have implemented an integrated antenna
on the same board. The assembly of distribuited sensors are integrated in a mesh network.
A mesh network is a generic name for a class of networked embedded systems that share
several characteristics including: Multi-Hop the capability of sending messages peer-to-peer
to a base station, thereby enabling scalable range extension; Self-Conguring capable of net-
work formation without human intervention; Self-Healing capable of adding and removing
network nodes automatically without having to reset the network; and Dynamic Routing
capable of adaptively determining the route based on dynamic network conditions (i.e., link
quality, hop-count, gradient, or other metric). Multihop protocol is a full featured multi-hop,
Fig. 3. Schematic overview of the system installed at prototype home
provider, placing a telephone call to a care provider, relative or health care service, or other
options. Figure 3 shows a schematic overview of the system.
The main idea consists in monitoring the person living alone in his home without interacting
with him. To start, it is needed to knowif he is at home in order to activate the ubiquitous cus-
todial care system. It is easy to know by the context if a resident is at home knowing that the
entrance door was opened and movement in the hall was detected. By means of distributed
sensors installed in each room at home we can know the activities and the elderly location.
The sensor boards developed by us allow to distinguish if movement event is induced by a
pet or person thanks to a dual passive infrared sensor. On the other hand, as the pressure
sensors are located in the bed and the favorite sofa in the living room, we can know more of
where he is even if he is not in movement. All this sensorial assembly will be ruled by an
articial intelligent software which will allow to learn of elderly diary activities. If the system
detects a suspicious event, i.e., movement in any room at 12 a.m and pressure in the bed, then
the system give an alert to the caregiver.
3.1 Assembly of Distributed Sensors
The terms sensor node, wireless node (WN), Smart Dust, mote, and COTS (commercial off-
the-shelf) mote are used somewhat interchangeably in the industry; the most general terms
used here are sensor node and WN. WSNs that combine physical sensing of parameters such
Radio Transceiver
M
u
l
t
i
s
e
n
s
o
r
U
A
R
T
B
U
S
I
/
O
I
n
t
e
r
f
a
c
e
8 Kb
1
2
8
K
B
F
l
a
s
h
M
e
m
o
r
y
Microcontroller
ADC
CPU
4 Kb
MOTEBOARD
Fig. 4. Sensor node scheme
as temperature, light, or others events with computation and networking capabilities are ex-
pected to become ubiquitous in the next future. The basic functionality of a WN generally
depends on the application and type of sensor device. Sensors are either passive or active
devices. Passive sensors in single-element form include, among others, seismic-, acoustic-,
strain-, humidity-, and temperature-measuring devices. Passive sensors in array forminclude
optical- (visible, infrared 1 mm, infrared 10 mm) and biochemical-measuring devices. Arrays
are geometrically regular clusters of WNs (i.e., following some topographical grid arrange-
ment). Passive sensors tend to be low-energy devices. Active sensors include radar and sonar;
these tend to be high-energy systems.
Activity monitoring can be benecial for elderly people who live alone at home. By means of
using electronic technologies to assist and monitor elderly, disabled, and chronically ill indi-
viduals in the home can improve quality of life, improve health outcomes, and help control
asistential care. This is done with mote devices developed ad hoc for this purpose which are
based on Iris mote from Crossbow (Horton & Suh, 2005). The mote board developed uses
a single channel 2.4 GHz radio to provide bi-directional communications at 40 kbit s
1
, and
an Atmel Atmega 1281 microcontroller running at 8 MHz controls the signal sampling and
data transmission. The wireless sensor node is powered by a pair conventional AA batteries
and a DC boost converter provides a stable voltage source. Figure 4 and 5 shows a schematic
overview of sensor node architecture and station base architecture.
This mote board was designed in order to provide basic environmental sensing, and expan-
sion for other sensing functionality. Actually, wearable sensors are also being included which
could measure and analyze the users health as biomedicals signals (ECG, heart rate, etc) and
activity such falls. Among other things because we have implemented an integrated antenna
on the same board. The assembly of distribuited sensors are integrated in a mesh network.
A mesh network is a generic name for a class of networked embedded systems that share
several characteristics including: Multi-Hop the capability of sending messages peer-to-peer
to a base station, thereby enabling scalable range extension; Self-Conguring capable of net-
work formation without human intervention; Self-Healing capable of adding and removing
network nodes automatically without having to reset the network; and Dynamic Routing
capable of adaptively determining the route based on dynamic network conditions (i.e., link
quality, hop-count, gradient, or other metric). Multihop protocol is a full featured multi-hop,
Fig. 5. Motes for PRO(totype)DIA project
ad-hoc, mesh networking protocol driven for events (Al-Karaki & Kamal, 2004; Li et al., 2008;
Sagduyu & Ephremides, 2004). This protocol is a modied protocol based on Xmesh de-
veloped by Crossbow for wireless networks. A multihop network protocol consists of WN
(Motes) that wirelessly communicate to each other and are capable of hopping radio mes-
sages to a base station where they are passed to a PC or other client. The hopping effectively
extends radio communication range and reduces the power required to transmit messages. By
hopping data in this way, our multihop protocol can provide two critical benets: improved
radio coverage and improved reliability. Two nodes do not need to be within direct radio
range of each other to communicate. A message can be delivered to one or more nodes in-
between which will route the data. Likewise, if there is a bad radio link between two nodes,
that obstacle can be overcome by rerouting around the area of bad service. Typically the nodes
run in a low power mode, spending most of their time in a sleep state, in order to achieve
multi-year battery life. On the other hand, the node is woke up when a event happened by
means of an interruption which is activated by sensor board when an event is detected. Also,
the mesh network protocol provides a networking service that is both self-organizing and self-
healing. It can route data fromnodes to a base station (upstream) or downstreamto individual
nodes. It can also broadcast within a single area of coverage or arbitrarily between any two
nodes in a cluster. QOS (Quality of Service) is provided by either a best effort (link level ac-
knowledgement) and guaranteed delivery (end-to-end acknowledgement). Also, XMesh can
be congured into various power modes including HP (high power), LP (low power), and
ELP (extended low power).
t
4
t
t
2
t
1
C
a
c
t
i
v
i
t
y
t
3
t
5
t
6
t
7
t
8
t
9
Int
x
Int
x
Int
x
Fig. 6. Composite interruption chronogram
3.2 Sensor Data Monitoring
Inside the sensor node, the microcontroller and the radio transceiver work in power save
mode most of the time. When a state change happens in the sensors (an event has happened),
an external interrupt wakes the microcontroller and the sensing process starts. The sensing is
made following the next sequence: rst, the external interrupt which has red the exception
is disabled for a 5 seconds interval; to save energy by preventing the same sensor ring con-
tinuously without relevant information. This is achieved by starting a 5 seconds timer which
we call the interrupt timer, when this timer is red the external interrupt is rearmed. For it,
there is a st of taking the data, the global interrupt bit is disabled until the data has been cap-
tured and the message has been sent. Third, the digital input is read using the TinyOS GPIO
management features. Fourth, battery level and temperature are read. The battery level and
temperature readings are made using routines based on TinyOS ADC library. At last, a mes-
sage is sent using the similar TinyOS routines. In this way, the message is sent to the sensor
parent in the mesh. The external led of the multisensor board is powered on when the sending
routine is started; and powered off when the sending process is nished. This external led can
be disabled via software in order to save battery power.
As an example, an events chronogram driven for interruption is shown in Figure 6, where
next thresholds was established: t
2
t
1
< 125 ms, t
3
t
1
< 5 s, t
4
t
1
< 5 s, t
5
t
1
= 5
s, t
6
t
5
< 1 ms, t
7
t
6
< 125 ms, t
8
t
6
= 5 s and t
9
t
8
< 1 ms. Figure 6 can be
descripted as follows: at t
1
an external interrupt Int
x
has occurred due to a change in a sensor.
The external interrupt Int
x
is disabled and the interrupt timer started. The sensor data is
taken. The message is sent and the external led of our multisensor board is powered on. At
t
2
the send process is nished. The external led is powered off. At t
3
, an external interrupt
Int
x
has occurred. The exception routine is not executed because the external interrupt Int
x
is disabled. The interrupt ag for Int
x
is raised. At t
4
, another interruption has occurred
but the interruption ag is already raised. At t
5
, the interrupt timer is red. The external
interrupt Int
x
is enabled. At t
6
, the exception routine is executed because the interrupt ag
is raised. The external interrupt Int
x
is disabled and the interrupt timer started. The sensor
data is taken. The message is sent and the external led powered on. At t
7
: The send process
has nished. The external led is powered off. At t
8
interrupt Int
x
is enabled.At t
9
, there are not more pending tasks.
3.3 Base Station
The event notications are sent from the sensors to the base station. Also commands are
sent from the gateway to the sensors. In short, the base station fuses the information and
Fig. 5. Motes for PRO(totype)DIA project
ad-hoc, mesh networking protocol driven for events (Al-Karaki & Kamal, 2004; Li et al., 2008;
Sagduyu & Ephremides, 2004). This protocol is a modied protocol based on Xmesh de-
veloped by Crossbow for wireless networks. A multihop network protocol consists of WN
(Motes) that wirelessly communicate to each other and are capable of hopping radio mes-
sages to a base station where they are passed to a PC or other client. The hopping effectively
extends radio communication range and reduces the power required to transmit messages. By
hopping data in this way, our multihop protocol can provide two critical benets: improved
radio coverage and improved reliability. Two nodes do not need to be within direct radio
range of each other to communicate. A message can be delivered to one or more nodes in-
between which will route the data. Likewise, if there is a bad radio link between two nodes,
that obstacle can be overcome by rerouting around the area of bad service. Typically the nodes
run in a low power mode, spending most of their time in a sleep state, in order to achieve
multi-year battery life. On the other hand, the node is woke up when a event happened by
means of an interruption which is activated by sensor board when an event is detected. Also,
the mesh network protocol provides a networking service that is both self-organizing and self-
healing. It can route data fromnodes to a base station (upstream) or downstreamto individual
nodes. It can also broadcast within a single area of coverage or arbitrarily between any two
nodes in a cluster. QOS (Quality of Service) is provided by either a best effort (link level ac-
knowledgement) and guaranteed delivery (end-to-end acknowledgement). Also, XMesh can
be congured into various power modes including HP (high power), LP (low power), and
ELP (extended low power).
t
4
t
t
2
t
1
C
a
c
t
i
v
i
t
y
t
3
t
5
t
6
t
7
t
8
t
9
Int
x
Int
x
Int
x
Fig. 6. Composite interruption chronogram
3.2 Sensor Data Monitoring
Inside the sensor node, the microcontroller and the radio transceiver work in power save
mode most of the time. When a state change happens in the sensors (an event has happened),
an external interrupt wakes the microcontroller and the sensing process starts. The sensing is
made following the next sequence: rst, the external interrupt which has red the exception
is disabled for a 5 seconds interval; to save energy by preventing the same sensor ring con-
tinuously without relevant information. This is achieved by starting a 5 seconds timer which
we call the interrupt timer, when this timer is red the external interrupt is rearmed. For it,
there is a st of taking the data, the global interrupt bit is disabled until the data has been cap-
tured and the message has been sent. Third, the digital input is read using the TinyOS GPIO
management features. Fourth, battery level and temperature are read. The battery level and
temperature readings are made using routines based on TinyOS ADC library. At last, a mes-
sage is sent using the similar TinyOS routines. In this way, the message is sent to the sensor
parent in the mesh. The external led of the multisensor board is powered on when the sending
routine is started; and powered off when the sending process is nished. This external led can
be disabled via software in order to save battery power.
As an example, an events chronogram driven for interruption is shown in Figure 6, where
next thresholds was established: t
2
t
1
< 125 ms, t
3
t
1
< 5 s, t
4
t
1
< 5 s, t
5
t
1
= 5
s, t
6
t
5
< 1 ms, t
7
t
6
< 125 ms, t
8
t
6
= 5 s and t
9
t
8
< 1 ms. Figure 6 can be
descripted as follows: at t
1
an external interrupt Int
x
has occurred due to a change in a sensor.
The external interrupt Int
x
is disabled and the interrupt timer started. The sensor data is
taken. The message is sent and the external led of our multisensor board is powered on. At
t
2
the send process is nished. The external led is powered off. At t
3
, an external interrupt
Int
x
has occurred. The exception routine is not executed because the external interrupt Int
x
is disabled. The interrupt ag for Int
x
is raised. At t
4
, another interruption has occurred
but the interruption ag is already raised. At t
5
interrupt Int
x
is enabled. At t
6
, the exception routine is executed because the interrupt ag
is raised. The external interrupt Int
x
is disabled and the interrupt timer started. The sensor
data is taken. The message is sent and the external led powered on. At t
7
: The send process
has nished. The external led is powered off. At t
8
interrupt Int
x
is enabled.At t
9
, there are not more pending tasks.
3.3 Base Station
The event notications are sent from the sensors to the base station. Also commands are
sent from the gateway to the sensors. In short, the base station fuses the information and
therefore is a central and special mote node in the network. This USB-based central node
was developed by us also. This provides different services to the wireless network. First, the
base station is the seed mote that forms the multihop network. It outputs route messages that
inform all nearby motes that it is the base station and has zero cost to forward any message.
Second, for downstreamcommunication the base station automatically routes messages down
the same path as the upstream communication from a mote. Third, it is compiled with a
large number of message buffers to handle more children than other motes in the network.
These messages are provided for TinyOS, a open-source low-power operative system. Fourth,
the base station forwards all messages upstream and downstream from the gateway using
a standard serial framer protocol. Five, the station base can periodically send a heartbeat
message to the client. If it does not get a response from the client within a predened time it
will assume the communication link has been lost and reset itself.
This base station is connected via USB to a gateway (miniPC) which is responsible of deter-
mining an appropriate response by means of an intelligent software in development now, i.e.
passive infra-red movement sensor might send an event at which point and moment towards
the gateway via base station for its processing. The application can monitor the events to de-
termine if a strange situation has occurred. Also, the application can ask to the sensors node
if the event has nished or was a malfunction of sensor. If normal behavior is detected by
the latter devices, then the event might just be recorded as an incident of interest, or the user
might be prompted to ask if they are alright. If, on the other hand, no normal behavior is
detected then the gateway might immediately query the user and send an emergency signal if
there is no response within a certain (short) period of time. With the emergency signal, access
would be granted to the remote care provider who could log in and via phone call.
3.4 Gateway
Our system has been designed considering the presence of a local gateway used to process
event patterns in situ and take decisions. This home gateway is provided with a java-based
intelligent software which is able to take decision about different events. In short, it has java
application for monitoring the elderly and ZigBee wireless connectivity provided by a USB
mote-based base station for our prototype. This layer stack form a global software archi-
tecture. The lowest layer is a hardware layer. In the context awareness layer, the software
obtains contextual information provided by sensors. The middle level software layer, model
of user behavior, obtains the actual state of attendee, detecting if the resident is in an emer-
gency situation which must be solved. The deep reasoning layer is being developed to solve
inconsistencies reached in the middle layer.
The gateway is based on a miniPC draws only 3-5 watts when running Linux (Ubuntu 7.10
(Gutsy) preloaded) consuming as little power as a standard PC does in stand-by mode. Ultra
small and ultra quiet, the gateway is about the size of a paperback book, is noiseless thanks
to a fanless design and gets barely warm. Gateway disposes a x86 architecture and integrated
hard disk. Fit-PC has dual 100 Mbps Ethernet making it a capable network computer. A
normal personal computer is too bulky, noisy and power hungry.
The motherboard of miniPC is a rugged embedded board having all components including
memory and CPU soldered on-board. The gateway is enclosed in an all-aluminum anodized
case that is splash and dust resistant. The case itself is used for heat removal- eliminating the
need for a fan and venting holes. Fit-PC has no moving parts other than the hard-disk. The
CPU is an AMD Geode LX800 500 MHz, the memory has 256 MB DDR 333 MHz soldered
on-board and the hard disk has 2.5" IDE 60 GB. To connect with base station, the gateway
Fig. 7. Gateway based on miniPC, Mote board and base station
disposes of 2 USB 2.0 HiSpeed 480 Mbps, also it has 2 RJ45 Ethernet ports 100 Mbps to
connect with Internet. Figure 7 shows the gateway ports base station and our mote board.
4. Results and Discussions
Figure 7 shows the hardware of the built wireless sensor node provides for mote board. In this
prototype, a variable and heterogeneous number of wireless sensor nodes are attached to mul-
tisensor boards in order to detect the activities of our elderly in the surrounding environment,
and they send their measurements to a base station when an event (change of state) is pro-
duced or when the gateway requires information in order to avoid inconsistencies. The base
station can transmit or receive data to or from the gateway by means of USB interface. It can
be seen that the sensor nodes of the prototype house detect the elderly activity. The infrared
passive, magnetic and pressure sensors have a high quality and sensitivity. Also, the low-
power multihop protocol works correctly. Therefore, the system can determine the location
and activity patterns of elderly, and in the close future when the intelligent software will learn
of elderly activities, the system will can take decisions about strange actions of elderly if they
are not stored in his history of activities. By now, the system knows some habitual patterns
of behavior and therefore it must be tuning in each particular case. Additionally, connectivity
between the gateway exists to the remote caregiver station via a local ethernet network. The
gateway currently receives streamed sensor data so that it can be used for analysis and al-
gorithm development for the intelligent software and the gateway is able potentially to send
data via ethernet to the caregiver station.
therefore is a central and special mote node in the network. This USB-based central node
was developed by us also. This provides different services to the wireless network. First, the
base station is the seed mote that forms the multihop network. It outputs route messages that
inform all nearby motes that it is the base station and has zero cost to forward any message.
Second, for downstreamcommunication the base station automatically routes messages down
the same path as the upstream communication from a mote. Third, it is compiled with a
large number of message buffers to handle more children than other motes in the network.
These messages are provided for TinyOS, a open-source low-power operative system. Fourth,
the base station forwards all messages upstream and downstream from the gateway using
a standard serial framer protocol. Five, the station base can periodically send a heartbeat
message to the client. If it does not get a response from the client within a predened time it
will assume the communication link has been lost and reset itself.
This base station is connected via USB to a gateway (miniPC) which is responsible of deter-
mining an appropriate response by means of an intelligent software in development now, i.e.
passive infra-red movement sensor might send an event at which point and moment towards
the gateway via base station for its processing. The application can monitor the events to de-
termine if a strange situation has occurred. Also, the application can ask to the sensors node
if the event has nished or was a malfunction of sensor. If normal behavior is detected by
the latter devices, then the event might just be recorded as an incident of interest, or the user
might be prompted to ask if they are alright. If, on the other hand, no normal behavior is
detected then the gateway might immediately query the user and send an emergency signal if
there is no response within a certain (short) period of time. With the emergency signal, access
would be granted to the remote care provider who could log in and via phone call.
3.4 Gateway
Our system has been designed considering the presence of a local gateway used to process
event patterns in situ and take decisions. This home gateway is provided with a java-based
intelligent software which is able to take decision about different events. In short, it has java
application for monitoring the elderly and ZigBee wireless connectivity provided by a USB
mote-based base station for our prototype. This layer stack form a global software archi-
tecture. The lowest layer is a hardware layer. In the context awareness layer, the software
obtains contextual information provided by sensors. The middle level software layer, model
of user behavior, obtains the actual state of attendee, detecting if the resident is in an emer-
gency situation which must be solved. The deep reasoning layer is being developed to solve
inconsistencies reached in the middle layer.
The gateway is based on a miniPC draws only 3-5 watts when running Linux (Ubuntu 7.10
(Gutsy) preloaded) consuming as little power as a standard PC does in stand-by mode. Ultra
small and ultra quiet, the gateway is about the size of a paperback book, is noiseless thanks
to a fanless design and gets barely warm. Gateway disposes a x86 architecture and integrated
hard disk. Fit-PC has dual 100 Mbps Ethernet making it a capable network computer. A
normal personal computer is too bulky, noisy and power hungry.
The motherboard of miniPC is a rugged embedded board having all components including
memory and CPU soldered on-board. The gateway is enclosed in an all-aluminum anodized
case that is splash and dust resistant. The case itself is used for heat removal- eliminating the
need for a fan and venting holes. Fit-PC has no moving parts other than the hard-disk. The
CPU is an AMD Geode LX800 500 MHz, the memory has 256 MB DDR 333 MHz soldered
on-board and the hard disk has 2.5" IDE 60 GB. To connect with base station, the gateway
Fig. 7. Gateway based on miniPC, Mote board and base station
disposes of 2 USB 2.0 HiSpeed 480 Mbps, also it has 2 RJ45 Ethernet ports 100 Mbps to
connect with Internet. Figure 7 shows the gateway ports base station and our mote board.
4. Results and Discussions
Figure 7 shows the hardware of the built wireless sensor node provides for mote board. In this
prototype, a variable and heterogeneous number of wireless sensor nodes are attached to mul-
tisensor boards in order to detect the activities of our elderly in the surrounding environment,
and they send their measurements to a base station when an event (change of state) is pro-
duced or when the gateway requires information in order to avoid inconsistencies. The base
station can transmit or receive data to or from the gateway by means of USB interface. It can
be seen that the sensor nodes of the prototype house detect the elderly activity. The infrared
passive, magnetic and pressure sensors have a high quality and sensitivity. Also, the low-
power multihop protocol works correctly. Therefore, the system can determine the location
and activity patterns of elderly, and in the close future when the intelligent software will learn
of elderly activities, the system will can take decisions about strange actions of elderly if they
are not stored in his history of activities. By now, the system knows some habitual patterns
of behavior and therefore it must be tuning in each particular case. Additionally, connectivity
between the gateway exists to the remote caregiver station via a local ethernet network. The
gateway currently receives streamed sensor data so that it can be used for analysis and al-
gorithm development for the intelligent software and the gateway is able potentially to send
data via ethernet to the caregiver station.
Fig. 8. Iris mote board and our rst Multisensor board prototype (2007)
As the transmission is digital, there is no noise in the signals. It represents an important
feature because noise effects commonly hardly affect telemedicine and assistence systems.
The baud rate allows the transmission of vital and activity signals without problems. The
discrete signals (movement, pressure and temperature, for example) are quickly transmitted.
Nevertheless, spending 5 s to transmit an signal sample or event does not represent a big
problem. Moreover, the system can interact with other applications based on information
technologies. Using standards represents an important step for integrating assisted living at
home systems. The systemwas implemented as previously we have described. As mentioned,
the system uses Java programming language in order to describe the activity of the elderly
and take a decision. The system guaranteed the transmission of a packet per less to 1 seconds,
e.g. the baud rate is 57 600 bit s
1
. Other signals, such as temperature, need the same time.
Furthermore, lost packets are tracked, once it is using a cyclic redundancy code (CRC). There
are a lot of sensors which can measure activities and environmental parameters unobtrusively.
Among them, just a few sensors are used in our prototype home. In the future, other useful
sensors will be used in experiments. For fall measurement (Sixsmith & Johnson, 2004b), a
method can be used applied using infrared vision. In addition, microphone/speaker sensors
can be used for tracking and ultrasound sensors also can be used for movement. Other sensors
can be easily incorporated into our system because we have already developed a small-size
multisensor board.
In this sense, we have decided design an accelerometer mote that is small and lightweight that
can be worn comfortably without obstructing normal activities. The wearable mote board has
mounted a 3-axis accelerometer with high resolution (13-bit) measurement at up to 16 g
(Analog Devices ADXL345). Digital output data is formatted as 16-bit twos complement and
is accessible through either a SPI (3- or 4-wire) (or I2C digital interface). The wearable mote
measures the static acceleration of gravity in tilt-sensing applications, as well as dynamic ac-
celeration resulting frommotion or shock. High resolution provided by ADXL345 (4 mg/LSB)
enables measurement of inclination changes less than 1.0
. Several special sensing functions

are provided. Activity and inactivity sensing detect the presence or lack of motion and if
the acceleration on any axis exceeds a user-set level. Tap sensing detects single and double
taps. Free-fall sensing detects if the device is falling. These functions can be mapped to one
of two interrupt output pins. An integrated, patent pending 32-level rst in, rst out (FIFO)
buffer can be used to store data to minimize host processor intervention. Low power modes
Fig. 9. Actor with accelerometer in his waist, log of data and accelometer sensor node proto-
type
enable intelligent motion-based power management with threshold sensing and active accel-
eration measurement at extremely low power dissipation. The mote ts inside a plastic box
measuring 441 cm, where the button battery is enclosed in the same package. Clearly, the
placement of the device on the body is of primary concern. Some of the criteria are that it
should be comfortable and that the device itself should not pose a threat to the wearer in the
event of a fall. For our experiments, we attached the mote to a belt worn around the waist.
We have not done sufcient experiments on elderly people. In this work, the experiments
should be considered preliminary and more data is needed. Figure 9 shows some pictures of
accelerometer sensor node and our proofs.
In the literature there is an absence of research data on a persons movement in his or her own
house that is not biased by self-report or by third party observation. We are in the process
of several threads of analysis that would provide more sophisticated capabilities for future
versions of the intelligent software. The assisted living system is a heterogenous wireless
network using and ZigBee radios to connect a diverse set of embedded sensor devices. These
devices and the wireless network can monitor the elderly activity in a secure and private
manner and issue alerts to the user, care givers or emergency services as necessary to provide
additional safety and security to the user. This system is being developed to provide this
safety and security so that elder citizens who might have to leave their own homes for a
group care facility will be able to extend their ability to remain at home longer. This will in
most cases provide themwith better quality of life and better health in a cost effective manner.
Fig. 8. Iris mote board and our rst Multisensor board prototype (2007)
As the transmission is digital, there is no noise in the signals. It represents an important
feature because noise effects commonly hardly affect telemedicine and assistence systems.
The baud rate allows the transmission of vital and activity signals without problems. The
discrete signals (movement, pressure and temperature, for example) are quickly transmitted.
Nevertheless, spending 5 s to transmit an signal sample or event does not represent a big
problem. Moreover, the system can interact with other applications based on information
technologies. Using standards represents an important step for integrating assisted living at
home systems. The systemwas implemented as previously we have described. As mentioned,
the system uses Java programming language in order to describe the activity of the elderly
and take a decision. The system guaranteed the transmission of a packet per less to 1 seconds,
e.g. the baud rate is 57 600 bit s
1
. Other signals, such as temperature, need the same time.
Furthermore, lost packets are tracked, once it is using a cyclic redundancy code (CRC). There
are a lot of sensors which can measure activities and environmental parameters unobtrusively.
Among them, just a few sensors are used in our prototype home. In the future, other useful
sensors will be used in experiments. For fall measurement (Sixsmith & Johnson, 2004b), a
method can be used applied using infrared vision. In addition, microphone/speaker sensors
can be used for tracking and ultrasound sensors also can be used for movement. Other sensors
can be easily incorporated into our system because we have already developed a small-size
multisensor board.
In this sense, we have decided design an accelerometer mote that is small and lightweight that
can be worn comfortably without obstructing normal activities. The wearable mote board has
mounted a 3-axis accelerometer with high resolution (13-bit) measurement at up to 16 g
(Analog Devices ADXL345). Digital output data is formatted as 16-bit twos complement and
is accessible through either a SPI (3- or 4-wire) (or I2C digital interface). The wearable mote
measures the static acceleration of gravity in tilt-sensing applications, as well as dynamic ac-
celeration resulting frommotion or shock. High resolution provided by ADXL345 (4 mg/LSB)
enables measurement of inclination changes less than 1.0
. Several special sensing functions

are provided. Activity and inactivity sensing detect the presence or lack of motion and if
the acceleration on any axis exceeds a user-set level. Tap sensing detects single and double
taps. Free-fall sensing detects if the device is falling. These functions can be mapped to one
of two interrupt output pins. An integrated, patent pending 32-level rst in, rst out (FIFO)
buffer can be used to store data to minimize host processor intervention. Low power modes
Fig. 9. Actor with accelerometer in his waist, log of data and accelometer sensor node proto-
type
enable intelligent motion-based power management with threshold sensing and active accel-
eration measurement at extremely low power dissipation. The mote ts inside a plastic box
measuring 441 cm, where the button battery is enclosed in the same package. Clearly, the
placement of the device on the body is of primary concern. Some of the criteria are that it
should be comfortable and that the device itself should not pose a threat to the wearer in the
event of a fall. For our experiments, we attached the mote to a belt worn around the waist.
We have not done sufcient experiments on elderly people. In this work, the experiments
should be considered preliminary and more data is needed. Figure 9 shows some pictures of
accelerometer sensor node and our proofs.
In the literature there is an absence of research data on a persons movement in his or her own
house that is not biased by self-report or by third party observation. We are in the process
of several threads of analysis that would provide more sophisticated capabilities for future
versions of the intelligent software. The assisted living system is a heterogenous wireless
network using and ZigBee radios to connect a diverse set of embedded sensor devices. These
devices and the wireless network can monitor the elderly activity in a secure and private
manner and issue alerts to the user, care givers or emergency services as necessary to provide
additional safety and security to the user. This system is being developed to provide this
safety and security so that elder citizens who might have to leave their own homes for a
group care facility will be able to extend their ability to remain at home longer. This will in
most cases provide themwith better quality of life and better health in a cost effective manner.
Fig. 10. Monitoring proofs with ssh communication at a patient residence
Also think that this assisted living system can be used in diagnostic because the activity data
can show indicators of illness. We think that changes in daily activity patterns can suggest
serious conditions and reveal abnormalities of the elderly resident. In summary, we think that
our Custodial Care system could be quite well-received by the elderly residents. We think
that the infrastructure will need to, i) deal robustly with a wide range of different homes and
scenarios, ii) be very reliable in diverse operating conditions, iii) communicate securely with
well-authenticated parties who are granted proper access to the information, iv) respect the
privacy of its users, and v) provide QoS even in the presence of wireless interference and
other environmental effects. We are continuing working on these issues. Figure 10 shows a
real scenario where we can see the log in the left when a resident is lying in the bed.
5. Summary
Assistence living at home care represents a growing eld in the social services. It reduces costs
and increases the quality of life of assisted citizen. As the modern life becomes more stressful
and acute diseases appear, prolonged assistence become more necessary. The same occurs
for the handicapped patients. Home care offers the possibility of assistence in the patients
house, with the assistance of the family. It reduces the need of transporting patients between
house and hospital. The assistence living at home routines can be switched by telemedicine
applications. Actually, this switch is also called telehomecare, which can be dened as the use
of information and communication technologies to enable effective delivery and management
of health services at a patients residence.
Summing up, we have reviewed the state of the art of technologies that allow the use of wire-
less sensor networks in AAL. More specically, technology based on the sensor nodes (WNs)
that conform it. We have proposed a wireless sensor network infrastructure for assisted liv-
ing at home using WSNs technology. These technologies can reduce or eliminate the need for
personal services in the home and can also improve treatment in residences for the elderly
and caregiver facilities. We have introduced its system architecture, power management, self-
conguration of network and routing. In this chapter, a multihop low-power network pro-
tocol has been presented for network conguration and routing since it can be considered
as a natural and appropriate choice for ZigBee networks. This network protocol is modied
of original protocol of Crossbow because our protocol is based in events and is not based
in timers. Moreover, it can give many advantages from the viewpoint of power network and
mediumaccess. Also, we have developed multisensors board for the nodes which can directly
drive events towards an USB base station with the help of our ZigBee multihop low-power
protocol. In this way, and by means of distributed sensors (motes) installed in each of rooms
in the home we can know the activities and the elderly location. A base station (a special mote
developed by us too) is connected to a gateway (miniPC) by means an USB connector which
is responsible of determining an appropriate response using an intelligent software, i.e. pas-
sive infra-red movement sensor might send an event at which point and moment towards the
gateway via base station for its processing. This software is in development in this moment
therefore is partially operative.
DIA project intends to be developed with participatory design between the users, care
providers and developers. With the WSN infrastructure in place, sensor devices will be iden-
tied for development and implemented as the system is expanded in a modular manner to
include a wide selection of devices. In conclusion, the non-invasive monitoring technologies
presented here could provide effective care coordination tools that, in our opinion, could be
accepted by elderly residents, and could have a positive impact on their quality of life. The
rst prototype home in which this is being tested is located in the Region de Murcia, Spain.
Follow these tests, the system will be shared with our partners for further evaluation in group
care facilities, hospitals and homes in our region.
6. Acknowledgments
The authors gratefully acknowledge the contribution of Spanish Ministry of Ciencia e Inno-
vacin (MICINN) and reviewers comments. This work was supported by the Spanish Min-
istry of Ciencia e Innovacin (MICINN) under grant TIN2009-14372-C03-02.
7. References
Al-Karaki, J. & Kamal, A. (2004). Routing techniques in wireless sensor networks: a survey,
11(6): 628.
Biemer, M. & Hampe, J. F. (2005). A mobile medical monitoring system: Concept, design and
deployment, ICMB 05: Proceedings of the International Conference on Mobile Business,
IEEE Computer Society, Washington, DC, USA, pp. 464471.
Bilstrup, U. & Wiberg, P.-A. (2004). An architecture comparison between a wireless sensor
network and an active rd system, Local Computer Networks, 2004. 29th Annual IEEE
International Conference on, pp. 583584.
Bota-Blaya, J., Palma, J., Villa, A., Prez, D. & Iborra, E. (2009). Ontology based approach to
the detection of domestic problems for independent senior people, IWINAC09, Inter-
national Work-Conference on the Interpalay Between Natural and Articial Compu-
tation, IWINAC, pp. 5564.
Cho, N., Song, S.-J., Kim, S., Kim, S. & Yoo, H.-J. (2005). A 5.1-w uhf rd tag chip integrated
with sensors for wireless environmental monitoring, Solid-State Circuits Conference,
2005. ESSCIRC 2005. Proceedings of the 31st European, pp. 279282.
Fernndez-Luque, F., Zapata, J., Ruiz, R. & Iborra, E. (2009). A wireless sensor network for
assisted living at home of elderly people, IWINAC 09: Proceedings of the 3rd Inter-
national Work-Conference on The Interplay Between Natural and Articial Computation,
Springer-Verlag, Berlin, Heidelberg, pp. 6574.
Fig. 10. Monitoring proofs with ssh communication at a patient residence
Also think that this assisted living system can be used in diagnostic because the activity data
can show indicators of illness. We think that changes in daily activity patterns can suggest
serious conditions and reveal abnormalities of the elderly resident. In summary, we think that
our Custodial Care system could be quite well-received by the elderly residents. We think
that the infrastructure will need to, i) deal robustly with a wide range of different homes and
scenarios, ii) be very reliable in diverse operating conditions, iii) communicate securely with
well-authenticated parties who are granted proper access to the information, iv) respect the
privacy of its users, and v) provide QoS even in the presence of wireless interference and
other environmental effects. We are continuing working on these issues. Figure 10 shows a
real scenario where we can see the log in the left when a resident is lying in the bed.
5. Summary
Assistence living at home care represents a growing eld in the social services. It reduces costs
and increases the quality of life of assisted citizen. As the modern life becomes more stressful
and acute diseases appear, prolonged assistence become more necessary. The same occurs
for the handicapped patients. Home care offers the possibility of assistence in the patients
house, with the assistance of the family. It reduces the need of transporting patients between
house and hospital. The assistence living at home routines can be switched by telemedicine
applications. Actually, this switch is also called telehomecare, which can be dened as the use
of information and communication technologies to enable effective delivery and management
of health services at a patients residence.
Summing up, we have reviewed the state of the art of technologies that allow the use of wire-
less sensor networks in AAL. More specically, technology based on the sensor nodes (WNs)
that conform it. We have proposed a wireless sensor network infrastructure for assisted liv-
ing at home using WSNs technology. These technologies can reduce or eliminate the need for
personal services in the home and can also improve treatment in residences for the elderly
and caregiver facilities. We have introduced its system architecture, power management, self-
conguration of network and routing. In this chapter, a multihop low-power network pro-
tocol has been presented for network conguration and routing since it can be considered
as a natural and appropriate choice for ZigBee networks. This network protocol is modied
of original protocol of Crossbow because our protocol is based in events and is not based
in timers. Moreover, it can give many advantages from the viewpoint of power network and
mediumaccess. Also, we have developed multisensors board for the nodes which can directly
drive events towards an USB base station with the help of our ZigBee multihop low-power
protocol. In this way, and by means of distributed sensors (motes) installed in each of rooms
in the home we can know the activities and the elderly location. A base station (a special mote
developed by us too) is connected to a gateway (miniPC) by means an USB connector which
is responsible of determining an appropriate response using an intelligent software, i.e. pas-
sive infra-red movement sensor might send an event at which point and moment towards the
gateway via base station for its processing. This software is in development in this moment
therefore is partially operative.
DIA project intends to be developed with participatory design between the users, care
providers and developers. With the WSN infrastructure in place, sensor devices will be iden-
tied for development and implemented as the system is expanded in a modular manner to
include a wide selection of devices. In conclusion, the non-invasive monitoring technologies
presented here could provide effective care coordination tools that, in our opinion, could be
accepted by elderly residents, and could have a positive impact on their quality of life. The
rst prototype home in which this is being tested is located in the Region de Murcia, Spain.
Follow these tests, the system will be shared with our partners for further evaluation in group
care facilities, hospitals and homes in our region.
6. Acknowledgments
The authors gratefully acknowledge the contribution of Spanish Ministry of Ciencia e Inno-
vacin (MICINN) and reviewers comments. This work was supported by the Spanish Min-
istry of Ciencia e Innovacin (MICINN) under grant TIN2009-14372-C03-02.
7. References
Al-Karaki, J. & Kamal, A. (2004). Routing techniques in wireless sensor networks: a survey,
11(6): 628.
Biemer, M. & Hampe, J. F. (2005). A mobile medical monitoring system: Concept, design and
deployment, ICMB 05: Proceedings of the International Conference on Mobile Business,
IEEE Computer Society, Washington, DC, USA, pp. 464471.
Bilstrup, U. & Wiberg, P.-A. (2004). An architecture comparison between a wireless sensor
network and an active rd system, Local Computer Networks, 2004. 29th Annual IEEE
International Conference on, pp. 583584.
Bota-Blaya, J., Palma, J., Villa, A., Prez, D. & Iborra, E. (2009). Ontology based approach to
the detection of domestic problems for independent senior people, IWINAC09, Inter-
national Work-Conference on the Interpalay Between Natural and Articial Compu-
tation, IWINAC, pp. 5564.
Cho, N., Song, S.-J., Kim, S., Kim, S. & Yoo, H.-J. (2005). A 5.1-w uhf rd tag chip integrated
with sensors for wireless environmental monitoring, Solid-State Circuits Conference,
2005. ESSCIRC 2005. Proceedings of the 31st European, pp. 279282.
Fernndez-Luque, F., Zapata, J., Ruiz, R. & Iborra, E. (2009). A wireless sensor network for
assisted living at home of elderly people, IWINAC 09: Proceedings of the 3rd Inter-
national Work-Conference on The Interplay Between Natural and Articial Computation,
Springer-Verlag, Berlin, Heidelberg, pp. 6574.
Horton, M. & Suh, J. (2005). A vision for wireless sensor networks, Proc. IEEE MTT-S Interna-
tional Microwave Symposium Digest, p. 4pp.
JLHLabs (2008). WSN Lab.
URL: http://www.jlhlabs.com/
Kahn, J. M., Katz, R. H. & Pister, K. S. J. (1999). Next century challenges: mobile network-
ing for "smart dust", MobiCom 99: Proceedings of the 5th annual ACM/IEEE interna-
tional conference on Mobile computing and networking, ACM Press, New York, NY, USA,
pp. 271278.
URL: http://dx.doi.org/10.1145/313451.313558
Li, Y., Thai, M. T. & Wu, W. (2008). Wireless Sensor Networks And Applications, Springer.
Lubrin, E., Lawrence, E. & Navarro, K. F. (2006). Motecare: an adaptive smart ban health
monitoring system, BioMed06: Proceedings of the 24th IASTED international conference
on Biomedical engineering, ACTA Press, Anaheim, CA, USA, pp. 6067.
Martin, H., Bernardos, A., Bergesio, L. & Tarrio, P. (2009). Analysis of key aspects to manage
wireless sensor networks in ambient assisted living environments, Applied Sciences in
Biomedical and Communication Technologies, 2009. ISABEL 2009. 2nd International Sym-
posium on, pp. 1 8.
MIT (2008). MIT WSN Research Group.
URL :http://mtlweb.mit.edu/researchgroups/icsystems/gallery.
html.
Pister, K. (2008). Smart dust, autonomous sensing and communication in a cubic millimeter.
URL: http://robotics.eecs.berkeley.edu/~pister/SmartDust/
Ross, P. (2004). Managing care through the air [remote health monitoring], Spectrum, IEEE
41(12): 2631.
Rowan, J. & Mynatt, E. D. (2005). Digital family portrait eld trial: Support for aging in place,
CHI 05: Proceedings of the SIGCHI conference on Human factors in computing systems,
ACM, New York, NY, USA, pp. 521530.
Sagduyu, Y. & Ephremides, A. (2004). The problem of medium access control in wireless
sensor networks, 11(6): 4453.
Sixsmith, A. & Johnson, N. (2004a). A smart sensor to detect the falls of the elderly, Pervasive
Computing, IEEE 3(2): 4247.
Sixsmith, A. & Johnson, N. (2004b). A smart sensor to detect the falls of the elderly, 3(2): 4247.
Sohraby, K., Minoli, D. & Znati, T. (2007). Wireless Sensor Networks: Technology, Protocols, and
Applications, John Wiley and Sons.
URL : http://www.wiley.com/WileyCDA/WileyTitle/
productCd-0471743003.html.
TinyOS (2009). Tinyos website.
URL: http://www.tinyos.net/
UCLA (2009). UCLA WSN Projects.
URL: http://nesl.ee.ucla.edu/projects/ahlos/mk2
Monitoring of human movements for fall detection and activities
recognition in elderly care using wireless sensor network: a survey 147
recognition in elderly care using wireless sensor network: a survey
Stefano Abbate, Marco Avvenuti, Paolo Corsini, Alessio Vecchio and Janet Light
0
Monitoring of human movements for fall
detection and activities recognition in elderly
care using wireless sensor network: a survey
Stefano Abbate
IMT Institute for Advanced Studies Lucca
Italy
Marco Avvenuti, Paolo Corsini and Alessio Vecchio
University of Pisa
Italy
Janet Light
University of New Brunswick
Canada
1. Introduction
The problem with accidental falls among elderly people has massive social and economic
impacts. Falls in elderly people are the main cause of admission and extended period of stay
in a hospital. It is the sixth cause of death for people over the age of 65, the second for people
between 65 and 75, and the rst for people over 75. Among people affected by Alzheimers
Disease, the probability of a fall increases by a factor of three.
Elderly care can be improved by using sensors that monitor the vital signs and activities of
patients, and remotely communicate this information to their doctors and caregivers. For
example, sensors installed in homes can alert caregivers when a patient falls. Research teams
in universities and industries are developing monitoring technologies for in-home elderly
care. They make use of a network of sensors including pressure sensors on chairs, cameras,
and RFID tags embedded throughout the home of the elderly people as well as in furniture
and clothing, which communicate with tag readers in oor mats, shelves, and walls.
Afall can occur not only when a person is standing, but also while sitting on a chair or lying on
a bed during sleep. The consequences of a fall can vary from scrapes to fractures and in some
cases lead to death. Even if there are no immediate consequences, the long-wait on the oor
for help increases the probability of death from the accident. This underlines the importance
of real-time monitoring and detection of a fall to enable rst-aid by relatives, paramedics or
caregivers as soon as possible.
Monitoring the activities of daily living (ADL) is often related to the fall problem and requires
a non-intrusive technology such as a wireless sensor network. An elderly with risk of fall
can be instrumented with (preferably) one wireless sensing device to capture and analyze the
9
body movements continuously, and the system triggers an alarm when a fall is detected. The
small size and the light weight make the sensor network an ideal candidate to handle the fall
problem.
The development of new techniques and technologies demonstrates that a major effort has
been taken during the past 30 years to address this issue. However, the researchers took many
different approaches to solve the problem without following any standard testing guidelines.
In some studies, they proposed their own guidelines.
In this Chapter, a contribution is made towards such a standardization by collecting the most
relevant parameters, data ltering techniques and testing approaches from the studies done
so far. State-of-the-art fall detection techniques were surveyed, highlighting the differences in
their effectiveness at fall detection. A standard database structure was created for fall study
that emphasizes the most important elements of a fall detection system that must be consid-
ered for designing a robust system, as well as addressing the constraints and challenges.
1.1 Denitions
A fall can be dened in different ways based on the aspects studied. The focus in this study
is on the kinematic analysis of the human movements. A a suitable denition of a fall is
Unintentionally coming to the ground or some lower level and other than as a consequence
of sustaining a violent blow, loss of consciousness, sudden onset of paralysis as in stroke or an
epileptic seizure. (Gibson et al., 1987). It is always possible to easily re-adapt this denition
to address the specic goals a researcher wants to pursue.
In terms of human anatomy, a fall usually occurs along one of two planes, called sagittal and
coronal planes. Figure 1(a) shows the sagittal plane, that is an X-Z imaginary plane that travels
vertically from the top to the bottom of the body, dividing it into left and right portions. In
this case a fall along the sagittal plane can occur forward or backward. Figure 1(b) shows the
coronal Y-Z plane, which divides the body into dorsal and ventral (back and front) portions.
The coronal plane is orthogonal to the sagittal plane and is therefore considered for lateral
falls (right or left). Note that if the person is standing without moving, that is, he or she is
in a static position, the fall occurs following in the down direction. The sense of x, y and z
are usually chosen in order to have positive z-values of the acceleration component when the
body is falling.
(a) Along sagittal plane (b) Along coronal plane
Fig. 1. Fall directions
Toppling simply refers to a loss in balance. Figure 2(a) shows the body from a kinematic point
of view. When the vertical line through the center of gravity lies outside the base of support
the body starts toppling. If there is no reaction to this loss of balance, the body falls on the
ground (Chapman, 2008).
Let us now consider the fall of a body from a stationary position at height h = H. Initially the
body has a potential energy mgh which is transformed into kinetic energy during the fall with
the highest value just before the impact on the oor (h = 0). During the impact the energy
is totally absorbed by the body and, after the impact, both potential and kinetic energy are
equal to zero. If the person is conscious the energy can be absorbed by the his muscles, for
example, using the arms (see Figure 2(b)), whereas if the person is unconscious it can lead to
sever injuries (see Figure 2(c)).
(a) Toppling
(b) Conscious fall (c) Unconscious fall
Fig. 2. Kinematic analysis of a fall
Strictly related to a fall is the posture, a conguration of the human body that is assumed inten-
tionally or habitually. Some examples are standing, sitting, bending and lying. A posture can
be determined by monitoring the tilt transition of the trunk and legs, the angular coordinates
of which are shown in Figure 3(a) and Figure 3(b) (Li et al., 2009; Yang & Hsu, 2007). The
ability to detect a posture helps to determine if there has been a fall.
(a) Trunk (b) Legs
Fig. 3. Angular coordinates
body movements continuously, and the system triggers an alarm when a fall is detected. The
small size and the light weight make the sensor network an ideal candidate to handle the fall
problem.
The development of new techniques and technologies demonstrates that a major effort has
been taken during the past 30 years to address this issue. However, the researchers took many
different approaches to solve the problem without following any standard testing guidelines.
In some studies, they proposed their own guidelines.
In this Chapter, a contribution is made towards such a standardization by collecting the most
relevant parameters, data ltering techniques and testing approaches from the studies done
so far. State-of-the-art fall detection techniques were surveyed, highlighting the differences in
their effectiveness at fall detection. A standard database structure was created for fall study
that emphasizes the most important elements of a fall detection system that must be consid-
ered for designing a robust system, as well as addressing the constraints and challenges.
1.1 Denitions
A fall can be dened in different ways based on the aspects studied. The focus in this study
is on the kinematic analysis of the human movements. A a suitable denition of a fall is
Unintentionally coming to the ground or some lower level and other than as a consequence
of sustaining a violent blow, loss of consciousness, sudden onset of paralysis as in stroke or an
epileptic seizure. (Gibson et al., 1987). It is always possible to easily re-adapt this denition
to address the specic goals a researcher wants to pursue.
In terms of human anatomy, a fall usually occurs along one of two planes, called sagittal and
coronal planes. Figure 1(a) shows the sagittal plane, that is an X-Z imaginary plane that travels
vertically from the top to the bottom of the body, dividing it into left and right portions. In
this case a fall along the sagittal plane can occur forward or backward. Figure 1(b) shows the
coronal Y-Z plane, which divides the body into dorsal and ventral (back and front) portions.
The coronal plane is orthogonal to the sagittal plane and is therefore considered for lateral
falls (right or left). Note that if the person is standing without moving, that is, he or she is
in a static position, the fall occurs following in the down direction. The sense of x, y and z
are usually chosen in order to have positive z-values of the acceleration component when the
body is falling.
(a) Along sagittal plane (b) Along coronal plane
Fig. 1. Fall directions
Toppling simply refers to a loss in balance. Figure 2(a) shows the body from a kinematic point
of view. When the vertical line through the center of gravity lies outside the base of support
the body starts toppling. If there is no reaction to this loss of balance, the body falls on the
ground (Chapman, 2008).
Let us now consider the fall of a body from a stationary position at height h = H. Initially the
body has a potential energy mgh which is transformed into kinetic energy during the fall with
the highest value just before the impact on the oor (h = 0). During the impact the energy
is totally absorbed by the body and, after the impact, both potential and kinetic energy are
equal to zero. If the person is conscious the energy can be absorbed by the his muscles, for
example, using the arms (see Figure 2(b)), whereas if the person is unconscious it can lead to
sever injuries (see Figure 2(c)).
(a) Toppling
(b) Conscious fall (c) Unconscious fall
Fig. 2. Kinematic analysis of a fall
Strictly related to a fall is the posture, a conguration of the human body that is assumed inten-
tionally or habitually. Some examples are standing, sitting, bending and lying. A posture can
be determined by monitoring the tilt transition of the trunk and legs, the angular coordinates
of which are shown in Figure 3(a) and Figure 3(b) (Li et al., 2009; Yang & Hsu, 2007). The
ability to detect a posture helps to determine if there has been a fall.
(a) Trunk (b) Legs
Fig. 3. Angular coordinates
1.2 Related Surveys of Research on Patient Monitoring Technologies
So far, a few surveys on fall detection systems have been written and extended. Some of them
propose their own standards and this is useful for people already working on the problem
of fall detection. This survey provides a comprehensive, if not exhaustive, guide from the
rst-hand approach of the problem, highlighting the best practices to merge valid but hetero-
geneous procedures.
The rst survey on fall detection by Noury et al. (2007) describes the systems, algorithms and
sensors used in the detection of a fall in elderly people. After an overview of the state-of-
the-art techniques, they discovered the lack of a common framework and hence proposed
some performance evaluation parameters in order to compare the different systems. These
parameters had to be evaluated for a set of falling scenarios that included real falls and actions
related to falls.
Yu (2008) focused on a classication of the approaches and principles of existing fall detection
methods. He also provided a classication of falls and a general framework of fall detection,
alert device and system schema.
The authors of Noury et al. (2007) described the in-depth sequence of falling (Noury et al.,
2008). They stated that it was difcult to compare academic studies because the conditions
of assessment are not always reported. This led to the evaluation of not only the above de-
scribed parameters and scenarios, but also of other objective criteria such as detection method,
usability and lifespan of a device.
In applications involving accelerometers, Kangas et al. (2007) used accelerometry-based pa-
rameters to determine thresholds for fall detection. The posture information was used to
distinguish between falls and activities of daily living. Their experiments showed the most
suitable placement for the sensor to be waist and the head, whereas placing the sensor on the
wrist gave rise to additional problems.
2. Fall risk factors
A person can be more or less prone to fall, depending on a number of risk factors and hence
a classication based on only age as a parameter is not enough. In fact, medical studies have
determined a set of so called risk factors:
Intrinsic:
Age (over 65)
Low mobility and bone fragility
Poor balance
Chronic disease
Cognitive and dementia problems
Parkinson disease
Sight problems
Use of drugs that affect the mind
Incorrect lifestyle (inactivity, use of alcohol, obesity)
Previous falls
Extrinsic:
Individual (incorrect use of shoes and clothes)
Drugs cocktail
Internal Environment:
Slipping oors
Stairs
Need to reach high objects
External Environment:
Damaged roads
Crowded places
Dangerous steps
Poor lighting
There is a clear correlation between the above list and the probability of fall. The number of
people that fall are as follows (Tinetti et al., 1988):
8% of people without any of risk factors
27% of people with only one risk factor
78% of people with four or more risk factors
The history of the falls is also important since people who have already fallen two times are
more at risk to fall again. This can be due to psychological (fear, shame, loss of self-esteem),
and/or physical (injuries, lack of exercise) reasons.
3. How, where and why people fall
Among elderly people that live at home, almost half of the falls take place near or inside the
house (Campbell et al., 1990; Lipsitz et al., 1991). Usually women fall in the kitchen whereas
men fall in the garden (Lord et al., 1993).
The rate of falls increases signicantly among elderly people living in nursing homes: at least
40% of the patients fell twice or more within 6 months. This rate is ve times more with
respect to the rate of fall when people live at home. This may be due to people having to
aquaint themselves with the new living enviroment and its obstacles.
3.1 Physical causes
The factors that lead to most of the falls in people over 65 are to stumble on obstacles or steps
and to slip on a smooth surface. The fall is usually caused by loss of balance due to dizziness.
Approximately 14% of people do not know why they fall and a smaller number of people
state that the fall is due to the fragility of the lower limbs (Lord et al., 1993).
Further researchers determined that traditional fall prevention measures such as bed rails can
make the fall worse (Masud & Morris, 2001).
3.2 Activities
Most of the falls happen during the activities of daily living (ADL) that involve a small loss of
balance such as standing or walking. Fewer falls happen during daily activities that involve
a more signicant movement such as sitting on a chair or climbing the stairs. Conversely,
activities usually dened dangerous, such as jogging or physical exercises are less likely to
increase the probability of a fall (Tinetti et al., 1988). There are more falls during the day than
during the night (Campbell et al., 1990).
1.2 Related Surveys of Research on Patient Monitoring Technologies
So far, a few surveys on fall detection systems have been written and extended. Some of them
propose their own standards and this is useful for people already working on the problem
of fall detection. This survey provides a comprehensive, if not exhaustive, guide from the
rst-hand approach of the problem, highlighting the best practices to merge valid but hetero-
geneous procedures.
The rst survey on fall detection by Noury et al. (2007) describes the systems, algorithms and
sensors used in the detection of a fall in elderly people. After an overview of the state-of-
the-art techniques, they discovered the lack of a common framework and hence proposed
some performance evaluation parameters in order to compare the different systems. These
parameters had to be evaluated for a set of falling scenarios that included real falls and actions
related to falls.
Yu (2008) focused on a classication of the approaches and principles of existing fall detection
methods. He also provided a classication of falls and a general framework of fall detection,
alert device and system schema.
The authors of Noury et al. (2007) described the in-depth sequence of falling (Noury et al.,
2008). They stated that it was difcult to compare academic studies because the conditions
of assessment are not always reported. This led to the evaluation of not only the above de-
scribed parameters and scenarios, but also of other objective criteria such as detection method,
usability and lifespan of a device.
In applications involving accelerometers, Kangas et al. (2007) used accelerometry-based pa-
rameters to determine thresholds for fall detection. The posture information was used to
distinguish between falls and activities of daily living. Their experiments showed the most
suitable placement for the sensor to be waist and the head, whereas placing the sensor on the
wrist gave rise to additional problems.
2. Fall risk factors
A person can be more or less prone to fall, depending on a number of risk factors and hence
a classication based on only age as a parameter is not enough. In fact, medical studies have
determined a set of so called risk factors:
Intrinsic:
Age (over 65)
Low mobility and bone fragility
Poor balance
Chronic disease
Cognitive and dementia problems
Parkinson disease
Sight problems
Use of drugs that affect the mind
Incorrect lifestyle (inactivity, use of alcohol, obesity)
Previous falls
Extrinsic:
Individual (incorrect use of shoes and clothes)
Drugs cocktail
Internal Environment:
Slipping oors
Stairs
Need to reach high objects
External Environment:
Damaged roads
Crowded places
Dangerous steps
Poor lighting
There is a clear correlation between the above list and the probability of fall. The number of
people that fall are as follows (Tinetti et al., 1988):
8% of people without any of risk factors
27% of people with only one risk factor
78% of people with four or more risk factors
The history of the falls is also important since people who have already fallen two times are
more at risk to fall again. This can be due to psychological (fear, shame, loss of self-esteem),
and/or physical (injuries, lack of exercise) reasons.
3. How, where and why people fall
Among elderly people that live at home, almost half of the falls take place near or inside the
house (Campbell et al., 1990; Lipsitz et al., 1991). Usually women fall in the kitchen whereas
men fall in the garden (Lord et al., 1993).
The rate of falls increases signicantly among elderly people living in nursing homes: at least
40% of the patients fell twice or more within 6 months. This rate is ve times more with
respect to the rate of fall when people live at home. This may be due to people having to
aquaint themselves with the new living enviroment and its obstacles.
3.1 Physical causes
The factors that lead to most of the falls in people over 65 are to stumble on obstacles or steps
and to slip on a smooth surface. The fall is usually caused by loss of balance due to dizziness.
Approximately 14% of people do not know why they fall and a smaller number of people
state that the fall is due to the fragility of the lower limbs (Lord et al., 1993).
Further researchers determined that traditional fall prevention measures such as bed rails can
make the fall worse (Masud & Morris, 2001).
3.2 Activities
Most of the falls happen during the activities of daily living (ADL) that involve a small loss of
balance such as standing or walking. Fewer falls happen during daily activities that involve
a more signicant movement such as sitting on a chair or climbing the stairs. Conversely,
activities usually dened dangerous, such as jogging or physical exercises are less likely to
increase the probability of a fall (Tinetti et al., 1988). There are more falls during the day than
during the night (Campbell et al., 1990).
3.3 Consequences
Accidental falls are the main cause of admission in a hospital and the sixth cause of death for
people over 65. For people aged between 65 and 75 accidental falls are the second cause of
death and the rst cause in those over 75 (Bradley et al., 2009).
3.3.1 Physical damage
Scratches and bruises are the soft injures due to a fall (Bradley et al., 2009). In the worst
cases the injuries are concentrated on the lower part of the body, mainly on the hip. On the
upper part of the body the head and the trunk injuries are the most frequent. About 66% of
admissions to an hospital are due to at least one fracture. The fracture of elbow and forearm
are more frequent but hip fracture is the most difcult to recover from. Such a fracture in fact
requires a long recovery period and involves the loss of independence and mobility.
Sometimes, when a person falls and is not able to stand up by himself, he lies down on the
oor for long time. This leads to additional health problems such as hypothermia, confusion,
complications and in extreme cases can cause death (Lord et al., 2001).
3.3.2 Psychological damage
A fall also involves hidden damages that affect the self-condence of a person (Lord et al.,
2001). Common consequences are fear, loss of independence, limited capabilities, low self-
esteem and generally, a lower quality of life.
3.3.3 Economic damage
The direct costs associated with falls are due to the medical examinations, hospital recoveries,
rehabilitation treatments, tools of aid (such as wheelchairs, canes etc.) and caregivers service
cost (Englander & Hodson, 1996).
Indirect costs concern the death of patients and their consequences. Recent studies have de-
termined that in the year 2000 alone fall-related expenses was above 19 billion dollars and it
is estimated to reach 54.9 billion in 2020. This shows that year by year, health costs due to the
falls are increasing dramatically (Massachusetts Department of Public Health, 2008).
3.4 Anatomy of a fall
A fall is generally the consequence of a normal activity of daily living and is triggered by a
hard-predictable event such as tripping over, slipping or loss of balance. Once the fall and
thus the impact on the oor occur, the subject usually lies down for some seconds or even
hours and then tries to recover by himself or with the help of someone else. Just before the
impact, the body of the subject is in a free-fall, its acceleration is the same as the gravitational
acceleration. Thus, it is possible to distinguish ve phases as depicted in Figure 4:
1. Activity of Daily Living
2. Hard-predictable event
3. Free-fall
4. Impact
5. Recovery (optional)
(a) Activity of
Daily Living
(b) Hard pre-
dictable event
(c) Free-fall (d) Impact (e) Recovery
Fig. 4. Anatomy of a fall
Note that there are activities of daily living that can be wrongly detected as falls, e.g. falling
on a chair.
4. Typical fall scenarios
The most important scenarios of falls are described by Yu (2008) in detail:
Fall from standing
1. It lasts from 1 to 2 seconds.
2. In the beginning the person is standing. At the end the head is stuck on the oor
for a certain amount of time.
3. A person falls along one direction and the head and the center of mass move along
a plane.
4. The height of the head varies from the height while standing and the height of the
oor.
5. During the fall the head is in free-fall.
6. After the fall the head lays in a virtual circle that is centered in the position of the
feet before the fall and has radius the height of the person.
Fall from chair
2. In the beginning the height of the head varies from the height of the chair to the
height of the oor.
4. After the fall the body is near the chair.
Fall from bed
2. In the beginning the person is lying.
3. The height of the body varies from the height of the bed to the height of the oor.
5. After the fall the body is near the bed.
3.3 Consequences
Accidental falls are the main cause of admission in a hospital and the sixth cause of death for
people over 65. For people aged between 65 and 75 accidental falls are the second cause of
death and the rst cause in those over 75 (Bradley et al., 2009).
3.3.1 Physical damage
Scratches and bruises are the soft injures due to a fall (Bradley et al., 2009). In the worst
cases the injuries are concentrated on the lower part of the body, mainly on the hip. On the
upper part of the body the head and the trunk injuries are the most frequent. About 66% of
admissions to an hospital are due to at least one fracture. The fracture of elbow and forearm
are more frequent but hip fracture is the most difcult to recover from. Such a fracture in fact
requires a long recovery period and involves the loss of independence and mobility.
Sometimes, when a person falls and is not able to stand up by himself, he lies down on the
oor for long time. This leads to additional health problems such as hypothermia, confusion,
complications and in extreme cases can cause death (Lord et al., 2001).
3.3.2 Psychological damage
A fall also involves hidden damages that affect the self-condence of a person (Lord et al.,
2001). Common consequences are fear, loss of independence, limited capabilities, low self-
esteem and generally, a lower quality of life.
3.3.3 Economic damage
The direct costs associated with falls are due to the medical examinations, hospital recoveries,
rehabilitation treatments, tools of aid (such as wheelchairs, canes etc.) and caregivers service
cost (Englander & Hodson, 1996).
Indirect costs concern the death of patients and their consequences. Recent studies have de-
termined that in the year 2000 alone fall-related expenses was above 19 billion dollars and it
is estimated to reach 54.9 billion in 2020. This shows that year by year, health costs due to the
falls are increasing dramatically (Massachusetts Department of Public Health, 2008).
3.4 Anatomy of a fall
A fall is generally the consequence of a normal activity of daily living and is triggered by a
hard-predictable event such as tripping over, slipping or loss of balance. Once the fall and
thus the impact on the oor occur, the subject usually lies down for some seconds or even
hours and then tries to recover by himself or with the help of someone else. Just before the
impact, the body of the subject is in a free-fall, its acceleration is the same as the gravitational
acceleration. Thus, it is possible to distinguish ve phases as depicted in Figure 4:
1. Activity of Daily Living
2. Hard-predictable event
3. Free-fall
4. Impact
5. Recovery (optional)
(a) Activity of
Daily Living
(b) Hard pre-
dictable event
(c) Free-fall (d) Impact (e) Recovery
Fig. 4. Anatomy of a fall
Note that there are activities of daily living that can be wrongly detected as falls, e.g. falling
on a chair.
4. Typical fall scenarios
The most important scenarios of falls are described by Yu (2008) in detail:
Fall from standing
2. In the beginning the person is standing. At the end the head is stuck on the oor
for a certain amount of time.
3. A person falls along one direction and the head and the center of mass move along
a plane.
4. The height of the head varies from the height while standing and the height of the
oor.
6. After the fall the head lays in a virtual circle that is centered in the position of the
feet before the fall and has radius the height of the person.
Fall from chair
2. In the beginning the height of the head varies from the height of the chair to the
height of the oor.
4. After the fall the body is near the chair.
Fall from bed
2. In the beginning the person is lying.
3. The height of the body varies from the height of the bed to the height of the oor.
5. After the fall the body is near the bed.
With the description of the main falls it is possible to simplify the complexity of a fall. This
enables in turn to focus on the resolution of the detection fall problem, rather than on the
reconstruction of a detailed scenario. The simplied and theoretical description often reects
the practical sequence of a fall.
5. Risk assessment tools
A risk assessment tool determines which people are at risk of falls that invoke specic coun-
termeasures, to avoid or at least reduce any injuries (Perell et al., 2001; Vassallo et al., 2008).
There are three fundamental types:
1. Medical exams performed by a geriatrician or other qualied people.
2. Risk factors evaluation performed in a hospital.
3. Evaluation of movement ability performed by a physiotherapist.
Medical exams take into account many parameters including the history of falls, drug therapy,
strength, balance, diet and chronic diseases. However, they are only descriptive tools and
hence do not provide numerical indexes. The risk factors evaluation is performed once a
patient is admitted to a hospital and is based on specic methods and indexes. The evaluation
is then periodically updated and is therefore more useful than the single assessment in the
previous category. The analysis of a person at home performed by a physiotherapist can be
more detailed but also more intrusive. Nevertheless, many researchers do not agree on the
validity of such tools. Oliver (2008) suggests the characteristics essential to an effective risk
assessment tool:
Short-time period to be completed
Parameters to address:
1. High-risk faller
2. Low-risk faller
3. Falls prediction probability
4. Non-falls prediction probability
5. Prediction accuracy
An integrated and on-line monitoring service would provide updated data about the condi-
tion of a patient, a condition that can vary frequently especially in elder people. A step further
from the monitoring of human movements is the monitoring of physiological parameters.
6. Technological approaches to fall detection
There are three main categories of devices based on the technology used:
Vision-based
Environmental
Wearable
A Vision-based approach uses xed cameras that continuously record the movement of the
patients. The acquired data is submitted to specic image algorithms that are able to recognize
the pattern of a fall to trigger an alarm. Vision-based approaches can be classied as:
1. Inactivity detection, based on the idea that after a fall, the patient lies on the oor without
moving.
2. Body shape change analysis, based on the change of posture after the fall.
3. 3D head motion analysis, based on the monitoring the position and velocity of the head.
The main limits of this approach are the time and cost of installation, the limited space of
application (only where there are the cameras) and privacy violation.
The use of Environmental devices is an approach based on the installation of sensors in the
places to be monitored. When people interact with the environment, infrared or pressure
sensors on the oor are able to detect a fall. The problemhere is the presence of false-negatives,
for example, a fall that occurs on a table is not detected.
Both Visual-based and Environmental device approaches require a pre-built infrastructure,
and this enables their use in hospitals and houses, but it is hard to use them outdoor.
In the Wearable approach, one or more wearable devices are worn by the patient. They are usu-
ally equipped with movement sensors such as accelerometers and gyroscopes, whose values
are transmitted via radio and analyzed. This solution offers advantages such as low installa-
tion cost (indoor and outdoor), small size and offers the possibility to also acquire physiolog-
ical data (blood pressure, ECG, EEG etc.).
7. Wireless sensor networks and general system architecture
A wireless sensor network is a set of spatially distributed sensing devices, also called nodes,
that are able to communicate with each other in a wireless ad-hoc network paradigm(Akyildiz
et al., 2002). Each device is usually battery-powered and can be instrumented with one or
more sensors which enable acquisition of physical data such as temperature, body acceleration
and so on. The nodes are able to organize themselves in order to create an ad-hoc routing tree,
whose root is represented by a sink node. The sink node is usually connected to a personal
computer, also called the base station, that will receive all the data sent by nodes (see Figure 5).
Besides the sensing and wireless communication capabilities, the nodes feature a processing
unit that enables local data treatment and ltering. This is important in order to reduce the
use of the radio communication which is the most energy expensive task performed by a node
with respect to sensing and processing.
Fig. 5. Wireless Sensor Network topology
With the description of the main falls it is possible to simplify the complexity of a fall. This
enables in turn to focus on the resolution of the detection fall problem, rather than on the
reconstruction of a detailed scenario. The simplied and theoretical description often reects
the practical sequence of a fall.
5. Risk assessment tools
A risk assessment tool determines which people are at risk of falls that invoke specic coun-
termeasures, to avoid or at least reduce any injuries (Perell et al., 2001; Vassallo et al., 2008).
There are three fundamental types:
1. Medical exams performed by a geriatrician or other qualied people.
2. Risk factors evaluation performed in a hospital.
3. Evaluation of movement ability performed by a physiotherapist.
Medical exams take into account many parameters including the history of falls, drug therapy,
strength, balance, diet and chronic diseases. However, they are only descriptive tools and
hence do not provide numerical indexes. The risk factors evaluation is performed once a
patient is admitted to a hospital and is based on specic methods and indexes. The evaluation
is then periodically updated and is therefore more useful than the single assessment in the
previous category. The analysis of a person at home performed by a physiotherapist can be
more detailed but also more intrusive. Nevertheless, many researchers do not agree on the
validity of such tools. Oliver (2008) suggests the characteristics essential to an effective risk
assessment tool:
Short-time period to be completed
Parameters to address:
1. High-risk faller
2. Low-risk faller
3. Falls prediction probability
4. Non-falls prediction probability
5. Prediction accuracy
An integrated and on-line monitoring service would provide updated data about the condi-
tion of a patient, a condition that can vary frequently especially in elder people. A step further
from the monitoring of human movements is the monitoring of physiological parameters.
6. Technological approaches to fall detection
There are three main categories of devices based on the technology used:
Vision-based
Environmental
Wearable
A Vision-based approach uses xed cameras that continuously record the movement of the
patients. The acquired data is submitted to specic image algorithms that are able to recognize
the pattern of a fall to trigger an alarm. Vision-based approaches can be classied as:
1. Inactivity detection, based on the idea that after a fall, the patient lies on the oor without
moving.
2. Body shape change analysis, based on the change of posture after the fall.
3. 3D head motion analysis, based on the monitoring the position and velocity of the head.
The main limits of this approach are the time and cost of installation, the limited space of
application (only where there are the cameras) and privacy violation.
The use of Environmental devices is an approach based on the installation of sensors in the
places to be monitored. When people interact with the environment, infrared or pressure
sensors on the oor are able to detect a fall. The problemhere is the presence of false-negatives,
for example, a fall that occurs on a table is not detected.
Both Visual-based and Environmental device approaches require a pre-built infrastructure,
and this enables their use in hospitals and houses, but it is hard to use them outdoor.
In the Wearable approach, one or more wearable devices are worn by the patient. They are usu-
ally equipped with movement sensors such as accelerometers and gyroscopes, whose values
are transmitted via radio and analyzed. This solution offers advantages such as low installa-
tion cost (indoor and outdoor), small size and offers the possibility to also acquire physiolog-
ical data (blood pressure, ECG, EEG etc.).
7. Wireless sensor networks and general system architecture
A wireless sensor network is a set of spatially distributed sensing devices, also called nodes,
that are able to communicate with each other in a wireless ad-hoc network paradigm(Akyildiz
et al., 2002). Each device is usually battery-powered and can be instrumented with one or
more sensors which enable acquisition of physical data such as temperature, body acceleration
and so on. The nodes are able to organize themselves in order to create an ad-hoc routing tree,
whose root is represented by a sink node. The sink node is usually connected to a personal
computer, also called the base station, that will receive all the data sent by nodes (see Figure 5).
Besides the sensing and wireless communication capabilities, the nodes feature a processing
unit that enables local data treatment and ltering. This is important in order to reduce the
use of the radio communication which is the most energy expensive task performed by a node
with respect to sensing and processing.
Fig. 5. Wireless Sensor Network topology
The light-weight characteristics of a wireless sensor network perfectly t the needs of a fall
detection system based on the wearable approach. The size, shape and weight of the nodes
enable them to be worn easily by a person. Moreover, many general purpose nodes are com-
mercially available at low-cost. According to the specic need of the study it is possible to
obtain customized hardware with reduced form factor still maintaining the same functional
characteristics. Figure 6(a) shows Tmote-Sky, a general purpose node that is able to sense tem-
perature, humidity and light (Polastre et al., 2005), whereas Figure 6(b) shows SHIMMER, a
smaller size version of the Tmote Sky which is more suitable to be worn by a person (Realtime
Technologies LTD, 2008). The SHIMMER is equipped with a tri-axial accelerometer for move-
ment monitoring and a Secure Digital (SD) slot to locally log a large amount of data. These
platforms enable addition of other sensors such as gyroscopes, in the same board.
(a) Tmote-Sky (b) SHIMMER
Fig. 6. Examples of nodes
Figure 7 shows the general architecture for a human movement monitoring system based on
a wireless sensor network. One or more sensing nodes are used to collect raw data. Analysis
of the data can be performed on the node or on the base station by a more powerful device
such as a smartphone or a laptop. The wireless connectivity standard between the nodes (e.g.
ZigBee) can be different from the one that connects the sink node with the base station (e.g.
Bluetooth). The base station in turn acts as a gateway to communicate with the caregivers
through wireless and/or wired data connection (e.g. Internet or other mobile phones).
Fig. 7. Traditional system architecture
7.1 Node sensors and position
A node for kinematic monitoring is typically instrumented with the following sensors:
Accelerometer, to measure the acceleration.
Gyroscope, to measure the angular velocity.
In particular, the gyroscope requires more energy than the accelerometer. If we connect the
acceleration of the movements with the position of the node worn by the patient, it would be
possible to detect the posture of a person.
The placement of one or more nodes on the body is the key to differentiate the inuences
of various fall detection algorithms. It is not possible to neglect the usability aspect, since it
strongly affects the effectiveness of the system. A node placed on the head gives an excellent
impact detection capability, but more hardware efforts are required to ensure its usability for
wearing the node continuously. The wrist is not recommended to be a good position, since
it is subject to many high acceleration movements that would increase the number of false
positives. The placement at the waist is more acceptable from the user point of the view, since
this option ts well in a belt and it is closer to the center of gravity of the body. There are
many other node locations selected by researchers, such as the armpit, the thigh or the trunk,
quoting their own advantages and disadvantages as explained later. Sometimes the nodes are
inserted in clothes, for example jackets, or in accessories such as watches or necklaces.
8. Performance evaluation parameters and scenarios
8.1 Indexes
A real working fall detection system requires to be sufciently accurate in order to be effective
and alleviate the work of the caregivers. The quality of the system is given by three indexes
that have been proposed based on the four possible situations shown in Table 1:
A fall occurs A fall does not occur
A fall is detected True Positive (TP) False Positive (FP)
A fall is not detected False Negative (FN) True Negative (TN)
Table 1. Possible outputs of a Fall Detection system
Sensitivity is the capacity to detect a fall. It is given by the ratio between the number of
detected falls and the total falls that occurred:
Sensitivity =
TP
TP +FN
(1)
Specicity is the capacity to avoid false positives. Intuitively it is the capacity to detect a
fall only if it really occurs:
Specicity =
TN
TN+FP
(2)
Accuracy is the ability to distinguish and detect both fall (TP) and non-fall movement
(TN):
Accuracy =
TP +TN
P +N
(3)
Where P and N are, respectively, the number of falls performed and the number of
non-falls performed.
The light-weight characteristics of a wireless sensor network perfectly t the needs of a fall
detection system based on the wearable approach. The size, shape and weight of the nodes
enable them to be worn easily by a person. Moreover, many general purpose nodes are com-
mercially available at low-cost. According to the specic need of the study it is possible to
obtain customized hardware with reduced form factor still maintaining the same functional
characteristics. Figure 6(a) shows Tmote-Sky, a general purpose node that is able to sense tem-
perature, humidity and light (Polastre et al., 2005), whereas Figure 6(b) shows SHIMMER, a
smaller size version of the Tmote Sky which is more suitable to be worn by a person (Realtime
Technologies LTD, 2008). The SHIMMER is equipped with a tri-axial accelerometer for move-
ment monitoring and a Secure Digital (SD) slot to locally log a large amount of data. These
platforms enable addition of other sensors such as gyroscopes, in the same board.
(a) Tmote-Sky (b) SHIMMER
Fig. 6. Examples of nodes
Figure 7 shows the general architecture for a human movement monitoring system based on
a wireless sensor network. One or more sensing nodes are used to collect raw data. Analysis
of the data can be performed on the node or on the base station by a more powerful device
such as a smartphone or a laptop. The wireless connectivity standard between the nodes (e.g.
ZigBee) can be different from the one that connects the sink node with the base station (e.g.
Bluetooth). The base station in turn acts as a gateway to communicate with the caregivers
through wireless and/or wired data connection (e.g. Internet or other mobile phones).
Fig. 7. Traditional system architecture
7.1 Node sensors and position
A node for kinematic monitoring is typically instrumented with the following sensors:
Accelerometer, to measure the acceleration.
Gyroscope, to measure the angular velocity.
In particular, the gyroscope requires more energy than the accelerometer. If we connect the
acceleration of the movements with the position of the node worn by the patient, it would be
possible to detect the posture of a person.
The placement of one or more nodes on the body is the key to differentiate the inuences
of various fall detection algorithms. It is not possible to neglect the usability aspect, since it
strongly affects the effectiveness of the system. A node placed on the head gives an excellent
impact detection capability, but more hardware efforts are required to ensure its usability for
wearing the node continuously. The wrist is not recommended to be a good position, since
it is subject to many high acceleration movements that would increase the number of false
positives. The placement at the waist is more acceptable from the user point of the view, since
this option ts well in a belt and it is closer to the center of gravity of the body. There are
many other node locations selected by researchers, such as the armpit, the thigh or the trunk,
quoting their own advantages and disadvantages as explained later. Sometimes the nodes are
inserted in clothes, for example jackets, or in accessories such as watches or necklaces.
8. Performance evaluation parameters and scenarios
8.1 Indexes
A real working fall detection system requires to be sufciently accurate in order to be effective
and alleviate the work of the caregivers. The quality of the system is given by three indexes
that have been proposed based on the four possible situations shown in Table 1:
A fall occurs A fall does not occur
A fall is detected True Positive (TP) False Positive (FP)
A fall is not detected False Negative (FN) True Negative (TN)
Table 1. Possible outputs of a Fall Detection system
Sensitivity is the capacity to detect a fall. It is given by the ratio between the number of
detected falls and the total falls that occurred:
Sensitivity =
TP
TP +FN
(1)
Specicity is the capacity to avoid false positives. Intuitively it is the capacity to detect a
fall only if it really occurs:
Specicity =
TN
TN+FP
(2)
Accuracy is the ability to distinguish and detect both fall (TP) and non-fall movement
(TN):
Accuracy =
TP +TN
P +N
(3)
Where P and N are, respectively, the number of falls performed and the number of
non-falls performed.
Accuracy (Equation 3) is a global index whereas sensitivity and specicity (Equations 1 and 2)
enable a better understanding of the some limits of a system.
A fall exhibits high acceleration or angular velocity which are not normally achievable during
the ADL. If we use a xed low threshold to detect a fall, the sensitivity is 100% but the speci-
city is low because there are fall-like movements like sitting quickly on a chair, a bed or a
sofa which might involve accelerations above that threshold.
8.2 Amplitude parameters
The logged data is sometimes pre-processed by applying some lters: a low-pass lter is used
to perform posture analysis and a high-pass lter is applied to execute motion analysis. How-
ever, this processing is not mandatory and it strongly depends on the fall detection algorithm.
The calibration of the sensors is sometimes neglected or not mentioned in research studies,
but it is an important element that ensures a stable behavior of the system over time.
Amplitude parameters are useful during specic phases of the fall (Dai et al., 2010; Kangas
et al., 2007; 2009).
The Total Sum Vector given in Equation 4 is used to establish the start of a fall:
SV
TOT
(t) =
(A
x
)
2
+ (A
y
)
2
+ (A
z
)
2
(4)
where A
x
, A
y
, A
z
are the gravitational accelerations along the x, y, z-axis.
The Dynamic Sum Vector is obtained using the Total Sum Vector formula applied to accelera-
tions that are ltered with a high-pass lter taking into account fast movements.
The MaxMin Sum Vector given in Equation 5 is used to detect fast changes in the acceleration
signal, which are the differences between the maximum and minimum acceleration values in
a xed-time (t = t
1
t
0
) sliding window for each axis.
SV
MaxMin
(t) = max
t
0
it
1
SV
TOT
(i) min
t
0
jt
1
SV
TOT
(j) (5)
Vertical acceleration given in Equation 6 is calculated considering the sumvectors SV
TOT
(t) and
SV
D
(t) and the gravitational acceleration G.
Z
2
=
SV
2
TOT
(t) SV
2
D
(t) G
2
2G
(6)
8.3 Fall Index
Fall Index in Equation 7 is proposed by (Yoshida et al., 2005). For any sample i in a xed time
window, the Fall Index can be calculated as:
FI
i
=
i19
((A
x
)
i
(A
x
)
i1
)
2
+
i
i19
((A
y
)
i
(A
y
)
i1
)
2
+
i
i19
((A
z
)
i
(A
z
)
i1
)
2
(7)
Since the Fall Index (FI) requires high sampling frequency and fast acceleration changes, it
will miss falls that happen slowly. Hence, FI is not used unless researchers want to compare
the performances of their systems with previous studies that have used it.
8.4 Standard trial scenarios and characteristics
Researcher should agree on a common set of trials in order to test and compare different fall
detection systems. In Table 2 we propose a set of actions for which a fall detection system
should always detect a fall. In Table 3 we propose a set of fall-like activities of daily living that
can lead the system to output false positives. In addition to performing tests on all the listed
36 actions, each research group can combine them in sequential protocols, called circuits (e.g.
sitting, standing, walking, falling).
# Name Symbol Direction Description
1 Front-lying FLY Forward From vertical going forward to the oor
2 Front-protecting-lying FPLY Forward From vertical going forward to the oor
with arm protection
3 Front-knees FKN Forward From vertical going down on the knees
4 Front-knees-lying FKLY Forward From vertical going down on the knees
and then lying on the oor
5 Front-right FR Forward From vertical going down on the oor,
ending in right lateral position
6 Front-left FL Forward From vertical going down on the oor,
ending in left lateral position
7 Front-quick-recovery FQR Forward From vertical going on the oor and
quick recovery
8 Front-slow-recovery FSR Forward From vertical going on the oor and
slow recovery
9 Back-sitting BS Backward From vertical going on the oor, ending
sitting
10 Back-lying BLY Backward From vertical going on the oor, ending
lying
11 Back-right BR Backward From vertical going on the oor, ending
lying in right lateral position
12 Back-left BL Backward From vertical going on the oor, ending
lying in left lateral position
13 Right-sideway RS Right From vertical going on the oor, ending
lying
14 Right-recovery RR Right From vertical going on the oor with
subsequent recovery
15 Left-sideway LS Left From vertical going on the oor, ending
lying
16 Left-recovery LR Left From vertical going on the oor with
subsequent recovery
17 Syncope SYD Down Fromstanding going on the oor follow-
ing a vertical trajectory
18 Syncope-wall SYW Down From standing going down slowly slip-
ping on a wall
19 Podium POD Down From vertical standing on a podium go-
ing on the oor
20 Rolling-out-bed ROBE Lateral From lying, rolling out of bed and going
on the oor
Table 2. Actions to be detected as falls
Accuracy (Equation 3) is a global index whereas sensitivity and specicity (Equations 1 and 2)
enable a better understanding of the some limits of a system.
A fall exhibits high acceleration or angular velocity which are not normally achievable during
the ADL. If we use a xed low threshold to detect a fall, the sensitivity is 100% but the speci-
city is low because there are fall-like movements like sitting quickly on a chair, a bed or a
sofa which might involve accelerations above that threshold.
8.2 Amplitude parameters
The logged data is sometimes pre-processed by applying some lters: a low-pass lter is used
to perform posture analysis and a high-pass lter is applied to execute motion analysis. How-
ever, this processing is not mandatory and it strongly depends on the fall detection algorithm.
The calibration of the sensors is sometimes neglected or not mentioned in research studies,
but it is an important element that ensures a stable behavior of the system over time.
Amplitude parameters are useful during specic phases of the fall (Dai et al., 2010; Kangas
et al., 2007; 2009).
The Total Sum Vector given in Equation 4 is used to establish the start of a fall:
SV
TOT
(t) =
(A
x
)
2
+ (A
y
)
2
+ (A
z
)
2
(4)
where A
x
, A
y
, A
z
are the gravitational accelerations along the x, y, z-axis.
The Dynamic Sum Vector is obtained using the Total Sum Vector formula applied to accelera-
tions that are ltered with a high-pass lter taking into account fast movements.
The MaxMin Sum Vector given in Equation 5 is used to detect fast changes in the acceleration
signal, which are the differences between the maximum and minimum acceleration values in
a xed-time (t = t
1
t
0
) sliding window for each axis.
SV
MaxMin
(t) = max
t
0
it
1
SV
TOT
(i) min
t
0
jt
1
SV
TOT
(j) (5)
Vertical acceleration given in Equation 6 is calculated considering the sumvectors SV
TOT
(t) and
SV
D
(t) and the gravitational acceleration G.
Z
2
=
SV
2
TOT
(t) SV
2
D
(t) G
2
2G
(6)
8.3 Fall Index
Fall Index in Equation 7 is proposed by (Yoshida et al., 2005). For any sample i in a xed time
window, the Fall Index can be calculated as:
FI
i
=
i19
((A
x
)
i
(A
x
)
i1
)
2
+
i
i19
((A
y
)
i
(A
y
)
i1
)
2
+
i
i19
((A
z
)
i
(A
z
)
i1
)
2
(7)
Since the Fall Index (FI) requires high sampling frequency and fast acceleration changes, it
will miss falls that happen slowly. Hence, FI is not used unless researchers want to compare
the performances of their systems with previous studies that have used it.
8.4 Standard trial scenarios and characteristics
Researcher should agree on a common set of trials in order to test and compare different fall
detection systems. In Table 2 we propose a set of actions for which a fall detection system
should always detect a fall. In Table 3 we propose a set of fall-like activities of daily living that
can lead the system to output false positives. In addition to performing tests on all the listed
36 actions, each research group can combine them in sequential protocols, called circuits (e.g.
sitting, standing, walking, falling).
1 Front-lying FLY Forward From vertical going forward to the oor
2 Front-protecting-lying FPLY Forward From vertical going forward to the oor
with arm protection
3 Front-knees FKN Forward From vertical going down on the knees
4 Front-knees-lying FKLY Forward From vertical going down on the knees
and then lying on the oor
5 Front-right FR Forward From vertical going down on the oor,
ending in right lateral position
6 Front-left FL Forward From vertical going down on the oor,
ending in left lateral position
7 Front-quick-recovery FQR Forward From vertical going on the oor and
quick recovery
8 Front-slow-recovery FSR Forward From vertical going on the oor and
slow recovery
9 Back-sitting BS Backward From vertical going on the oor, ending
sitting
10 Back-lying BLY Backward From vertical going on the oor, ending
lying
11 Back-right BR Backward From vertical going on the oor, ending
lying in right lateral position
12 Back-left BL Backward From vertical going on the oor, ending
lying in left lateral position
13 Right-sideway RS Right From vertical going on the oor, ending
lying
14 Right-recovery RR Right From vertical going on the oor with
subsequent recovery
15 Left-sideway LS Left From vertical going on the oor, ending
lying
16 Left-recovery LR Left From vertical going on the oor with
subsequent recovery
17 Syncope SYD Down Fromstanding going on the oor follow-
ing a vertical trajectory
18 Syncope-wall SYW Down From standing going down slowly slip-
ping on a wall
19 Podium POD Down From vertical standing on a podium go-
ing on the oor
20 Rolling-out-bed ROBE Lateral From lying, rolling out of bed and going
on the oor
Table 2. Actions to be detected as falls
21 Lying-bed LYBE Lateral From vertical lying on the bed
22 Rising-bed RIBE Lateral From lying to sitting
23 Sit-bed SIBE Backward Fromvertical sitting with a certain accel-
eration on a bed (soft surface)
24 Sit-chair SCH Backward Fromvertical sitting with a certain accel-
eration on a chair (hard surface)
25 Sit-sofa SSO Backward Fromvertical sitting with a certain accel-
eration on a sofa (soft surface)
26 Sit-air SAI Backward From vertical sitting in the air exploiting
the muscles of legs
27 Walking WAF Forward Walking
28 Jogging JOF Forward Running
29 Walking WAB Backward Walking
30 Bending BEX Forward Bending of about X degrees (0-90)
31 Bending-pick-up BEP Forward Bending to pick up an object on the oor
32 Stumble STU Forward Stumbling with recovery
33 Limp LIM Forward Walking with a limp
34 Squatting-down SQD Down Going down, then up
35 Trip-over TRO Forward Bending while walking and than con-
tinue walking
36 Coughing-sneezing COSN - -
Table 3. Activities that must not be detected as falls
8.4.1 Participant characteristics
Different people have different physical characteristics and therefore it is extremely important
to specify, for each trial, the following ve parameters:
Gender
Age
Weight
Height
Body Mass Index
1
8.4.2 Hardware characteristics
Variation among the technology of the nodes depends on their level of the development and
manufacturing cost. It is therefore important to dene some basic characteristics for the hard-
ware used in trials:
Model
Sampling frequency
Update rate
Movement detection delay time
Range of measurement
Size
1
Body mass index (BMI) is a measure of body fat based on height and weight that applies to adult men
and women.
Weight
Wired/wireless communication protocol
9. Falls study database
Data acquisition is probably the most difcult and time-consuming portion in a fall-detection
study. In the best case, log les of fall trials contain raw accelerations measured during the
simulation of an action (fall or ADL). If other researchers want to access and use such raw ac-
celerations, it is necessary to provide an accurate description of the trials. Moreover, previous
studies generally describe the tests performed and the results obtained, but the acceleration
data is usually not publicly made available. This points out the need for a database with a
standard structure to store all the logs. Such a database is intended to be available to the
scientic community and has two main advantages: on one hand the possibility of storing
and sharing data coming from sensors following a standard format; on the other hand, the
availability of raw sensed data before, during, and after a fall or an activity of daily living that
enables the researchers to test and validate fall detection algorithms using the same test-beds.
A trial or experiment is described in terms of the action performed, the conguration used for
the wearable device and the users prole. Human actions under study are all characterized
by the following aspects: i) posture: users have a particular body orientation before and after
the action is performed; ii) surface: users body is supported by a particular kind of surface
before and after the action is performed. A conguration establishes a particular way to sense
kinematic data, and it can be described in terms of the following: i) position: the device is
worn at some body position; ii) device used: the type of sensor node adopted for the collection
of data. The Entity-Relationship model depicted in Figure 8 is derived from the previous
considerations.
Posture Surface Device
Action
Experiment User
Config
Position
Fig. 8. Database Entity-Relationship diagram
A possible structure of the table is the following:
Postures (ID, posture)
Surfaces (ID, surface)
Action (ID, starting_posture, starting_surface, ending_posture,
ending_surface, description)
21 Lying-bed LYBE Lateral From vertical lying on the bed
22 Rising-bed RIBE Lateral From lying to sitting
23 Sit-bed SIBE Backward Fromvertical sitting with a certain accel-
eration on a bed (soft surface)
24 Sit-chair SCH Backward Fromvertical sitting with a certain accel-
eration on a chair (hard surface)
25 Sit-sofa SSO Backward Fromvertical sitting with a certain accel-
eration on a sofa (soft surface)
26 Sit-air SAI Backward From vertical sitting in the air exploiting
the muscles of legs
27 Walking WAF Forward Walking
28 Jogging JOF Forward Running
29 Walking WAB Backward Walking
30 Bending BEX Forward Bending of about X degrees (0-90)
31 Bending-pick-up BEP Forward Bending to pick up an object on the oor
32 Stumble STU Forward Stumbling with recovery
33 Limp LIM Forward Walking with a limp
34 Squatting-down SQD Down Going down, then up
35 Trip-over TRO Forward Bending while walking and than con-
tinue walking
36 Coughing-sneezing COSN - -
Table 3. Activities that must not be detected as falls
8.4.1 Participant characteristics
Different people have different physical characteristics and therefore it is extremely important
to specify, for each trial, the following ve parameters:
Gender
Age
Weight
Height
Body Mass Index
1
8.4.2 Hardware characteristics
Variation among the technology of the nodes depends on their level of the development and
manufacturing cost. It is therefore important to dene some basic characteristics for the hard-
ware used in trials:
Model
Sampling frequency
Update rate
Movement detection delay time
Range of measurement
Size
1
Body mass index (BMI) is a measure of body fat based on height and weight that applies to adult men
and women.
Weight
Wired/wireless communication protocol
9. Falls study database
Data acquisition is probably the most difcult and time-consuming portion in a fall-detection
study. In the best case, log les of fall trials contain raw accelerations measured during the
simulation of an action (fall or ADL). If other researchers want to access and use such raw ac-
celerations, it is necessary to provide an accurate description of the trials. Moreover, previous
studies generally describe the tests performed and the results obtained, but the acceleration
data is usually not publicly made available. This points out the need for a database with a
standard structure to store all the logs. Such a database is intended to be available to the
scientic community and has two main advantages: on one hand the possibility of storing
and sharing data coming from sensors following a standard format; on the other hand, the
availability of raw sensed data before, during, and after a fall or an activity of daily living that
enables the researchers to test and validate fall detection algorithms using the same test-beds.
A trial or experiment is described in terms of the action performed, the conguration used for
the wearable device and the users prole. Human actions under study are all characterized
by the following aspects: i) posture: users have a particular body orientation before and after
the action is performed; ii) surface: users body is supported by a particular kind of surface
before and after the action is performed. A conguration establishes a particular way to sense
kinematic data, and it can be described in terms of the following: i) position: the device is
worn at some body position; ii) device used: the type of sensor node adopted for the collection
of data. The Entity-Relationship model depicted in Figure 8 is derived from the previous
considerations.
Posture Surface Device
Action
Experiment User
Config
Position
Fig. 8. Database Entity-Relationship diagram
A possible structure of the table is the following:
Postures (ID, posture)
Surfaces (ID, surface)
Action (ID, starting_posture, starting_surface, ending_posture,
ending_surface, description)
Position (ID, position)
Device (ID, manufacturer, model, description, characteristics)
Configuration (ID, record_content, Mote, scale_G, sample_frequency,
Body_position, x_direction, y_direction, z_direction)
Users (ID, age, gender, height_cm, weight_kg, body_mass_index)
Experiments (ID, Configuration, Action, User, content)
Note that we decided to collect, represent, and store extra information, such as the posture
of the user before and after a potential fall, the separate acceleration values and acceleration
magnitude as-well. This has been done to foster the reuse of the collected data and to enable
the evaluation of future techniques on the same sets of data.
10. Overview of fall detection algorithms
From what has been explained so far, many different approaches have been taken to solve the
fall detection problem using accelerometers. The basic and trivial system uses a threshold to
establish if a person falls, which is subject to many false positives. Some researchers have tried
to introduce computationally-hard type of intensive algorithms but the goal has been always
to nd a trade-off between the system accuracy and the cost.
Depeursinge et al. (2001) used a two-level neural network algorithm to analyze the accelera-
tions given by two sensors placed in distinct parts of the body. Such accelerations are trans-
lated into spatial coordinates and fed into the algorithm. The output of the system represents
the probability that a fall is happening: if the probability is low, the system continues moni-
toring whereas if the probability is medium or high, the system generates an alarm unless the
person presses a button.
Clifford et al. (2007) developed a system composed of a series of accelerometers, a processor
and a wireless transceiver. The acquired acceleration data is constantly compared with some
standard values. If there is a fall event, the processor sends an alarm signal to a remote re-
ceiver. Asimilar approach is given by Lee et al. (2007) using a sensor module and an algorithm
to detect posture, activity and fall. For long range communication with the base station, there
are intermediate nodes that act as repeaters. The sensitivity was 93.2%.
Lindemann et al. (2005) used an acoustic device on the rear side of the ear, to measure velocity
and acceleration. Also Wang et al. (2008) used a sensor on the head of the patient since it
increases the accuracy of the detection.
The Inescapable Smart Impact detection System ISIS (Prado-Velasco et al., 2008) used a sensor
with an accelerometer and a smartphone as base station. Moving the processing to the smart-
phone extended the lifetime of the batteries and the usability of the sensor. They achieved
100% sensitivity with reduction in specicity.
Other methods are based on the body posture and use more than one sensor. Some researchers
divided the human activities into two parts: static position and dynamic transition (Li et al.,
2009). They used two sensors both with an accelerometer and a gyroscope, one placed on the
chest and the other on the thigh. The gyroscope helped to decrease the false positives.
Noury et al. (2003) used a sensor with two accelerometers, one orthogonal to the other and
placed under the armpit. The fall is detected on the basis of the inclination of the chest and
its velocity. The alarm is not raised if the patient presses a button on time, avoiding thus false
alarms. An experimental evaluation showed levels of sensitivity and specicity equal to 81%.
In a similar study researchers used a device with three different sensors for body posture
detection, vibration detection and to measure vertical acceleration (Noury et al., 2000). Data
was processed by the base station. The sensitivity and specicity here were 85%.
Other researchers developed a real-time algorithm for automatic recognition of physical ac-
tivities and their intensities (Tapia et al., 2007). They used ve accelerometers placed on the
wrist, the ankle, the upper arm, the upper thigh and the hip. In addition, they used a heart rate
monitor placed on the chest. Trials have been conducted on 21 people for 30 different physical
activities such as lying down, standing, walking, cycling, running and using the stairs. Data
analyzed both in time and frequency domain were classied using the Naive Bayes classi-
er. Results showed an accuracy of 94.6% for a person using the training set of that person,
whereas the accuracy was 56.3% using the training sets of all the other people.
Another research work exploited an accelerometer placed on the waist (Mathie et al., 2001).
The device was so small that it tted in a belt. The authors analyzed the duration, velocity,
angle of a movement and its energy consumption to distinguish between activity and rest. The
processing of the information was conducted by a base station. The authors used a threshold
of 2.5G to detect a fall under the assumption that the subjects are not in good health and
therefore unable to perform actions with acceleration above that threshold. This means that,
to avoid false positives, they had to reduce the activity recognition capability of the system.
Hwang et al. (2004) used a node placed on the chest featuring an accelerometer, a gyroscope,
a tilt sensor, a processing unit and a Bluetooth transmitter. The accelerometer measured the
kinetic force whereas the tilt sensor and the gyroscope estimated the body posture. The goal
was to detect some activities of daily living and falls. The authors experimented on three
people, aged over 26 years, studying the four activities: forward fall, backward fall, lateral fall
and sit-stand. In this study, the system could distinguish between fall and daily activities. The
accuracy of fall detection was 96.7%.
Recently, smartphones with embedded accelerometers have been used to act both as fall de-
tector and as gateway to alert the caregivers (Dai et al., 2010; Sposaro & Tyson, 2009). The
problems associated with this approach are related to the device placement (in a xed posi-
tion or not) and to the short battery lifetime. Usually in these applications there is a trivial fall
detection algorithm and to avoid false positives, the user should press a button to dismiss the
alarm when there is no real fall.
11. Issues and challenges for designing a robust system
The review of the above proposed solutions shows some pitfalls for a real implementation.
The system found more promising is the one that takes into account postures given by the
accelerometers and gyroscopes to reduce false positives (Li et al., 2009). But the authors used
two nodes and did not detect activities of daily living such a falling on a chair or a bed. The
reported sensitivity is 92% and specicity 91%.
Hence the rst challenge is to improve the performance of systems, to assist the patient only
when there is a real fall. If we imagine to deploy the system in a hospital, it would be very
annoying to run frequently to a patient because of false alarms.
The next challenge is to take into account the usability. The ideal system should be based on
only one wearable sensor with small form factor, possibly placed in a comfortable place such
as a belt. This may complicate the posture detection. Moreover the energy consumption must
be low to extend the battery lifetime. This requires careful management of radio communica-
tions (the activity with the highest consumption of energy), ash storage and data sampling
Position (ID, position)
Device (ID, manufacturer, model, description, characteristics)
Configuration (ID, record_content, Mote, scale_G, sample_frequency,
Body_position, x_direction, y_direction, z_direction)
Users (ID, age, gender, height_cm, weight_kg, body_mass_index)
Experiments (ID, Configuration, Action, User, content)
Note that we decided to collect, represent, and store extra information, such as the posture
of the user before and after a potential fall, the separate acceleration values and acceleration
magnitude as-well. This has been done to foster the reuse of the collected data and to enable
the evaluation of future techniques on the same sets of data.
10. Overview of fall detection algorithms
From what has been explained so far, many different approaches have been taken to solve the
fall detection problem using accelerometers. The basic and trivial system uses a threshold to
establish if a person falls, which is subject to many false positives. Some researchers have tried
to introduce computationally-hard type of intensive algorithms but the goal has been always
to nd a trade-off between the system accuracy and the cost.
Depeursinge et al. (2001) used a two-level neural network algorithm to analyze the accelera-
tions given by two sensors placed in distinct parts of the body. Such accelerations are trans-
lated into spatial coordinates and fed into the algorithm. The output of the system represents
the probability that a fall is happening: if the probability is low, the system continues moni-
toring whereas if the probability is medium or high, the system generates an alarm unless the
person presses a button.
Clifford et al. (2007) developed a system composed of a series of accelerometers, a processor
and a wireless transceiver. The acquired acceleration data is constantly compared with some
standard values. If there is a fall event, the processor sends an alarm signal to a remote re-
ceiver. Asimilar approach is given by Lee et al. (2007) using a sensor module and an algorithm
to detect posture, activity and fall. For long range communication with the base station, there
are intermediate nodes that act as repeaters. The sensitivity was 93.2%.
Lindemann et al. (2005) used an acoustic device on the rear side of the ear, to measure velocity
and acceleration. Also Wang et al. (2008) used a sensor on the head of the patient since it
increases the accuracy of the detection.
The Inescapable Smart Impact detection System ISIS (Prado-Velasco et al., 2008) used a sensor
with an accelerometer and a smartphone as base station. Moving the processing to the smart-
phone extended the lifetime of the batteries and the usability of the sensor. They achieved
100% sensitivity with reduction in specicity.
Other methods are based on the body posture and use more than one sensor. Some researchers
divided the human activities into two parts: static position and dynamic transition (Li et al.,
2009). They used two sensors both with an accelerometer and a gyroscope, one placed on the
chest and the other on the thigh. The gyroscope helped to decrease the false positives.
Noury et al. (2003) used a sensor with two accelerometers, one orthogonal to the other and
placed under the armpit. The fall is detected on the basis of the inclination of the chest and
its velocity. The alarm is not raised if the patient presses a button on time, avoiding thus false
alarms. An experimental evaluation showed levels of sensitivity and specicity equal to 81%.
In a similar study researchers used a device with three different sensors for body posture
detection, vibration detection and to measure vertical acceleration (Noury et al., 2000). Data
was processed by the base station. The sensitivity and specicity here were 85%.
Other researchers developed a real-time algorithm for automatic recognition of physical ac-
tivities and their intensities (Tapia et al., 2007). They used ve accelerometers placed on the
wrist, the ankle, the upper arm, the upper thigh and the hip. In addition, they used a heart rate
monitor placed on the chest. Trials have been conducted on 21 people for 30 different physical
activities such as lying down, standing, walking, cycling, running and using the stairs. Data
analyzed both in time and frequency domain were classied using the Naive Bayes classi-
er. Results showed an accuracy of 94.6% for a person using the training set of that person,
whereas the accuracy was 56.3% using the training sets of all the other people.
Another research work exploited an accelerometer placed on the waist (Mathie et al., 2001).
The device was so small that it tted in a belt. The authors analyzed the duration, velocity,
angle of a movement and its energy consumption to distinguish between activity and rest. The
processing of the information was conducted by a base station. The authors used a threshold
of 2.5G to detect a fall under the assumption that the subjects are not in good health and
therefore unable to perform actions with acceleration above that threshold. This means that,
to avoid false positives, they had to reduce the activity recognition capability of the system.
Hwang et al. (2004) used a node placed on the chest featuring an accelerometer, a gyroscope,
a tilt sensor, a processing unit and a Bluetooth transmitter. The accelerometer measured the
kinetic force whereas the tilt sensor and the gyroscope estimated the body posture. The goal
was to detect some activities of daily living and falls. The authors experimented on three
people, aged over 26 years, studying the four activities: forward fall, backward fall, lateral fall
and sit-stand. In this study, the system could distinguish between fall and daily activities. The
accuracy of fall detection was 96.7%.
Recently, smartphones with embedded accelerometers have been used to act both as fall de-
tector and as gateway to alert the caregivers (Dai et al., 2010; Sposaro & Tyson, 2009). The
problems associated with this approach are related to the device placement (in a xed posi-
tion or not) and to the short battery lifetime. Usually in these applications there is a trivial fall
detection algorithm and to avoid false positives, the user should press a button to dismiss the
alarm when there is no real fall.
11. Issues and challenges for designing a robust system
The review of the above proposed solutions shows some pitfalls for a real implementation.
The system found more promising is the one that takes into account postures given by the
accelerometers and gyroscopes to reduce false positives (Li et al., 2009). But the authors used
two nodes and did not detect activities of daily living such a falling on a chair or a bed. The
reported sensitivity is 92% and specicity 91%.
Hence the rst challenge is to improve the performance of systems, to assist the patient only
when there is a real fall. If we imagine to deploy the system in a hospital, it would be very
annoying to run frequently to a patient because of false alarms.
The next challenge is to take into account the usability. The ideal system should be based on
only one wearable sensor with small form factor, possibly placed in a comfortable place such
as a belt. This may complicate the posture detection. Moreover the energy consumption must
be low to extend the battery lifetime. This requires careful management of radio communica-
tions (the activity with the highest consumption of energy), ash storage and data sampling
and processing. To support clinical requirements battery lifetime is a major concern: the min-
imum battery lifetime should be at least one day, in order to avoid stressing the caregivers
with the tasks of recharging and replacing the devices, considering that longer the battery life
better the continuity and the effectiveness of the system.
12. Conclusion
The development of a fall detection system requires a non-negligible warm-up time to fully
understand the problem of falls. In this survey the basics of the fall-problem together with
the most relevant approaches have been described. The aim is to provide guidelines to speed-
up the design process of a new fall detection system by compiling the merits of efforts taken
during the past 30 years in developing a fall detection system. The researchers took many
different approaches to solve the problem of falls among elderly with the lack of any standard
testing guidelines. They proposed their own guidelines but they did not cover the problem
from the beginning. The review shows the different approaches and presents a standard pro-
cedure by collecting the most relevant parameters, data ltering and testing protocols. This
study also provided a standard structure for a database considering the issues and challenges
of a fall detection system.
A step further from the detection is the prediction of non-accidental falls. Some papers left
prediction as a future work, suggesting consideration of the physiological state of elderly. The
rst problem to face is the selection of the physiological measurements that are relevant to
a fall and the ways to measure them. It is presumable that even if the complexity of such a
predictive system increases, the advantages are much more.
13. References
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y. &Cayirci, E. (2002). Wireless sensor networks:
a survey, Computer Networks 38: 393422.
Bradley, C., Pointer, S., of Health, A. I., Welfare. & of South Australia., F. U. (2009). Hospi-
talisations due to falls by older people, Australia, 2005-06 / Clare Bradley, Sophie Pointer,
Australian Institute of Health and Welfare, Canberra :.
Campbell, A. J., Borrie, M. J., Spears, G. F., Jackson, S. L., Brown, J. S. & Fitzgerald, J. L. (1990).
Circumstances and Consequences of Falls Experienced by a Community Population
70 Years and over during a Prospective Study, Age Ageing 19(2): 136141.
Chapman, A. (2008). Biomechanical Analysis of Fundamental Human Movements, Human Kinet-
ics, United States.
Clifford, M. A., Borras, R. L. & Gomez, L. (2007). System and method for human body fall
detection.
URL: http://www.freepatentsonline.com/7248172.html
Dai, J., Bai, X., Yang, Z., Shen, Z. &Xuan, D. (2010). Perfalld: Apervasive fall detection system
using mobile phones, Pervasive Computing and Communications Workshops (PERCOM
Workshops), 2010 8th IEEE International Conference on, pp. 292 297.
Depeursinge, Y., Krauss, J. & El-khoury, M. (2001). Device for monitoring the activity of a
person and/or detecting a fall, in particular with a view to providing help in the
event of an incident hazardous to life or limb.
Englander, F.and Terregrossa, R. & Hodson, T. (1996). Economic dimensions of slip and fall
injuries, Journal of Forensic Sciences (JOFS) 41(05): 733746.
Gibson, M., Andres, R., Isaacs, B., Radebaugh, T. & Worm-Petersen, J. (1987). The prevention
of falls in later life. a report of the kellogg international work group on the prevention
of falls by the elderly., Danish Medical Bulletin 34(4): 124.
Hwang, J., Kang, J., Jang, Y. & Kim, H. (2004). Development of novel algorithm and real-
time monitoring ambulatory system using bluetooth module for fall detection in the
elderly, Engineering in Medicine and Biology Society, 2004. IEMBS 04. 26th Annual In-
ternational Conference of the IEEE, Vol. 1, pp. 2204 2207.
Kangas, M., Konttila, A., Winblad, I. &Jamsa, T. (2007). Determination of simple thresholds for
accelerometry-based parameters for fall detection, Engineering in Medicine and Biology
Society, 2007. EMBS 2007. 29th Annual International Conference of the IEEE, pp. 1367
1370.
Kangas, M., Vikman, I., Wiklander, J., Lindgren, P., Nyberg, L. & Jms, T. (2009). Sensitivity
and specicity of fall detection in people aged 40 years and over., Gait Posture .
Lee, Y., Kim, J., Son, M. & Lee, M. (2007). Implementation of accelerometer sensor module
and fall detection monitoring system based on wireless sensor network, Engineering
in Medicine and Biology Society, 2007. EMBS 2007. 29th Annual International Conference
of the IEEE, pp. 23152318.
Li, Q., Stankovic, J. A., Hanson, M. A., Barth, A. T., Lach, J. & Zhou, G. (2009). Accurate, fast
fall detection using gyroscopes and accelerometer-derived posture information, BSN
09: Proceedings of the 2009 Sixth International Workshop on Wearable and Implantable
Body Sensor Networks, IEEE Computer Society, Washington, DC, USA, pp. 138143.
Lindemann, U., Hock, A., Stuber, M. & Keck, W. (2005). Evaluation of a fall detector based
on accelerometers: A pilot study, Medical and Biological Engineering and Computing
43(5): 548551.
Lipsitz, L. A., Jonsson, P. V., Kelley, M. M. & Koestner, J. S. (1991). Causes and Correlates of
Recurrent Falls in Ambulatory Frail Elderly, Journal of Gerontology 46(4): M114M122.
Lord, S., C., S. & Menz, H. (2001). Falls in older people: Risk factors and strategies for preven-
tion, Ageing and Society 21(05): 667675.
Lord, S., Ward, J., Williams, P. & Anstey, K. (1993). An epidemiological study of falls in
older community-dwelling women: the Randwick falls and fractures study, Aus-
tralian Journal of Public Health 17(3): 240245.
Massachusetts Department of Public Health (2008). Unintentional Fall-Related Injuries among
Massachusetts Older Adults, Commonwealth of Massachusetts, Department of Public
Health.
Masud, T. & Morris, R. (2001). Epidemiology of falls, Age Ageing 30(4): 37.
Mathie, M., Basilakis, J. & Celler, B. (2001). A system for monitoring posture and physical
activity using accelerometers, Engineering in Medicine and Biology Society, 2001. Pro-
ceedings of the 23rd Annual International Conference of the IEEE, Vol. 4, pp. 3654 3657
vol.4.
Noury, N., Barralon, P., Virone, G., Boissy, P., Hamel, M. & Rumeau, P. (2003). A smart sensor
based on rules and its evaluation in daily routines, Engineering in Medicine and Biology
Society, 2003. Proceedings of the 25th Annual International Conference of the IEEE, Vol. 4,
pp. 32863289 Vol.4.
Noury, N., Fleury, A., Rumeau, P., Bourke, A., Laighin, G., Rialle, V. & Lundy, J. (2007). Fall
detection - principles and methods, Engineering in Medicine and Biology Society, 2007.
EMBS 2007. 29th Annual International Conference of the IEEE, pp. 16631666.
and processing. To support clinical requirements battery lifetime is a major concern: the min-
imum battery lifetime should be at least one day, in order to avoid stressing the caregivers
with the tasks of recharging and replacing the devices, considering that longer the battery life
better the continuity and the effectiveness of the system.
12. Conclusion
The development of a fall detection system requires a non-negligible warm-up time to fully
understand the problem of falls. In this survey the basics of the fall-problem together with
the most relevant approaches have been described. The aim is to provide guidelines to speed-
up the design process of a new fall detection system by compiling the merits of efforts taken
during the past 30 years in developing a fall detection system. The researchers took many
different approaches to solve the problem of falls among elderly with the lack of any standard
testing guidelines. They proposed their own guidelines but they did not cover the problem
from the beginning. The review shows the different approaches and presents a standard pro-
cedure by collecting the most relevant parameters, data ltering and testing protocols. This
study also provided a standard structure for a database considering the issues and challenges
of a fall detection system.
A step further from the detection is the prediction of non-accidental falls. Some papers left
prediction as a future work, suggesting consideration of the physiological state of elderly. The
rst problem to face is the selection of the physiological measurements that are relevant to
a fall and the ways to measure them. It is presumable that even if the complexity of such a
predictive system increases, the advantages are much more.
13. References
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y. &Cayirci, E. (2002). Wireless sensor networks:
a survey, Computer Networks 38: 393422.
Bradley, C., Pointer, S., of Health, A. I., Welfare. & of South Australia., F. U. (2009). Hospi-
talisations due to falls by older people, Australia, 2005-06 / Clare Bradley, Sophie Pointer,
Australian Institute of Health and Welfare, Canberra :.
Campbell, A. J., Borrie, M. J., Spears, G. F., Jackson, S. L., Brown, J. S. & Fitzgerald, J. L. (1990).
Circumstances and Consequences of Falls Experienced by a Community Population
70 Years and over during a Prospective Study, Age Ageing 19(2): 136141.
Chapman, A. (2008). Biomechanical Analysis of Fundamental Human Movements, Human Kinet-
ics, United States.
Clifford, M. A., Borras, R. L. & Gomez, L. (2007). System and method for human body fall
detection.
Dai, J., Bai, X., Yang, Z., Shen, Z. &Xuan, D. (2010). Perfalld: Apervasive fall detection system
using mobile phones, Pervasive Computing and Communications Workshops (PERCOM
Workshops), 2010 8th IEEE International Conference on, pp. 292 297.
Depeursinge, Y., Krauss, J. & El-khoury, M. (2001). Device for monitoring the activity of a
person and/or detecting a fall, in particular with a view to providing help in the
event of an incident hazardous to life or limb.
Englander, F.and Terregrossa, R. & Hodson, T. (1996). Economic dimensions of slip and fall
injuries, Journal of Forensic Sciences (JOFS) 41(05): 733746.
Gibson, M., Andres, R., Isaacs, B., Radebaugh, T. & Worm-Petersen, J. (1987). The prevention
of falls in later life. a report of the kellogg international work group on the prevention
of falls by the elderly., Danish Medical Bulletin 34(4): 124.
Hwang, J., Kang, J., Jang, Y. & Kim, H. (2004). Development of novel algorithm and real-
time monitoring ambulatory system using bluetooth module for fall detection in the
elderly, Engineering in Medicine and Biology Society, 2004. IEMBS 04. 26th Annual In-
ternational Conference of the IEEE, Vol. 1, pp. 2204 2207.
Kangas, M., Konttila, A., Winblad, I. &Jamsa, T. (2007). Determination of simple thresholds for
accelerometry-based parameters for fall detection, Engineering in Medicine and Biology
1370.
Kangas, M., Vikman, I., Wiklander, J., Lindgren, P., Nyberg, L. & Jms, T. (2009). Sensitivity
and specicity of fall detection in people aged 40 years and over., Gait Posture .
Lee, Y., Kim, J., Son, M. & Lee, M. (2007). Implementation of accelerometer sensor module
and fall detection monitoring system based on wireless sensor network, Engineering
in Medicine and Biology Society, 2007. EMBS 2007. 29th Annual International Conference
of the IEEE, pp. 23152318.
Li, Q., Stankovic, J. A., Hanson, M. A., Barth, A. T., Lach, J. & Zhou, G. (2009). Accurate, fast
fall detection using gyroscopes and accelerometer-derived posture information, BSN
09: Proceedings of the 2009 Sixth International Workshop on Wearable and Implantable
Body Sensor Networks, IEEE Computer Society, Washington, DC, USA, pp. 138143.
Lindemann, U., Hock, A., Stuber, M. & Keck, W. (2005). Evaluation of a fall detector based
on accelerometers: A pilot study, Medical and Biological Engineering and Computing
43(5): 548551.
Lipsitz, L. A., Jonsson, P. V., Kelley, M. M. & Koestner, J. S. (1991). Causes and Correlates of
Recurrent Falls in Ambulatory Frail Elderly, Journal of Gerontology 46(4): M114M122.
Lord, S., C., S. & Menz, H. (2001). Falls in older people: Risk factors and strategies for preven-
tion, Ageing and Society 21(05): 667675.
Lord, S., Ward, J., Williams, P. & Anstey, K. (1993). An epidemiological study of falls in
older community-dwelling women: the Randwick falls and fractures study, Aus-
tralian Journal of Public Health 17(3): 240245.
Massachusetts Department of Public Health (2008). Unintentional Fall-Related Injuries among
Massachusetts Older Adults, Commonwealth of Massachusetts, Department of Public
Health.
Masud, T. & Morris, R. (2001). Epidemiology of falls, Age Ageing 30(4): 37.
Mathie, M., Basilakis, J. & Celler, B. (2001). A system for monitoring posture and physical
activity using accelerometers, Engineering in Medicine and Biology Society, 2001. Pro-
ceedings of the 23rd Annual International Conference of the IEEE, Vol. 4, pp. 3654 3657
vol.4.
Noury, N., Barralon, P., Virone, G., Boissy, P., Hamel, M. & Rumeau, P. (2003). A smart sensor
based on rules and its evaluation in daily routines, Engineering in Medicine and Biology
Society, 2003. Proceedings of the 25th Annual International Conference of the IEEE, Vol. 4,
pp. 32863289 Vol.4.
Noury, N., Fleury, A., Rumeau, P., Bourke, A., Laighin, G., Rialle, V. & Lundy, J. (2007). Fall
detection - principles and methods, Engineering in Medicine and Biology Society, 2007.
EMBS 2007. 29th Annual International Conference of the IEEE, pp. 16631666.
Noury, N., Herve, T., Rialle, V., Virone, G., Mercier, E., Morey, G., Moro, A. & Porcheron, T.
(2000). Monitoring behavior in home using a smart fall sensor and position sensors,
Microtechnologies in Medicine and Biology, 1st Annual International, Conference On. 2000,
pp. 607610.
Noury, N., Rumeau, P., Bourke, A., Laighin, G. & Lundy, J. (2008). A proposal for the classi-
cation and evaluation of fall detectors, IRBM 29(6): 340 349.
Oliver, D. (2008). Falls riskprediction tools for hospital inpatients. time to put them to bed?,
Age Ageing 37(3): 248250.
Perell, K., Nelson, A., R.L., G., Luther, S., Prieto-Lewis, N. & Rubenstein, L. (2001). Fall risk
assessment measures: an analytic review, Journal of Gerontology 56A(12): M761M766.
Polastre, J., Szewczyk, R. & Culler, D. (2005). Telos: enabling ultra-low power wireless re-
search, IPSN05: Proceedings of the 4th international symposiumon Information processing
in sensor networks, IEEE Press, Piscataway, NJ, USA.
Prado-Velasco, M., del Rio-Cidoncha, M. G. & Ortiz-Marin, R. (2008). The inescapable smart
impact detection system (isis): An ubiquitous and personalized fall detector based
on a distributed ?divide and conquer strategy?, Engineering in Medicine and Biology
3335.
Realtime Technologies LTD (2008). Shimmer: Sensing health with intelligence modularity,
mobility and experimental reusability.
Sposaro, F. & Tyson, G. (2009). ifall: An android application for fall monitoring and response,
Engineering in Medicine and Biology Society, 2009. EMBC2009. Annual International Con-
ference of the IEEE, pp. 6119 6122.
Tapia, E. M., Intille, S. S., Haskell, W., Larson, K., Wright, J., King, A. & Friedman, R. (2007).
Real-time recognition of physical activities and their intensities using wireless ac-
celerometers and a heart rate monitor, ISWC 07: Proceedings of the 2007 11th IEEE
International Symposium on Wearable Computers, IEEE Computer Society, Washington,
DC, USA, pp. 14.
Tinetti, M. E., Speechley, M. & Ginter, S. F. (1988). Risk factors for falls among elderly persons
living in the community, New England Journal of Medicine 319(26): 17011707.
Vassallo, M., Poynter, L., Sharma, J. C., Kwan, J. &Allen, S. C. (2008). Fall riskassessment tools
compared with clinical judgment: an evaluation in a rehabilitation ward, Age Ageing
37(3): 277281.
Wang, C.-C., Chiang, C.-Y., Lin, P.-Y., Chou, Y.-C., Kuo, I.-T., Huang, C.-N. & Chan, C.-T.
(2008). Development of a fall detecting system for the elderly residents, Bioinformat-
ics and Biomedical Engineering, 2008. ICBBE 2008. The 2nd International Conference on,
pp. 13591362.
Yang, C.-C. & Hsu, Y.-L. (2007). Algorithm design for real-time physical activity identi-
cation with accelerometry measurement, Industrial Electronics Society, 2007. IECON
2007. 33rd Annual Conference of the IEEE, pp. 2996 3000.
Yoshida, T., Mizuno, F., Hayasaka, T., Tsubota, K., Wada, S. & Yamaguchi, T. (2005). A wear-
able computer system for a detection and prevention of elderly users from falling,
The 12th International Conference on Biomedical Engineering .
Yu, X. (2008). Approaches and principles of fall detection for elderly and patient, e-health Net-
working, Applications and Services, 2008. HealthCom 2008. 10th International Conference
on, pp. 4247.
Odor Recognition and Localization Using Sensor Networks 167
Odor Recognition and Localization Using Sensor Networks
Rabie A. Ramadan
X

Odor Recognition and Localization
Using Sensor Networks

Rabie A. Ramadan
Cairo University
Cairo, Egypt

1. Introduction
Odor usually quantified by five parameters which are 1) intensity, 2) degree of
offensiveness, 3) character, 4) frequency, and 5) duration. It has different forms including
Gas, Chemical, Radiation, Organic Compounds, and Water odors including different water
contaminations. For such odors, there are many of the traditional methods that have been
used for a number of years. However, these methods suffer from different problems
including the detection cost, the long time taken for analysis and detection, and exposing
human to danger. On the other hand, the advances in sensing technology lead to the usage
of sensor networks in many applications. For instance, sensors have been used to monitor
animals in habitat areas and monitor patients health. In addition, sensor networks have
been used to monitor critical infrastructures such as gas, transportation, energy, and water
pipelines as well as important buildings. Sensors are tiny devices that can be included in
small areas. At the same time, they are capable of capturing different phenomena from the
environment, analyze the collected data, and take decisions. In addition, sensors are able to
form unattended wireless ad hoc network that can survive for long time. Such features
enable wireless sensor networks (WSN) to play an essential role in odor detection. In fact,
odor detection became one of the important applications due to the terrorist attack that
started by the one occurred at Tokyo Subway in 1995. Since this time, odor detection and
localization is considered as one of the important applications. Researchers believe that
sensors and sensor networks will play an important role in odor detection and localization.
In this chapter, we generalize the term odor to include the radiation detection and
localization since the radiation in most of the recent work is considered as an odor.

Odor detection in human depends on the smell sense; therefore, it is considered as other
senses such as vision and hearing that have a theory behind it. The theory behind the smell
is explained in (Al-Bastaki, 2009) where olfactory systems simulate the neurobiological
information processing systems (biological neural networks) as shown in Figure 1. The
collected olfactory information is processed in both the olfactory bulb and in the olfactory
cortex. The function of the cortex, then, is to perform the pattern classification and
recognition of the odors. Once the odor is identified, its information is transmitted to
10

hippocampus, limbic system and the cerebral cortex. At this moment, the conscious
perception of the odor and how to act on it takes place.

Fig. 1. The major processes of the olfactory system (Al-Bastaki, 2009)

To simulate such process, electronic noses have been developed. As can be understood, the
main components of such noses are the sensing and the pattern recognition components.
The first part consists of many of the sensors including gas, chemical, and many other
sensors. The term chemical sensors refer to a set of sensors that respond to a particular
analyte in a selective way through a chemical reaction. The second part, pattern recognition,
is the science of discovering regular and irregular patterns out of a given materials. Many
Artificial Intelligence (AI) algorithms and techniques are utilized in this part. Some of these
techniques will be explained later in this chapter. To simplify the idea of the electronic
noses, Figure 2 shows the basic components of an electronic nose. The figure shows that an
electronic nose must contain a processor and a memory for analyzing the received digital
data. At the same time, it has to have the appropriate set of sensors that identifies the smell
print of an odor.

Once the odor is detected, its source has to be localized and contaminated if it is dangerous
such as chemicals or radiations. There are different localization methods including the one
that use mobile robots as well as different AI algorithms. Therefore, for odor manipulation,
we have three phases as shown in figure 3 which are odor sensing, recognition, and
localization. In each phase different techniques and algorithms are used. In the following
sections we explore some of the detection and localization methods. Then, we propose a
hybrid odor localization method that is based on Genetic Algorithms (GA), Fuzzy Logic
Controller (FLC), and Swarm Intelligence. The initial results showed some significant results
in localizing the odor sources.

Fig. 2. Basic components of an electronic nose (Gardner et al., 1999)

Fig. 3. Odor manipulation phases

2. Odor Sensing
As mentioned, there are many types of odors including the odor in environment pollution.
In fact, the need for detecting odors that causes pollution and the need for clean
environment are leading the research in this field. Reliable real time detection techniques are
urgently required especially with the increasing of new diseases that are caused by these
odors. Sampling and analytical procedures are no longer on call due to the availability of
other techniques that can produce the results on demand.

Different sensors have been used to detect odors including Conductivity, Piezoelectric,
Metal-oxide-silicon field-effect-transistor (MOSFET), Optical Fiber, and Spectrometry-Based
Sensors. The idea behind the operation of these sensors are explained in (Korotkaya, 2010)
and summarized as follows:
The conductivity sensors exhibit a change in resistance when exposed to volatile organic
compounds. Such sensors respond to water vapor, humidity difference, but not too sensitive
for specific odorants. The Piezoelectric sensors are used to measure temperature, mass
changes, pressure, force, and acceleration. The main idea behind these sensors is that the gas
sample is adsorbed at the surface of the polymer, increasing the mass of the disk-polymer
device and thereby reducing the resonance frequency. The reduction is inversely
proportional to odorant mass adsorbed by the polymer. The MOSFET odor-sensing devices
are based on the principle that volatile odor components in contact with a catalytic metal
can produce a reaction in the metal. The reactions products can diffuse through the gate of
a MOSFET to change the electrical properties of the device. Optical-fiber sensors utilize glass
fibers with a thin chemically active material coating on their slides or ends. A light source at
Odor Sensing Recognition Localization

hippocampus, limbic system and the cerebral cortex. At this moment, the conscious
perception of the odor and how to act on it takes place.

Fig. 1. The major processes of the olfactory system (Al-Bastaki, 2009)

To simulate such process, electronic noses have been developed. As can be understood, the
main components of such noses are the sensing and the pattern recognition components.
The first part consists of many of the sensors including gas, chemical, and many other
sensors. The term chemical sensors refer to a set of sensors that respond to a particular
analyte in a selective way through a chemical reaction. The second part, pattern recognition,
is the science of discovering regular and irregular patterns out of a given materials. Many
Artificial Intelligence (AI) algorithms and techniques are utilized in this part. Some of these
techniques will be explained later in this chapter. To simplify the idea of the electronic
noses, Figure 2 shows the basic components of an electronic nose. The figure shows that an
electronic nose must contain a processor and a memory for analyzing the received digital
data. At the same time, it has to have the appropriate set of sensors that identifies the smell
print of an odor.

Once the odor is detected, its source has to be localized and contaminated if it is dangerous
such as chemicals or radiations. There are different localization methods including the one
that use mobile robots as well as different AI algorithms. Therefore, for odor manipulation,
we have three phases as shown in figure 3 which are odor sensing, recognition, and
localization. In each phase different techniques and algorithms are used. In the following
sections we explore some of the detection and localization methods. Then, we propose a
hybrid odor localization method that is based on Genetic Algorithms (GA), Fuzzy Logic
Controller (FLC), and Swarm Intelligence. The initial results showed some significant results
in localizing the odor sources.

Fig. 2. Basic components of an electronic nose (Gardner et al., 1999)

Fig. 3. Odor manipulation phases

2. Odor Sensing
As mentioned, there are many types of odors including the odor in environment pollution.
In fact, the need for detecting odors that causes pollution and the need for clean
environment are leading the research in this field. Reliable real time detection techniques are
urgently required especially with the increasing of new diseases that are caused by these
odors. Sampling and analytical procedures are no longer on call due to the availability of
other techniques that can produce the results on demand.

Different sensors have been used to detect odors including Conductivity, Piezoelectric,
Metal-oxide-silicon field-effect-transistor (MOSFET), Optical Fiber, and Spectrometry-Based
Sensors. The idea behind the operation of these sensors are explained in (Korotkaya, 2010)
and summarized as follows:
The conductivity sensors exhibit a change in resistance when exposed to volatile organic
compounds. Such sensors respond to water vapor, humidity difference, but not too sensitive
for specific odorants. The Piezoelectric sensors are used to measure temperature, mass
changes, pressure, force, and acceleration. The main idea behind these sensors is that the gas
sample is adsorbed at the surface of the polymer, increasing the mass of the disk-polymer
device and thereby reducing the resonance frequency. The reduction is inversely
proportional to odorant mass adsorbed by the polymer. The MOSFET odor-sensing devices
are based on the principle that volatile odor components in contact with a catalytic metal
can produce a reaction in the metal. The reactions products can diffuse through the gate of
a MOSFET to change the electrical properties of the device. Optical-fiber sensors utilize glass
fibers with a thin chemically active material coating on their slides or ends. A light source at
Odor Sensing Recognition Localization

a single frequency (or at a narrow band of frequencies) is used to interrogate the active
material, which in turn responds with a change in color to the presence of the odorant to be
detected and measured. Finally, Spectrometry-Based sensors use the principle that each
molecular has a distinct infrared spectrum

The state-of-the-art method for detecting odor emissions is the classical olfactometry. In this
method, the odor detection is based on a group of people (panelists) with 95% probability of
average odor sensitive. The results depend on the smelling capabilities of the panel
members. This method is expensive and it is not feasible where continuous and online
monitoring is required. Devices like electronic noses that utilize different odor sensors,
stated previously, are more effective and much cheaper. The following are some examples
on the usage of electronic noses in odors detection.

In (Staples & Viswanathan, 2005), the authors used an electronic nose named zNose that
simulates an almost unlimited number of specific virtual chemical sensors, and produces
olfactory images based upon aroma chemistry. Figure 3 shows the zNose device and its
componenets. Using zNose odor detection is done in seconds. The nose uses a handheld air
sampler that consists of a battery operated sample pump and a tenax filled probe. This
enables organic compounds associated with odors to be remotely collected.

Fig. 4. zNose device (zNose, 2010)

Another example of electronic noses is in detecting the human body odor. One may think
that it is not that important to use an electronic device for human body odor detection.
However, the argument is if we succeeded in detecting the human body odor, the same
device and experience will be applicable for other important applications such as healthcare
monitoring, biometrics and cosmetics. Further to these applications, human odor could be
used to uniquely identify a person which might be important in other fields such as in
security. In addition, the electronic noses could be used to diagnose the urine odor of the
patients with kidney disorders (Natale et al., 1999). Moreover, the human odor detection,
armpit odor, is studied in (Chatchawal et al., 2009) using an electronic nose. The authors
reported two problems they face when trying to detect the human body odors; the first
problem is that the human may have different sweat at different time and environment;
therefore, the humidity might affect the odor detection procedure in which it affects the gas
sensors quality. The second problem people may use deodorants to reduce unpleasant body
odor. The authors controlled the humidity using two methods including hardware-based
and software-based methods through adding some of the sweat thresholds.

Electronic noses also used to detect explosives to help fighting international terrorism. The
main idea behind a nose that detects explosives is to emulate the dog nose capabilities
avoiding some of the dogs drawbacks such as rigorous training, testing, and validation
exercises in various operational scenarios with different types of explosives. Various types
of noses have been developed; some of them are based on sensor arrays to detect different
combinations of explosions. An example on this type of nose is the one developed by Walt
and his group where they developed an expensive senor array of fiber optics cable. Some
others are based on vapour sensors to detect different vapors such as DNT and TNT
emanating from landmines. A good source on reviewing about the electronic noses that are
used for explosive detection could be found in (Jehuda, 2003).

A network of the previous electronic noses might be efficient in detecting the odors or
explosives in a specific area. However, the difficulties of building a network of different
noses involve the wireless media problems, the data analysis coming from all of these
nodes, and the deployment of these nodes. One prototype of wireless sensor (e-noses)
networks is mentioned in (Jianfeng el at., 2009) where the authors developed a network to
monitor odorant gases and accurately estimating odor strength in and around livestock
farms. Each nose consists of four metal-oxide semiconductor (MOS) gas sensors. Figure 5
shows the e-nose with gas chamber, pump and sensor array. For communication and
networking purposes, this node (e-nose) is mounted on a MICAZ board (MICAZ, 2010).
The sink node uses a modified Kalman filter for the data filtering.

Fig. 5. E-nose with gas chamber, pump and sensor array exerted from (Jianfeng et al. , 2009)

3. Odor Recognition
Once the odors are sensed, they have to be analyzed for decision making. However, such
decision is not easy since there are many factors that affect for instance (Roppel and Wilson,
2000) (i) sensors have some overlapping specificities, (ii) not all the sensors will have the
same performance in sensing the odors, and (iii) no general agreement has been reached on
what constitutes the fundamental components of odor space. Therefore, recognizing the
odor and taking the appropriate decision still a problem to many of the applications. In the
following paragraph, we will go through some of the artificial intelligence techniques that
try to solve such problem including neural networks and fuzzy logic.
In (Linder et al., 2005), the authors use a standard feedforward network for odor intensity
recognition for multiclass classification problem. As shown in Figure 6, the network consists
of the following inputs: (1) peaks D1 (acetone), (2) D2 (ethanol), (3) C1+D3 (first peak of

a single frequency (or at a narrow band of frequencies) is used to interrogate the active
material, which in turn responds with a change in color to the presence of the odorant to be
detected and measured. Finally, Spectrometry-Based sensors use the principle that each
molecular has a distinct infrared spectrum

The state-of-the-art method for detecting odor emissions is the classical olfactometry. In this
method, the odor detection is based on a group of people (panelists) with 95% probability of
average odor sensitive. The results depend on the smelling capabilities of the panel
members. This method is expensive and it is not feasible where continuous and online
monitoring is required. Devices like electronic noses that utilize different odor sensors,
stated previously, are more effective and much cheaper. The following are some examples
on the usage of electronic noses in odors detection.

In (Staples & Viswanathan, 2005), the authors used an electronic nose named zNose that
simulates an almost unlimited number of specific virtual chemical sensors, and produces
olfactory images based upon aroma chemistry. Figure 3 shows the zNose device and its
componenets. Using zNose odor detection is done in seconds. The nose uses a handheld air
sampler that consists of a battery operated sample pump and a tenax filled probe. This
enables organic compounds associated with odors to be remotely collected.

Fig. 4. zNose device (zNose, 2010)

Another example of electronic noses is in detecting the human body odor. One may think
that it is not that important to use an electronic device for human body odor detection.
However, the argument is if we succeeded in detecting the human body odor, the same
device and experience will be applicable for other important applications such as healthcare
monitoring, biometrics and cosmetics. Further to these applications, human odor could be
used to uniquely identify a person which might be important in other fields such as in
security. In addition, the electronic noses could be used to diagnose the urine odor of the
patients with kidney disorders (Natale et al., 1999). Moreover, the human odor detection,
armpit odor, is studied in (Chatchawal et al., 2009) using an electronic nose. The authors
reported two problems they face when trying to detect the human body odors; the first
problem is that the human may have different sweat at different time and environment;
therefore, the humidity might affect the odor detection procedure in which it affects the gas
sensors quality. The second problem people may use deodorants to reduce unpleasant body
odor. The authors controlled the humidity using two methods including hardware-based
and software-based methods through adding some of the sweat thresholds.

Electronic noses also used to detect explosives to help fighting international terrorism. The
main idea behind a nose that detects explosives is to emulate the dog nose capabilities
avoiding some of the dogs drawbacks such as rigorous training, testing, and validation
exercises in various operational scenarios with different types of explosives. Various types
of noses have been developed; some of them are based on sensor arrays to detect different
combinations of explosions. An example on this type of nose is the one developed by Walt
and his group where they developed an expensive senor array of fiber optics cable. Some
others are based on vapour sensors to detect different vapors such as DNT and TNT
emanating from landmines. A good source on reviewing about the electronic noses that are
used for explosive detection could be found in (Jehuda, 2003).

A network of the previous electronic noses might be efficient in detecting the odors or
explosives in a specific area. However, the difficulties of building a network of different
noses involve the wireless media problems, the data analysis coming from all of these
nodes, and the deployment of these nodes. One prototype of wireless sensor (e-noses)
networks is mentioned in (Jianfeng el at., 2009) where the authors developed a network to
monitor odorant gases and accurately estimating odor strength in and around livestock
farms. Each nose consists of four metal-oxide semiconductor (MOS) gas sensors. Figure 5
shows the e-nose with gas chamber, pump and sensor array. For communication and
networking purposes, this node (e-nose) is mounted on a MICAZ board (MICAZ, 2010).
The sink node uses a modified Kalman filter for the data filtering.

Fig. 5. E-nose with gas chamber, pump and sensor array exerted from (Jianfeng et al. , 2009)

3. Odor Recognition
Once the odors are sensed, they have to be analyzed for decision making. However, such
decision is not easy since there are many factors that affect for instance (Roppel and Wilson,
2000) (i) sensors have some overlapping specificities, (ii) not all the sensors will have the
same performance in sensing the odors, and (iii) no general agreement has been reached on
what constitutes the fundamental components of odor space. Therefore, recognizing the
odor and taking the appropriate decision still a problem to many of the applications. In the
following paragraph, we will go through some of the artificial intelligence techniques that
try to solve such problem including neural networks and fuzzy logic.
In (Linder et al., 2005), the authors use a standard feedforward network for odor intensity
recognition for multiclass classification problem. As shown in Figure 6, the network consists
of the following inputs: (1) peaks D1 (acetone), (2) D2 (ethanol), (3) C1+D3 (first peak of

matrix, i. e. lemon oil, and isopropanol) , (4) C2+D4 (second peak of matrix and isoamyl
acetate), and (5) C3 (third matrix peak). The output of the network is three classes which are
weak, distinct, strong odour. The network has been trained based on some of the medical
data prepared by the authors. The authors claimed that such network was very successful in
detecting different odors intensity.

Fig. 6. Structure of a standard feedforward network that could serve for the assignment of
different classes of odour intensity (Linder et al., 2005).

In (Bekiret al., 2007) , the authors combined the fuzzy logic and neural networks for better
odor recognition. The sensing is done by a handheld odor meter, OMX-GR sensor which is
a commercial product. Figure 7 shows the used system diagram where a gas sensor array is
used to differentiate between 11 types of gases. Once the sensors collect the odors data, the
data is transferred to the unsupervised feature extraction block where a fuzzy c-mean
algorithm is applied for data clustering. As it is known, fuzzy c-mean divides the data into
fuzzy partitions, which overlap with each other. Therefore, the containment of each data to
each cluster is defined by a membership grade in (0, 1). This clustering reduces the number
of inputs given to the neural networks given in the following block. The function of the
neural network is to classify the sensor array output into odor categories. The authors
trained the network using 16 types of perfumes with 20 samples of each. The accuracy of the
developed algorithm claimed by the authors was 93.75%.

Fig. 7. Fuzzy Neural system for odor recognition (Bekiret et al., 2007)

Another AI technique is used for odor recognition is proposed in (hsan et al. , 2009). The
authors use Cerebellar Model Articulation Controller (CMAC) based neural networks. The
CMAC concept is not new; The CMAC was firstly proposed during the 1970s by James
Albus, whose idea was based on a model of the cerebellum which is a part of the brain
responsible for learning process (Cui et al., 2004). The CMAC can generally be described as a

transformation device that transforms given input vectors into associated output vectors
(Wisnu et al., 2008). It plays an important role in nonlinear function approximation and
system modeling. CMAC is extremely fast learning technique compared to multiple layer
perceptron (MLP) neural networks. As shown in Figure 8, the CMAC basically consists of
three layers which are the normalized input space, basis functions, and weight vector. The
output of the network could be considered as associative memory that holds the odor
detection decision.

Fig. 8. CMAC layers

4. Odor Localization
Odor localization has gained a lot of attention after some of the terrorist attacks such as the
one occurred at Tokyo Subway in 1995. Odor localization problem falls into one of two main
approaches which are Forward problems and Inverse problems. In the forward problems ,
the state of the odor is estimated in advance while in the inverse problems the prior state of
the odor is estimated based on the its current state. Odor localization usually done through
using robots which is an emulation to the usage of dogs to find bombs, mines, and drugs.
However, the dog has the capability to sense, recognize, and analyze the data; then naturally
takes the decision. For robots, there is a need for localization algorithms. In the following
paragraphs, we review the main ideas behind the work done in this area. Then, in the next
section, we introduce our approach for odor localization.

Odor localization using robots has attracted many of the researchers and most of the work
done in this area is trying to imitate the animals, birds, and swarms behaviors. For instance,
insects fly in a zigzag pattern towards their matting partners. Ants use the same strategy to
follow the pheromones trails. Therefore, robots could also do the same to locate the odor
sources (Wei et al. , 2001). On the other hand, animals may follow spiral surge algorithm to
locate the odor sources (Hayes et al. , 2002) where animals move in spiral path until it
perceives a certain concentration and then moves straight upwind. As soon as it loses the
scent, it starts spiraling again. A simple example on the zigzag and spiral procedures is
shown in Figure 9. A comparison between the spiral and the zigzag algorithms are
presented in (Lochmatter el at., 2008). The results show that the spiral is more efficient than
the zigzag algorithm.

matrix, i. e. lemon oil, and isopropanol) , (4) C2+D4 (second peak of matrix and isoamyl
acetate), and (5) C3 (third matrix peak). The output of the network is three classes which are
weak, distinct, strong odour. The network has been trained based on some of the medical
data prepared by the authors. The authors claimed that such network was very successful in
detecting different odors intensity.

Fig. 6. Structure of a standard feedforward network that could serve for the assignment of
different classes of odour intensity (Linder et al., 2005).

In (Bekiret al., 2007) , the authors combined the fuzzy logic and neural networks for better
odor recognition. The sensing is done by a handheld odor meter, OMX-GR sensor which is
a commercial product. Figure 7 shows the used system diagram where a gas sensor array is
used to differentiate between 11 types of gases. Once the sensors collect the odors data, the
data is transferred to the unsupervised feature extraction block where a fuzzy c-mean
algorithm is applied for data clustering. As it is known, fuzzy c-mean divides the data into
fuzzy partitions, which overlap with each other. Therefore, the containment of each data to
each cluster is defined by a membership grade in (0, 1). This clustering reduces the number
of inputs given to the neural networks given in the following block. The function of the
neural network is to classify the sensor array output into odor categories. The authors
trained the network using 16 types of perfumes with 20 samples of each. The accuracy of the
developed algorithm claimed by the authors was 93.75%.

Fig. 7. Fuzzy Neural system for odor recognition (Bekiret et al., 2007)

Another AI technique is used for odor recognition is proposed in (hsan et al. , 2009). The
authors use Cerebellar Model Articulation Controller (CMAC) based neural networks. The
CMAC concept is not new; The CMAC was firstly proposed during the 1970s by James
Albus, whose idea was based on a model of the cerebellum which is a part of the brain
responsible for learning process (Cui et al., 2004). The CMAC can generally be described as a

transformation device that transforms given input vectors into associated output vectors
(Wisnu et al., 2008). It plays an important role in nonlinear function approximation and
system modeling. CMAC is extremely fast learning technique compared to multiple layer
perceptron (MLP) neural networks. As shown in Figure 8, the CMAC basically consists of
three layers which are the normalized input space, basis functions, and weight vector. The
output of the network could be considered as associative memory that holds the odor
detection decision.

Fig. 8. CMAC layers

4. Odor Localization
Odor localization has gained a lot of attention after some of the terrorist attacks such as the
one occurred at Tokyo Subway in 1995. Odor localization problem falls into one of two main
approaches which are Forward problems and Inverse problems. In the forward problems ,
the state of the odor is estimated in advance while in the inverse problems the prior state of
the odor is estimated based on the its current state. Odor localization usually done through
using robots which is an emulation to the usage of dogs to find bombs, mines, and drugs.
However, the dog has the capability to sense, recognize, and analyze the data; then naturally
takes the decision. For robots, there is a need for localization algorithms. In the following
paragraphs, we review the main ideas behind the work done in this area. Then, in the next
section, we introduce our approach for odor localization.

Odor localization using robots has attracted many of the researchers and most of the work
done in this area is trying to imitate the animals, birds, and swarms behaviors. For instance,
insects fly in a zigzag pattern towards their matting partners. Ants use the same strategy to
follow the pheromones trails. Therefore, robots could also do the same to locate the odor
sources (Wei et al. , 2001). On the other hand, animals may follow spiral surge algorithm to
locate the odor sources (Hayes et al. , 2002) where animals move in spiral path until it
perceives a certain concentration and then moves straight upwind. As soon as it loses the
scent, it starts spiraling again. A simple example on the zigzag and spiral procedures is
shown in Figure 9. A comparison between the spiral and the zigzag algorithms are
presented in (Lochmatter el at., 2008). The results show that the spiral is more efficient than
the zigzag algorithm.

Fig. 9. Zigzagging (a) and spiral surge (b) are two bio-inspired odor source localization
algorithms for single-robot systems (Lochmatter et al. , 2007).

Another method is reported in (Duckett et al., 2001) where the mobile robot turns around
for 360
to point to the direction of the odor. Then it moves in the odor source direction. A
similar technique has been used in (Loutfi and Coradeschi, 2002) where two gas sensors are
mounted on the front and back of a Koala mobile robot given in Figure 10. The robot has
programmed to use a wall following algorithm to avoid obstacles. The odor tracking
algorithm used in (Loutfi and Coradeschi, 2002) is very simple. The default move direction
for the robot is to move forward. If the front senor has a strong reaction to the odor, the
robot turns 180
and move forward. This turn is used to avoid sensors saturation. If the rear
sensor has stronger reaction to the odor than the front sensor, that means the robot has
passed the odor source and it has to turn around. If both sensors have the same reaction, the
robot is adjusted to move forward.

Fig. 10. Koala mobile robot

In (Wisnu et al. , 2008) , the authors try to localize the odor sources using swarm intelligence
by a group of robots. The main challenge in the swarm intelligence in the authors design is
the interaction. The interaction process is divided into three phases which are encoding,
synchronization, and comparison. The output of the overall interaction process is one of two
outputs according to the robot that has the interaction box and the one is communicating
with it. The encoding phase deals with formatting the received signals in a suitable format
to the robot processing component. The synchronization phase defines which robot the

holder of the interaction box can communicate with. In the third phase, comparison, the
robot compares its sensed data to the received ones. Figure 11 shows the interaction state
diagram for a robot. As stated in (Wisnu et al. , 2008), a robot is wandering in the arena is in
the state of wandering. If it perceives some robots who are in beaconing state, it selects one
of them randomly to make a contact. The robot compares the odor concentrations received
by the two robots. If the concentration received by its mate is higher, the robot approaches
its mate; otherwise it keeps wandering. To get out of the wandering state, either the robot
loses its mate signal or its concentration indicates inverse result. If no way the robot changes
its state to attraction state, after certain period of time, the robot changes its beaconing state.
After another period of time, the robot goes back to wandering state. If two robots move
close to each other, they should avoid collision which is the obstacle avoidance state. The
overall designed swarm behavior follows spatio-temporal model based on Eulerian
framework.

Fig. 11. Interaction state diagram for a robot (Wisnu et al. , 2008).

A swarm based fuzzy logic controller approach is also proposed (Cui el at. , 2004) for odor
localization. The swarms are assumed to be the mobile robots deployed in the odor field.
Robots form a wireless network to exchange their findings using table-driven routing
protocol (Royer and TohToh, 1999). The monitored odor field is divided into a grid of cells
that are marked differently by the robots as given in Figure 12. To avoid collision and
wasting of time and energy, each robot avoids exploration of a cell occupied by other robot.
To keep the robots connected, the swarm cohesion property is utilized for this purpose. The
expansion cell in the Figure is the cell in the grid map that is unexplored and unoccupied.
The fuzzy logic controller is used to avoid the uncertainties in the collected information. The
rule-based fuzzy logic controller is used as shown in Figure 13 where the robot movement
direction is decided by the controller based on the collected information from other robots.
The directions are identified by 8 linguistic variables shown in Figure 14. As shown in the
Figure, the direction limited to only 45
only.


Fig. 9. Zigzagging (a) and spiral surge (b) are two bio-inspired odor source localization
algorithms for single-robot systems (Lochmatter et al. , 2007).

Another method is reported in (Duckett et al., 2001) where the mobile robot turns around
for 360
to point to the direction of the odor. Then it moves in the odor source direction. A
similar technique has been used in (Loutfi and Coradeschi, 2002) where two gas sensors are
mounted on the front and back of a Koala mobile robot given in Figure 10. The robot has
programmed to use a wall following algorithm to avoid obstacles. The odor tracking
algorithm used in (Loutfi and Coradeschi, 2002) is very simple. The default move direction
for the robot is to move forward. If the front senor has a strong reaction to the odor, the
robot turns 180
and move forward. This turn is used to avoid sensors saturation. If the rear
sensor has stronger reaction to the odor than the front sensor, that means the robot has
passed the odor source and it has to turn around. If both sensors have the same reaction, the
robot is adjusted to move forward.

Fig. 10. Koala mobile robot

In (Wisnu et al. , 2008) , the authors try to localize the odor sources using swarm intelligence
by a group of robots. The main challenge in the swarm intelligence in the authors design is
the interaction. The interaction process is divided into three phases which are encoding,
synchronization, and comparison. The output of the overall interaction process is one of two
outputs according to the robot that has the interaction box and the one is communicating
with it. The encoding phase deals with formatting the received signals in a suitable format
to the robot processing component. The synchronization phase defines which robot the

holder of the interaction box can communicate with. In the third phase, comparison, the
robot compares its sensed data to the received ones. Figure 11 shows the interaction state
diagram for a robot. As stated in (Wisnu et al. , 2008), a robot is wandering in the arena is in
the state of wandering. If it perceives some robots who are in beaconing state, it selects one
of them randomly to make a contact. The robot compares the odor concentrations received
by the two robots. If the concentration received by its mate is higher, the robot approaches
its mate; otherwise it keeps wandering. To get out of the wandering state, either the robot
loses its mate signal or its concentration indicates inverse result. If no way the robot changes
its state to attraction state, after certain period of time, the robot changes its beaconing state.
After another period of time, the robot goes back to wandering state. If two robots move
close to each other, they should avoid collision which is the obstacle avoidance state. The
overall designed swarm behavior follows spatio-temporal model based on Eulerian
framework.

Fig. 11. Interaction state diagram for a robot (Wisnu et al. , 2008).

A swarm based fuzzy logic controller approach is also proposed (Cui el at. , 2004) for odor
localization. The swarms are assumed to be the mobile robots deployed in the odor field.
Robots form a wireless network to exchange their findings using table-driven routing
protocol (Royer and TohToh, 1999). The monitored odor field is divided into a grid of cells
that are marked differently by the robots as given in Figure 12. To avoid collision and
wasting of time and energy, each robot avoids exploration of a cell occupied by other robot.
To keep the robots connected, the swarm cohesion property is utilized for this purpose. The
expansion cell in the Figure is the cell in the grid map that is unexplored and unoccupied.
The fuzzy logic controller is used to avoid the uncertainties in the collected information. The
rule-based fuzzy logic controller is used as shown in Figure 13 where the robot movement
direction is decided by the controller based on the collected information from other robots.
The directions are identified by 8 linguistic variables shown in Figure 14. As shown in the
Figure, the direction limited to only 45
only.


Fig. 12. Monitored field cells and their marking (Cui el at. , 2004)

Fig. 13. Robots Fuzzy Logic Controller (Cui el at. , 2004).

Fig. 14. Fuzzy Logic used linguistics (Cui el at. , 2004).

A combination of a static Wireless Sensor Network (WSN) and mobile robots are also used
for muli-odor sources localization (Zhen el at. , 2010). The WSN is assumed to be deployed
in the monitored area to collect information about the odor and the wind speed and

direction. The WSN works as backbone to the robots where nodes may sense the
environment odor, transmit the data to other nodes, sense and transmit at the same time,
and may be nodes could be in non-active state. for multi-hop routing , the authors uses
GPSR-based protocol. Robots on the other side are assumed powerful machines where they
are mounted with odor and vision sensors, Global Positioning System (GPS), and may be
some of the tools to prevent odor from spreading. Robots could communicate to each other
or to the WSN nodes using their wireless devices. However, robots are assumed to form an
ad hoc network with one as a manager and others as workers. The manager collects the
sensed information from the robots and the sensors; then it takes the decision to move the
workers towards the correct direction. Again, the monitored field is divided into small
zones and based on the historical data, the manager can assign different number of robots.

5. Localization and Hybrid Approach
In the following subsections, we introduce a new hybrid solution for odor localization. The
solution uses Genetic algorithms for local search, fuzzy logic to select the suitable direction
and the swarm intelligence to keep robots connected. This work is ongoing work and it is
not finalized yet; however, the initial results seem promising and outperform the previous
solutions.

5.1. The Problem
The problem that we tackle in this section is to find a single source of odor in a certain space
R. The space is described as a 2D environment with known dimensions. R is assumed to
contain a dangerous odor source(s) in unknown location x
s
and y
s
,where s refers to the
hazardous source. Here, we investigate single odor source localization only; multiple odor
sources localizations problem is considered as part of our future work. The space R could
be divided into a grid of cells with known dimensions. In order to save the human rescue
lives as well as to speed up the search in the contaminated space, a number of robots with
appropriate sensors are used to locate the odor source.

Robots are assumed to have a wireless communication that enables them to form an ad hoc
wireless network. The communication range does not have to be large enough to cover all of
the search space. In addition, robots are assumed having the capability of running different
algorithms such as Genetic Algorithms (GA) , Fuzzy Logic Controller (FLC), and Swarm
Intelligence. Robots are also assumed to be capable of identifying their location from a priori
given reference point. Moreover, we assume that the robots will have enough energy
sources to complete their task. Robots are initially deployed close to each other to guarantee
their connectivity. However, the deployment points could be anywhere in the search space.

Another fact that needs to be considered during the localization process is the wind speed
which might cause sensors readings uncertainties. The wind speed has its most effect at
outdoor environments. However, indoor environments might have another source of
readings noise. Therefore, an appropriate method is required to handle the readings
uncertainties. We propose a fuzzy logic approach to solve this problem.

Fig. 12. Monitored field cells and their marking (Cui el at. , 2004)

Fig. 13. Robots Fuzzy Logic Controller (Cui el at. , 2004).

Fig. 14. Fuzzy Logic used linguistics (Cui el at. , 2004).

A combination of a static Wireless Sensor Network (WSN) and mobile robots are also used
for muli-odor sources localization (Zhen el at. , 2010). The WSN is assumed to be deployed
in the monitored area to collect information about the odor and the wind speed and

direction. The WSN works as backbone to the robots where nodes may sense the
environment odor, transmit the data to other nodes, sense and transmit at the same time,
and may be nodes could be in non-active state. for multi-hop routing , the authors uses
GPSR-based protocol. Robots on the other side are assumed powerful machines where they
are mounted with odor and vision sensors, Global Positioning System (GPS), and may be
some of the tools to prevent odor from spreading. Robots could communicate to each other
or to the WSN nodes using their wireless devices. However, robots are assumed to form an
ad hoc network with one as a manager and others as workers. The manager collects the
sensed information from the robots and the sensors; then it takes the decision to move the
workers towards the correct direction. Again, the monitored field is divided into small
zones and based on the historical data, the manager can assign different number of robots.

5. Localization and Hybrid Approach
In the following subsections, we introduce a new hybrid solution for odor localization. The
solution uses Genetic algorithms for local search, fuzzy logic to select the suitable direction
and the swarm intelligence to keep robots connected. This work is ongoing work and it is
not finalized yet; however, the initial results seem promising and outperform the previous
solutions.

5.1. The Problem
The problem that we tackle in this section is to find a single source of odor in a certain space
R. The space is described as a 2D environment with known dimensions. R is assumed to
contain a dangerous odor source(s) in unknown location x
s
and y
s
,where s refers to the
hazardous source. Here, we investigate single odor source localization only; multiple odor
sources localizations problem is considered as part of our future work. The space R could
be divided into a grid of cells with known dimensions. In order to save the human rescue
lives as well as to speed up the search in the contaminated space, a number of robots with
appropriate sensors are used to locate the odor source.

Robots are assumed to have a wireless communication that enables them to form an ad hoc
wireless network. The communication range does not have to be large enough to cover all of
the search space. In addition, robots are assumed having the capability of running different
algorithms such as Genetic Algorithms (GA) , Fuzzy Logic Controller (FLC), and Swarm
Intelligence. Robots are also assumed to be capable of identifying their location from a priori
given reference point. Moreover, we assume that the robots will have enough energy
sources to complete their task. Robots are initially deployed close to each other to guarantee
their connectivity. However, the deployment points could be anywhere in the search space.

Another fact that needs to be considered during the localization process is the wind speed
which might cause sensors readings uncertainties. The wind speed has its most effect at
outdoor environments. However, indoor environments might have another source of
readings noise. Therefore, an appropriate method is required to handle the readings
uncertainties. We propose a fuzzy logic approach to solve this problem.

5.2 Hybrid Computational Intelligence Algorithm (H-CIA) for odor Source Localization
In this section, we explain the details of our algorithm for odor source localization. The
algorithm consists of three phases; the first phase allows the robot to locally search its local
area and come up with the best location to start from. This phase utilizes the GA for
identifying the best location (x, y) to start from. A space search is proposed in the second
phase where robots benefits from the neighbors robots to identify the moving direction.
Since the neighbors readings might not be accurate and involve uncertain information, a
fuzzy logic approach is proposed to identify the moving direction. In the third phase, to
keep the sensors together and get the best of their readings, swarm intelligence is exploited.
Through the next subsections, we explain the details of each phase.

5.2.1 Phase One: Local Search
As mentioned in the problem statement, the search space is divided into a grid of cells.
These cells are assumed equal in terms of their area. Also, robots are initially loaded with
the space map and its cells dimensions. Therefore, a robot i is assumed capable of localizing
itself and identifying its location in the cell. A certain point Pr(x
r
, y
1
) is selected to be the
robots localization reference. Once robots are deployed in the search space, they start to
localize themselves and determine their deployed cells. Our proposal in this phase is to
allow the sensor to locally search their cell for the odor source. Our methodology in this
search is the usage of GA. GA is a well know optimization algorithm and proven to be
efficient in many of the search problems. Therefore, we exploit the GA in the local search
phase and we propose our own GA representation to the problem. This representation
showed good results from among other representations.

First the robots chromosome is its position in terms of x and y coordinates regarding the
given reference point in the search space. A certain set of chromosomes C are initially
generated to represent the GA initial population. This set of chromosomes is limited to the
robots current cell dimensions to limit the GA search space. The crossover in done between
two chromosomes in two steps: 1) randomly select x or y from the chromosome to exchange
and 2) exchange the selected coordinate. The mutation is done based on a probability Pm
where 0< Pm >1. If Pm is greater than 0.5 , a random chromosome is selected from the initial
population and its x coordinate replaces x coordinate in one the current used chromosomes.
On the next iterations, if the evaluation function enhanced, we keep replacing the x
s

coordinates; otherwise we exchange y
s
coordinates instead. The evaluation function is
represented by the measured odor concentration value at each new location which in our
case are two locations based on the current generated chromosomes. The odor distribution
is assumed to follow the Gaussian distribution. The GA algorithm runs for certain number
of iterations and terminates. During these iterations, the best chromosome is stored and it
will be the final robots location in this phase.

5.2.2 Phase Two: Space Search
In this phase of the solution, we try to move the robot towards the odor source and avoid
the GA local minima. At the same time, we make the most out of robots readings so far. As
stated previously, robots readings are uncertain due to the wind speed in outdoor
environments or other noise in indoor environments. In this case Fuzzy logic comes to to

handle the robots readings uncertainties. The objective of fuzzy logic is to identify the
correct direction of the robot to move to based on the current sensors hazardous
concentration readings as well as neighbors robots positions.

Therefore, for the fuzzy logic to work, there are two inputs with two membership functions.
The first input is the robots neighbors readings and the second is their positions in terms of
x and y coordinates. These coordinates are converted to a certain linguistics that relates the
robots positions to the reference point. Here, we assume the reference point is located at the
center of the space and each robot could identify its location to the reference points in terms
of the following 8 directions: 1) Left (L) , 2) Right (R), 3) Top (T), 4) Down (D), 5) Right Top
(RT), 6) Left Top (LT), 7) Left Down (LD), and 8) Right Down (RD). The membership of the
readings input is chosen to be dynamic based on the readings values. Five linguistics are
chosen for this input which are Very-Low, Low, Medium, High, and Very-High. Each one
falls in almost 20% of the inputs range. For the defuzzification, it seems that the center of
gravity (COG) of fuzzy sets is an essential feature that concurrently reflects the location and
shape of the fuzzy sets concerned. Therefore, we use COG as our defuzzification process as
shown in equation (1).

) (
* ) (
a
a a
COG
A
A
(1)
Where, ( )
A
a is the membership function of set A.

The output membership function produces the same eight directions as presented in the
input. Once the robot decides on the direction, it moves according to the next phase
restrictions. When it reaches a new zone, it starts again the GA.

5.2.3 Phase Three: Swarm Movement Control
Once the direction is identified, if the robots move freely, another problem might occur in
which the robots might get disconnected from each other. A similar Swarm intelligence
model presented in (Fei el at., 2008) is used to keep robots together. Probability Particle
Swarm Optimization (P-PSO) algorithm uses probability to express the local and global
fitness functions. More detailed description about the P-PSO algorithm can be found in (Fei
el at., 2008). The coherence characteristic of the P-PSO moves only the robot that does not
affect the robots connectivity.

It is worth mentioning that the initial results based on simulation setup shows that 85% of
the time the robots reach the odor source. However, the cell size has to be small to prevent
the robot in moving in a zigzag form and taken long time to converge. Currently we prepare
a real robot experiments to check the performance of the hybrid approach in odor
localization. Several parameters might not easy to handle in real environments such as the
effect of obstacles, robots speed, robots batteries, etc.


5.2 Hybrid Computational Intelligence Algorithm (H-CIA) for odor Source Localization
In this section, we explain the details of our algorithm for odor source localization. The
algorithm consists of three phases; the first phase allows the robot to locally search its local
area and come up with the best location to start from. This phase utilizes the GA for
identifying the best location (x, y) to start from. A space search is proposed in the second
phase where robots benefits from the neighbors robots to identify the moving direction.
Since the neighbors readings might not be accurate and involve uncertain information, a
fuzzy logic approach is proposed to identify the moving direction. In the third phase, to
keep the sensors together and get the best of their readings, swarm intelligence is exploited.
Through the next subsections, we explain the details of each phase.

5.2.1 Phase One: Local Search
As mentioned in the problem statement, the search space is divided into a grid of cells.
These cells are assumed equal in terms of their area. Also, robots are initially loaded with
the space map and its cells dimensions. Therefore, a robot i is assumed capable of localizing
itself and identifying its location in the cell. A certain point Pr(x
r
, y
1
) is selected to be the
robots localization reference. Once robots are deployed in the search space, they start to
localize themselves and determine their deployed cells. Our proposal in this phase is to
allow the sensor to locally search their cell for the odor source. Our methodology in this
search is the usage of GA. GA is a well know optimization algorithm and proven to be
efficient in many of the search problems. Therefore, we exploit the GA in the local search
phase and we propose our own GA representation to the problem. This representation
showed good results from among other representations.

First the robots chromosome is its position in terms of x and y coordinates regarding the
given reference point in the search space. A certain set of chromosomes C are initially
generated to represent the GA initial population. This set of chromosomes is limited to the
robots current cell dimensions to limit the GA search space. The crossover in done between
two chromosomes in two steps: 1) randomly select x or y from the chromosome to exchange
and 2) exchange the selected coordinate. The mutation is done based on a probability Pm
where 0< Pm >1. If Pm is greater than 0.5 , a random chromosome is selected from the initial
population and its x coordinate replaces x coordinate in one the current used chromosomes.
On the next iterations, if the evaluation function enhanced, we keep replacing the x
s

coordinates; otherwise we exchange y
s
coordinates instead. The evaluation function is
represented by the measured odor concentration value at each new location which in our
case are two locations based on the current generated chromosomes. The odor distribution
is assumed to follow the Gaussian distribution. The GA algorithm runs for certain number
of iterations and terminates. During these iterations, the best chromosome is stored and it
will be the final robots location in this phase.

5.2.2 Phase Two: Space Search
In this phase of the solution, we try to move the robot towards the odor source and avoid
the GA local minima. At the same time, we make the most out of robots readings so far. As
stated previously, robots readings are uncertain due to the wind speed in outdoor
environments or other noise in indoor environments. In this case Fuzzy logic comes to to

handle the robots readings uncertainties. The objective of fuzzy logic is to identify the
correct direction of the robot to move to based on the current sensors hazardous
concentration readings as well as neighbors robots positions.

Therefore, for the fuzzy logic to work, there are two inputs with two membership functions.
The first input is the robots neighbors readings and the second is their positions in terms of
x and y coordinates. These coordinates are converted to a certain linguistics that relates the
robots positions to the reference point. Here, we assume the reference point is located at the
center of the space and each robot could identify its location to the reference points in terms
of the following 8 directions: 1) Left (L) , 2) Right (R), 3) Top (T), 4) Down (D), 5) Right Top
(RT), 6) Left Top (LT), 7) Left Down (LD), and 8) Right Down (RD). The membership of the
readings input is chosen to be dynamic based on the readings values. Five linguistics are
chosen for this input which are Very-Low, Low, Medium, High, and Very-High. Each one
falls in almost 20% of the inputs range. For the defuzzification, it seems that the center of
gravity (COG) of fuzzy sets is an essential feature that concurrently reflects the location and
shape of the fuzzy sets concerned. Therefore, we use COG as our defuzzification process as
shown in equation (1).

) (
* ) (
a
a a
COG
A
A
(1)
Where, ( )
A
a is the membership function of set A.

The output membership function produces the same eight directions as presented in the
input. Once the robot decides on the direction, it moves according to the next phase
restrictions. When it reaches a new zone, it starts again the GA.

5.2.3 Phase Three: Swarm Movement Control
Once the direction is identified, if the robots move freely, another problem might occur in
which the robots might get disconnected from each other. A similar Swarm intelligence
model presented in (Fei el at., 2008) is used to keep robots together. Probability Particle
Swarm Optimization (P-PSO) algorithm uses probability to express the local and global
fitness functions. More detailed description about the P-PSO algorithm can be found in (Fei
el at., 2008). The coherence characteristic of the P-PSO moves only the robot that does not
affect the robots connectivity.

It is worth mentioning that the initial results based on simulation setup shows that 85% of
the time the robots reach the odor source. However, the cell size has to be small to prevent
the robot in moving in a zigzag form and taken long time to converge. Currently we prepare
a real robot experiments to check the performance of the hybrid approach in odor
localization. Several parameters might not easy to handle in real environments such as the
effect of obstacles, robots speed, robots batteries, etc.


6. Conclusion
In this chapter, we explored some of techniques and algorithms used for odor recognition
and localization. We started by introducing different types of sensors that are currently used
for odor sensing. We focused on one of the famous devices which is the electronic nose.
Then, we reviewed some of the concepts used for odor recognition including neural
networks and swarm intelligence. This phase is followed by odor localization using single
and multi-robots. Finally, we proposed a new hybrid method based on Genetic Algorithms,
Fuzzy Logic, and Swarm Intelligence. The Genetic Algorithm is used for local search and the
fuzzy logic identifies the direction of the robots movement. Then, the Swarm Intelligence is
used for robots cohesion and connectivity.

7. References
Al-Bastaki, Y. (2009). An Artificial Neural Networks-Based on-Line Monitoring Odor
Bekir, K. & Kemal , Y. (2007) Fuzzy Clustering Neural Networks for Real-Time Odor
Recognition System. Journal of Automated Methods and Management in
Chemistry, Vol. 2007.
Chatchawal, W. ; Mario, L.; & Teerakiat, K. (2009). Detection and Classification of Human
Body Odor Using an Electronic Nose. Sensors 2009, Vol. 9, pp.7234-7249.
Cui , X.; Hardin , T. ; Ragade, R.; Elmaghraby, A. (2004). A Swarm-based Fuzzy Logic
Control Mobile Sensor Network for Hazardous Contaminants Localization. In
Proceedings of the IEEE International Conference on Mobile Ad-hoc and Sensor
Systems (MASS04).
Duckett, T.; Axelsson, M.; Saffiotti, A. (2001) Learning to locate an odour source with a
mobile robot. In Proc. ICRA-2001, IEEE International Conference on Robotics and
Automation, pp. 2126.
Fei, L.; Qing-Hao, M.; Shuang, B.; Ji-Gong, L.; Dorin, P. (2008). Probability-PSO Algorithm
for Multi-robot Based Odor Source Localization in Ventilated Indoor. Intelligent
Robotics and Applications, Lecture Notes in Computer Science Springer, Vol.
5314,pp. 1206-1215.
Gardner, J. & Bartlett, P. (1999). Electronic Noses Principles and Applications. Oxford
University Press. Oxford, UK.
Hayes, A.; Martinoli, A.; Goodman, R. (2002) Distributed Odor Source Localization. IEEE
Sensors Journal, Vol. 2, No. 3, pp. 260-271.
http://www.it.lut.fi/kurssit/03-04/010970000/seminars/Korotkaya.pdf as visited
on 06/09/2010
hsan, . & Bekir, K. (2009) . Hazardous Odor Recognition by CMAC Based Neural
Networks. Sensor 2009, Vol. 9, pp. 7308-7319;
Jehuda, Y. (2003). Detection of Explosives by Electronic Noses. Analytical Chemistry , Vol. 75,
No. 5, pp. 98 A-105 A.
Jianfeng, Q.; Yi, C.; & Simon, X. (2009). A Real-Time De-Noising Algorithm for E-Noses in a
Wireless Sensor Network. Sensor 2009. Vol. 9, pp.895-908;
Korotkaya, Z. Biometric Person Authentication: Odor, Pages: 1 6,
Linder, R. ; Zamelczyk, M. ; Pppl, R. ; Komider, J. (2005). Polish Journal of Environmental
Studies. Vol. 14, No. 4, pp. 477-481

Lochmatter, L.; Raemy,X.; Matthey, L.; Indra, S.; Martinoli, A. (2008). A Comparison of
Casting and Spiraling Algorithms for Odor Source Localization in Laminar Flow. In
Proceedings of the 2008 IEEE International Conference on Robotics and
Automation (ICRA 2008), pp. 1138-1143.
Lochmatter, T.; Raemy, X.; Martinoli, A. (2007). Odor Source Localization with Mobile
Robots. Bulletin of the Swiss Society for Automatic Control, Vol. 46: pp.11-14.
Loutfi, A. & Coradeschi , S. (2002). Relying on an electronic nose for odor localization," IEEE
International Symposium on Virtual and Intelligent Measurement Systems. VIMS
'02.
MICAZ, http://www.openautomation.net/uploadsproductos/micaz_datasheet.pdf as
visited on 06/09/2010
Natale, C.D.; Mantini, A.; Macagnano, A.; Antuzzi, D.; Paolesse, R.; D'Amico, A. (1999).
Electronic Nose Analysis of Urine Samples Containing Blood. Phys. Meas., pp. 377
384.
Roppel, T. & Wilson, D. (2000). Biologically-inspired pattern recognition for odor detection.
Pattern Recogn. Lett. 21, No.3, pp. 213-219.
Royer, E. & TohToh, C. (1999). A review of current routing protocols for ad hoc mobile,
IEEE Personal Communication, Vol. 6, No.2, pp. 4655.
Sensing System, Journal of Computer Science , Vol. 5, No. 11, pp. 878-882,
Staples, E. & Viswanathan,S. (2005). Odor Detection and Analysis using GC/SAW zNose.
Electronic Sensor Technology, School of Engineering and Technology, National
University.
Wei, L.; Jay, A.; Ring, T. (2001). Tracking of Fluid-Advected Odor Plumes: Strategies
Inspired by Insect Orientation to Pheromone. Journal of Adaptive Behavior, Vol. 9,
No. 3-4, pp. 143-170.
Wisnu, J. ; Petrus, M. ; Benyamin, K. ; Kosuke, S. ; Toshio, F. (2008) Modified PSO
Algorithm Based on Flow of Wind for Odor Source Localization Problems in
Dynamic Environments. In Wseas transactions on Systems, Vol. 7.
Zhen, F.; Zhan, Z. ; Xunxue, C. ; Daoqu, G. ; Yundong, X. ; LiDong, D. ; ShaoHua, W. (2010).
Multi-odor Sources Localization and Tracking with Wireless Sensor Network and
Mobile Robots. 1st IET international Conference on Wireless Sensor Networks. IET-
WSN-2010.
zNose device available at http://www.estcal.com/ as visited on 06/09/2010.


6. Conclusion
In this chapter, we explored some of techniques and algorithms used for odor recognition
and localization. We started by introducing different types of sensors that are currently used
for odor sensing. We focused on one of the famous devices which is the electronic nose.
Then, we reviewed some of the concepts used for odor recognition including neural
networks and swarm intelligence. This phase is followed by odor localization using single
and multi-robots. Finally, we proposed a new hybrid method based on Genetic Algorithms,
Fuzzy Logic, and Swarm Intelligence. The Genetic Algorithm is used for local search and the
fuzzy logic identifies the direction of the robots movement. Then, the Swarm Intelligence is
used for robots cohesion and connectivity.

7. References
Al-Bastaki, Y. (2009). An Artificial Neural Networks-Based on-Line Monitoring Odor
Bekir, K. & Kemal , Y. (2007) Fuzzy Clustering Neural Networks for Real-Time Odor
Recognition System. Journal of Automated Methods and Management in
Chemistry, Vol. 2007.
Chatchawal, W. ; Mario, L.; & Teerakiat, K. (2009). Detection and Classification of Human
Body Odor Using an Electronic Nose. Sensors 2009, Vol. 9, pp.7234-7249.
Cui , X.; Hardin , T. ; Ragade, R.; Elmaghraby, A. (2004). A Swarm-based Fuzzy Logic
Control Mobile Sensor Network for Hazardous Contaminants Localization. In
Proceedings of the IEEE International Conference on Mobile Ad-hoc and Sensor
Systems (MASS04).
Duckett, T.; Axelsson, M.; Saffiotti, A. (2001) Learning to locate an odour source with a
mobile robot. In Proc. ICRA-2001, IEEE International Conference on Robotics and
Automation, pp. 2126.
Fei, L.; Qing-Hao, M.; Shuang, B.; Ji-Gong, L.; Dorin, P. (2008). Probability-PSO Algorithm
for Multi-robot Based Odor Source Localization in Ventilated Indoor. Intelligent
Robotics and Applications, Lecture Notes in Computer Science Springer, Vol.
5314,pp. 1206-1215.
Gardner, J. & Bartlett, P. (1999). Electronic Noses Principles and Applications. Oxford
University Press. Oxford, UK.
Hayes, A.; Martinoli, A.; Goodman, R. (2002) Distributed Odor Source Localization. IEEE
Sensors Journal, Vol. 2, No. 3, pp. 260-271.
http://www.it.lut.fi/kurssit/03-04/010970000/seminars/Korotkaya.pdf as visited
on 06/09/2010
hsan, . & Bekir, K. (2009) . Hazardous Odor Recognition by CMAC Based Neural
Networks. Sensor 2009, Vol. 9, pp. 7308-7319;
Jehuda, Y. (2003). Detection of Explosives by Electronic Noses. Analytical Chemistry , Vol. 75,
No. 5, pp. 98 A-105 A.
Jianfeng, Q.; Yi, C.; & Simon, X. (2009). A Real-Time De-Noising Algorithm for E-Noses in a
Wireless Sensor Network. Sensor 2009. Vol. 9, pp.895-908;
Korotkaya, Z. Biometric Person Authentication: Odor, Pages: 1 6,
Linder, R. ; Zamelczyk, M. ; Pppl, R. ; Komider, J. (2005). Polish Journal of Environmental
Studies. Vol. 14, No. 4, pp. 477-481

Lochmatter, L.; Raemy,X.; Matthey, L.; Indra, S.; Martinoli, A. (2008). A Comparison of
Casting and Spiraling Algorithms for Odor Source Localization in Laminar Flow. In
Proceedings of the 2008 IEEE International Conference on Robotics and
Automation (ICRA 2008), pp. 1138-1143.
Lochmatter, T.; Raemy, X.; Martinoli, A. (2007). Odor Source Localization with Mobile
Robots. Bulletin of the Swiss Society for Automatic Control, Vol. 46: pp.11-14.
Loutfi, A. & Coradeschi , S. (2002). Relying on an electronic nose for odor localization," IEEE
International Symposium on Virtual and Intelligent Measurement Systems. VIMS
'02.
MICAZ, http://www.openautomation.net/uploadsproductos/micaz_datasheet.pdf as
visited on 06/09/2010
Natale, C.D.; Mantini, A.; Macagnano, A.; Antuzzi, D.; Paolesse, R.; D'Amico, A. (1999).
Electronic Nose Analysis of Urine Samples Containing Blood. Phys. Meas., pp. 377
384.
Roppel, T. & Wilson, D. (2000). Biologically-inspired pattern recognition for odor detection.
Pattern Recogn. Lett. 21, No.3, pp. 213-219.
Royer, E. & TohToh, C. (1999). A review of current routing protocols for ad hoc mobile,
IEEE Personal Communication, Vol. 6, No.2, pp. 4655.
Sensing System, Journal of Computer Science , Vol. 5, No. 11, pp. 878-882,
Staples, E. & Viswanathan,S. (2005). Odor Detection and Analysis using GC/SAW zNose.
Electronic Sensor Technology, School of Engineering and Technology, National
University.
Wei, L.; Jay, A.; Ring, T. (2001). Tracking of Fluid-Advected Odor Plumes: Strategies
Inspired by Insect Orientation to Pheromone. Journal of Adaptive Behavior, Vol. 9,
No. 3-4, pp. 143-170.
Wisnu, J. ; Petrus, M. ; Benyamin, K. ; Kosuke, S. ; Toshio, F. (2008) Modified PSO
Algorithm Based on Flow of Wind for Odor Source Localization Problems in
Dynamic Environments. In Wseas transactions on Systems, Vol. 7.
Zhen, F.; Zhan, Z. ; Xunxue, C. ; Daoqu, G. ; Yundong, X. ; LiDong, D. ; ShaoHua, W. (2010).
Multi-odor Sources Localization and Tracking with Wireless Sensor Network and
Mobile Robots. 1st IET international Conference on Wireless Sensor Networks. IET-
WSN-2010.
zNose device available at http://www.estcal.com/ as visited on 06/09/2010.

Communication and Networking Technologies
Part 2
Communication and Networking Technologies
Modelling Underwater Wireless Sensor Networks
X

Modelling Underwater Wireless
Sensor Networks

Universidad Miguel Hernndez de Elche
Spain

1. Introduction
The study of Underwater Wireless Networks as a research field has grown significantly in
recent years offering a multitude of proposal to resolve the communication between the
nodes and protocols for information exchange networks. Acoustics has been used by nature
for years to communicate in the underwater environment using it as a language, dolphins
and whales for instance are able to use it to send information between their groups. The first
reference to the underwater sound propagation can be found in what Leonardo Da Vinci
wrote in 1490: "If you cause your ship to stop and place the head of a long tube in the water
and place the outer extremity to your ear, you will hear ships at a great distance from you".

Years later in 1826 the first scientific studies were done by picking real data measures
(Colladon, 1893). The physicist Jean-Daniel Colladon, and his partner Charles-Francois
Sturm a mathematic, made the first recorded attempt at Lake Geneva, Switzerland, to find
out the speed of sound in water. After experimenting with an underwater bell with ignition
of gunpowder on a first boat, the sound of the bell and flash from the gunpowder were
observed 10 miles away on a second boat. With this collection of data of the time between
the gunpowder flash and the reception of the sound reaching to the second boat they were
able to establish a pretty accurate value for the speed of the sound in water, tested with this
empirical method.

In the early XX century in 1906, the first sonar type was developed for military purpose by
Lewis Nixon; there was a great interest in this technology during World War I so as to be
able to detect submarines. It was in 1915, when the "echo location to detect submarines" was
released by physicist Paul Langvin and the engineer Constantine Chilowski, device capable
for detecting submarines using the piezoelectric properties of the quartz. It was not useful
for the war as it arrived too late, but it established the roots of the upcoming design for
sonar devices.

The first targets in which the developing of underwater sound technology was involved
were to determine the distance to the shore or to other ships. After experimenting it was
quickly discovered by researchers that pointing the sound device down towards the
11

seafloor, the depth could also be collected with enough precision. Then, by picking a lot of
values it was used for new purposes like measuring the relief of the ocean (bathymetry),
seafloor shape registering, search for geological resources (i.e. oil, gas, etc.), detecting and
tracking fish banks, submarine archaeology, etc.

These were the main underwater acoustic application mainly use for the exploration of
seafloor and fishery with sonar devices. In the 90s the researchers became aware of a new
feature applicable to underwater communications, multipoint connections could be capable
of translating the networked communication technology to the underwater environment.
One of the former deployments was the Autonomous Oceanographic Surveillance Network
(AOSN), supported by the US Office of Naval Research (ONR) (Curtin et al, 1993). It calls for
a system of moorings, surface buoys, underwater sensor nodes and Autonomous
Underwater Vehicles (AUVs) to coordinate their sampling via an acoustic telemetry
network.

Wireless terrestrial networking technologies have experienced a considerably development
in the last fifteen years, not only in the standardization areas but also in the market
deployment of a bunch of devices, services and applications. Among all these wireless
products, wireless sensor networks are exhibiting an incredible boom, being one of the
technological areas with greater scientific and industrial development step (Akyildiz et al,
2002).

The interest and opportunity in working on wireless sensor network technologies is
endorsed by (a) technological indicators like the ones published by MIT (Massachusetts
Institute of Technology) in 2003 (Werff, 2003) where wireless sensor network technology
was defined as one of the 10 technologies that will change the world, and (b) economic and
market forecasts published by different economic magazines like (Rosenbush et al, 2004),
where investment in Wireless Sensor Network (WSN) ZigBee technology was estimated
over 3.500 Million dollars during 2007.

Recently, wireless sensor networks have been proposed for their deployment in underwater
environments where many of applications such us aquiculture, pollution monitoring,
offshore exploration, etc. would benefit from this technology (Cui et al, 2006). Despite
having a very similar functionality, Underwater Wireless Sensor Networks (UWSNs) exhibit
several architectural differences with respect to the terrestrial ones, which are mainly due to
the transmission medium characteristics (sea water) and the signal employed to transmit
data (acoustic ultrasound signals) (Akyildiz et al, 2006).

Then, the design of appropriate network architecture for UWSNs is seriously hardened by
the conditions of the communication system and, as a consequence, what is valid for
terrestrial WSNs is perhaps not valid for UWSNs. So, a general review of the overall
network architecture is required in order to supply an appropriate network service for the
demanding applications in such an unfriendly submarine communication environment.

Major challenges in the design of underwater acoustic networks (Llor & Malumbres 2009)
are:

Battery power is limited and usually batteries cannot be recharged because solar energy
cannot be exploited;
The available bandwidth is severely limited;
The channel suffers from long and variable propagation delays, multi-path and fading
problems;
Bit error rates are typically very high;
Underwater sensors are prone to frequent failures because of fouling, corrosion, etc.

This chapter will give an overview of underwater wireless networks going-through all the
layers with emphasis on the physical layer and how it behaves in different and changing
environment conditions. Besides a brief outline of the most outstanding MAC layer
protocols as the ones of the routing layer algorithms will be presented. Also the main
application are presented and finally the conclusions.

In the next section, we briefly describe the main issues in the design of efficient underwater
wireless sensor networks. Following a bottom-to-top approach, we will review the network
architecture, highlighting some critical design parameters at each of the different network
layers, and overcoming the limitations and problems introduced by UWSN environments.

2. Topology
In (Partan et al, 2006), taxonomy of UWSN regimes is outlined. They classify different UWSNs
in terms of both spatial coverage and node density. For every kind of network topology,
different architectural approaches have to be considered in order to improve the network
performance (throughput, delay, power consumption, packet loss, etc.). So, it is important to
design the network architecture taking into account the intended network topology.

3. Physical Layer: Acoustic Link
The most common way to send data in underwater environments is by means of acoustic
signals, dolphins and whales use it to communicate. Radio frequency signals have serious
problems to propagate in sea water, being operative for radio-frequency only at very short
ranges (up to 10 meters) and with low-bandwidth modems (terms of Kbps). When using
optical signals the light is strongly scattered and absorbed underwater, so only in very clear
water conditions (often very deep) does the range go up to 100 meters with high bandwidth
modems (several Mbps).

The theory of the sound propagation is according to the description by Urick (Urick &
Robert, 1983), a regular molecular movement in an elastic substance that propagates to
adjacent particles. A sound wave can be considered as the mechanical energy that is
transmitted by the source from particle to particle, being propagated through the ocean at
the sound speed. The propagation of such waves will refract upwards or downwards in
agreement with the changes in salinity, temperature and the pressure that have a great
impact on the sound speed, ranging from 1450 to 1540 m/s.

seafloor, the depth could also be collected with enough precision. Then, by picking a lot of
values it was used for new purposes like measuring the relief of the ocean (bathymetry),
seafloor shape registering, search for geological resources (i.e. oil, gas, etc.), detecting and
tracking fish banks, submarine archaeology, etc.

These were the main underwater acoustic application mainly use for the exploration of
seafloor and fishery with sonar devices. In the 90s the researchers became aware of a new
feature applicable to underwater communications, multipoint connections could be capable
of translating the networked communication technology to the underwater environment.
One of the former deployments was the Autonomous Oceanographic Surveillance Network
(AOSN), supported by the US Office of Naval Research (ONR) (Curtin et al, 1993). It calls for
a system of moorings, surface buoys, underwater sensor nodes and Autonomous
Underwater Vehicles (AUVs) to coordinate their sampling via an acoustic telemetry
network.

Wireless terrestrial networking technologies have experienced a considerably development
in the last fifteen years, not only in the standardization areas but also in the market
deployment of a bunch of devices, services and applications. Among all these wireless
products, wireless sensor networks are exhibiting an incredible boom, being one of the
technological areas with greater scientific and industrial development step (Akyildiz et al,
2002).

The interest and opportunity in working on wireless sensor network technologies is
endorsed by (a) technological indicators like the ones published by MIT (Massachusetts
Institute of Technology) in 2003 (Werff, 2003) where wireless sensor network technology
was defined as one of the 10 technologies that will change the world, and (b) economic and
market forecasts published by different economic magazines like (Rosenbush et al, 2004),
where investment in Wireless Sensor Network (WSN) ZigBee technology was estimated
over 3.500 Million dollars during 2007.

Recently, wireless sensor networks have been proposed for their deployment in underwater
environments where many of applications such us aquiculture, pollution monitoring,
offshore exploration, etc. would benefit from this technology (Cui et al, 2006). Despite
having a very similar functionality, Underwater Wireless Sensor Networks (UWSNs) exhibit
several architectural differences with respect to the terrestrial ones, which are mainly due to
the transmission medium characteristics (sea water) and the signal employed to transmit
data (acoustic ultrasound signals) (Akyildiz et al, 2006).

Then, the design of appropriate network architecture for UWSNs is seriously hardened by
the conditions of the communication system and, as a consequence, what is valid for
terrestrial WSNs is perhaps not valid for UWSNs. So, a general review of the overall
network architecture is required in order to supply an appropriate network service for the
demanding applications in such an unfriendly submarine communication environment.

Major challenges in the design of underwater acoustic networks (Llor & Malumbres 2009)
are:

Battery power is limited and usually batteries cannot be recharged because solar energy
cannot be exploited;
The channel suffers from long and variable propagation delays, multi-path and fading
problems;
Bit error rates are typically very high;
Underwater sensors are prone to frequent failures because of fouling, corrosion, etc.

This chapter will give an overview of underwater wireless networks going-through all the
layers with emphasis on the physical layer and how it behaves in different and changing
environment conditions. Besides a brief outline of the most outstanding MAC layer
protocols as the ones of the routing layer algorithms will be presented. Also the main
application are presented and finally the conclusions.

In the next section, we briefly describe the main issues in the design of efficient underwater
wireless sensor networks. Following a bottom-to-top approach, we will review the network
architecture, highlighting some critical design parameters at each of the different network
layers, and overcoming the limitations and problems introduced by UWSN environments.

2. Topology
In (Partan et al, 2006), taxonomy of UWSN regimes is outlined. They classify different UWSNs
in terms of both spatial coverage and node density. For every kind of network topology,
different architectural approaches have to be considered in order to improve the network
performance (throughput, delay, power consumption, packet loss, etc.). So, it is important to
design the network architecture taking into account the intended network topology.

3. Physical Layer: Acoustic Link
The most common way to send data in underwater environments is by means of acoustic
signals, dolphins and whales use it to communicate. Radio frequency signals have serious
problems to propagate in sea water, being operative for radio-frequency only at very short
ranges (up to 10 meters) and with low-bandwidth modems (terms of Kbps). When using
optical signals the light is strongly scattered and absorbed underwater, so only in very clear
water conditions (often very deep) does the range go up to 100 meters with high bandwidth
modems (several Mbps).

The theory of the sound propagation is according to the description by Urick (Urick &
Robert, 1983), a regular molecular movement in an elastic substance that propagates to
adjacent particles. A sound wave can be considered as the mechanical energy that is
transmitted by the source from particle to particle, being propagated through the ocean at
the sound speed. The propagation of such waves will refract upwards or downwards in
agreement with the changes in salinity, temperature and the pressure that have a great
impact on the sound speed, ranging from 1450 to 1540 m/s.

Fig. 1. This diagram offers a basic illustration of the depth at which different colors of light
penetrate ocean waters. Water absorbs warm colors like reds and oranges and scatters the
cooler colors

Fig. 2. Temperature variation depending on latitude and season

Depth (m) Salinity (ppm)
0 37.45
50 36.02
100 35.34
500 35.11
1000 34.90
1500 34.05
Table 1. Salinity depending on the depth
0
200
400
600
800
1000
1200
1400
1600
1800
2000
0 5 10 15 20 25 30
LOWLAT
MIDLATSUMMER
MIDLATWINTER
HIGHLAT
Temperature (C)
Depth (m)

The transmission loss (TL) is defined as the decrease of the sound intensity through the path
from the sender to the receiver. There have been developed diverse empirical expressions to
measure the transmission loss. Thorp formula (Urick & Robert, 1983) defines the signal
transmission loss as:

0.11 f
1 f

44 f
4100 f

0

10

(1)

where f is frequency in kHz, r is the range in meters; SS is the spherical spreading factor and
is the attenuation factor. Then a more accurate expression for the attenuation factor was
presented, the one proposed in the Thorp formula in (Berkhovskikh & Laysanov, 1982):

0.11 f
1 f

44 f
4100 f
. 10
0.00
(2)

Since acoustic signals are mainly used in UWSNs, it is necessary to take into account the
main aspects involved in the propagation of acoustic signals in underwater environments,
including: (1) the propagation speed of sound underwater is around 1500 m/s (5 orders of
magnitude slower than the speed of light), and so the communication links will suffer from
large and variable propagation delays and relatively large motion-induced Doppler effects;
(2) phase and magnitude fluctuations lead to higher bit error rates compared with radio
channels behaviour, this makes necessary the use of forward error correction codes (FEC);
(3) as frequency increases, the attenuation observed in the acoustic channel also increases,
which is a serious bandwidth constraint; (4) multipath interference in underwater acoustic
communications is severe due mainly to the surface waves or vessel activity, that are an
important issue to attain good bandwidth efficiency.

Several works in the literature propose models for an acoustic underwater link, taking into
account environment parameters as salinity degree, temperature, depth, environmental
interference, etc. Other physical aspects of the ocean as noise in the medium (Coates, 1989),
the wind, thermal noise, the turbulence and the ship noise are included by these formulas,
depending on the frequency and this factors:

10
1 0
10
40 0 0.
10
0 .
0 40 0.4
10
1 0

(3)


Fig. 1. This diagram offers a basic illustration of the depth at which different colors of light
penetrate ocean waters. Water absorbs warm colors like reds and oranges and scatters the
cooler colors

Fig. 2. Temperature variation depending on latitude and season

Depth (m) Salinity (ppm)
0 37.45
50 36.02
100 35.34
500 35.11
1000 34.90
1500 34.05
Table 1. Salinity depending on the depth
0
200
400
600
800
1000
1200
1400
1600
1800
2000
0 5 10 15 20 25 30
LOWLAT
MIDLATSUMMER
MIDLATWINTER
HIGHLAT
Temperature (C)
Depth (m)

The transmission loss (TL) is defined as the decrease of the sound intensity through the path
from the sender to the receiver. There have been developed diverse empirical expressions to
measure the transmission loss. Thorp formula (Urick & Robert, 1983) defines the signal
transmission loss as:

0.11 f
1 f

44 f
4100 f

0

10

(1)

where f is frequency in kHz, r is the range in meters; SS is the spherical spreading factor and
is the attenuation factor. Then a more accurate expression for the attenuation factor was
presented, the one proposed in the Thorp formula in (Berkhovskikh & Laysanov, 1982):

0.11 f
1 f

44 f
4100 f
. 10
0.00
(2)

Since acoustic signals are mainly used in UWSNs, it is necessary to take into account the
main aspects involved in the propagation of acoustic signals in underwater environments,
including: (1) the propagation speed of sound underwater is around 1500 m/s (5 orders of
magnitude slower than the speed of light), and so the communication links will suffer from
large and variable propagation delays and relatively large motion-induced Doppler effects;
(2) phase and magnitude fluctuations lead to higher bit error rates compared with radio
channels behaviour, this makes necessary the use of forward error correction codes (FEC);
(3) as frequency increases, the attenuation observed in the acoustic channel also increases,
which is a serious bandwidth constraint; (4) multipath interference in underwater acoustic
communications is severe due mainly to the surface waves or vessel activity, that are an
important issue to attain good bandwidth efficiency.

Several works in the literature propose models for an acoustic underwater link, taking into
account environment parameters as salinity degree, temperature, depth, environmental
interference, etc. Other physical aspects of the ocean as noise in the medium (Coates, 1989),
the wind, thermal noise, the turbulence and the ship noise are included by these formulas,
depending on the frequency and this factors:

10
1 0
10
40 0 0.
10
0 .
0 40 0.4
10
1 0

(3)


where N
t
is the noise due to turbulence, N
s
is the noise due to shipping, N
w
is the noise due
to wind, and N
th
represents the thermal noise. The overall noise power spectral density for a
given frequency f is then:

(4)

In (Xie & Gibson, 2006) the authors present the Monterey-Miami Parabolic Equation to
describe the behavior of the propagation of the sound. In (Porter & Liu 2010) Bellhop, a ray
tracing tool shows how the physical environment conditions and terrain shapes have a great
impact in the sound attenuation.

3.1 MMPE
Monterey-Miami Parabolic Equation (MMPE) model is used to predict underwater acoustic
propagation using a parabolic equation which is closer to the Helmholzt equation (wave
equation), this equation is based on Fourier analysis. The sound pressure is calculated in
small increments changes in range and depth, forming a grid. If we increase the step size,
we can obtain better performance. The propagation loss formula based on the MMPE model:

t f, s, d
A
, d
B
t (5)
where:
PL(t): propagation loss while transmitting from node A to node B.
m(): propagation loss without random and periodic components; obtained from regression
using MMPE data.
f: frequency of transmitted acoustic signals (in kHz).
d
A
: senders depth (in meters).
d
B
: receivers depth (in meters).
r: horizontal distance between A and B nodes, called range in MMPE model (in meters).
s: Euclidean distance between A and B nodes (in meters).
w(t): periodic function to approximate signal loss due to wave movement.
e(): signal loss due to random noise or error.

The m() function represents the propagation loss provided by the MMPE model. According
to the logarithmic nature of the data, a nonlinear regression is the best option to provide an
approach to the model based on the coefficients supplied by the preliminary model. The
proposed expression to calculate this function is the following one:

, ,

0.914
0.00 0.00

(6)

The w() function considers the movement of a particle that will oscillate around its location
in a sinusoidal way. That movement is represented as circular oscillations that reduce their
radius as the depth of the particle increases. The length of that radius is dependent of the

energy of the wave and is related to the height of the wave. The common waves have
hundreds of meters of wavelength and have an effect up to 50 meters of depth

For the calculation of the effects of the wave we will consider:

(7)

where
w(t): periodic function to approximate the lost signal by the wave movement.
h(): scale factor function.
l
W
: ocean wave length (meters).
d
B
: depth of the receiver node.
h
W
: wave height (meters).
T
W
: wave period (seconds).
E(): function of wave effects in nodes.

This function contains the elements that are resembled the node movement, first calculating
the scale factor h() and then the wave effect in a particular phase of the movement. The
calculation of the scale factor is as follows:

0.5

2mod T

(8)

The e() function represents a random term to explain background noise. As the number of
sound sources is large and undetermined, this random noise follows a Gaussian distribution
and is modeled to have a maximum of 20dB at the furthest distance. This function is
calculated by the following equation:

20

R
N

(9)
where:

e(): random noise function
s: distance between the sender and receiver (in meters).
s
max
: maximum distance (transmission range)
h
w
: height of the wave (in meters).
R
N
: random number, Gaussian distribution centered in 0 and with variance 1.


where N
t
is the noise due to turbulence, N
s
is the noise due to shipping, N
w
is the noise due
to wind, and N
th
represents the thermal noise. The overall noise power spectral density for a
given frequency f is then:

(4)

In (Xie & Gibson, 2006) the authors present the Monterey-Miami Parabolic Equation to
describe the behavior of the propagation of the sound. In (Porter & Liu 2010) Bellhop, a ray
tracing tool shows how the physical environment conditions and terrain shapes have a great
impact in the sound attenuation.

3.1 MMPE
Monterey-Miami Parabolic Equation (MMPE) model is used to predict underwater acoustic
propagation using a parabolic equation which is closer to the Helmholzt equation (wave
equation), this equation is based on Fourier analysis. The sound pressure is calculated in
small increments changes in range and depth, forming a grid. If we increase the step size,
we can obtain better performance. The propagation loss formula based on the MMPE model:

t f, s, d
A
, d
B
t (5)
where:
PL(t): propagation loss while transmitting from node A to node B.
m(): propagation loss without random and periodic components; obtained from regression
using MMPE data.
f: frequency of transmitted acoustic signals (in kHz).
d
A
: senders depth (in meters).
d
B
: receivers depth (in meters).
r: horizontal distance between A and B nodes, called range in MMPE model (in meters).
s: Euclidean distance between A and B nodes (in meters).
w(t): periodic function to approximate signal loss due to wave movement.
e(): signal loss due to random noise or error.

The m() function represents the propagation loss provided by the MMPE model. According
to the logarithmic nature of the data, a nonlinear regression is the best option to provide an
approach to the model based on the coefficients supplied by the preliminary model. The
proposed expression to calculate this function is the following one:

, ,

0.914
0.00 0.00

(6)

The w() function considers the movement of a particle that will oscillate around its location
in a sinusoidal way. That movement is represented as circular oscillations that reduce their
radius as the depth of the particle increases. The length of that radius is dependent of the

energy of the wave and is related to the height of the wave. The common waves have
hundreds of meters of wavelength and have an effect up to 50 meters of depth

For the calculation of the effects of the wave we will consider:

(7)

where
w(t): periodic function to approximate the lost signal by the wave movement.
h(): scale factor function.
l
W
: ocean wave length (meters).
d
B
: depth of the receiver node.
h
W
: wave height (meters).
T
W
: wave period (seconds).
E(): function of wave effects in nodes.

This function contains the elements that are resembled the node movement, first calculating
the scale factor h() and then the wave effect in a particular phase of the movement. The
calculation of the scale factor is as follows:

0.5

2mod T

(8)

The e() function represents a random term to explain background noise. As the number of
sound sources is large and undetermined, this random noise follows a Gaussian distribution
and is modeled to have a maximum of 20dB at the furthest distance. This function is
calculated by the following equation:

20

R
N

(9)
where:

e(): random noise function
s: distance between the sender and receiver (in meters).
s
max
: maximum distance (transmission range)
h
w
: height of the wave (in meters).
R
N
: random number, Gaussian distribution centered in 0 and with variance 1.


Fig. 3. MMPE Transmission Loss (db)

3.2 Bellhop
Ray tracing requires the solution of the ray equations to determine the ray coordinates.
Amplitude and acoustic pressure requires the solution of the dynamic ray equations. For a
system with cylindrical symmetry the ray equations can be written:

dr
ds
cs ,
d
ds

1
c
c
r

(10)

where r(s) and z(s) represent the ray coordinates in cylindrical coordinates and s is the
arclenght along the ray; the pair c(s) [ (s),(s)] represents the tangent versor along the ray.
Initial conditions for and r(s), and z(s) , (s) and (s) are

r0 r
, z0 z
, 0
cos
, 0
sin

(11)

Fig. 4. Bellhop ray trace

where
s
represents the launching angle, ( r
s
, z
s
) is the source position, and c
s
is the sound
speed at the source position. The coordinates are sufficient to obtain the ray travel time:

ds
s

(12)

which is calculated along the curve, [ r , z
s
].

Fig. 5. Bellhop pressure


Fig. 3. MMPE Transmission Loss (db)

3.2 Bellhop
Ray tracing requires the solution of the ray equations to determine the ray coordinates.
Amplitude and acoustic pressure requires the solution of the dynamic ray equations. For a
system with cylindrical symmetry the ray equations can be written:

dr
ds
cs ,
d
ds

1
c
c
r

(10)

where r(s) and z(s) represent the ray coordinates in cylindrical coordinates and s is the
arclenght along the ray; the pair c(s) [ (s),(s)] represents the tangent versor along the ray.
Initial conditions for and r(s), and z(s) , (s) and (s) are

r0 r
, z0 z
, 0
cos
, 0
sin

(11)

Fig. 4. Bellhop ray trace

where
s
represents the launching angle, ( r
s
, z
s
) is the source position, and c
s
is the sound
speed at the source position. The coordinates are sufficient to obtain the ray travel time:

ds
s

(12)

which is calculated along the curve, [ r , z
s
].

Fig. 5. Bellhop pressure


4. Mac Layer
The main task of MAC protocols is to provide efficient and reliable access to the shared
physical medium in terms of throughput, delay, error rates and energy consumption.
However, due to the different nature of the underwater environment, there are several
drawbacks with respect to the suitability of the existing terrestrial MAC solutions for the
underwater environment. In fact, channel access control in UWSNs poses additional
challenges due to the aforementioned peculiarities of underwater channels (Molins and
Stojanovic, 2006).

In the MAC Layer we have different types of protocols and ones of the latest examples
presented in each one of the categories. The first ones are the non handshaking protocol
such as Aloha-like, an on the opposite side we have the handshaking protocols like DACAP
(Distance Aware Collision Avoidance Protocol) a collision avoidance protocol based on
virtual Medium Access Control carrier sensing. A number of adaptations have been
proposed to adopt MACA (Multiple Access with Collision Avoidanc) (Karn, 1990),
MACAW (Media Access Protocol for Wireless LAN's) (Bharghavan et al, 1994), and FAMA
(Floor Acquisition Multiple Access) (Fullmer & Luna-Acebes, 1995) for underwater
networks in (Molins & Stojanovic, 2006). But also new protocols such as T-Lohi (between
handshake and non-handshake) are becoming more popular as they have also a great
efficiency in terms of the battery use.

4.1 Aloha-like
In Aloha like mode the source node sends its data frames as soon as it receives a packet from
the upper-layer protocol. It does not check the medium to see if it is busy and so it does not
perform any back-off, it. The node that receives the data will answer with and acknowledge
data frame, if there was no problem at the reception such as a collision or packet lost during
the transmission (i.e. when there is overlapping of the receiving periods of two or more
frames at the destination location, or the receiver was transmitting).

If the source, does not receive an ACK because either the frame was not correctly delivered
or the ACK was lost, the sender will timeout, wait a random period (back-off) and
retransmit the frame. This protocol follows the stop-and-wait paradigm. That is, the source
must receive an acknowledgement for each data frame before the next frame can be sent. In
addition, after a successful frame transmission, the sender will perform a back off, even if it
has additional frames to send from the same packet or from a new packet.

There different version called Aloha-based protocols, a couple of proposals modification for
underwater networks of the protocol can be found in (Chirdchoo, 2007) that presents Aloha
with Collision Avoidance (Aloha-CA) and Aloha with Advance Notification (Aloha AN).

Aloha-CA: Pays close attention to every packet it overhears, picking the information of
who are the sender and receiver. With this information it can easily calculate the busy
duration due to the packet at every one of the nodes. Each node will store in a database
table the information of the monitored packets with the busy durations of every node of
the neighbourhood.

Aloha-AN: This protocol has all the features of Aloha-CA and adds the sending of a
small advanced notification packet with the necessary information to let the other nodes
build the databases tables. The sender will wait after this packet for a lag time before
sending the actual data packet. Whenever a node has a packet to transmit, it will check
the database table to ensure that the packet does not result in a collision at any other
reachable nodes.

4.2 DACAP
DACAP is a handshaking protocol Distance Aware Collision Avoidance Protocol (Peleato &
Stojanovic 2007) for Ad Hoc Underwater Acoustic Sensor Networks. This protocol is a non-
synchronized data access that uses control messages to decide when to start the
communication Request-To-Send (RTS) / Clear-To-Send (CTS) handshake. If a collision
happens during the handshake, the transmitter will not receive a CTS that enables him to
send the data packet.

The protocol is describe in Figure 6 and Figure 7 and explained in the following steps:

RTS reception: After this event, the node sends a CTS to the sender, and waits for a data
packet. If another RTS is received the node sends a warning short packet to give and
advice to the sender of the last one that the medium is in use.

Fig. 6. Transmission in DACAP

CTS reception: When the transmitter receives this message it waits for those nodes
whose transmissions are still happening to avoid any collisions. If it happens to receive
another CTS or a warning packet reception the current data packet to send will be
deferred for a random back-off time, if the waiting time expires the transmission
proceeds normally.


4. Mac Layer
The main task of MAC protocols is to provide efficient and reliable access to the shared
physical medium in terms of throughput, delay, error rates and energy consumption.
However, due to the different nature of the underwater environment, there are several
drawbacks with respect to the suitability of the existing terrestrial MAC solutions for the
underwater environment. In fact, channel access control in UWSNs poses additional
challenges due to the aforementioned peculiarities of underwater channels (Molins and
Stojanovic, 2006).

In the MAC Layer we have different types of protocols and ones of the latest examples
presented in each one of the categories. The first ones are the non handshaking protocol
such as Aloha-like, an on the opposite side we have the handshaking protocols like DACAP
(Distance Aware Collision Avoidance Protocol) a collision avoidance protocol based on
virtual Medium Access Control carrier sensing. A number of adaptations have been
proposed to adopt MACA (Multiple Access with Collision Avoidanc) (Karn, 1990),
MACAW (Media Access Protocol for Wireless LAN's) (Bharghavan et al, 1994), and FAMA
(Floor Acquisition Multiple Access) (Fullmer & Luna-Acebes, 1995) for underwater
networks in (Molins & Stojanovic, 2006). But also new protocols such as T-Lohi (between
handshake and non-handshake) are becoming more popular as they have also a great
efficiency in terms of the battery use.

4.1 Aloha-like
In Aloha like mode the source node sends its data frames as soon as it receives a packet from
the upper-layer protocol. It does not check the medium to see if it is busy and so it does not
perform any back-off, it. The node that receives the data will answer with and acknowledge
data frame, if there was no problem at the reception such as a collision or packet lost during
the transmission (i.e. when there is overlapping of the receiving periods of two or more
frames at the destination location, or the receiver was transmitting).

If the source, does not receive an ACK because either the frame was not correctly delivered
or the ACK was lost, the sender will timeout, wait a random period (back-off) and
retransmit the frame. This protocol follows the stop-and-wait paradigm. That is, the source
must receive an acknowledgement for each data frame before the next frame can be sent. In
addition, after a successful frame transmission, the sender will perform a back off, even if it
has additional frames to send from the same packet or from a new packet.

There different version called Aloha-based protocols, a couple of proposals modification for
underwater networks of the protocol can be found in (Chirdchoo, 2007) that presents Aloha
with Collision Avoidance (Aloha-CA) and Aloha with Advance Notification (Aloha AN).

Aloha-CA: Pays close attention to every packet it overhears, picking the information of
who are the sender and receiver. With this information it can easily calculate the busy
duration due to the packet at every one of the nodes. Each node will store in a database
table the information of the monitored packets with the busy durations of every node of
the neighbourhood.

Aloha-AN: This protocol has all the features of Aloha-CA and adds the sending of a
small advanced notification packet with the necessary information to let the other nodes
build the databases tables. The sender will wait after this packet for a lag time before
sending the actual data packet. Whenever a node has a packet to transmit, it will check
the database table to ensure that the packet does not result in a collision at any other
reachable nodes.

4.2 DACAP
DACAP is a handshaking protocol Distance Aware Collision Avoidance Protocol (Peleato &
Stojanovic 2007) for Ad Hoc Underwater Acoustic Sensor Networks. This protocol is a non-
synchronized data access that uses control messages to decide when to start the
communication Request-To-Send (RTS) / Clear-To-Send (CTS) handshake. If a collision
happens during the handshake, the transmitter will not receive a CTS that enables him to
send the data packet.

The protocol is describe in Figure 6 and Figure 7 and explained in the following steps:

RTS reception: After this event, the node sends a CTS to the sender, and waits for a data
packet. If another RTS is received the node sends a warning short packet to give and
advice to the sender of the last one that the medium is in use.

Fig. 6. Transmission in DACAP

CTS reception: When the transmitter receives this message it waits for those nodes
whose transmissions are still happening to avoid any collisions. If it happens to receive
another CTS or a warning packet reception the current data packet to send will be
deferred for a random back-off time, if the waiting time expires the transmission
proceeds normally.


Fig. 7. Reception in DACAP

DACAP is a collision avoidance protocol with an easy scalable adaptation to big networks
involving more nodes and a greater area. The protocol is aware of power consumption by
avoiding collisions at the same time that maximizes the throughput. It minimizes the
handshake time using the tolerance to interference of the receiver node, when this one is
close to the limit of the range reception range. It works with a half-duplex communication
link, the nodes do not need to be synchronized and it supports mobile nodes.

The throughput with this protocol is several times higher than the one achieved with Slotted
FAMA, while offering similar protection to collisions, i.e. savings in energy. Although CS-
ALOHA offers higher throughput in most cases, it wastes too much power on collisions.

4.3 T-Lohi
Tone-Lohi is a contention-based Mac protocol that uses short packet as wake up tone to
reserve the medium. It is a full distributed reservation process and one of its main features is
the power consumption, the nodes will be in an idle mode with low energy requirement
until it receives the wake up tone.

Fig. 8. T-Lohi Protocol Scheme

The main goals of T-Lohi are to make an efficient use of channel utilization, achieve a stable
throughput, and save as much energy as possible without having an impact in the
performance. This energy conservation is approached in two ways: (1) the reservation to
prevent data packet collisions or at least reduce them, (2) and the usage of wake-up tones for
the receivers to keep them in low power while in listening mode.

5. Network Layer
This layer is mainly responsible of routing packets to the proper destinations. So, a routing
protocol is required when a packet must go through several hops to reach its destination. It
is responsible for finding a route for the packet and making sure it is forwarded through the
appropriate path. The way paths are selected for every source destination pair will have a
direct impact on the overall network performance.

Most of the routing proposals for UWSN are based on the ones developed for terrestrial ad-
hoc and wireless sensor networks. Some of the protocols designed exclusively for
underwater wireless networks are:

5.1 DBR
Depth-Based Routing (Xie et al, 2008). It can handle network dynamics efficiently without
the assistance of a localization service, it needs only local depth information. It is a greedy
algorithm that tries to deliver a packet from a source node to sinks.

Fig. 9. Multiple sink underwater sensor network architecture

The source nodes send the data packets seeking for the sinks, in this process as the packet
hops from one node to another the depth decreases as it gets closer to the final sink receiver.
Finally the packet can achieve to reach to the surface. The decision that is taken in each one
of the nodes during the transmission is based on its own depth and the depth of the
previous sender.

When a node receives a packet it extracts the information of the depth of the previous node
and compares against its own depth. After comparing the node will have two behaviours:

Fig. 7. Reception in DACAP

DACAP is a collision avoidance protocol with an easy scalable adaptation to big networks
involving more nodes and a greater area. The protocol is aware of power consumption by
avoiding collisions at the same time that maximizes the throughput. It minimizes the
handshake time using the tolerance to interference of the receiver node, when this one is
close to the limit of the range reception range. It works with a half-duplex communication
link, the nodes do not need to be synchronized and it supports mobile nodes.

The throughput with this protocol is several times higher than the one achieved with Slotted
FAMA, while offering similar protection to collisions, i.e. savings in energy. Although CS-
ALOHA offers higher throughput in most cases, it wastes too much power on collisions.

4.3 T-Lohi
Tone-Lohi is a contention-based Mac protocol that uses short packet as wake up tone to
reserve the medium. It is a full distributed reservation process and one of its main features is
the power consumption, the nodes will be in an idle mode with low energy requirement
until it receives the wake up tone.

Fig. 8. T-Lohi Protocol Scheme

The main goals of T-Lohi are to make an efficient use of channel utilization, achieve a stable
throughput, and save as much energy as possible without having an impact in the
performance. This energy conservation is approached in two ways: (1) the reservation to
prevent data packet collisions or at least reduce them, (2) and the usage of wake-up tones for
the receivers to keep them in low power while in listening mode.

5. Network Layer
This layer is mainly responsible of routing packets to the proper destinations. So, a routing
protocol is required when a packet must go through several hops to reach its destination. It
is responsible for finding a route for the packet and making sure it is forwarded through the
appropriate path. The way paths are selected for every source destination pair will have a
direct impact on the overall network performance.

Most of the routing proposals for UWSN are based on the ones developed for terrestrial ad-
hoc and wireless sensor networks. Some of the protocols designed exclusively for
underwater wireless networks are:

5.1 DBR
Depth-Based Routing (Xie et al, 2008). It can handle network dynamics efficiently without
the assistance of a localization service, it needs only local depth information. It is a greedy
algorithm that tries to deliver a packet from a source node to sinks.

Fig. 9. Multiple sink underwater sensor network architecture

The source nodes send the data packets seeking for the sinks, in this process as the packet
hops from one node to another the depth decreases as it gets closer to the final sink receiver.
Finally the packet can achieve to reach to the surface. The decision that is taken in each one
of the nodes during the transmission is based on its own depth and the depth of the
previous sender.

When a node receives a packet it extracts the information of the depth of the previous node
and compares against its own depth. After comparing the node will have two behaviours:

(1) The node is closer to the surface d
c
< d
p
so it will forward the packet. (2) If the current
node depth is greater d
c
> d
p,
it will discard the packet as it comes from a node with a better
position.

Probably specially at the beginning of the transmission in the first hops, a lot of receivers of
the packet will decide to forward it. To avoid the collisions that these retransmissions would
bring and the high power requirements needed in the network, the number of forwarded
messages must be controlled.

Also it can happen that as it is been using a multiple omnidirectional path algorithm to
route the packets, that a node receive several times the same packet and in the same way
forward it the same number of times. In order to save energy a node will know which
packets have already sent so as not to send a packet more than one time.

Fig. 10. DBR Packet Format

The Packet Format will be divided in:

Sender ID: is the identifier of the source node.
Packet Sequence Number: is a unique sequence number assigned by the source node to
the packet. Together with Sender ID, Packet Sequence Number is used to differentiate
packets in later data forwarding.
Depth: is the depth information of the recent forwarder, which is updated hop-by-hop
when the packet is forwarded.

As mention before there is need to reduce power consumption forwarding only the
necessary packets. To achieve this protocol uses the Redundant Packet Suppression, which
consist of two features for avoiding redundant packet. One is that multiple paths are
naturally used to forward packets. The other is that a node may send a packet many times.
Although multiple paths in DBR cannot be completely eliminated, a priority queue is
created to reduce the number of forwarding nodes, and thus control the number of
forwarding paths. To solve the second problem, a packet sent buffer is used in DBR to
ensures that a node forwards the same packet only once in a certain time interval.

5.2 VBF
Vector-Based Forwarding protocol (Xie et al, 2006) is an algorithm that allows the nodes to
weigh the benefit to forward packets and reduce energy consumption by discarding the low
benefit packets. One of the main factors in underwater wireless networks is to safe power so
as not to let nodes run out of batteries, not being able to recharge them during long period
of times. This protocol tries to focus its features in this direction. To aim this target each

packet will include the information of the location of the sender the final receiver and the
one of next hop of the packet.

To be able to run this protocol it is assumed that every node have the capacity of measuring
the distance and angle of arrival (AOA) of the signal. The route of the packet is compute in
the sender and included in the packet. When a node receives a packet it calculates its
relative position towards the target. This works recursively in all the nodes during the
transmission. If the node knows that is close enough to the routing vector (it will be under
the threshold value established for this purpose) it will include its position and forward the
packet, in other case it drops the packet. In this way, all the packet forwarders in the sensor
network form a routing pipe: the sensor nodes in this pipe are eligible for packet
forwarding, and those which are out do not forward.

Fig. 11. High Level view of VBR for UWSN.

The figure represents the nodes that are within the routing pipe that forward the packet,
w the threshold used to measure the width of the pipe. And the nodes that are out of this
path discard the packets. This protocol is scalable to the size of the network. This kind of
forwarding path (specified by the routing vector) involving nodes for packet routing has as
result the energy of the network.

5.3 FBR
Focus Beam Routing, a routing protocol (Jornet & Stojanovic 2008) based on location is
presented as a way to find the path between two nodes in a random deployed network. The
figure 8 shows a simple two-dimensional network to explain the protocol, although it works
in the same way in three-dimensional scenarios.


(1) The node is closer to the surface d
c
< d
p
so it will forward the packet. (2) If the current
node depth is greater d
c
> d
p,
it will discard the packet as it comes from a node with a better
position.

Probably specially at the beginning of the transmission in the first hops, a lot of receivers of
the packet will decide to forward it. To avoid the collisions that these retransmissions would
bring and the high power requirements needed in the network, the number of forwarded
messages must be controlled.

Also it can happen that as it is been using a multiple omnidirectional path algorithm to
route the packets, that a node receive several times the same packet and in the same way
forward it the same number of times. In order to save energy a node will know which
packets have already sent so as not to send a packet more than one time.

Fig. 10. DBR Packet Format

The Packet Format will be divided in:

Sender ID: is the identifier of the source node.
Packet Sequence Number: is a unique sequence number assigned by the source node to
the packet. Together with Sender ID, Packet Sequence Number is used to differentiate
packets in later data forwarding.
Depth: is the depth information of the recent forwarder, which is updated hop-by-hop
when the packet is forwarded.

As mention before there is need to reduce power consumption forwarding only the
necessary packets. To achieve this protocol uses the Redundant Packet Suppression, which
consist of two features for avoiding redundant packet. One is that multiple paths are
naturally used to forward packets. The other is that a node may send a packet many times.
Although multiple paths in DBR cannot be completely eliminated, a priority queue is
created to reduce the number of forwarding nodes, and thus control the number of
forwarding paths. To solve the second problem, a packet sent buffer is used in DBR to
ensures that a node forwards the same packet only once in a certain time interval.

5.2 VBF
Vector-Based Forwarding protocol (Xie et al, 2006) is an algorithm that allows the nodes to
weigh the benefit to forward packets and reduce energy consumption by discarding the low
benefit packets. One of the main factors in underwater wireless networks is to safe power so
as not to let nodes run out of batteries, not being able to recharge them during long period
of times. This protocol tries to focus its features in this direction. To aim this target each

packet will include the information of the location of the sender the final receiver and the
one of next hop of the packet.

To be able to run this protocol it is assumed that every node have the capacity of measuring
the distance and angle of arrival (AOA) of the signal. The route of the packet is compute in
the sender and included in the packet. When a node receives a packet it calculates its
relative position towards the target. This works recursively in all the nodes during the
transmission. If the node knows that is close enough to the routing vector (it will be under
the threshold value established for this purpose) it will include its position and forward the
packet, in other case it drops the packet. In this way, all the packet forwarders in the sensor
network form a routing pipe: the sensor nodes in this pipe are eligible for packet
forwarding, and those which are out do not forward.

Fig. 11. High Level view of VBR for UWSN.

The figure represents the nodes that are within the routing pipe that forward the packet,
w the threshold used to measure the width of the pipe. And the nodes that are out of this
path discard the packets. This protocol is scalable to the size of the network. This kind of
forwarding path (specified by the routing vector) involving nodes for packet routing has as
result the energy of the network.

5.3 FBR
Focus Beam Routing, a routing protocol (Jornet & Stojanovic 2008) based on location is
presented as a way to find the path between two nodes in a random deployed network. The
figure 8 shows a simple two-dimensional network to explain the protocol, although it works
in the same way in three-dimensional scenarios.


Fig. 12. Nodes within the transmitters cone are candidate relays.

Assuming a communication between node A and node B, node A will send and Request to
Send (RTS) multicast message to all the reachable neighbours. This packet will include the
information of the source (node A) and final receiver (node B). As the protocol works with
power levels, the first try is done at the lowest level and it increases if there is a need
because it receives no answers within a wait time established for each power level.

This request is a short control packet that contains the location of the source node (A) and of
the final destination (B). Note that this is in fact a multicast request. The initial transaction is
performed at the lowest power level and the power is increased only if necessary. Power
control is performed as an integral part of routing and medium access control.

Each power level will have a radius and though it will reach to a certain number of nodes.
This will be the nodes that receive the RTS and its information that will be used to calculate
the relative position to the AB line. This is done to know if the node is a candidate to be a
relay node. Candidate nodes are inside a cone of /2 from the line AB. Every candidate
node will answer to with a Clear to Send (CTS) to the transmitter; the nodes out of the
candidate zone will stay in silence.

After sending the RTS the transmitter will wait to receive CTS messages from other nodes,
three possible things can occur: (1) The transmitter receive no answers, the RTS has not
reach any neighbour, therefore the transmitter increases the power level an tries again as it
is shown in the example. (2) The transmitter receives one CTS, the sender of this message is
selected as a relay for the next hop, sending him the DATA message. (3) The transmitter
receives more than one CTS message, looking at the location information of the candidates
included in CTS message the node that is closer to the final destination is selected as relay
receiving the DATA message. After sending data the transmitter will wait for an
acknowledgement message. This process will continue until we reach the final destination.

Packet collisions can happen but always will involve short packet as the link is safe for data
packets which have no risk of collisions. Although the chances of collision are small, if the

source node detects a collision, it will detect signal but it will not decode the information of
the data, it will resend the RTS once again, without increasing the power level.

6. Applications
As established in the introduction, underwater acoustic wireless sensor networks can be
used in a wide range of different applications, as it is done by radio frequency air networks.
Ones of the main places we can use UWSN are:

Environmental Monitoring. Pollution is nowadays one of the greatest problems, oil spills
from ships or broken tubes can make a lot of harm to the marine biological activity, the
industry and tourist places. Monitoring ecosystems can help understanding and
predicting the human and climate or weather effect in underwater environment.
Prevention of natural disaster. By measuring the seismic activity from different remote
location the sensors could alert to the coast places by detecting tsunami or submarine
earthquakes alarms.
Underwater Navigation. The sensor can be placed to make routing, identifying hazards
on the seafloor, rocks or shoals in shallow water,
Assisted Navigation. Sensors can be used to identify hazards on the seabed, locate
dangerous rocks or shoals in shallow waters, mooring positions and drawing the
bathymetry profile of the area.
Underwater Discovery. Underwater wireless sensor networks can be used to find
oilfields or reservoirs, locate routes for placing connections for intercontinental
submarine cables. Also they could seek for shipwrecks or archaeology or lost sink cities.
Underwater Autonomous Vehicles (UAVs). Distributed sensor in movement can help
monitoring area for surveillance, recognition and intrusion detection.

7. Future Trends
A lot of advantages can be achieved by using underwater sensor networks, but a lot of
research must be done in the next years. The developing of this technology will have a great
impact in the industry.

It is necessary to improve the physical layer performance in terms of efficiency, building low
power acoustic modem that are able to make a make best use of the bandwidth, reducing
the error rate with forward error correcting coders.

Although there many proposal in the MAC layer, it seems that the collision avoidance
protocols in the top for been chosen for underwater networks. In any case this decision can
vary according to the application or topology type. These protocols should be also aware of
power consumption, making one of their main objectives.

Currently there is a lot of works related to MAC layer proposals since this is one of the more
sensible parts of the UWSN architecture. It seems that distributed CDMA-based schemes are
the candidates for underwater environments, but it depends of many factors such as the

Fig. 12. Nodes within the transmitters cone are candidate relays.

Assuming a communication between node A and node B, node A will send and Request to
Send (RTS) multicast message to all the reachable neighbours. This packet will include the
information of the source (node A) and final receiver (node B). As the protocol works with
power levels, the first try is done at the lowest level and it increases if there is a need
because it receives no answers within a wait time established for each power level.

This request is a short control packet that contains the location of the source node (A) and of
the final destination (B). Note that this is in fact a multicast request. The initial transaction is
performed at the lowest power level and the power is increased only if necessary. Power
control is performed as an integral part of routing and medium access control.

Each power level will have a radius and though it will reach to a certain number of nodes.
This will be the nodes that receive the RTS and its information that will be used to calculate
the relative position to the AB line. This is done to know if the node is a candidate to be a
relay node. Candidate nodes are inside a cone of /2 from the line AB. Every candidate
node will answer to with a Clear to Send (CTS) to the transmitter; the nodes out of the
candidate zone will stay in silence.

After sending the RTS the transmitter will wait to receive CTS messages from other nodes,
three possible things can occur: (1) The transmitter receive no answers, the RTS has not
reach any neighbour, therefore the transmitter increases the power level an tries again as it
is shown in the example. (2) The transmitter receives one CTS, the sender of this message is
selected as a relay for the next hop, sending him the DATA message. (3) The transmitter
receives more than one CTS message, looking at the location information of the candidates
included in CTS message the node that is closer to the final destination is selected as relay
receiving the DATA message. After sending data the transmitter will wait for an
acknowledgement message. This process will continue until we reach the final destination.

Packet collisions can happen but always will involve short packet as the link is safe for data
packets which have no risk of collisions. Although the chances of collision are small, if the

source node detects a collision, it will detect signal but it will not decode the information of
the data, it will resend the RTS once again, without increasing the power level.

6. Applications
As established in the introduction, underwater acoustic wireless sensor networks can be
used in a wide range of different applications, as it is done by radio frequency air networks.
Ones of the main places we can use UWSN are:

Environmental Monitoring. Pollution is nowadays one of the greatest problems, oil spills
from ships or broken tubes can make a lot of harm to the marine biological activity, the
industry and tourist places. Monitoring ecosystems can help understanding and
predicting the human and climate or weather effect in underwater environment.
Prevention of natural disaster. By measuring the seismic activity from different remote
location the sensors could alert to the coast places by detecting tsunami or submarine
earthquakes alarms.
Underwater Navigation. The sensor can be placed to make routing, identifying hazards
on the seafloor, rocks or shoals in shallow water,
Assisted Navigation. Sensors can be used to identify hazards on the seabed, locate
dangerous rocks or shoals in shallow waters, mooring positions and drawing the
bathymetry profile of the area.
Underwater Discovery. Underwater wireless sensor networks can be used to find
oilfields or reservoirs, locate routes for placing connections for intercontinental
submarine cables. Also they could seek for shipwrecks or archaeology or lost sink cities.
Underwater Autonomous Vehicles (UAVs). Distributed sensor in movement can help
monitoring area for surveillance, recognition and intrusion detection.

7. Future Trends
A lot of advantages can be achieved by using underwater sensor networks, but a lot of
research must be done in the next years. The developing of this technology will have a great
impact in the industry.

It is necessary to improve the physical layer performance in terms of efficiency, building low
power acoustic modem that are able to make a make best use of the bandwidth, reducing
the error rate with forward error correcting coders.

Although there many proposal in the MAC layer, it seems that the collision avoidance
protocols in the top for been chosen for underwater networks. In any case this decision can
vary according to the application or topology type. These protocols should be also aware of
power consumption, making one of their main objectives.

Currently there is a lot of works related to MAC layer proposals since this is one of the more
sensible parts of the UWSN architecture. It seems that distributed CDMA-based schemes are
the candidates for underwater environments, but it depends of many factors such as the

application and network topology. Also, MAC protocols should be designed taking energy
consumption into account as a main design parameter.

According to the routing layer protocols, there is a need to be able to adapt to the changing
conditions, to include mobility patterns and also be capable of saving energy. Most of the
routing protocols need to know the location of all the nodes, geographically-based
algorithms may be appropriate for underwater networks. They have to include methods to
avoid errors, deal with shadow zones or disconnections or failures and mobility. Cross layer
communication between the layers should be required to share the information and adjust
the parameters depending on the environment condition variation.

8. Conclusions
Underwater Acoustic Wireless Sensor Networks is still growing and following the path of
Radio Frequency in Terrestrial Networks, although having a very different environment
with a lot of challenges to achieve in changing conditions. There are many potential research
fields in which it can be applied, and needs to solve some open issues so as to be able to
provide a reliable and efficient way to communicate in the network.

The development of new modem and the incorporation of companies and research to the
UWSN technology will come with new commercial products and solutions to take
advantage of the possibilities that underwater bring us, building and industry around
submarine technologies.

9. References
Akyildiz, F., Su, W., Sankarasubramaniam, Y., and Cayirci, E. (2002). Wireless sensor
networks: A survey. Computer Networks, 38(4):393422.
Akyildiz, F., Pompili, D., and Melodia, T. (2006). State of the Art in Protocol Research for
Underwater Acoustic Sensor Network. Proceedings of the ACM International
Workshop on UnderWater Networks (WUWNet), Los Angeles CA, September 25th, pp.
7-16.
Berkhovskikh. L., Laysanov, Y. (1982). Fundamentals of Ocean Acoustics, N.Y: Springer
Bharghavan, V., Demers, A., Shenker, S., Zhang, L. (1994). MACAW: A Media Access
Protocol for Wireless LAN's. Proceedings of the ACM SIGCOMM Conference, London,
UK, August 31- September 2, pp. 212-225.
Chirdchoo, N., Soh, W., Chua, K. C. (2007). The IEEE Conference on Computer Communications
(infocom) , Anchorage, Alaska, USA, May
Coates, R. (1989), Underwater Acoustic Systems, Wiley, New York.
Colladon, J. D. (1893). Souvenirs et Memoires. Geneva, CH: Albert-Schuchardt.
Cui, J-H., Kong, J., Gerla, M. and Zhou, S. (2006). Challenges: Building scalable mobile
underwater wireless sensor networks for aquatic applications. IEEE Network 3:12-
18.
Curtin, T. B., Bellingham, J. G., Catipovic, J., and Webb, D. (1993). Autonomous
oceanographic sampling networks. Oceanography 6:8694.

Freitag, L., Stojanovic, M., Kifoyle, D., and Preisig, J. (2004). High-Rate Phase-Coherent
Acoustic communication: A Review of a Decade of Research and a Perspective on
Future Challenges. Proceedings of the 7th European Conf. on Underwater Acoustics,
Delft, HL, 5-8 July.
Fullmer, C.L., Garca-Luna-Acebes, J.J. (1995). Floor acquisition multiple access (FAMA) for
packet-radio networks. Computer Communication Review 25(4):262-273.
Jornet, J.M., Stojanovic, M. (2008). Focused beam routing protocol for underwater acoustic
networks. Proceedings of the third ACM international workshop on Underwater
Networks, San Francisco, California, USA, September 15.
Karn, P. (1990). MACA - A new channel access method for packet radio. Proceedings of the
ARRL 9th Computer Networking Conference, London, Ontario (CA), September 22, pp.
134-140.
Molins, M., Stojanovic, M. (2006). Slotted FAMA: A MAC Protocol for Underwater Acoustic
Networks. Proceedings of the IEEE OCEANS, Singapore, 16-19 May, pp. 1-7.
Llor, J., Malumbres M.P. (2009). Analyzing the Behaviour of Acoustic Link Models in
Underwater Wireless Sensor Networks. PM2HW2N
Partan, J., Kurose, J., Neil Levine, B. (2006). A Survey of Practical Issues in Underwater
Networks. Proceedings of the ACM International Workshop on UnderWater
Networks (WUWNet), Los Angeles CA, September 25th, pp. 17-24.
Peleato, B., Stojanovic, M. (2007). Distance Aware Collision Avoidance Protocol for Ad Hoc
Underwater Acoustic Sensor Networks, IEEE Comm. Letters
Porter, M., Liu, Yong-Chun Bellhop (2010). [Online]. Code and manual available at:
http://oalib.hlsresearch.com/Rays/index.html
Rosenbush, S., Crockett, R. O., Yang, C. (2004). Sin cables, sin normas y a bajo precio.
Dinero: Inteligencia empresarial 932:70-73.
Sozer, Ethem M., Stojanovic, Milica and Proakis, John g. (1999), Design and Simulation of an
Underwater Acoustic Local Area Network, Northeastern University,
Communications and Digital Signal Processing Center, 409 Dana Research Building,
Boston, Massachusetts.
Urick, Robert J. (1983). Principles of Underwater Sound. NewYork, USA: McGraw-Hill (3rd
edition).
van der Werff, T.J. (2003). 10 Emerging Technologies That Will Change the World.
Technology Review (MIT). [Online].
http://www.technologyreview.com/read_article.aspx?id=13060&ch=infotech
Xie, G, Gibson, J, Diaz-Gonzalez, L, (2006). Incorporating Realistic Acoustic Propagation
Models in Simulation of Underwater Acoustic Networks: A Statistical Approach,
Proceedings of MTS/IEEE Oceans Conference, Boston, September 2006.
Xie, P., Cui, J.H., Lao L. (2006). VBF: Vector-Based Forwarding Protocol for Underwater
Sensor Networks. Proceedings of IFIP Networking, May 2006.
Yan, H., Shi, Z., Cui, J.H., (2008). DBR: Depth-Based Routing for Underwater Sensor
Networks. Proceedings of Networking'08, Singapore. , May 5-9


application and network topology. Also, MAC protocols should be designed taking energy
consumption into account as a main design parameter.

According to the routing layer protocols, there is a need to be able to adapt to the changing
conditions, to include mobility patterns and also be capable of saving energy. Most of the
routing protocols need to know the location of all the nodes, geographically-based
algorithms may be appropriate for underwater networks. They have to include methods to
avoid errors, deal with shadow zones or disconnections or failures and mobility. Cross layer
communication between the layers should be required to share the information and adjust
the parameters depending on the environment condition variation.

8. Conclusions
Underwater Acoustic Wireless Sensor Networks is still growing and following the path of
Radio Frequency in Terrestrial Networks, although having a very different environment
with a lot of challenges to achieve in changing conditions. There are many potential research
fields in which it can be applied, and needs to solve some open issues so as to be able to
provide a reliable and efficient way to communicate in the network.

The development of new modem and the incorporation of companies and research to the
UWSN technology will come with new commercial products and solutions to take
advantage of the possibilities that underwater bring us, building and industry around
submarine technologies.

9. References
Akyildiz, F., Su, W., Sankarasubramaniam, Y., and Cayirci, E. (2002). Wireless sensor
networks: A survey. Computer Networks, 38(4):393422.
Akyildiz, F., Pompili, D., and Melodia, T. (2006). State of the Art in Protocol Research for
Underwater Acoustic Sensor Network. Proceedings of the ACM International
Workshop on UnderWater Networks (WUWNet), Los Angeles CA, September 25th, pp.
7-16.
Berkhovskikh. L., Laysanov, Y. (1982). Fundamentals of Ocean Acoustics, N.Y: Springer
Bharghavan, V., Demers, A., Shenker, S., Zhang, L. (1994). MACAW: A Media Access
Protocol for Wireless LAN's. Proceedings of the ACM SIGCOMM Conference, London,
UK, August 31- September 2, pp. 212-225.
Chirdchoo, N., Soh, W., Chua, K. C. (2007). The IEEE Conference on Computer Communications
(infocom) , Anchorage, Alaska, USA, May
Coates, R. (1989), Underwater Acoustic Systems, Wiley, New York.
Colladon, J. D. (1893). Souvenirs et Memoires. Geneva, CH: Albert-Schuchardt.
Cui, J-H., Kong, J., Gerla, M. and Zhou, S. (2006). Challenges: Building scalable mobile
underwater wireless sensor networks for aquatic applications. IEEE Network 3:12-
18.
Curtin, T. B., Bellingham, J. G., Catipovic, J., and Webb, D. (1993). Autonomous
oceanographic sampling networks. Oceanography 6:8694.

Freitag, L., Stojanovic, M., Kifoyle, D., and Preisig, J. (2004). High-Rate Phase-Coherent
Acoustic communication: A Review of a Decade of Research and a Perspective on
Future Challenges. Proceedings of the 7th European Conf. on Underwater Acoustics,
Delft, HL, 5-8 July.
Fullmer, C.L., Garca-Luna-Acebes, J.J. (1995). Floor acquisition multiple access (FAMA) for
packet-radio networks. Computer Communication Review 25(4):262-273.
Jornet, J.M., Stojanovic, M. (2008). Focused beam routing protocol for underwater acoustic
networks. Proceedings of the third ACM international workshop on Underwater
Networks, San Francisco, California, USA, September 15.
Karn, P. (1990). MACA - A new channel access method for packet radio. Proceedings of the
ARRL 9th Computer Networking Conference, London, Ontario (CA), September 22, pp.
134-140.
Molins, M., Stojanovic, M. (2006). Slotted FAMA: A MAC Protocol for Underwater Acoustic
Networks. Proceedings of the IEEE OCEANS, Singapore, 16-19 May, pp. 1-7.
Llor, J., Malumbres M.P. (2009). Analyzing the Behaviour of Acoustic Link Models in
Underwater Wireless Sensor Networks. PM2HW2N
Partan, J., Kurose, J., Neil Levine, B. (2006). A Survey of Practical Issues in Underwater
Networks. Proceedings of the ACM International Workshop on UnderWater
Networks (WUWNet), Los Angeles CA, September 25th, pp. 17-24.
Peleato, B., Stojanovic, M. (2007). Distance Aware Collision Avoidance Protocol for Ad Hoc
Underwater Acoustic Sensor Networks, IEEE Comm. Letters
Porter, M., Liu, Yong-Chun Bellhop (2010). [Online]. Code and manual available at:
http://oalib.hlsresearch.com/Rays/index.html
Rosenbush, S., Crockett, R. O., Yang, C. (2004). Sin cables, sin normas y a bajo precio.
Dinero: Inteligencia empresarial 932:70-73.
Sozer, Ethem M., Stojanovic, Milica and Proakis, John g. (1999), Design and Simulation of an
Underwater Acoustic Local Area Network, Northeastern University,
Communications and Digital Signal Processing Center, 409 Dana Research Building,
Boston, Massachusetts.
Urick, Robert J. (1983). Principles of Underwater Sound. NewYork, USA: McGraw-Hill (3rd
edition).
van der Werff, T.J. (2003). 10 Emerging Technologies That Will Change the World.
Technology Review (MIT). [Online].
http://www.technologyreview.com/read_article.aspx?id=13060&ch=infotech
Xie, G, Gibson, J, Diaz-Gonzalez, L, (2006). Incorporating Realistic Acoustic Propagation
Models in Simulation of Underwater Acoustic Networks: A Statistical Approach,
Proceedings of MTS/IEEE Oceans Conference, Boston, September 2006.
Xie, P., Cui, J.H., Lao L. (2006). VBF: Vector-Based Forwarding Protocol for Underwater
Sensor Networks. Proceedings of IFIP Networking, May 2006.
Yan, H., Shi, Z., Cui, J.H., (2008). DBR: Depth-Based Routing for Underwater Sensor
Networks. Proceedings of Networking'08, Singapore. , May 5-9

Prospects and Problems of Optical Diffuse Wireless
Communication for Underwater Wireless Sensor Networks (UWSNs) 205
Prospects and Problems of Optical Diffuse Wireless Communication for
Underwater Wireless Sensor Networks (UWSNs)
Davide Anguita, Davide Brizzolara and Giancarlo Parodi
0
Prospects and Problems of Optical
Diffuse Wireless Communication for Underwater
Wireless Sensor Networks (UWSNs)
Davide Anguita, Davide Brizzolara, Giancarlo Parodi
University of Genoa
Italy
1. Introduction
The aim of this paper is to highlight the prospects and problems of optical wireless com-
munication for applications in the eld of Underwater Wireless Sensor Networks (UWSNs).
The necessity of wireless underwater connections, especially in UWSN, has dramatically in-
creased in the last few years for a wide range of applications, from environmental monitoring
to surveillance. The problem of wireless underwater communication is a challenging eld
of investigation since common terrestrial devices, equipped with radio wireless links, cannot
be employed underwater and current underwater available technologies, mainly based on
acoustical solutions, still pose problems due to low propagation speed of sound in water and
low data-rate.
In the last years, the interest towards optical wireless communication has increased both for
terrestrial (Akella et al., 2005) than for underwater applications (Doniec et al., 2010) since
it allows to overcome some of the problems of acoustical communication (Akyildiz et al.,
2005). In fact low data rate and low propagation speed require specic solutions and make it
difcult the integration of underwater technologies with current available terrestrial modules.
Transferring the paradigmof Smart Dust (V. Hsu &Pister, 1998) to UWSNs, largely unaffected
by WSN revolution, is a challenge that can open great opportunities since current underwater
solutions are still of large dimension, expensive and power-consuming.
Fig. 1. Wireless Node Structure equipped with FPGA
12
This paper highlight the relatively less explored and innovative possibility of optical under-
water communication as an interesting and feasible alternative solution for underwater net-
working targeting small, low-cost and low-power devices. In particular it focuses both the
problems related to specic characteristics of the underwater channel targeting the interface
with current terrestrial technologies available for Wireless Sensor Network (WSN) such as the
one developed at DIBE WiseLab (Akyildiz et al., 2005) depicted in Fig.1 .
The goal of our work is to build a prototype of an UWSN based on optical communication
among nodes. The optical PHY and MAC Layers have been developed considering the char-
acteristics of IEEE 802.11 Infrared (IR) PHY layer and the compatibility with the current IEEE
802.15.4 protocol for terrestrial Wireless Sensor Networks (WSNs). This can be considered as
a rst step which, of course, should be followed by modications and adaptation of the upper
layers.
The network architecture can be organized as a three or bi-dimensional underwater net-
work in which each node oats at different depths trying to maintain a xed position with
a good approximation, for instance by an anchorage to the sea bed (as detailed in Chapter
11). The nodes should be designed of slight dimensions (between 15 and 20 cm), they should
be densely deployed, maintaining a distance between 10 and 30 meters from each other. As
detailed in the next paragraph, each node should be able to communicate, by using only a
wireless optical communication link, to the neighbors and, by using a multi hop path, to a
base station placed on the water surface.
The interesting aspects which are reported in this work are related to:
the use of optical communication (the adaptation of current available protocols, the
choice of modulation, the adaptation to water channel variability, etc ...);
the interface with current technology for WSN;
the design and implementation of circuits for diffuse generation and reception of light
impulses targeting low-cost and low-power components;
Fig. 2. Underwater Attenuation of Light from experimental data
Acoustic Electromagnetic Optical
Nominal speed (m/s) 1, 500 light speed light speed
Power Loss 0.1dB/m/Hz 28dB/1km/100MHz turbidity
Bandwidth kHz MHz 10 150MHz
Frequency band KHz MHz 10
14
10
15
Hz
Antenna size 0.1m 0.5m 0.1m
Effective range km 10m 10 100m
Table 1. Comparison of Different Technologies for Underwater Wireless Communication
2. Underwater Wireless Communication: physical aspects
Due to the impossibility of using Radio Frequencies (RF), traditionally wireless underwater
communication employs acoustic waves because sound propagates well in water and its range
can be very long (km). However, it has several disadvantages such as narrow bandwidth
and latency in communication due to the slow speed of acoustic wave in water. For instance,
at ranges of less than 100 m the data transmission rates of these systems in shallow littoral
waters are 10 kb/s.
Experimental tests have shown that an alternative feasible solution is optical communica-
tion especially in blue/green light wavelengths, even if limited to short distances (up to 100
m)(Lanbo et al., 2008). Compared to acoustic communication it offers a practical choice for
high-bandwidth communication and it propagates faster in the water (2.255 x 10
8
).
Nevertheless it is affected by different factors to take into account for an efcient design. The
attenuation of a light beam between two points can be described as in 1 where d
1
and d
2
are
the positions of the points.
A = e
k(d
1
d
2
)
d
1
d
2
2
(1)
In the rst term, k is dened as k = a() + b() and it is dependent by the wavelength: a
is the term related to the absorption of water while b models the scattering which depends
both on light wavelength and turbidity. The second term, instead, models the quadratic at-
tenuation. A comparison between different wireless underwater technologies for underwater
communication is illustrated in Table 1.
3. Wireless Underwater Communication Systems: analysis of the state of art
In this chapter will be briey investigated the state of art related to underwater wireless com-
munication.
Currently the use of wireless communication is very common in a wide range of terrestrial
devices. In particular, one of the most innovative application is related to Wireless Sensor Net-
works (WSNs), as well detailed in (Akyldiz et al., 2001) (Puccinelli & Haenggi, 2005), where a
large number of small nodes communicate and work by using an optical wireless link.
3.1 Underwater Acoustical Communication
As discussed in the introduction, although there are many recently developed solutions for
WSNs, the unique characteristics of the underwater acoustic communication channel, such
as limited bandwidth capacity and variable delays, require very efcient and reliable new
data communication protocols and UWSNs nodes are still an open eld of research (Akyldiz
This paper highlight the relatively less explored and innovative possibility of optical under-
water communication as an interesting and feasible alternative solution for underwater net-
working targeting small, low-cost and low-power devices. In particular it focuses both the
problems related to specic characteristics of the underwater channel targeting the interface
with current terrestrial technologies available for Wireless Sensor Network (WSN) such as the
one developed at DIBE WiseLab (Akyildiz et al., 2005) depicted in Fig.1 .
The goal of our work is to build a prototype of an UWSN based on optical communication
among nodes. The optical PHY and MAC Layers have been developed considering the char-
acteristics of IEEE 802.11 Infrared (IR) PHY layer and the compatibility with the current IEEE
802.15.4 protocol for terrestrial Wireless Sensor Networks (WSNs). This can be considered as
a rst step which, of course, should be followed by modications and adaptation of the upper
layers.
The network architecture can be organized as a three or bi-dimensional underwater net-
work in which each node oats at different depths trying to maintain a xed position with
a good approximation, for instance by an anchorage to the sea bed (as detailed in Chapter
11). The nodes should be designed of slight dimensions (between 15 and 20 cm), they should
be densely deployed, maintaining a distance between 10 and 30 meters from each other. As
detailed in the next paragraph, each node should be able to communicate, by using only a
wireless optical communication link, to the neighbors and, by using a multi hop path, to a
base station placed on the water surface.
The interesting aspects which are reported in this work are related to:
the use of optical communication (the adaptation of current available protocols, the
choice of modulation, the adaptation to water channel variability, etc ...);
the interface with current technology for WSN;
the design and implementation of circuits for diffuse generation and reception of light
impulses targeting low-cost and low-power components;
Fig. 2. Underwater Attenuation of Light from experimental data
Acoustic Electromagnetic Optical
Nominal speed (m/s) 1, 500 light speed light speed
Power Loss 0.1dB/m/Hz 28dB/1km/100MHz turbidity
Bandwidth kHz MHz 10 150MHz
Frequency band KHz MHz 10
14
10
15
Hz
Antenna size 0.1m 0.5m 0.1m
Effective range km 10m 10 100m
Table 1. Comparison of Different Technologies for Underwater Wireless Communication
2. Underwater Wireless Communication: physical aspects
Due to the impossibility of using Radio Frequencies (RF), traditionally wireless underwater
communication employs acoustic waves because sound propagates well in water and its range
can be very long (km). However, it has several disadvantages such as narrow bandwidth
and latency in communication due to the slow speed of acoustic wave in water. For instance,
at ranges of less than 100 m the data transmission rates of these systems in shallow littoral
waters are 10 kb/s.
Experimental tests have shown that an alternative feasible solution is optical communica-
tion especially in blue/green light wavelengths, even if limited to short distances (up to 100
m)(Lanbo et al., 2008). Compared to acoustic communication it offers a practical choice for
high-bandwidth communication and it propagates faster in the water (2.255 x 10
8
).
Nevertheless it is affected by different factors to take into account for an efcient design. The
attenuation of a light beam between two points can be described as in 1 where d
1
and d
2
are
the positions of the points.
A = e
k(d
1
d
2
)
d
1
d
2
2
(1)
In the rst term, k is dened as k = a() + b() and it is dependent by the wavelength: a
is the term related to the absorption of water while b models the scattering which depends
both on light wavelength and turbidity. The second term, instead, models the quadratic at-
tenuation. A comparison between different wireless underwater technologies for underwater
communication is illustrated in Table 1.
3. Wireless Underwater Communication Systems: analysis of the state of art
In this chapter will be briey investigated the state of art related to underwater wireless com-
munication.
Currently the use of wireless communication is very common in a wide range of terrestrial
devices. In particular, one of the most innovative application is related to Wireless Sensor Net-
works (WSNs), as well detailed in (Akyldiz et al., 2001) (Puccinelli & Haenggi, 2005), where a
large number of small nodes communicate and work by using an optical wireless link.
3.1 Underwater Acoustical Communication
As discussed in the introduction, although there are many recently developed solutions for
WSNs, the unique characteristics of the underwater acoustic communication channel, such
as limited bandwidth capacity and variable delays, require very efcient and reliable new
data communication protocols and UWSNs nodes are still an open eld of research (Akyldiz
et al., 2001) (Akyildiz et al., 2005) (Lanbo et al., 2008). The main limitations due to acoustic
communication, which will be described can be summarized as follows:
The underwater channel is severely impaired, especially due to multi-path and fading;
Propagation delay in underwater is ve orders of magnitude higher than in radio fre-
quency (RF) terrestrial channels, and extremely variable;
High bit error rates and temporary losses of connectivity (shadow zones) can be expe-
rienced, due to the extreme characteristics of the underwater channel.
Considering low-power and low-cost devices and targeting short-mediumdistances, the most
important available products are described in (Yan et al., 2007) (Singh et al., 2006) and they
can reach data-rate in the order of Kb/s.
3.2 Underwater Electromagnetic Communication
As regards Electromagnetic Communication, extremely low frequency radio signals have
been used in military applications: Germans pioneered electromagnetic communication in
radio frequency for submarines during World War II, where the antenna was capable of out-
putting up to 1 to 2 Mega-Watt (MW) of power. An extremely low frequency (ELF) signal,
typically around 80 Hz at much lower power, has been used to communicate with naval sub-
marines globally today. This is possible mainly because most of the transmission paths are
through the atmosphere (Shelley, 2005).
In (Al-Shammaa et al., 2004) a theoretical analysis and experiments show that radio waves
within a frequency range 1 to 20MHz is able to propagate over distances up to 100 m by using
dipole radiation with transmission powers in the order of 100 W. This will yield high data
rates beyond 1 Mbps which allows video images to be propagated at standard camera frame
rates (25Hz) (Lucas et al., 2004). The antenna design in such case is very different from that
of the antennas used for conventional service in the atmosphere: in fact, instead of having
direct contact with seawater, the metal transmitting and receiving aerials are surrounded by
waterproof electrically insulating materials. This way, an EM signal can be launched from the
transmitter into a body of seawater and picked up by a distant receiver.
Recently, in September 2006, the rst commercial underwater radio-frequency (RF) modem in
the world, model S1510, was released by Wireless Fibre Systems (Fibre, 2008). Its data rate is
100 bps, and communication range is about several tens of meters. In January 2007, a broad-
band underwater RF modem, model S5510, came into birth. It supports 1-10 Mbps within 1
meter range (Fibre, 2008). Due to the propagation property of EM waves, EMCOMM is an
appealing choice only for very short range applications. One example is the communication
between autonomous underwater vehicles (AUVs) and base stations, where the AUVs can
move within the communication range of a base station to ofoad data and receive further
instructions (Shelley, 2005).
3.3 Optical Underwater Wireless Communication
Underwater optical communication has been investigated both through theoretical studies
(Jaruwatanadilok, 2008) (Lanbo et al., 2008) (Kedar, 2007) and in experimental tests (Feng Lu,
2009)(Hanson & Radic, 2008), mainly developed in USA, Canada and Australia.
Currently, there are not many research activities on underwater optical communication, and
few commercial optical modems are available specically for underwater communication.
As well detailed in the next paragraphs, recent interests in underwater sensor networks and
sea oor observatories have greatly stimulated the interest in short-range high-rate optical
underwater communication.
3.3.1 Point-to-point Communication
In (Tivey et al., 2004) a low power and low cost underwater optical communication system is
proposed by using inespensive components. It is based on IrDa protocol and the adaptation to
underwater channel is performed by replacing Infrared Communication with light generated
by LEDs. Also in (Shill et al., 2004) an underwater communication system has been imple-
mented for a swarm of submersibles. It combines the IrDA physical layer with 3 Watt high
power light emitting diodes, emitting light in the green and blue part of the visible spectrum.
As in the previous example, the approach is to use the IrDAphysical layer modulation replac-
ing the infrared light emitting diodes (LEDs) with high power green or blue LEDs, and also
the photodiode with a type which is sensitive in the visible part of the spectrum. The pro-
totype transceiver costs approximately $45 per unit and, on this low hardware level, no link
management or higher level error correction is done. The wide angular coverage, the uniform
emission footprint and very high light intensity allow for either omnidirectional coverage up
to 2 metre radius with only ve transmitters, when using simultaneously transmitting expen-
sive LEDs that consume 2W or, with additional lenses, long range directional links having a
collimated beam (up to 5 meters).
In (Feng Lu, 2009) and (Lee et al., 2008) a low-cost (in the order of $10)medium-range optical
underwater modem is proposed. A sophisticated detection algorithm is exploited (based on
spread spectrum) to maximize the communication range. Tests have been performed up to 10
meters and a data-rate of 310bps has been achieved.
In (Vasilescu et al., 2007) and (Vasilescu et al., 2005) the underwater wireless sensor network
AquaNodes is described. In this application the use of optical communication is shown as an
efcient solution for data muling in the network proposed by CSIRO ICT Centre (Australia)
and MIT CSIAL (USA). This research proposes a network where data exchange is performed
by connecting, both optically and acoustically, previously deployed static nodes and mobile
(AUVs) vehicles, so that the characteristics of the two communication strategies can be ex-
ploited. Each device is provided with an optical modem, which can performa transmission up
to 300 kb/s in a range below 8 meters. This approach clearly shows that the use of an optical
communication system allows for a considerable reduction in terms of energy consumption
with respect to a multi-hop acoustic transmission and a resulting increase of operational life
of the system. Moreover, considering this work, the use of optical modems instead of acoustic
ones for short range communications leads to a remarkable reduction of costs, since the cost of
an optical modem is of the order of $50 node against approximately $3000 node of an acoustic
modem.
In (Ito et al., 2008a) is illustrated a careful analysis by modeling the underwater channel based
on underwater optics. Through this analysis, it is showed that a single color LED is very
weak in a wavelength dependent underwater channel and, to overcome this problem, a multi-
wavelength adaptive scheme combined with rate adaptive transmission is proposed taking
inspiration from already developed algorithms. The proposed system can adapt to the chan-
nel by considering the change in power for each wavelength band, and controlling the data
rate.
In (Hanson &Radic, 2008) the work is motivated by the need to demonstrate error-free under-
water communication at qualitatively higher data rates than previously reported with either
optical or acoustic methods. In this case a Laser source is used and an error-free underwater
et al., 2001) (Akyildiz et al., 2005) (Lanbo et al., 2008). The main limitations due to acoustic
communication, which will be described can be summarized as follows:
The underwater channel is severely impaired, especially due to multi-path and fading;
Propagation delay in underwater is ve orders of magnitude higher than in radio fre-
quency (RF) terrestrial channels, and extremely variable;
High bit error rates and temporary losses of connectivity (shadow zones) can be expe-
rienced, due to the extreme characteristics of the underwater channel.
Considering low-power and low-cost devices and targeting short-mediumdistances, the most
important available products are described in (Yan et al., 2007) (Singh et al., 2006) and they
can reach data-rate in the order of Kb/s.
3.2 Underwater Electromagnetic Communication
As regards Electromagnetic Communication, extremely low frequency radio signals have
been used in military applications: Germans pioneered electromagnetic communication in
radio frequency for submarines during World War II, where the antenna was capable of out-
putting up to 1 to 2 Mega-Watt (MW) of power. An extremely low frequency (ELF) signal,
typically around 80 Hz at much lower power, has been used to communicate with naval sub-
marines globally today. This is possible mainly because most of the transmission paths are
through the atmosphere (Shelley, 2005).
In (Al-Shammaa et al., 2004) a theoretical analysis and experiments show that radio waves
within a frequency range 1 to 20MHz is able to propagate over distances up to 100 m by using
dipole radiation with transmission powers in the order of 100 W. This will yield high data
rates beyond 1 Mbps which allows video images to be propagated at standard camera frame
rates (25Hz) (Lucas et al., 2004). The antenna design in such case is very different from that
of the antennas used for conventional service in the atmosphere: in fact, instead of having
direct contact with seawater, the metal transmitting and receiving aerials are surrounded by
waterproof electrically insulating materials. This way, an EM signal can be launched from the
transmitter into a body of seawater and picked up by a distant receiver.
Recently, in September 2006, the rst commercial underwater radio-frequency (RF) modem in
the world, model S1510, was released by Wireless Fibre Systems (Fibre, 2008). Its data rate is
100 bps, and communication range is about several tens of meters. In January 2007, a broad-
band underwater RF modem, model S5510, came into birth. It supports 1-10 Mbps within 1
meter range (Fibre, 2008). Due to the propagation property of EM waves, EMCOMM is an
appealing choice only for very short range applications. One example is the communication
between autonomous underwater vehicles (AUVs) and base stations, where the AUVs can
move within the communication range of a base station to ofoad data and receive further
instructions (Shelley, 2005).
3.3 Optical Underwater Wireless Communication
Underwater optical communication has been investigated both through theoretical studies
(Jaruwatanadilok, 2008) (Lanbo et al., 2008) (Kedar, 2007) and in experimental tests (Feng Lu,
2009)(Hanson & Radic, 2008), mainly developed in USA, Canada and Australia.
Currently, there are not many research activities on underwater optical communication, and
few commercial optical modems are available specically for underwater communication.
As well detailed in the next paragraphs, recent interests in underwater sensor networks and
sea oor observatories have greatly stimulated the interest in short-range high-rate optical
underwater communication.
3.3.1 Point-to-point Communication
In (Tivey et al., 2004) a low power and low cost underwater optical communication system is
proposed by using inespensive components. It is based on IrDa protocol and the adaptation to
underwater channel is performed by replacing Infrared Communication with light generated
by LEDs. Also in (Shill et al., 2004) an underwater communication system has been imple-
mented for a swarm of submersibles. It combines the IrDA physical layer with 3 Watt high
power light emitting diodes, emitting light in the green and blue part of the visible spectrum.
As in the previous example, the approach is to use the IrDAphysical layer modulation replac-
ing the infrared light emitting diodes (LEDs) with high power green or blue LEDs, and also
the photodiode with a type which is sensitive in the visible part of the spectrum. The pro-
totype transceiver costs approximately $45 per unit and, on this low hardware level, no link
management or higher level error correction is done. The wide angular coverage, the uniform
emission footprint and very high light intensity allow for either omnidirectional coverage up
to 2 metre radius with only ve transmitters, when using simultaneously transmitting expen-
sive LEDs that consume 2W or, with additional lenses, long range directional links having a
collimated beam (up to 5 meters).
In (Feng Lu, 2009) and (Lee et al., 2008) a low-cost (in the order of $10)medium-range optical
underwater modem is proposed. A sophisticated detection algorithm is exploited (based on
spread spectrum) to maximize the communication range. Tests have been performed up to 10
meters and a data-rate of 310bps has been achieved.
In (Vasilescu et al., 2007) and (Vasilescu et al., 2005) the underwater wireless sensor network
AquaNodes is described. In this application the use of optical communication is shown as an
efcient solution for data muling in the network proposed by CSIRO ICT Centre (Australia)
and MIT CSIAL (USA). This research proposes a network where data exchange is performed
by connecting, both optically and acoustically, previously deployed static nodes and mobile
(AUVs) vehicles, so that the characteristics of the two communication strategies can be ex-
ploited. Each device is provided with an optical modem, which can performa transmission up
to 300 kb/s in a range below 8 meters. This approach clearly shows that the use of an optical
communication system allows for a considerable reduction in terms of energy consumption
with respect to a multi-hop acoustic transmission and a resulting increase of operational life
of the system. Moreover, considering this work, the use of optical modems instead of acoustic
ones for short range communications leads to a remarkable reduction of costs, since the cost of
an optical modem is of the order of $50 node against approximately $3000 node of an acoustic
modem.
In (Ito et al., 2008a) is illustrated a careful analysis by modeling the underwater channel based
on underwater optics. Through this analysis, it is showed that a single color LED is very
weak in a wavelength dependent underwater channel and, to overcome this problem, a multi-
wavelength adaptive scheme combined with rate adaptive transmission is proposed taking
inspiration from already developed algorithms. The proposed system can adapt to the chan-
nel by considering the change in power for each wavelength band, and controlling the data
rate.
In (Hanson &Radic, 2008) the work is motivated by the need to demonstrate error-free under-
water communication at qualitatively higher data rates than previously reported with either
optical or acoustic methods. In this case a Laser source is used and an error-free underwater
optical transmission measurements at 1 Gbit/s over a 2 m path in a laboratory water pipe.
In (Chancey, 2007) a short range underwater optical communication link is established by
using an open source free space modem proposed by the project RONJA. Unfortunately, the
used equipment is very expensive and difcult to use in real applications; as reported, the
experimental results achieved 10 Mb/s at a distance up to 5 meters.
A commercial product targeting optical underwater communication, from Ambalux (Am-
balux, 2008), includes high-bandwidth transceivers, which allow a point-to-point transmis-
sion at a data rate of 10Mbs up to 40m, depending on environmental conditions. A set of
tests based on this commercial products has been described in (Baiden & Bissiri, 2007), where
a solution for omni-directional communication is also proposed in order to improve the ef-
cacy of the connection with an underwater untethered vehicle. This approach, however, is not
suitable for a dense UWSN or small and low power devices, where nodes are deployed with
low accuracy and concerns on system size and energy consumption are addressed.
Recently an underwater optical wireless modem (AquaOptical)(Doniec et al., 2009) has been
implemented. It is designed for integration in robotic application and allows to support a
wireless communication of 2 Mb/s up to a range of 30m.
3.4 Diffusive underwater optical communication
In these paragraph are reported some researches related to the use of diffusive underwater
optical communication for different applications.
In (Baiden et al., 2009) the point-to-point optical communication system developed by Am-
balux (Ambalux, 2008) is used to perform a wide number of tests with the aim of paving the
way for a future underwater omni-directional wireless optical communication systems. The
LEDs used in the test emitted light in the green and blue light spectrum and were tested in
a pool and in a tank lled with lake water. The primary objective of these tests was to get
proles of the behaviors of such communication systems with respect to water characteristics
such as turbidity levels, prior to building an omni-directional optical communication. The
results of the tests indicated that turbidity level, viewing angle and separation distance plays
a signicant role in the behavior of blue light in water. In particular, the aim was to dene a
threshold viewing angle (TVA), the minimum viewing angle at which communication is lost
when one of the communication devices is rotated with respect to a virtual axis that contains
the segment represented by the center of gravities of the two devices when they are aligned.
On the basis of the previous studies, two geometric forms were modelled: icosahedron and
spherical hexagon. The icosahedron was retained because of its simplicity in geometry and its
ability to provide complete free space coverage using the selected LED.
In (Baiden & Bissiri, 2007), the previous illustrated conguration has been tested to imple-
ment a high bandwidth optical networking for underwater untethered tele-robotic operation.
A rate of 4 Mb/s was achieved in hemispherical conguration up to 5 meters. Experimenta-
tion in the eld achieved initially 115 Kb/s over a distance of 15 metres. Asecond experiment,
after the modication of the transmitter and the receiver software, was attempted: in this case
the transmission was increased to 1.5 Mb/s and the rst wireless underwater video pictures
were transmitted. Despite the originally anticipated different kinds of turbidity (sh, plank-
ton, rocks or other objects) in the water, the experiment still received video feedback of the
anchor of a oating laboratory at around 15 metres deep in the bottom of the lake.
In (Fair et al., 2006) an optical modem technology for seaoor observatories is illustrated and
problems related to design and implementation of a system for underwater optical wireless
communication are focused. The idea is to implement an optical modemsystemwhich should
provide sufcient bandwidth to allow transmission of compressed high resolution video (i.e.,
2-10 Mbit/s for studio quality video), allowing nearly unrestricted motion on the part of the
mobile sensor (UUV), and working at ranges below 100 m. In this case, the design of the sys-
tem have been developed considering the following aspects: (1) selection of a transmission
light source (wavelength, power, beamwidth), (2) selection of a detector (eld of view, quan-
tum efciency, gain), and (3) selection of an aiming and tracking strategy. As regards trans-
mission light source, LEDs are suggested to be used for the transmitters as they are switchable
at high rates, can be reasonably collimated, and are easily arrayed to increase transmit power.
High intensity blue LEDs such as those fabricated from Gallium Indium Nitride (InGaN) on
a silicon carbide (SiC) substrate can provide on the order of 10 mW each and they can be
congured in arrays to increase the total radiant ux (optical power). The light from one or
more LEDs can be collimated with a lens to focus its beam. While their bandwidth is low by
optical standards (30 ns rise time), a 100 ns bit duration meets the design goal of 10 Mbit/s.
LED transmitters provide signicant exibility as well: both OOK and PPM are possible with
the same hardware. As regards the receiver, in this work a photomultiplier tube (PMT) is
chosen for the detector because it provides higher sensitivity and less noise than photodiodes
(including avalanche or PIN types) although the tube size can be a limiting factor in certain
conditions. Considering the aiming and tracking strategies, three possible congurations for
optical communications between a xed node and a free-swimming vehicle are suggested in
this paper:
1. Pointed transmission (Tx) and reception (Rx) with acoustic or optical aiming is the most
efcient scenario from an optical transmission standpoint. The transmitter consists of
several collimated LEDs or a laser diode, and the receiver is a PMT. This method re-
quires a search and acquisition mode by both the Rx and Tx which consumes time and
energy (regardless of whether the aiming is acoustic or optical);
2. Directional-Tx to omni-directional-Rx scenario can be accomplished with a hemispher-
ical PMT detector and either a laser diode or a hemispherical array of LEDs. In this
case, the aiming problem is one-sided and could be accomplished via acoustics or with
optics;
3. Omni-directional Tx and Rx which is the mechanically simplest solution. This can be ac-
complished using the large area PMT detector with a moderate output-power blue laser
diode or LEDand diffusing optics. The diffusing optics can be done either with discrete
reective and refractive elements or with a high transmission scattering medium.
Taking into account the previous proposed theoretical studies, some optical systems have
been implemented and tested, as reported in (Farr et al., 2005). The omni-directional light
source used was blue-green (470nm) to take advantage of the low attenuation in seawater at
those wavelengths, and produced a uniform light eld over a 2 steradian hemisphere. Six
commercially available 470nm (blue) light emitting diodes (LED) were arranged in a hemi-
spherical geometry and encapsulated in a weekly scattering potting material to provide some
diffusion of the light eld over the full hemisphere of operation. The receiver consisted of a
large-aperture, hemispherical photomultiplier tube (PMT) chosen for its high sensitivity, low
noise and high speed. Apertures and mirrors have been used to obtain a 91 m path length in
a 15 m deep pool and to prevent reections from the surrounding walls from contaminating
the results. A 100 meter bench test has been performed to verify the range and geometry cal-
culation using neutral density lters to simulate the attenuation of water.
In the same work (Farr et al., 2005) different tests are performed to validate the concept of
optical transmission measurements at 1 Gbit/s over a 2 m path in a laboratory water pipe.
In (Chancey, 2007) a short range underwater optical communication link is established by
using an open source free space modem proposed by the project RONJA. Unfortunately, the
used equipment is very expensive and difcult to use in real applications; as reported, the
experimental results achieved 10 Mb/s at a distance up to 5 meters.
A commercial product targeting optical underwater communication, from Ambalux (Am-
balux, 2008), includes high-bandwidth transceivers, which allow a point-to-point transmis-
sion at a data rate of 10Mbs up to 40m, depending on environmental conditions. A set of
tests based on this commercial products has been described in (Baiden & Bissiri, 2007), where
a solution for omni-directional communication is also proposed in order to improve the ef-
cacy of the connection with an underwater untethered vehicle. This approach, however, is not
suitable for a dense UWSN or small and low power devices, where nodes are deployed with
low accuracy and concerns on system size and energy consumption are addressed.
Recently an underwater optical wireless modem (AquaOptical)(Doniec et al., 2009) has been
implemented. It is designed for integration in robotic application and allows to support a
wireless communication of 2 Mb/s up to a range of 30m.
3.4 Diffusive underwater optical communication
In these paragraph are reported some researches related to the use of diffusive underwater
optical communication for different applications.
In (Baiden et al., 2009) the point-to-point optical communication system developed by Am-
balux (Ambalux, 2008) is used to perform a wide number of tests with the aim of paving the
way for a future underwater omni-directional wireless optical communication systems. The
LEDs used in the test emitted light in the green and blue light spectrum and were tested in
a pool and in a tank lled with lake water. The primary objective of these tests was to get
proles of the behaviors of such communication systems with respect to water characteristics
such as turbidity levels, prior to building an omni-directional optical communication. The
results of the tests indicated that turbidity level, viewing angle and separation distance plays
a signicant role in the behavior of blue light in water. In particular, the aim was to dene a
threshold viewing angle (TVA), the minimum viewing angle at which communication is lost
when one of the communication devices is rotated with respect to a virtual axis that contains
the segment represented by the center of gravities of the two devices when they are aligned.
On the basis of the previous studies, two geometric forms were modelled: icosahedron and
spherical hexagon. The icosahedron was retained because of its simplicity in geometry and its
ability to provide complete free space coverage using the selected LED.
In (Baiden & Bissiri, 2007), the previous illustrated conguration has been tested to imple-
ment a high bandwidth optical networking for underwater untethered tele-robotic operation.
A rate of 4 Mb/s was achieved in hemispherical conguration up to 5 meters. Experimenta-
tion in the eld achieved initially 115 Kb/s over a distance of 15 metres. Asecond experiment,
after the modication of the transmitter and the receiver software, was attempted: in this case
the transmission was increased to 1.5 Mb/s and the rst wireless underwater video pictures
were transmitted. Despite the originally anticipated different kinds of turbidity (sh, plank-
ton, rocks or other objects) in the water, the experiment still received video feedback of the
anchor of a oating laboratory at around 15 metres deep in the bottom of the lake.
In (Fair et al., 2006) an optical modem technology for seaoor observatories is illustrated and
problems related to design and implementation of a system for underwater optical wireless
communication are focused. The idea is to implement an optical modemsystemwhich should
provide sufcient bandwidth to allow transmission of compressed high resolution video (i.e.,
2-10 Mbit/s for studio quality video), allowing nearly unrestricted motion on the part of the
mobile sensor (UUV), and working at ranges below 100 m. In this case, the design of the sys-
tem have been developed considering the following aspects: (1) selection of a transmission
light source (wavelength, power, beamwidth), (2) selection of a detector (eld of view, quan-
tum efciency, gain), and (3) selection of an aiming and tracking strategy. As regards trans-
mission light source, LEDs are suggested to be used for the transmitters as they are switchable
at high rates, can be reasonably collimated, and are easily arrayed to increase transmit power.
High intensity blue LEDs such as those fabricated from Gallium Indium Nitride (InGaN) on
a silicon carbide (SiC) substrate can provide on the order of 10 mW each and they can be
congured in arrays to increase the total radiant ux (optical power). The light from one or
more LEDs can be collimated with a lens to focus its beam. While their bandwidth is low by
optical standards (30 ns rise time), a 100 ns bit duration meets the design goal of 10 Mbit/s.
LED transmitters provide signicant exibility as well: both OOK and PPM are possible with
the same hardware. As regards the receiver, in this work a photomultiplier tube (PMT) is
chosen for the detector because it provides higher sensitivity and less noise than photodiodes
(including avalanche or PIN types) although the tube size can be a limiting factor in certain
conditions. Considering the aiming and tracking strategies, three possible congurations for
optical communications between a xed node and a free-swimming vehicle are suggested in
this paper:
1. Pointed transmission (Tx) and reception (Rx) with acoustic or optical aiming is the most
efcient scenario from an optical transmission standpoint. The transmitter consists of
several collimated LEDs or a laser diode, and the receiver is a PMT. This method re-
quires a search and acquisition mode by both the Rx and Tx which consumes time and
energy (regardless of whether the aiming is acoustic or optical);
2. Directional-Tx to omni-directional-Rx scenario can be accomplished with a hemispher-
ical PMT detector and either a laser diode or a hemispherical array of LEDs. In this
case, the aiming problem is one-sided and could be accomplished via acoustics or with
optics;
3. Omni-directional Tx and Rx which is the mechanically simplest solution. This can be ac-
complished using the large area PMT detector with a moderate output-power blue laser
diode or LEDand diffusing optics. The diffusing optics can be done either with discrete
reective and refractive elements or with a high transmission scattering medium.
Taking into account the previous proposed theoretical studies, some optical systems have
been implemented and tested, as reported in (Farr et al., 2005). The omni-directional light
source used was blue-green (470nm) to take advantage of the low attenuation in seawater at
those wavelengths, and produced a uniform light eld over a 2 steradian hemisphere. Six
commercially available 470nm (blue) light emitting diodes (LED) were arranged in a hemi-
spherical geometry and encapsulated in a weekly scattering potting material to provide some
diffusion of the light eld over the full hemisphere of operation. The receiver consisted of a
large-aperture, hemispherical photomultiplier tube (PMT) chosen for its high sensitivity, low
noise and high speed. Apertures and mirrors have been used to obtain a 91 m path length in
a 15 m deep pool and to prevent reections from the surrounding walls from contaminating
the results. A 100 meter bench test has been performed to verify the range and geometry cal-
culation using neutral density lters to simulate the attenuation of water.
In the same work (Farr et al., 2005) different tests are performed to validate the concept of
omni-directional free water optical communication in the range of 10 meters. In particular,
the power spectrum of the received signal during the transmission of a pulse train at different
repetition rates is reported: 1.25 MHz, 5 MHz and 10 MHz.
This work focuses in particular the communication between Underwater Mobile Vehicles and
a xed node targeting application in sea oor observatories. The achieved distances and data-
rate are interesting but it could be difcult to apply these results to low-cost and small devices
where transmitters and receivers should be placed on the same small surface.
In (Liu & Ge, 2006) the concept of underwater laser sensor network is illustrated as a new
approach for broadband communication in the underwater environment based on the blue-
green laser has been proposed. In this paper the applications of the underwater sensor net-
work in the undersea exploration are discussed with the difculties in the traditional the un-
derwater acoustic sensor network. A basic of prototype of underwater laser sensor network is
described: it includes the architecture of laser sensor node and protocol stack for underwater
laser sensor network, but it does not shown implementation or interesting practical results.
The here described devices show that the use of optical communication can be a feasible so-
lutions for underwater wireless communication. In the comparison with the here proposed
works, the aim of this paper is to illustrate the problems and possible solutions for the ap-
plication of optical wireless communication in order to support a UWSN of dense deployed
nodes as detailed in Chapter 1 and 11.
4. Application of underwater optical wireless communication
Optical underwater communication is an effective alternative to current underwater technol-
ogy especially in some particular environments such as, for instance, shallow, coastal and
fresh inland waters where the use of this approach is useful to overcome all the shortcomings
related to the use of acoustic communication and to allow a wide adoption of underwater
monitoring systems. In particular the possibility of transferring high amount of data in a lim-
ited amount of time reducing power consumption can support the transmission of short video
and pictures for a reliable monitoring and surveillance. Small dimensions and low-cost com-
ponents allow to establish a dense deployed networks performing an effective ne grained
sampling in the area of interest. It could be possible, for instance, to perform pollution mon-
itoring and frequent data collection (water temperature, specic conductivity, pH, turbidity,
and possibly oxygen concentration) and, by using a high-data rate optical link, periodically
deliver data reducing the time devoted to transmission and network congestion.
5. Design consideration for a diffuse optical underwater communication
The design of underwater optical systems for underwater networking should takes into ac-
count different aspects which are illustrated in the next paragraphs:
the Physical(PHY) Layer to manage the trasmission and reception of data by using de-
voted circuits;
the Medium Access Control(MAC) Layer to manage the possibility of more than one
devices to communicate each others;
the aspects related to the circuits design and implementation (such as choice of the Led
and photodiodes and their collocation on a surface).
In the next paragraph some design consideration for each of the previous aspect will be pre-
sented and our design and implementation results will be briey illustrated. The here de-
scribed system is an based on a previous implementation illustrated in (Anguita et al., 2010)
(Anguita et al., 2008) (Anguita et al., 2009) and it implements PHY and MAC Layer modules
which have been described by using an Hardware Description Language (HDL), in particular
VHDL (Very High Speed Hardware Description Language), as suggested in (Pang et al., 2007)
(Dasalukunte & Owall, 2008) and well detailed in Chapter 6.3.
Fig. 3. General block diagram of the Optical Transmission System
6. The Physical Layer
The Physical Layer has to support:
1. the transmission and reception of data by using optical communication;
2. the interface with a MAC Layer of the communication stack by providind the services
required by the upper layers.
In particular, the design should take into account the peculiarity of underwater channel and
focus on the following elements:
1. efcient wireless optical modulation techniques;
2. adaptation to underwater channel variability;
3. adaptation to data-rate communication requirements;
4. interface with current standard used for WSN (i.e. IEEE 802.15.4).
omni-directional free water optical communication in the range of 10 meters. In particular,
the power spectrum of the received signal during the transmission of a pulse train at different
repetition rates is reported: 1.25 MHz, 5 MHz and 10 MHz.
This work focuses in particular the communication between Underwater Mobile Vehicles and
a xed node targeting application in sea oor observatories. The achieved distances and data-
rate are interesting but it could be difcult to apply these results to low-cost and small devices
where transmitters and receivers should be placed on the same small surface.
In (Liu & Ge, 2006) the concept of underwater laser sensor network is illustrated as a new
approach for broadband communication in the underwater environment based on the blue-
green laser has been proposed. In this paper the applications of the underwater sensor net-
work in the undersea exploration are discussed with the difculties in the traditional the un-
derwater acoustic sensor network. A basic of prototype of underwater laser sensor network is
described: it includes the architecture of laser sensor node and protocol stack for underwater
laser sensor network, but it does not shown implementation or interesting practical results.
The here described devices show that the use of optical communication can be a feasible so-
lutions for underwater wireless communication. In the comparison with the here proposed
works, the aim of this paper is to illustrate the problems and possible solutions for the ap-
plication of optical wireless communication in order to support a UWSN of dense deployed
nodes as detailed in Chapter 1 and 11.
4. Application of underwater optical wireless communication
Optical underwater communication is an effective alternative to current underwater technol-
ogy especially in some particular environments such as, for instance, shallow, coastal and
fresh inland waters where the use of this approach is useful to overcome all the shortcomings
related to the use of acoustic communication and to allow a wide adoption of underwater
monitoring systems. In particular the possibility of transferring high amount of data in a lim-
ited amount of time reducing power consumption can support the transmission of short video
and pictures for a reliable monitoring and surveillance. Small dimensions and low-cost com-
ponents allow to establish a dense deployed networks performing an effective ne grained
sampling in the area of interest. It could be possible, for instance, to perform pollution mon-
itoring and frequent data collection (water temperature, specic conductivity, pH, turbidity,
and possibly oxygen concentration) and, by using a high-data rate optical link, periodically
deliver data reducing the time devoted to transmission and network congestion.
5. Design consideration for a diffuse optical underwater communication
The design of underwater optical systems for underwater networking should takes into ac-
count different aspects which are illustrated in the next paragraphs:
the Physical(PHY) Layer to manage the trasmission and reception of data by using de-
voted circuits;
the Medium Access Control(MAC) Layer to manage the possibility of more than one
devices to communicate each others;
the aspects related to the circuits design and implementation (such as choice of the Led
and photodiodes and their collocation on a surface).
In the next paragraph some design consideration for each of the previous aspect will be pre-
sented and our design and implementation results will be briey illustrated. The here de-
scribed system is an based on a previous implementation illustrated in (Anguita et al., 2010)
(Anguita et al., 2008) (Anguita et al., 2009) and it implements PHY and MAC Layer modules
which have been described by using an Hardware Description Language (HDL), in particular
VHDL (Very High Speed Hardware Description Language), as suggested in (Pang et al., 2007)
(Dasalukunte & Owall, 2008) and well detailed in Chapter 6.3.
Fig. 3. General block diagram of the Optical Transmission System
6. The Physical Layer
The Physical Layer has to support:
1. the transmission and reception of data by using optical communication;
2. the interface with a MAC Layer of the communication stack by providind the services
required by the upper layers.
In particular, the design should take into account the peculiarity of underwater channel and
focus on the following elements:
1. efcient wireless optical modulation techniques;
2. adaptation to underwater channel variability;
3. adaptation to data-rate communication requirements;
4. interface with current standard used for WSN (i.e. IEEE 802.15.4).
Modulation Implementation
Complexity
Sensitivity to multi-
path delay
Required
transmitted
power
Bandwidth
OOK simple general P
ROOK
2R
B
FSK most complex most anti-sensitive
1
2
P
ROOK
| f
1
f
0
| + 2R
B
(2
FSK)
DPSK more complex more sensitive
1
8
ln2
P
ROOK
2R
B
(2 DPSK)
PPM simple general
2
P
ROOK
L
log
2
L
R
B
(L-PPM)
Table 2. Comparison of Various Modulations Techniques
6.1 Wireless optical modulation techniques
As well detailed in literature (Ghassemlooy et al., 2007) (Park & Barry, 1995) (Kahn & Barry,
1997) basic modulation techniques mainly contain three formats, such as Amplitude Modula-
tion (ASK), Frequency Modulation (FSK) and Phase Modulation (PSK).
Taking into account previous reported studies (Lee et al., 2008), Pulse-Position Modulation
(PPM) can be considered in order to further improve the anti-disturbance capacity of infor-
mation transmission and it proposed in different systems both for terrestrial (IEEE 802.11 IR)
than for underwater systems (Lee et al., 2008).
Pulse Position modulation is a well-known orthogonal modulation technique, in which M
message bits are encoded by transmitting a single pulse in one of 2
M
possible time-shift. PPM
scheme, at each time interval is T
s
and L = 2
M
time-shifts constitute a PPM frame. At the
transmitter end, the signal will be launched by the light pulse to form a specic time slot,
and in the receiver end, the photoelectric diode detects the light pulse and then to judge its
time slot to evaluate his position and resume the signal. In L-PPM system, a block of input
bits is mapped on to one of distinct waveforms, each including one on chip and M-1 o f f
chips. A pulse is transmitted during the on chip, it can be dened as (Meihong, Xinsheng &
Zhangguo, 2009):
P
m
=
i
s
t [m1]T
f
/L, mT
f
/L
0 else
(2)
Considering the characteristics of the different modulation the Table 2 can be proposed for a
comparison between different techniques (R
B
is the base-band signal bandwidth).
Since the characteristics of PPM appeared to be interesting for a wide range of applications
some alternative and modied PPM modulations can be considered such as Differential Pulse
Position Modulation (DPPM), Pulse Interval Modulation (PIM) and Dual Header Pulse Inter-
val Modulation (DH-PIM).
6.1.1 BER vs. SNR
Considering the previous described modulations it is interesting to evaluate the Bit Error Rate
versus SNR. The following results are mainly based on (Meihong, Xinsheng & Zhangguo,
2009) (Meihong, Xinsheng & Fengli, 2009). For OOK demodulation format a threshold is used
to compare the received voltage to decide "1" or "0". In AWGN channel model, the received
voltage is:
p(t) =
i
s
+ n
c(t)
, "1"
n
c(t)
, 0"
(3)
where n
c(t)
is the Gaussian process. For the data "1", the probability density of x(t) is:
p
1
(x) =
1
2
exp(
(x i
s
)
2
2
2
) (4)
for the data bit "0" the probability density of x(t) is:
p
0
(x) =
1
2
exp(
x
2
2
2
) (5)
By xing the the judgment threshold to
1
2
i, the bit error rate is dened as:
p
e
(ook) =
1
2
er f c
i
s
2
2
=
1
2
er f c
S
2
2
(6)
where erfc is the complementary error function. According to (Lee & Kahn, 1999), in additive
Gaussian noise channel, the bit error rate of 2FSK coherent modulation and 2DPSK coherent
modulation are given by the following equations:
p
e
(FSK) =
1
2
er f c
S
4
(7)
p
e
(DFSK) = er f c
S
2
(1
1
2
S
2
) (8)
For the L-PPM modulation in the Gaussian white noise channel, there are many performance
evaluation methods for the BER as suggested in (C.X.Fan et al., 2001) (Ma, 2003) (Malik et al.,
1996). The following formula is used in (Meihong, Xinsheng & Fengli, 2009):
p
e
(L PPM) =
1
L
[
1
2
er f c(
1 k
2
LS
) +
L 1
2
er f c(
k
2
LS)] (9)
in order to show the Bit Error Rate in different SNR conditions. The 8-PPM is reported to be a
good choice to reduce the BER.
6.1.2 Power Consumption
Also problems related to power consumption has been investigated in literature (Tivey et al.,
2004) (Meihong, Xinsheng & Zhangguo, 2009) since they are crucial in underwater devices.
Small and light systems are desirable in order to improve underwater system performance
and considerations about power consumption are important in the choice of modulation. The
following considerations can be carried out for the different modulations:
1. for the Frequency Shift Key (FSK) modulation a specic frequency carrier wave for
digital 1, and a different frequency carrier wave for digital 0 are generated. In
this case the optical transmit power required as the transmitting duration is always on
and it appears to be inefcient for applications on power-constrained devices.
Modulation Implementation
Complexity
Sensitivity to multi-
path delay
Required
transmitted
power
Bandwidth
OOK simple general P
ROOK
2R
B
FSK most complex most anti-sensitive
1
2
P
ROOK
| f
1
f
0
| + 2R
B
(2
FSK)
DPSK more complex more sensitive
1
8
ln2
P
ROOK
2R
B
(2 DPSK)
PPM simple general
2
P
ROOK
L
log
2
L
R
B
(L-PPM)
Table 2. Comparison of Various Modulations Techniques
6.1 Wireless optical modulation techniques
As well detailed in literature (Ghassemlooy et al., 2007) (Park & Barry, 1995) (Kahn & Barry,
1997) basic modulation techniques mainly contain three formats, such as Amplitude Modula-
tion (ASK), Frequency Modulation (FSK) and Phase Modulation (PSK).
Taking into account previous reported studies (Lee et al., 2008), Pulse-Position Modulation
(PPM) can be considered in order to further improve the anti-disturbance capacity of infor-
mation transmission and it proposed in different systems both for terrestrial (IEEE 802.11 IR)
than for underwater systems (Lee et al., 2008).
Pulse Position modulation is a well-known orthogonal modulation technique, in which M
message bits are encoded by transmitting a single pulse in one of 2
M
possible time-shift. PPM
scheme, at each time interval is T
s
and L = 2
M
time-shifts constitute a PPM frame. At the
transmitter end, the signal will be launched by the light pulse to form a specic time slot,
and in the receiver end, the photoelectric diode detects the light pulse and then to judge its
time slot to evaluate his position and resume the signal. In L-PPM system, a block of input
bits is mapped on to one of distinct waveforms, each including one on chip and M-1 o f f
chips. A pulse is transmitted during the on chip, it can be dened as (Meihong, Xinsheng &
Zhangguo, 2009):
P
m
=
i
s
t [m1]T
f
/L, mT
f
/L
0 else
(2)
Considering the characteristics of the different modulation the Table 2 can be proposed for a
comparison between different techniques (R
B
is the base-band signal bandwidth).
Since the characteristics of PPM appeared to be interesting for a wide range of applications
some alternative and modied PPM modulations can be considered such as Differential Pulse
Position Modulation (DPPM), Pulse Interval Modulation (PIM) and Dual Header Pulse Inter-
val Modulation (DH-PIM).
6.1.1 BER vs. SNR
Considering the previous described modulations it is interesting to evaluate the Bit Error Rate
versus SNR. The following results are mainly based on (Meihong, Xinsheng & Zhangguo,
2009) (Meihong, Xinsheng & Fengli, 2009). For OOK demodulation format a threshold is used
to compare the received voltage to decide "1" or "0". In AWGN channel model, the received
voltage is:
p(t) =
i
s
+ n
c(t)
, "1"
n
c(t)
, 0"
(3)
where n
c(t)
is the Gaussian process. For the data "1", the probability density of x(t) is:
p
1
(x) =
1
2
exp(
(x i
s
)
2
2
2
) (4)
for the data bit "0" the probability density of x(t) is:
p
0
(x) =
1
2
exp(
x
2
2
2
) (5)
By xing the the judgment threshold to
1
2
i, the bit error rate is dened as:
p
e
(ook) =
1
2
er f c
i
s
2
2
=
1
2
er f c
S
2
2
(6)
where erfc is the complementary error function. According to (Lee & Kahn, 1999), in additive
Gaussian noise channel, the bit error rate of 2FSK coherent modulation and 2DPSK coherent
modulation are given by the following equations:
p
e
(FSK) =
1
2
er f c
S
4
(7)
p
e
(DFSK) = er f c
S
2
(1
1
2
S
2
) (8)
For the L-PPM modulation in the Gaussian white noise channel, there are many performance
evaluation methods for the BER as suggested in (C.X.Fan et al., 2001) (Ma, 2003) (Malik et al.,
1996). The following formula is used in (Meihong, Xinsheng & Fengli, 2009):
p
e
(L PPM) =
1
L
[
1
2
er f c(
1 k
2
LS
) +
L 1
2
er f c(
k
2
LS)] (9)
in order to show the Bit Error Rate in different SNR conditions. The 8-PPM is reported to be a
good choice to reduce the BER.
6.1.2 Power Consumption
Also problems related to power consumption has been investigated in literature (Tivey et al.,
2004) (Meihong, Xinsheng & Zhangguo, 2009) since they are crucial in underwater devices.
Small and light systems are desirable in order to improve underwater system performance
and considerations about power consumption are important in the choice of modulation. The
following considerations can be carried out for the different modulations:
1. for the Frequency Shift Key (FSK) modulation a specic frequency carrier wave for
digital 1, and a different frequency carrier wave for digital 0 are generated. In
this case the optical transmit power required as the transmitting duration is always on
and it appears to be inefcient for applications on power-constrained devices.
OOK FSK DPSK 4-PPM 8-PPM
Maximum rate
1
2P
1
2P
1
2P
1
2P
3
8P
Transmit power Middle Higher Highest Low Lowest
Complexity of modula-
tion
Low Higher Highest Lower Lower
Table 3. Transfer rate versus implementation complexity from (Meihong, Xinsheng & Fengli,
2009)
2. the Phase Shift Key (PSK) modulator generates an in phase signal for digital 1 and an
out of phase signal for a digital 0. In this case the PSK demodulator are complex since
the different coherent demodulation are needed to compare the current phase with the
previous phase and it can be inefcient for power-constrained devices.
3. the On-Off Keying (OOK) and Pulse Position Modulation (PPM) do not use completely
the frequency or phase information, but the design of the receiver and the transmitter
is simple. The data throughput of PPM modulation is smaller than OOK modulation,
but the required receive power is just:
1
L
2
log
2
L
of OOK modulation at the same error rate performance. It means that PPM could trans-
mit longer distance than OOK at the same transmitting power condition.
If P represent the smallest pulse width, the comparison of different modulation techniques,
taking into account the Maximumrate, the Transmit power and the Complexity of modulation
is shown in Table 3.
6.2 Adaptation to underwater channel
As well detailed in theoretical studies (Jaruwatanadilok, 2008) (Giles & Bankman, 2005)
(Smart, 2005) optical underwater transmission is deeply inuenced by turbidity and the re-
ceived power is highly dependent by the wavelenght of transmitted signal and the dimen-
sions of the particles and substances dissolved in water.
To overcome this problemthe use of different wavelenghts for transmission can improve SNR.
By using visible optical communication, this results in having different colours which can be
used to generate the transmitted signal. This solution has been successfully explored in (Ito
et al., 2008a) where it is shown, by comparing theoretical and experimental results, that the
better wavelength lies around 420 nm (blue) and increase in presence of turbidity (Chancey,
2007).
The design of the PHY Layer should take into account the possibility of switching from one
colour to another considering underwater conditions; the design can be based on different
parameter such as the evaluation of decrease or improvement of SNR or comparison between
different transmission lines. By using a system equipped with different wavelength LEDs for
transmission it could be possible to evaluate in parallel the performance of each transmission
line (Diana & Kahn, 1999). The modules described in the following paragraphs, as for the
channels in radio based transmission, allow the management of different transmission lines
by using an ad-hoc parameter. Currently the evaluation of an automatic system is under
evaluation taking into account the previous reported works.
The transmission system has to consider also the adaptation of the transmission date-rate
which can be motivated both by the evaluation of underwater channel conditions and by the
information which has to be transmitted. If a PPM modulation is used for the transmission
the impulse duration has also a strong inuence on the power comsuption since power con-
sumption increase when LEDs for transmission are activated.
An approximate evaluation of power consumption (Energy/bit) can be based on the following
formula:
E = P t
pulse
pulses/bit (10)
with P the input power to the LED, t
pulse
the time for each pulse and taking into account the
number of pulses per bits.
6.3 PHY Layer HDL modules implementation
The PHY Layer has been completely described by using an Hardware Description Language
(HDL), while a small subset of functions of the MAC Layer for testing purposes have been
implemented by a software interface. The description of each module has been carried out by
using an Hardware Description Language because:
1. it could easily integrated in WSNnodes 1 developed for terrestrial application equipped
by a Field Programmable Gate Array (FPGA), such as the one developed by the WISI
Laboratory at DIBE (WiseLab, 2010);
2. it could be used for the implementation of an ASIC/ASIP specically devoted to man-
age the transmission and reception of data by using the optical. In this case the im-
plementation of a prototype on FPGA is to be considered as a passage for simulation
and optimization whose aim is then to build a specic CHIP for the management of the
optical communication.
The used HDL Language is VHDL (Very High Speed Hardware Description Language). It is
dened in IEEE as a tool of creation of electronics systembecause it supports the development,
verication, synthesis and testing of hardware design, the communication of hardware design
data and the maintenance, modication and procurement of hardware.
The here proposed work has the aim to implement and design modules for the management
of optical communication in a Underwater Wireless Sensor Network (UWSN), targeting the
interface with current terrestrial technologies (in particular those based on IEEE 802.15.4).
The functions of the Transmitter are in synthesis:
1. the generation of a synchronization signal which is used by the transmitter in order to
synchronize the impulse duration which can dened by the user;
2. a transmission signal based on PPMmodulation (4 or 16 PPM)in which bits are encoded
by the position of the light pulse in time slots.
The synchronization is performed by alternating 32 presences and absences of a pulse in con-
secutive time slots, allowing the receiver to calculate the pulse duration chosen for the follow-
ing transmission of data.
The choice of PPM is based on the evaluation of the performance of different modulation
schemes for underwater optical wireless communication illustrated in the previous chapter.
Although most reported underwater communication systems use OOKmodulation technique
because of its simplicity for implementation, the proposed results argue that OOK has the
disadvantages in power efciency and control capacity of the error rate for underwater optical
OOK FSK DPSK 4-PPM 8-PPM
Maximum rate
1
2P
1
2P
1
2P
1
2P
3
8P
Transmit power Middle Higher Highest Low Lowest
Complexity of modula-
tion
Low Higher Highest Lower Lower
Table 3. Transfer rate versus implementation complexity from (Meihong, Xinsheng & Fengli,
2009)
2. the Phase Shift Key (PSK) modulator generates an in phase signal for digital 1 and an
out of phase signal for a digital 0. In this case the PSK demodulator are complex since
the different coherent demodulation are needed to compare the current phase with the
previous phase and it can be inefcient for power-constrained devices.
3. the On-Off Keying (OOK) and Pulse Position Modulation (PPM) do not use completely
the frequency or phase information, but the design of the receiver and the transmitter
is simple. The data throughput of PPM modulation is smaller than OOK modulation,
but the required receive power is just:
1
L
2
log
2
L
of OOK modulation at the same error rate performance. It means that PPM could trans-
mit longer distance than OOK at the same transmitting power condition.
If P represent the smallest pulse width, the comparison of different modulation techniques,
taking into account the Maximumrate, the Transmit power and the Complexity of modulation
is shown in Table 3.
6.2 Adaptation to underwater channel
As well detailed in theoretical studies (Jaruwatanadilok, 2008) (Giles & Bankman, 2005)
(Smart, 2005) optical underwater transmission is deeply inuenced by turbidity and the re-
ceived power is highly dependent by the wavelenght of transmitted signal and the dimen-
sions of the particles and substances dissolved in water.
To overcome this problemthe use of different wavelenghts for transmission can improve SNR.
By using visible optical communication, this results in having different colours which can be
used to generate the transmitted signal. This solution has been successfully explored in (Ito
et al., 2008a) where it is shown, by comparing theoretical and experimental results, that the
better wavelength lies around 420 nm (blue) and increase in presence of turbidity (Chancey,
2007).
The design of the PHY Layer should take into account the possibility of switching from one
colour to another considering underwater conditions; the design can be based on different
parameter such as the evaluation of decrease or improvement of SNR or comparison between
different transmission lines. By using a system equipped with different wavelength LEDs for
transmission it could be possible to evaluate in parallel the performance of each transmission
line (Diana & Kahn, 1999). The modules described in the following paragraphs, as for the
channels in radio based transmission, allow the management of different transmission lines
by using an ad-hoc parameter. Currently the evaluation of an automatic system is under
evaluation taking into account the previous reported works.
The transmission system has to consider also the adaptation of the transmission date-rate
which can be motivated both by the evaluation of underwater channel conditions and by the
information which has to be transmitted. If a PPM modulation is used for the transmission
the impulse duration has also a strong inuence on the power comsuption since power con-
sumption increase when LEDs for transmission are activated.
An approximate evaluation of power consumption (Energy/bit) can be based on the following
formula:
E = P t
pulse
pulses/bit (10)
with P the input power to the LED, t
pulse
the time for each pulse and taking into account the
number of pulses per bits.
6.3 PHY Layer HDL modules implementation
The PHY Layer has been completely described by using an Hardware Description Language
(HDL), while a small subset of functions of the MAC Layer for testing purposes have been
implemented by a software interface. The description of each module has been carried out by
using an Hardware Description Language because:
1. it could easily integrated in WSNnodes 1 developed for terrestrial application equipped
by a Field Programmable Gate Array (FPGA), such as the one developed by the WISI
Laboratory at DIBE (WiseLab, 2010);
2. it could be used for the implementation of an ASIC/ASIP specically devoted to man-
age the transmission and reception of data by using the optical. In this case the im-
plementation of a prototype on FPGA is to be considered as a passage for simulation
and optimization whose aim is then to build a specic CHIP for the management of the
optical communication.
The used HDL Language is VHDL (Very High Speed Hardware Description Language). It is
dened in IEEE as a tool of creation of electronics systembecause it supports the development,
verication, synthesis and testing of hardware design, the communication of hardware design
data and the maintenance, modication and procurement of hardware.
The here proposed work has the aim to implement and design modules for the management
of optical communication in a Underwater Wireless Sensor Network (UWSN), targeting the
interface with current terrestrial technologies (in particular those based on IEEE 802.15.4).
The functions of the Transmitter are in synthesis:
1. the generation of a synchronization signal which is used by the transmitter in order to
synchronize the impulse duration which can dened by the user;
2. a transmission signal based on PPMmodulation (4 or 16 PPM)in which bits are encoded
by the position of the light pulse in time slots.
The synchronization is performed by alternating 32 presences and absences of a pulse in con-
secutive time slots, allowing the receiver to calculate the pulse duration chosen for the follow-
ing transmission of data.
The choice of PPM is based on the evaluation of the performance of different modulation
schemes for underwater optical wireless communication illustrated in the previous chapter.
Although most reported underwater communication systems use OOKmodulation technique
because of its simplicity for implementation, the proposed results argue that OOK has the
disadvantages in power efciency and control capacity of the error rate for underwater optical
channel. DPSK has good error control capability and high bandwidth, but it consumes large
power and it is more complex to implement in embedded devices.
The frame format is organized as described in 4. As usual it is composed by different elds
which have the following functions:
1. a Preamble composed by:
a eld SYNC which is used to perform the synchronization, composed by 64 slots
(32 alternating impulse of a prexed duration), and SFR (Start Frame Delimiter)
to indicate the end of the synchronization.
a eld DR which is used to identify which modulation is chosen (in this case 4-
PPM or 16-PPM).
2. a Frame Length used to communicate the number of byte in the PSDU (Physical layer
Standard Data Unit) transmitted by using PPM modulation;
3. a Physical layer Standard Data Unit (PSDU) which is composed by the data transmitted
to MAC Layer;
Fig. 4. Frame Format
The Receiver has been designed to synchronize automatically with the transmitter baud rate:
it determines how many clock cycles lasts a time slot for the transmitter by counting 32 tran-
sitions of the input. This value is the output prescaler and it is used to decode the following
PPM transmission. The receiver detects the modulation (4 or 16 PPM) and performs also a
clock correction in order to maintain synchronization by using two different methods. The
rst one is based on clock modication considering the truncating error made during mean
calculation of the time slot duration. The second one modies system operations by verify-
ing that the input value is constant for the entire duration of the time slot. These methods
are carefully described in the following paragraphs. The possibility of modify the duration
of a single impulse can support the adaptation to different external circuits and the automatic
variation of impulse duration.
Two different algorithms to maintain the synchronization have been implemented, since re-
ceiver and transmitter clocks could not be perfectly syncronised:
1. the automatic correction of time slot duration during the reception of data on the basis
of the truncating error in the calculation of the time-slot period.
2. the dynamic correction by evaluating that the sampling should always be in the center
of the transmitted impulse and modifying the sampling clock if this condition is not
reached.
The Physical Layer Management Entity (PLME) provides the layer management service in-
terfaces through which layer management functions may be invoked.
In this optical PHY Layer the PLME can manage the following services:
1. PLME-ED (Energy Detection) which reply by sending informations about the detect
energy of the received signal (in this case the information can be received as input or a
default value can be set if the detection is not supported);
2. PLME-GET read settings of the PHY Layer;
3. PLME-SET modify the settings of the PHY Layer;
4. PLME-CCA performs the CCA (clear channel assessment) and send the collected infor-
mation to the MAC Layer;
5. PLME-SETTRX- STATE allows to change the internal operating state of the transceiver;
6. PLME-PPM is a specic service has been added to support the optical communication
and the choice of a transmission based on 4 or 16-PPM;
The choice of the primitives, inspired to those of RF PHY Layer, have been adapted to optical
communication in particular considering:
1. the management of the CCA (Clear Channel Assessment) which evaluates the current
state of reception and the current state of the channel: the evaluation of this parameter
is based on the state of the receiver, since, while the system is receiving, a transmission
is not allowed, or on the evaluation of a period of time previously dened if;
2. the management of modulation by adding the possibility of choosing between 4-PPM
and 16-PPM.
3. the adaptation of some attributes related to the state and function of the PHY Layer.
The Physical Data (PD) service enables the transmission and reception of PHY protocol data
units (PPDUs) across the physical channel.
It is implemented by using 3 different sub-modules: the PDTR for the transmission, the
PDREC for the reception and the PDSAP which allows coordination among internal sub-
modules and the rest of the PHY Layer by implementing FIFO buffers. In addition it is in-
terfaced to the MAC Layer. The PD manages four I/O interfaces to communicate through
four FIFO buffers: two for upper layer (reading and writing) and two for PDTR and PDREC.
This PD communicates to MAC Layer by using the same primitives structure performed in
the PLME, with different codes, The operations performed by the PD does not require shared
resources since the 3 modules, which have been easily optimized, can work in parallel.
The implemented PHY Layer are interfaced to the MAC modules described in the next para-
graph.
7. The MAC Layer
7.1 Mac Layer Design Consideration
The previous described optical PHY Layer can be interfaced with an appropriate Multiple Ac-
cess Control (MAC) module for providing addressing and channel access control mechanisms
that make it possible for several network nodes to communicate within a multi-point network.
As detailed in advance the idea is to support the interface with current terrestrial technologies
developed for Wireless Sensor Networks based on IEEE802.15.4.
channel. DPSK has good error control capability and high bandwidth, but it consumes large
power and it is more complex to implement in embedded devices.
The frame format is organized as described in 4. As usual it is composed by different elds
which have the following functions:
1. a Preamble composed by:
a eld SYNC which is used to perform the synchronization, composed by 64 slots
(32 alternating impulse of a prexed duration), and SFR (Start Frame Delimiter)
to indicate the end of the synchronization.
a eld DR which is used to identify which modulation is chosen (in this case 4-
PPM or 16-PPM).
2. a Frame Length used to communicate the number of byte in the PSDU (Physical layer
Standard Data Unit) transmitted by using PPM modulation;
3. a Physical layer Standard Data Unit (PSDU) which is composed by the data transmitted
to MAC Layer;
Fig. 4. Frame Format
The Receiver has been designed to synchronize automatically with the transmitter baud rate:
it determines how many clock cycles lasts a time slot for the transmitter by counting 32 tran-
sitions of the input. This value is the output prescaler and it is used to decode the following
PPM transmission. The receiver detects the modulation (4 or 16 PPM) and performs also a
clock correction in order to maintain synchronization by using two different methods. The
rst one is based on clock modication considering the truncating error made during mean
calculation of the time slot duration. The second one modies system operations by verify-
ing that the input value is constant for the entire duration of the time slot. These methods
are carefully described in the following paragraphs. The possibility of modify the duration
of a single impulse can support the adaptation to different external circuits and the automatic
variation of impulse duration.
Two different algorithms to maintain the synchronization have been implemented, since re-
ceiver and transmitter clocks could not be perfectly syncronised:
1. the automatic correction of time slot duration during the reception of data on the basis
of the truncating error in the calculation of the time-slot period.
2. the dynamic correction by evaluating that the sampling should always be in the center
of the transmitted impulse and modifying the sampling clock if this condition is not
reached.
The Physical Layer Management Entity (PLME) provides the layer management service in-
terfaces through which layer management functions may be invoked.
In this optical PHY Layer the PLME can manage the following services:
1. PLME-ED (Energy Detection) which reply by sending informations about the detect
energy of the received signal (in this case the information can be received as input or a
default value can be set if the detection is not supported);
2. PLME-GET read settings of the PHY Layer;
3. PLME-SET modify the settings of the PHY Layer;
4. PLME-CCA performs the CCA (clear channel assessment) and send the collected infor-
mation to the MAC Layer;
5. PLME-SETTRX- STATE allows to change the internal operating state of the transceiver;
6. PLME-PPM is a specic service has been added to support the optical communication
and the choice of a transmission based on 4 or 16-PPM;
The choice of the primitives, inspired to those of RF PHY Layer, have been adapted to optical
communication in particular considering:
1. the management of the CCA (Clear Channel Assessment) which evaluates the current
state of reception and the current state of the channel: the evaluation of this parameter
is based on the state of the receiver, since, while the system is receiving, a transmission
is not allowed, or on the evaluation of a period of time previously dened if;
2. the management of modulation by adding the possibility of choosing between 4-PPM
and 16-PPM.
3. the adaptation of some attributes related to the state and function of the PHY Layer.
The Physical Data (PD) service enables the transmission and reception of PHY protocol data
units (PPDUs) across the physical channel.
It is implemented by using 3 different sub-modules: the PDTR for the transmission, the
PDREC for the reception and the PDSAP which allows coordination among internal sub-
modules and the rest of the PHY Layer by implementing FIFO buffers. In addition it is in-
terfaced to the MAC Layer. The PD manages four I/O interfaces to communicate through
four FIFO buffers: two for upper layer (reading and writing) and two for PDTR and PDREC.
This PD communicates to MAC Layer by using the same primitives structure performed in
the PLME, with different codes, The operations performed by the PD does not require shared
resources since the 3 modules, which have been easily optimized, can work in parallel.
The implemented PHY Layer are interfaced to the MAC modules described in the next para-
graph.
7. The MAC Layer
7.1 Mac Layer Design Consideration
The previous described optical PHY Layer can be interfaced with an appropriate Multiple Ac-
cess Control (MAC) module for providing addressing and channel access control mechanisms
that make it possible for several network nodes to communicate within a multi-point network.
As detailed in advance the idea is to support the interface with current terrestrial technologies
developed for Wireless Sensor Networks based on IEEE802.15.4.
In this paragraph some considerations for the design of an hardware friendly MAC Layer are
proposed taking into account the previous implemented optical PHY Layer and the current
terrestrial technologies for WSN.
The current IEEE 802.15.4 standard proposes different MAC frame format (Beacon Frame for-
mat, Data and Acknowledgment frame format, MAC command frame format). The possi-
bility of choosing the duration of the Mac Protocol Data Unit (MSPU) in the structure of the
PHY Layer can easily allows the adaptation to the different format. The creation of the MSPU
can be performed by using the hardware modules similar to those described in the previous
paragraph (for instance in the description of the PD module).
The MAC Layer hardware implementation can be based on a modular approach similar to
that used for PHY Layer implementation. The MAC Layer handles the access to the PHY
Layer through the PD-SAP and the PLME-SAP, while the MAC Management Entity (MLME)
should manage the different services for the Upper Layers, similarly to the PLME in the PHY
Layer.
The use of optical communication allows to overcome the problems related to low propaga-
tion speed of acoustical communication which required complex access algorithms.
Considering power consuption, it is unlike to have a continuous communication between
nodes and the communication activity is not regular between the node. Agood choice appears
to be a competition for channel access based on a CSMA/CAprotocol in which a device, when
it wishes to transmit data, waits for a random number of back-off periods before sensing the
channel. If the channel is busy, the device increases the number of attempts by one and checks
if the maximum number of attempts has been reached. If the limit is exceeded, the device
generates a channel access error and reports this event to upper layers. If the number of
attempts is below the limit, the device reiterates this procedures until it either captures the
channel successfully or the number of attempts exceeds the limit.
In particular, as concerns the parameters specically related to the optical communication two
main aspects will be considered:
1. the possibility of changing the wavelength of the emitter, by choosing different colors
(blue, red or green) if water turbidity varies. It can be implemented, for instance, by
using a mechanism such as the one described in (Ito et al., 2008b) by managing the
corresponding parameters in the PHY Layer;
2. the possibility of choosing between a 4 or 16 PPM;
3. the support of an higher data rate in comparison with the Radio based technologies
for WSN. For instance, the IEEE 802.15.4 can achieve 250 Kb/s and the implementation
of the Optical Management modules on a dedicated FPGA or by using an ASIP/ASIC
should target the synchronization with the Upper Layers of the node.
If the transmission and reception of data should be encrypted for reasons of security, and
encryption module can be easily integrated in the design as suggested in (Dasalukunte &
Owall, 2008).
7.2 MAC Layer HDL implemented modules
The MAC Layer hardware design is based on a modular approach similar to that used for
PHY Layer implementation.
The MAC Layer handles the access to the PHY Layer through the PD and the PLME, while it
manages different services for the Upper Layers, similarly to the PLME in the PHY Layer.
Currently the following functions have been implemented:
Fig. 5. CSMA-CA implementation
1. Packet parsing and addresses verication;
2. Cyclic Redundancy Check (CRC): to check the integrity of transferred data;
3. a module to manage the transmission and reception of the MAC Payload to the upper
layer;
4. the CSMA/CA mechanism for channel access.
The previous described CSMA/CAmechanismhas been implemented in hardware by using a
dedicated module composed by a randomnumber generator, for the calculation of the random
delay, a counter and an other module, based on a FSM, to manage the CSMA/CA algorithm
as depicted in Figure 5.
8. Implementation Results for PHY and MAC Layer
As a rst step, the previous described modules have been synthesized and implemented on a
Xilinx Spartan-3 FPGAin order to use a previously implemented testbed (Anguita et al., 2010).
The second step, reported in this paper, has been the implementation an optimization of the
previously described modules on an Actel IGLOO low-power FPGA in order to integrate
the system in a WSN node (WiseLab, 2010)(Figure 1). A dedicated program in C# has been
integrated from the previous version (Anguita et al., 2010) in order to manage the interface
and to perform tests of transmission by setting different parameters of the PHY and MAC
Layer (Figure 6).
Fig. 6. User interface to optical PHY and MAC Layer
In Table 4 the results of the implementation on Actel IGLOO AGL250 are reported in terms of
core cells. The maximum frequency that can be achieved by the system is 20 MHz which can
support a transmission in the order of Mb/s.
In this paragraph some considerations for the design of an hardware friendly MAC Layer are
proposed taking into account the previous implemented optical PHY Layer and the current
terrestrial technologies for WSN.
The current IEEE 802.15.4 standard proposes different MAC frame format (Beacon Frame for-
mat, Data and Acknowledgment frame format, MAC command frame format). The possi-
bility of choosing the duration of the Mac Protocol Data Unit (MSPU) in the structure of the
PHY Layer can easily allows the adaptation to the different format. The creation of the MSPU
can be performed by using the hardware modules similar to those described in the previous
paragraph (for instance in the description of the PD module).
The MAC Layer hardware implementation can be based on a modular approach similar to
that used for PHY Layer implementation. The MAC Layer handles the access to the PHY
Layer through the PD-SAP and the PLME-SAP, while the MAC Management Entity (MLME)
should manage the different services for the Upper Layers, similarly to the PLME in the PHY
Layer.
The use of optical communication allows to overcome the problems related to low propaga-
tion speed of acoustical communication which required complex access algorithms.
Considering power consuption, it is unlike to have a continuous communication between
nodes and the communication activity is not regular between the node. Agood choice appears
to be a competition for channel access based on a CSMA/CAprotocol in which a device, when
it wishes to transmit data, waits for a random number of back-off periods before sensing the
channel. If the channel is busy, the device increases the number of attempts by one and checks
if the maximum number of attempts has been reached. If the limit is exceeded, the device
generates a channel access error and reports this event to upper layers. If the number of
attempts is below the limit, the device reiterates this procedures until it either captures the
channel successfully or the number of attempts exceeds the limit.
In particular, as concerns the parameters specically related to the optical communication two
main aspects will be considered:
1. the possibility of changing the wavelength of the emitter, by choosing different colors
(blue, red or green) if water turbidity varies. It can be implemented, for instance, by
using a mechanism such as the one described in (Ito et al., 2008b) by managing the
corresponding parameters in the PHY Layer;
2. the possibility of choosing between a 4 or 16 PPM;
3. the support of an higher data rate in comparison with the Radio based technologies
for WSN. For instance, the IEEE 802.15.4 can achieve 250 Kb/s and the implementation
of the Optical Management modules on a dedicated FPGA or by using an ASIP/ASIC
should target the synchronization with the Upper Layers of the node.
If the transmission and reception of data should be encrypted for reasons of security, and
encryption module can be easily integrated in the design as suggested in (Dasalukunte &
Owall, 2008).
7.2 MAC Layer HDL implemented modules
The MAC Layer hardware design is based on a modular approach similar to that used for
PHY Layer implementation.
The MAC Layer handles the access to the PHY Layer through the PD and the PLME, while it
manages different services for the Upper Layers, similarly to the PLME in the PHY Layer.
Currently the following functions have been implemented:
Fig. 5. CSMA-CA implementation
1. Packet parsing and addresses verication;
2. Cyclic Redundancy Check (CRC): to check the integrity of transferred data;
3. a module to manage the transmission and reception of the MAC Payload to the upper
layer;
4. the CSMA/CA mechanism for channel access.
The previous described CSMA/CAmechanismhas been implemented in hardware by using a
dedicated module composed by a randomnumber generator, for the calculation of the random
delay, a counter and an other module, based on a FSM, to manage the CSMA/CA algorithm
as depicted in Figure 5.
8. Implementation Results for PHY and MAC Layer
As a rst step, the previous described modules have been synthesized and implemented on a
Xilinx Spartan-3 FPGAin order to use a previously implemented testbed (Anguita et al., 2010).
The second step, reported in this paper, has been the implementation an optimization of the
previously described modules on an Actel IGLOO low-power FPGA in order to integrate
the system in a WSN node (WiseLab, 2010)(Figure 1). A dedicated program in C# has been
integrated from the previous version (Anguita et al., 2010) in order to manage the interface
and to perform tests of transmission by setting different parameters of the PHY and MAC
Layer (Figure 6).
Fig. 6. User interface to optical PHY and MAC Layer
In Table 4 the results of the implementation on Actel IGLOO AGL250 are reported in terms of
core cells. The maximum frequency that can be achieved by the system is 20 MHz which can
support a transmission in the order of Mb/s.
Module Core Cells %
PHY LAYER 2405 40
Trasmitter 347 6
Receiver 735 12
PLME 442 7
PD 881 15
MAC Layer 962 16
CSMA-CA 601 10
Packet Parsing 238 4
Others 123 2
Table 4. Implementation on IGLOO Devices (AGLN250)
Wavelength of
emitted light
Photodiode
Sensitivity
Absorption Coef-
cient
450nm (blue) 0.37 A/W 9.2 x 10
5
cm
1
532nm (green) 0.39 A/W 4.4 x 10
4
cm
1
650nm (red) 0.42 A/W 3.4 x 10
3
cm
1
890nm (infrared) 0.63 A/W 6.0 x 10
2
cm
1
Table 5. Sensitivity for PDB-S5971 High-speed photodiodes and absorption coefcients
9. Wireless optical communication circuits
The implementation of an efcient diffuse underwater communication require the implemen-
tation of circuits for transmission and reception of light impulse. Considering the application
of the communication system in a USWN the focus should be on using low-cost and low-
power components targeting a low/medium distance up to 30 m. They are the basis to imple-
ment a diffuse optical communication which could support a dense network of small nodes
able to exchange data at high data-rate (in the order of Mb/s).
As regards the transceiver components, a careful evaluation should be performed. For the
transmitter the use of LED seems to be more interesting for low cost and low power applica-
tion, offering small, exible, cheap devices which are easy to array. For the receiver, the use of
PIN-photodiodes could be attractive taking into account their fast response, but also the use
of photomultiplier tubes (PMTs) and avalanche photodiodes (APDs) could be consider.
A presentation of both point-to-point than planar diffuse optical circuits, which have been
implemented and tested, is proposed.
9.1 Point-to-point communication
The transmitter generates an impulse of light of a xed duration (250 ns) which allows to
support a transmission at 1Mb/s in the case of an 16-PPM modulation or at 2Mb/s in the case
of a 4-PPM. The choice of LED wavelength has been done to maximize the power of received
signal as illustrated below (Figure 8).
Particular attention has been posed on the circuit for the reception, since the reciprocal dis-
tance of the underwater devices cannot be xed in advance and the receiver has to maintain
his functioning in all the coverage area of the transmitter.
The receiver has been designed considering different blocks: a photodiode; a transresistance
amplier, to have a conversion from current to voltage; a bandpass lter to eliminate noise
Fig. 7. Experimental set-up for point-to-point tests (a) - Bidirectional Transceiver (b)
below 10 kHz and above 20 MHz; an Automatic Gain Control (AGC), based on a Linear Tech-
nology LT1006, used to amplify the signal received by the rst part of the circuit and to au-
tomatically increase or decrease the gain according to the signal amplitude; a comparator, to
determine the output value by xing a threshold.
The receiver has been implemented by using the following component: Si PIN photodiode
Hamamatsu S5971 - high-speed photodiodes, with 1mm
2
surface area. To evaluate the better
wavelength for the transmitter Table 5 has been compiled considering the absorption coef-
cients for clear water and the photodiode sensitivity reported in the component datasheet.
The output current of the photodiode is proportional to:
S e
k(d)
P (11)
where S is sensitivity an P is the power in watts, d the distance and k the absorption coefcient.
Figure 8 shows how the output current varies according to distance for different wavelengths
of light, by using the equation 11.
Due to the severe attenuation in water, the output current for infrared is less than for
blue/green light at 10m, even if the sensitivity of the photodiode is higher for infrared. It
is possible to note that red light outperforms green light up to 1.5/2 m but blue and green are
better beyond. Taking into account the previous results, also considering that the attenuation
is strongly inuenced by turbidity, the better choice appear to be blue or green light. These
considerations are very important because the system performance is determined by the de-
tector when signal attenuation along a wireless link is considered. It is crucial for the receiver
to detect low-level optical signals maintaining a Signal-to-Noise Ratio (SNR) sufciently large
to yield an acceptable Bit Error Rate (BER).
The receiver circuit has been tested in air and in water. Tests in clear water allows to receive
correctly the transmitted sequence generated by the system (Figure 7) up to 2 meters: a more
accurate evaluation of the BER is planned for turbid water. Considering the limitation of the
available testbed, tests in air been performed at different distances: a reception of the trans-
mitted sequence is possible starting from few centimeters, since the AGC avoids saturation,
up to 10 meters, while below light impulses are not clearly detected. Even if the AGC stage
has to be modied to allow the reception in case of higher distances, considering that our
Module Core Cells %
PHY LAYER 2405 40
Trasmitter 347 6
Receiver 735 12
PLME 442 7
PD 881 15
MAC Layer 962 16
CSMA-CA 601 10
Packet Parsing 238 4
Others 123 2
Table 4. Implementation on IGLOO Devices (AGLN250)
Wavelength of
emitted light
Photodiode
Sensitivity
Absorption Coef-
cient
450nm (blue) 0.37 A/W 9.2 x 10
5
cm
1
532nm (green) 0.39 A/W 4.4 x 10
4
cm
1
650nm (red) 0.42 A/W 3.4 x 10
3
cm
1
890nm (infrared) 0.63 A/W 6.0 x 10
2
cm
1
Table 5. Sensitivity for PDB-S5971 High-speed photodiodes and absorption coefcients
9. Wireless optical communication circuits
The implementation of an efcient diffuse underwater communication require the implemen-
tation of circuits for transmission and reception of light impulse. Considering the application
of the communication system in a USWN the focus should be on using low-cost and low-
power components targeting a low/medium distance up to 30 m. They are the basis to imple-
ment a diffuse optical communication which could support a dense network of small nodes
able to exchange data at high data-rate (in the order of Mb/s).
As regards the transceiver components, a careful evaluation should be performed. For the
transmitter the use of LED seems to be more interesting for low cost and low power applica-
tion, offering small, exible, cheap devices which are easy to array. For the receiver, the use of
PIN-photodiodes could be attractive taking into account their fast response, but also the use
of photomultiplier tubes (PMTs) and avalanche photodiodes (APDs) could be consider.
A presentation of both point-to-point than planar diffuse optical circuits, which have been
implemented and tested, is proposed.
9.1 Point-to-point communication
The transmitter generates an impulse of light of a xed duration (250 ns) which allows to
support a transmission at 1Mb/s in the case of an 16-PPM modulation or at 2Mb/s in the case
of a 4-PPM. The choice of LED wavelength has been done to maximize the power of received
signal as illustrated below (Figure 8).
Particular attention has been posed on the circuit for the reception, since the reciprocal dis-
tance of the underwater devices cannot be xed in advance and the receiver has to maintain
his functioning in all the coverage area of the transmitter.
The receiver has been designed considering different blocks: a photodiode; a transresistance
amplier, to have a conversion from current to voltage; a bandpass lter to eliminate noise
Fig. 7. Experimental set-up for point-to-point tests (a) - Bidirectional Transceiver (b)
below 10 kHz and above 20 MHz; an Automatic Gain Control (AGC), based on a Linear Tech-
nology LT1006, used to amplify the signal received by the rst part of the circuit and to au-
tomatically increase or decrease the gain according to the signal amplitude; a comparator, to
determine the output value by xing a threshold.
The receiver has been implemented by using the following component: Si PIN photodiode
Hamamatsu S5971 - high-speed photodiodes, with 1mm
2
surface area. To evaluate the better
wavelength for the transmitter Table 5 has been compiled considering the absorption coef-
cients for clear water and the photodiode sensitivity reported in the component datasheet.
The output current of the photodiode is proportional to:
S e
k(d)
P (11)
where S is sensitivity an P is the power in watts, d the distance and k the absorption coefcient.
Figure 8 shows how the output current varies according to distance for different wavelengths
of light, by using the equation 11.
Due to the severe attenuation in water, the output current for infrared is less than for
blue/green light at 10m, even if the sensitivity of the photodiode is higher for infrared. It
is possible to note that red light outperforms green light up to 1.5/2 m but blue and green are
better beyond. Taking into account the previous results, also considering that the attenuation
is strongly inuenced by turbidity, the better choice appear to be blue or green light. These
considerations are very important because the system performance is determined by the de-
tector when signal attenuation along a wireless link is considered. It is crucial for the receiver
to detect low-level optical signals maintaining a Signal-to-Noise Ratio (SNR) sufciently large
to yield an acceptable Bit Error Rate (BER).
The receiver circuit has been tested in air and in water. Tests in clear water allows to receive
correctly the transmitted sequence generated by the system (Figure 7) up to 2 meters: a more
accurate evaluation of the BER is planned for turbid water. Considering the limitation of the
available testbed, tests in air been performed at different distances: a reception of the trans-
mitted sequence is possible starting from few centimeters, since the AGC avoids saturation,
up to 10 meters, while below light impulses are not clearly detected. Even if the AGC stage
has to be modied to allow the reception in case of higher distances, considering that our
Fig. 8. Comparison between different wavelengths
target is up to 10-15 meters underwater, the tested circuit is a good starting point for further
improvements.
The cost of the components is less than 30 euros, very cheap in comparison with some acoustic
modem.
10. Design Consideration and Implementation of Planar Optical Circuits
The possibility of targeting the connection between more than two devices leads to the neces-
sity of implementing a directional, up to omni-directional, transceiver able to send and detect
optical impulse. Taking into account some previous works (Akella et al., 2005), in this para-
graph some considerations regarding the design and implementation of a circular transceiver
are reported with a preliminary implementation description.
Fig. 9. Transmitter disposed to cover a 2-D circular area with (a) and without overlap (b)
In particular the collocation of components (LED or photodiodes) on a 2D structure is con-
sidered. Assuming that n transmitter are placed at equal distance gaps on the circular node
(radius r), considering that the diameter of a transmitter is 2:
=
2r 2n
n
(12)
The angular difference between any two neighboring transceiver is given as:
= 360
0

2r
(13)
The coverage area L of a single transmitter can be given by (Akella et al., 2005):
L = R
2
tan() +0.5Rtan()
2
(14)
Two cases can happen for the effective coverage area C of a single transmitter, based on the
value of , , R, and r:
1. if coverage area of the neighbor transmitter do not overlap (Figure 9):
Rtan() (R + r)tan(0.5) (15)
In this case, the effective coverage area is equivalent to the coverage area, i.e. C=L.
2. Coverage area of the neighbor transmitter overlap (Figure 9):
Rtan() > (R + r)tan(0.5) (16)
In this case, the effective coverage area is equivalent to the coverage area excluding
the area that interferes with the neighbor transceiver. If I is the interference area that
overlaps with the neighbor transmitters coverage, then C = L-I. Considering that the
target of our work is to have a directional or at least omni-directional a good design
approach should minimize the interfere area.
If the target is to have a transmission and reception of data by using all the elements placed
on the circular structure, the overlap is not a problem for the communication. But, since the
nal idea is to add the possibility of supporting also a directional communication in which a
single LED could be activated, the idea is to minimize the interference between adjacent LEDs
and receivers.
Taking into account the previous circuits, a 2-d transceiver has been implemented. The LEDs
disposition on a 2-d structure has been performed using 12 Ledman LL1503PLBL1-301 blue
LED (with 30
0
of FOV) disposed on a circular disk of 10 cm diameter. The reduction of the
overlap between the signal generated by each LEDhas been targeted so to allowthe possibility
of supporting also a directional transmission, by using only one or a reduced set of LEDs. The
same approach has been considered for the placement of the photodiodes for the 2-d receiver.
Tests of the transmitter have been carried out by using a single receiver, equipped with a
SFH-2013P photodiode. Different measures have been considered at different distances. The
prole of the received optical signal, determined by considering the maximum value in a line-
of-sight condition and the minimum value which is measured between a LED and another,
show that the generated light impulses is uniformly distributed on the surface. Tests on the
receiver have shown a performance decrease in comparison with the point-to-point circuit
due to the noise generated by the photodiodes which are not directly exposed to the light
impulses. Nevertheless a reception up to 4 meters in air can be achieved.
Fig. 8. Comparison between different wavelengths
target is up to 10-15 meters underwater, the tested circuit is a good starting point for further
improvements.
The cost of the components is less than 30 euros, very cheap in comparison with some acoustic
modem.
10. Design Consideration and Implementation of Planar Optical Circuits
The possibility of targeting the connection between more than two devices leads to the neces-
sity of implementing a directional, up to omni-directional, transceiver able to send and detect
optical impulse. Taking into account some previous works (Akella et al., 2005), in this para-
graph some considerations regarding the design and implementation of a circular transceiver
are reported with a preliminary implementation description.
Fig. 9. Transmitter disposed to cover a 2-D circular area with (a) and without overlap (b)
In particular the collocation of components (LED or photodiodes) on a 2D structure is con-
sidered. Assuming that n transmitter are placed at equal distance gaps on the circular node
(radius r), considering that the diameter of a transmitter is 2:
=
2r 2n
n
(12)
The angular difference between any two neighboring transceiver is given as:
= 360
0

2r
(13)
The coverage area L of a single transmitter can be given by (Akella et al., 2005):
L = R
2
tan() +0.5Rtan()
2
(14)
Two cases can happen for the effective coverage area C of a single transmitter, based on the
value of , , R, and r:
1. if coverage area of the neighbor transmitter do not overlap (Figure 9):
Rtan() (R + r)tan(0.5) (15)
In this case, the effective coverage area is equivalent to the coverage area, i.e. C=L.
2. Coverage area of the neighbor transmitter overlap (Figure 9):
Rtan() > (R + r)tan(0.5) (16)
In this case, the effective coverage area is equivalent to the coverage area excluding
the area that interferes with the neighbor transceiver. If I is the interference area that
overlaps with the neighbor transmitters coverage, then C = L-I. Considering that the
target of our work is to have a directional or at least omni-directional a good design
approach should minimize the interfere area.
If the target is to have a transmission and reception of data by using all the elements placed
on the circular structure, the overlap is not a problem for the communication. But, since the
nal idea is to add the possibility of supporting also a directional communication in which a
single LED could be activated, the idea is to minimize the interference between adjacent LEDs
and receivers.
Taking into account the previous circuits, a 2-d transceiver has been implemented. The LEDs
disposition on a 2-d structure has been performed using 12 Ledman LL1503PLBL1-301 blue
LED (with 30
0
of FOV) disposed on a circular disk of 10 cm diameter. The reduction of the
overlap between the signal generated by each LEDhas been targeted so to allowthe possibility
of supporting also a directional transmission, by using only one or a reduced set of LEDs. The
same approach has been considered for the placement of the photodiodes for the 2-d receiver.
Tests of the transmitter have been carried out by using a single receiver, equipped with a
SFH-2013P photodiode. Different measures have been considered at different distances. The
prole of the received optical signal, determined by considering the maximum value in a line-
of-sight condition and the minimum value which is measured between a LED and another,
show that the generated light impulses is uniformly distributed on the surface. Tests on the
receiver have shown a performance decrease in comparison with the point-to-point circuit
due to the noise generated by the photodiodes which are not directly exposed to the light
impulses. Nevertheless a reception up to 4 meters in air can be achieved.
11. Future Research Directions
Starting from the previous described results, the creation of an innovative long survival op-
tical Underwater Wireless Sensor Network (UWSN) could be targeted. It will able to sense,
compute, communicate and cooperate in an underwater environment by using long-survival,
low-cost and eventually disposable nodes. Innovative approaches could be developed to ad-
dress:
1. wireless, adaptive and low-power optical underwater communication;
2. underwater energy scavenging and harvesting;
3. autonomous and self-governing behavior of the nodes in the underwater environment,
including intelligent energy storing and utilization, to guarantee long-term operation
under a wide range of conditions and avoid frequent and costly rescue procedures.
The rst innovation will address the communication capabilities. Efcient optical communi-
cation could be implemented thank to a low-cost, efcient and omni-directional communica-
tion system, based on short wavelength LEDs, and able to exploit the minimum absorption
wavelength window, shifting toward longer wavelengths in turbid waters. The optical com-
munication, despite reaching shorter distances respect to acoustic communication, will allow
to achieve high data rates with lower energy requirements.
Fig. 10. optical UWSN Concept
Finally, the implementation of adaptive directionality, by activating only some of the transduc-
ers will allow to spare energy and optimize the communication efciency. Online nonlinear
modeling and adaptation to the time-varying aquatic channel characteristics of the optical sys-
tem can be used to advance the state-of-the-art and achieve effective and efcient Free Space
Omni-directional Optical (FSOO) underwater communication.
The second innovative eld of research could be underwater energy scavenging for self-
supply or for increasing the lifetime of both each node and the entire network. While solar
power can be exploited on the water surface and in shallow but clear waters, other techniques
must be explored for producing and storing energy in a general underwater environment.
Terrestrial energy scavenging for articial artifacts and autonomous sensors relies mainly on
exploiting environmental vibrations, converting mechanical energy in electrical energy. In the
static underwater nodes, random movements forced by the water ow, underwater currents
and noise, can be exploited to perform underwater energy scavenging (D.Zhu, 2010). The
delicate equilibrium between energy harvesting, storage and consumption is to addressed by
developing adaptive behavior, to optimize the survivability of both the nodes and the entire
network.
The third eld of research should address the development of low-power, low-cost miniatur-
ized nodes able to sense, compute, communicate and cooperate in an aquatic environment.
The development of new nodes, with volumes that are orders of magnitude smaller than cur-
rent generation equipments, will allow the development of new applications, where tiny and,
eventually, disposable nodes will be able to perform 4D monitoring of aquatic environments.
The increased density of the network, which will be possible thank to the miniaturization and
low-cost of the nodes, will allow to compensate for the relatively short range of the under-
water optical communication channel, which is, as shown, well below 100m. Furthermore,
in settings where environmental issues are of limited concern, the use of low-cost disposable
nodes will avoid frequent and costly rescue procedures, by simply adding new nodes to areas
not covered by the network or to overcome the malfunctioning of old nodes.
The development of the previous described elds of research could lead to the implementation
of low-cost optically communicating nodes, able to be deployed with lowaccuracy on the area
of interest, and capable of self-conguring as a sensing network as depicted in Figure 10.
12. References
Akella, J., Liu, C., Partyka, D., Yuksel, M., Kalyanaraman, S. & Dutta, P. (2005). Building
blocks for mobile free-space-optical networks, Second IFIP International Conference on
Wireless and Optical Communications Networks, 2005. WOCN 2005., pp. 164 168.
Akyildiz, I. F., Pompili, D. & Melodia, T. (2005). Underwater acoustic sensor networks: re-
search challenges, Ad Hoc Networks 3(3): 257 279.
Akyldiz, I. F., Su, W., Sankarasubramaniam, Y. & Cayirci, E. (2001). A survey on sensor net-
work.
Al-Shammaa, A., Shaw, A. & Saman, S. (2004). Propagation of electromagnetic waves at
mhz frequencies through seawater, Antennas and Propagation, IEEE Transactions on
52(11): 28432849.
Ambalux (2008). http://www.ambalux.com.
Anguita, D., Brizzolara, D., Ghio, A. & Parodi, G. (2008). Smart plankton: a nature inspired
underwater wireless sensor network, Natural Computation, 2008. ICNC 08. Fourth In-
ternational Conference on 7: 701 705.
Anguita, D., Brizzolara, D. & Parodi, G. (2009). Building an underwater wireless sensor net-
work based on optical communication: Research challenges and current results, Sen-
sor Technologies and Applications, 2009. SENSORCOMM 09. Third International Confer-
ence on pp. 476 479.
Anguita, D., Brizzolara, D. & Parodi, G. (2010). Design and implementation of hdl mod-
ules and circuits for underwater optical wireless communication, Proceedings of the
9th WSEAS International Conference on TELECOMMUNICATIONS and INFORMAT-
ICS pp. 132 137.
11. Future Research Directions
Starting from the previous described results, the creation of an innovative long survival op-
tical Underwater Wireless Sensor Network (UWSN) could be targeted. It will able to sense,
compute, communicate and cooperate in an underwater environment by using long-survival,
low-cost and eventually disposable nodes. Innovative approaches could be developed to ad-
dress:
1. wireless, adaptive and low-power optical underwater communication;
2. underwater energy scavenging and harvesting;
3. autonomous and self-governing behavior of the nodes in the underwater environment,
including intelligent energy storing and utilization, to guarantee long-term operation
under a wide range of conditions and avoid frequent and costly rescue procedures.
The rst innovation will address the communication capabilities. Efcient optical communi-
cation could be implemented thank to a low-cost, efcient and omni-directional communica-
tion system, based on short wavelength LEDs, and able to exploit the minimum absorption
wavelength window, shifting toward longer wavelengths in turbid waters. The optical com-
munication, despite reaching shorter distances respect to acoustic communication, will allow
to achieve high data rates with lower energy requirements.
Fig. 10. optical UWSN Concept
Finally, the implementation of adaptive directionality, by activating only some of the transduc-
ers will allow to spare energy and optimize the communication efciency. Online nonlinear
modeling and adaptation to the time-varying aquatic channel characteristics of the optical sys-
tem can be used to advance the state-of-the-art and achieve effective and efcient Free Space
Omni-directional Optical (FSOO) underwater communication.
The second innovative eld of research could be underwater energy scavenging for self-
supply or for increasing the lifetime of both each node and the entire network. While solar
power can be exploited on the water surface and in shallow but clear waters, other techniques
must be explored for producing and storing energy in a general underwater environment.
Terrestrial energy scavenging for articial artifacts and autonomous sensors relies mainly on
exploiting environmental vibrations, converting mechanical energy in electrical energy. In the
static underwater nodes, random movements forced by the water ow, underwater currents
and noise, can be exploited to perform underwater energy scavenging (D.Zhu, 2010). The
delicate equilibrium between energy harvesting, storage and consumption is to addressed by
developing adaptive behavior, to optimize the survivability of both the nodes and the entire
network.
The third eld of research should address the development of low-power, low-cost miniatur-
ized nodes able to sense, compute, communicate and cooperate in an aquatic environment.
The development of new nodes, with volumes that are orders of magnitude smaller than cur-
rent generation equipments, will allow the development of new applications, where tiny and,
eventually, disposable nodes will be able to perform 4D monitoring of aquatic environments.
The increased density of the network, which will be possible thank to the miniaturization and
low-cost of the nodes, will allow to compensate for the relatively short range of the under-
water optical communication channel, which is, as shown, well below 100m. Furthermore,
in settings where environmental issues are of limited concern, the use of low-cost disposable
nodes will avoid frequent and costly rescue procedures, by simply adding new nodes to areas
not covered by the network or to overcome the malfunctioning of old nodes.
The development of the previous described elds of research could lead to the implementation
of low-cost optically communicating nodes, able to be deployed with lowaccuracy on the area
of interest, and capable of self-conguring as a sensing network as depicted in Figure 10.
12. References
Akella, J., Liu, C., Partyka, D., Yuksel, M., Kalyanaraman, S. & Dutta, P. (2005). Building
blocks for mobile free-space-optical networks, Second IFIP International Conference on
Wireless and Optical Communications Networks, 2005. WOCN 2005., pp. 164 168.
Akyildiz, I. F., Pompili, D. & Melodia, T. (2005). Underwater acoustic sensor networks: re-
search challenges, Ad Hoc Networks 3(3): 257 279.
Akyldiz, I. F., Su, W., Sankarasubramaniam, Y. & Cayirci, E. (2001). A survey on sensor net-
work.
Al-Shammaa, A., Shaw, A. & Saman, S. (2004). Propagation of electromagnetic waves at
mhz frequencies through seawater, Antennas and Propagation, IEEE Transactions on
52(11): 28432849.
Ambalux (2008). http://www.ambalux.com.
Anguita, D., Brizzolara, D., Ghio, A. & Parodi, G. (2008). Smart plankton: a nature inspired
underwater wireless sensor network, Natural Computation, 2008. ICNC 08. Fourth In-
ternational Conference on 7: 701 705.
Anguita, D., Brizzolara, D. & Parodi, G. (2009). Building an underwater wireless sensor net-
work based on optical communication: Research challenges and current results, Sen-
sor Technologies and Applications, 2009. SENSORCOMM 09. Third International Confer-
ence on pp. 476 479.
Anguita, D., Brizzolara, D. & Parodi, G. (2010). Design and implementation of hdl mod-
ules and circuits for underwater optical wireless communication, Proceedings of the
9th WSEAS International Conference on TELECOMMUNICATIONS and INFORMAT-
ICS pp. 132 137.
Baiden, G. &Bissiri, Y. (2007). High bandwidth optical networking for underwater untethered
telerobotic operation, OCEANS 2007, pp. 19.
Baiden, G., Bissiri, Y. & Masoti, A. (2009). Paving the way for a future underwater omni-
directional wireless optical communication systems, Ocean Engineering 36(9-10): 633
640.
Chancey, M. A. (2007). Degree of master of science: Short range underwater optical commu-
nication links.
C.X.Fan, F.Zhang, Xu, B. & Wu, C. (2001). Communication theory, Defense Industry Press.
Dasalukunte, D. & Owall, V. (2008). A generic hardware mac for wireless personal area net-
work platforms, Proceedings of the 11th International Symposium on Wireless Personal
Multimedia Communication (WPMC08) .
Diana, L. & Kahn, J. (1999). Rate-adaptive modulation techniques for infrared wireless com-
munication, Proceedings of ICC 99 1: 597603.
Doniec, M., Detweiler, C., Vasilescu, I. & Rus, D. (2010). Using optical communication for
remote underwater robot operation.
Doniec, M., Vasilescu, I., Chitre, M., Detweiler, C., Hoffmann-Kuhnt, M. & Rus, D. (2009).
Aquaoptical: A lightweight device for high-rate long-range underwater point-to-
point communication, OCEANS 2009, MTS/IEEE Biloxi - Marine Technology for Our
Future: Global and Local Challenges pp. 1 6.
D.Zhu, M.J.Tudor, S. (2010). Strategies for increasing the operating frequency range of vibra-
tion energy harvesters: a review, Measurement Science and Technology .
Fair, N., Chave, A., Freitag, L., Preisig, J., White, S., Yoerger, D. &Sonnichsen, F. (2006). Optical
modem technology for seaoor observatories, OCEANS 2006 pp. 16.
Farr, N., Chave, A., Freitag, L., Preisig, J., White, S., Yoerger, D. & Titterton, P. (2005). Op-
tical modem technology for seaoor observatories, OCEANS, 2005. Proceedings of
MTS/IEEE pp. 928934 Vol. 1.
Feng Lu, Sammy Lee, J. M. C. S. (2009). Low-cost medium-range optical underwater modem,
WUWNET 2009, pp. 16.
Fibre, W. (2008). http://www.wirelessbre.co.uk/.
Ghassemlooy, Z., Popoola, W., Rajbhandari, S., Amiri, M. & Hashemi, S. (2007). A synopsis of
modulation techniques for wireless infrared communication, ICTON Mediterranean
Winter Conference, 2007. ICTON-MW 2007, pp. 16.
Giles, J. & Bankman, I. (2005). Underwater optical communications systems. part 2: ba-
sic design considerations, IEEE Military Communications Conference. MILCOM 2005.
pp. 17001705 Vol. 3.
Hanson, F. & Radic, S. (2008). High bandwidth underwater optical communication, Applied
Optics 36: 277283.
Ito, Y., Haruyama, S. & Nakagawa, M. (2008a). Short-range underwater wireless communica-
tion using visible light leds.
Ito, Y., Haruyama, S. & Nakagawa, M. (2008b). Short-range underwater wireless communica-
Jaruwatanadilok, S. (2008). Underwater wireless optical communication channel modeling
and performance evaluation using vector radiative transfer theory, Selected Areas in
Communications, IEEE Journal on 26(9): 16201627.
Kahn, J. &Barry, J. (1997). Wireless infrared communications, Proceedings of the IEEE 85(2): 265
298.
Kedar, D. (2007). Underwater sensor network using optical wireless communication, SPIE
Newsroom - The International Society for Optical Engineering .
Lanbo, L., Shengli, Z. & Jun-Hong, C. (2008). Prospects and problems of wireless communica-
tion for underwater sensor networks, Wirel. Commun. Mob. Comput. 8(8): 977994.
Lee, D. & Kahn, J. (1999). Coding and equalization for ppm on wireless infrared channels,
IEEE Transaction on Communication 47: 255260.
Lee, S., Mounzer, J., Mirza, D. & Schurgers, C. (2008). Demo abstract: Low cost, medium
range optical, communication for underwater test beds, The Third ACM International
Workshop on UnderWater Networks (WUWNet).
Liu, Y. & Ge, X. (2006). Underwater laser sensor network: A new approach for broadband
communication in the underwater, Proceedings of the 5th WSEAS International Confer-
ence on Telecommunications and Informatics, pp. 421425.
Lucas, J., Al-Shammaa, A., Seim, J., Loehr, W., Puchbauer, G. & McGregor, D. (2004). Under-
water communications using electromagnetic waves (emcomms), Vol. Proceeding of
European Conference on Marine Science and Ocean Technology.
Ma, H. (2003). Research on optical modulation and demodulation techniques in mobile atmo-
spheric laser communication, National University of Defense Technology.
Malik, D., Joseph, M. & John, R. (1996). Performance of pulse-position on measured non-
directed indoor infrared channels, IEEE Transaction on Communications 44: 173177.
Meihong, S., Xinsheng, Y. & Fengli, Z. (2009). The evaluation of modulation techniques for
underwater wireless optical communications, Communication Software and Networks,
International Conference on 0: 138142.
Meihong, S., Xinsheng, Y. & Zhangguo, Z. (2009). The modied ppm modulation for under-
water wireless optical communication, Communication Software and Networks, Interna-
tional Conference on 0: 173177.
Pang, W., Chew, W., Choong, F. & Teoh, E. (2007). Vhdl modeling of the ieee802.11b dcf mac,
Proceedings of the 6th WSEAS International Conference on Instrumentation, Measurement,
Circuits and Systems .
Park, H. & Barry, J. (1995). Modulation analysis for wireless infrared communications, Com-
munications, 1995. ICC 95 Seattle, Gateway to Globalization, 1995 IEEE International
Conference on, Vol. 2, pp. 11821186 vol.2.
Puccinelli, D. & Haenggi, M. (2005). Wireless sensor networks: applications and challenges of
ubiquitous sensing, IEEE, Circuits and Systems Magazine.
Shelley, T. (2005). Radio waves transmit information underwater, Eureka Magazine.
Shill, F., Zimmer, U. R. & Trumpi, J. (2004). Visible spectrum optical communication and
distance sensing for underwater applications.
Singh, S., Grund, M., Bingham, B., Eustice, R., Singh, H. & Freitag, L. (2006). Underwater
acoustic navigation with the whoi micro-modem, OCEANS 2006, pp. 14.
Smart, J. (2005). Underwater optical communications systems part 1: variability of water op-
tical parameters, IEEE Military Communications Conference. MILCOM 2005. pp. 1140
1146 Vol. 2.
Tivey, M., Fucile, P. & Sichel, E. (2004). A low power, low cost, underwater optical communi-
cation system, Ridge 2000 Events pp. 2729.
V. Hsu, J. M. K. & Pister, K. S. J. (1998). Wireless communications for smart dust, Electronics
Research Laboratory Technical Memorandum Number M98/2.
Baiden, G. &Bissiri, Y. (2007). High bandwidth optical networking for underwater untethered
telerobotic operation, OCEANS 2007, pp. 19.
Baiden, G., Bissiri, Y. & Masoti, A. (2009). Paving the way for a future underwater omni-
directional wireless optical communication systems, Ocean Engineering 36(9-10): 633
640.
Chancey, M. A. (2007). Degree of master of science: Short range underwater optical commu-
nication links.
C.X.Fan, F.Zhang, Xu, B. & Wu, C. (2001). Communication theory, Defense Industry Press.
Dasalukunte, D. & Owall, V. (2008). A generic hardware mac for wireless personal area net-
work platforms, Proceedings of the 11th International Symposium on Wireless Personal
Multimedia Communication (WPMC08) .
Diana, L. & Kahn, J. (1999). Rate-adaptive modulation techniques for infrared wireless com-
munication, Proceedings of ICC 99 1: 597603.
Doniec, M., Detweiler, C., Vasilescu, I. & Rus, D. (2010). Using optical communication for
remote underwater robot operation.
Doniec, M., Vasilescu, I., Chitre, M., Detweiler, C., Hoffmann-Kuhnt, M. & Rus, D. (2009).
Aquaoptical: A lightweight device for high-rate long-range underwater point-to-
point communication, OCEANS 2009, MTS/IEEE Biloxi - Marine Technology for Our
Future: Global and Local Challenges pp. 1 6.
D.Zhu, M.J.Tudor, S. (2010). Strategies for increasing the operating frequency range of vibra-
tion energy harvesters: a review, Measurement Science and Technology .
Fair, N., Chave, A., Freitag, L., Preisig, J., White, S., Yoerger, D. &Sonnichsen, F. (2006). Optical
modem technology for seaoor observatories, OCEANS 2006 pp. 16.
Farr, N., Chave, A., Freitag, L., Preisig, J., White, S., Yoerger, D. & Titterton, P. (2005). Op-
tical modem technology for seaoor observatories, OCEANS, 2005. Proceedings of
MTS/IEEE pp. 928934 Vol. 1.
Feng Lu, Sammy Lee, J. M. C. S. (2009). Low-cost medium-range optical underwater modem,
WUWNET 2009, pp. 16.
Fibre, W. (2008). http://www.wirelessbre.co.uk/.
Ghassemlooy, Z., Popoola, W., Rajbhandari, S., Amiri, M. & Hashemi, S. (2007). A synopsis of
modulation techniques for wireless infrared communication, ICTON Mediterranean
Winter Conference, 2007. ICTON-MW 2007, pp. 16.
Giles, J. & Bankman, I. (2005). Underwater optical communications systems. part 2: ba-
sic design considerations, IEEE Military Communications Conference. MILCOM 2005.
pp. 17001705 Vol. 3.
Hanson, F. & Radic, S. (2008). High bandwidth underwater optical communication, Applied
Optics 36: 277283.
Ito, Y., Haruyama, S. & Nakagawa, M. (2008a). Short-range underwater wireless communica-
Ito, Y., Haruyama, S. & Nakagawa, M. (2008b). Short-range underwater wireless communica-
Jaruwatanadilok, S. (2008). Underwater wireless optical communication channel modeling
and performance evaluation using vector radiative transfer theory, Selected Areas in
Communications, IEEE Journal on 26(9): 16201627.
Kahn, J. &Barry, J. (1997). Wireless infrared communications, Proceedings of the IEEE 85(2): 265
298.
Kedar, D. (2007). Underwater sensor network using optical wireless communication, SPIE
Newsroom - The International Society for Optical Engineering .
Lanbo, L., Shengli, Z. & Jun-Hong, C. (2008). Prospects and problems of wireless communica-
tion for underwater sensor networks, Wirel. Commun. Mob. Comput. 8(8): 977994.
Lee, D. & Kahn, J. (1999). Coding and equalization for ppm on wireless infrared channels,
IEEE Transaction on Communication 47: 255260.
Lee, S., Mounzer, J., Mirza, D. & Schurgers, C. (2008). Demo abstract: Low cost, medium
range optical, communication for underwater test beds, The Third ACM International
Workshop on UnderWater Networks (WUWNet).
Liu, Y. & Ge, X. (2006). Underwater laser sensor network: A new approach for broadband
communication in the underwater, Proceedings of the 5th WSEAS International Confer-
ence on Telecommunications and Informatics, pp. 421425.
Lucas, J., Al-Shammaa, A., Seim, J., Loehr, W., Puchbauer, G. & McGregor, D. (2004). Under-
water communications using electromagnetic waves (emcomms), Vol. Proceeding of
European Conference on Marine Science and Ocean Technology.
Ma, H. (2003). Research on optical modulation and demodulation techniques in mobile atmo-
spheric laser communication, National University of Defense Technology.
Malik, D., Joseph, M. & John, R. (1996). Performance of pulse-position on measured non-
directed indoor infrared channels, IEEE Transaction on Communications 44: 173177.
Meihong, S., Xinsheng, Y. & Fengli, Z. (2009). The evaluation of modulation techniques for
underwater wireless optical communications, Communication Software and Networks,
International Conference on 0: 138142.
Meihong, S., Xinsheng, Y. & Zhangguo, Z. (2009). The modied ppm modulation for under-
water wireless optical communication, Communication Software and Networks, Interna-
tional Conference on 0: 173177.
Pang, W., Chew, W., Choong, F. & Teoh, E. (2007). Vhdl modeling of the ieee802.11b dcf mac,
Proceedings of the 6th WSEAS International Conference on Instrumentation, Measurement,
Circuits and Systems .
Park, H. & Barry, J. (1995). Modulation analysis for wireless infrared communications, Com-
munications, 1995. ICC 95 Seattle, Gateway to Globalization, 1995 IEEE International
Conference on, Vol. 2, pp. 11821186 vol.2.
Puccinelli, D. & Haenggi, M. (2005). Wireless sensor networks: applications and challenges of
ubiquitous sensing, IEEE, Circuits and Systems Magazine.
Shelley, T. (2005). Radio waves transmit information underwater, Eureka Magazine.
Shill, F., Zimmer, U. R. & Trumpi, J. (2004). Visible spectrum optical communication and
distance sensing for underwater applications.
Singh, S., Grund, M., Bingham, B., Eustice, R., Singh, H. & Freitag, L. (2006). Underwater
acoustic navigation with the whoi micro-modem, OCEANS 2006, pp. 14.
Smart, J. (2005). Underwater optical communications systems part 1: variability of water op-
tical parameters, IEEE Military Communications Conference. MILCOM 2005. pp. 1140
1146 Vol. 2.
Tivey, M., Fucile, P. & Sichel, E. (2004). A low power, low cost, underwater optical communi-
cation system, Ridge 2000 Events pp. 2729.
V. Hsu, J. M. K. & Pister, K. S. J. (1998). Wireless communications for smart dust, Electronics
Research Laboratory Technical Memorandum Number M98/2.
Vasilescu, I., Detweiler, C. & Rus, D. (2007). Aquanodes: an underwater sensor network,
WuWNet 07: Proceedings of the second workshop on Underwater networks, ACM, New
York, NY, USA, pp. 8588.
Vasilescu, I., Kotay, K., Rus, D., Dunbabin, M. & Corke, P. (2005). Data collection, storage,
and retrieval with an underwater sensor network, SenSys 05: Proceedings of the 3rd
international conference on Embedded networked sensor systems, ACM, New York, NY,
USA, pp. 154165.
WiseLab (2010). http://www.wise-laboratory.it/test/index.htm, DIBE: Department of Biophys-
ical and Electronic Engineering .
Yan, H., Zhou, S., Shi, Z. J. & Li, B. (2007). A dsp implementation of ofdm acoustic modem,
WuWNet 07: Proceedings of the second workshop on Underwater networks, ACM, New
York, NY, USA, pp. 8992.
Estimation of Propagation Characteristics along Random Rough Surface for Sensor Networks 231
Estimation of Propagation Characteristics along Random Rough Surface
for Sensor Networks
0
Estimation of Propagation Characteristics along
Random Rough Surface for Sensor Networks
Fukuoka Institute of Technology
Japan
1. Introduction
The main focus in the development of wireless communications engineering is providing
higher data rates, using lower transmission power, and maintaining quality of services in
complicated physical environments, such as an urban area with high-rise buildings, a ran-
domly proled terrestrial ground and so on. In order to achieve these goals, there has been
substantial progress in the development of low-power circuits, digital algorithms for modula-
tion and coding, networking controls, and circuit simulators in recent years [Aryanfar,2007].
However, insufcient improvement has been made in wireless channel modeling which is one
of the most basic and signicant engineering problems corresponding to the physical layer of
the OSI model.
Recently, the sensor network technologies have attracted many researchers interest especially
in the elds of wireless communications engineering as well as in the elds of sensor engineer-
ing. The sensor devices are usually located on the terrestrial surfaces such as dessert, hilly ter-
rain, forest, sea surface and so on, of which proles are considered to be statistically random.
In this context, it is very important to investigate the propagation characteristics of electro-
magnetic waves traveling along random rough surfaces (RRSs) and construct an efcient as
well as reliable sensor network over terrestrial grounds with RRS-like proles [Uchida,2007],
[Uchida,2008], [Uchida,2009], [Honda,2010].
In the early years of our investigations, we applied the nite volume time domain (FVTD)
method to estimate the electromagnetic propagation characteristics along one-dimensional
(1D) RRSs [Honda,2006], [Uchida,2007]. The FVTD method, however, requires too much com-
puter memory and computation time to deal with relatively long RRSs necessary for a sensor
network in the realistic situation. To overcome this difculty, we have introduced the dis-
crete ray tracing method (DRTM) based on the theory of geometrical optics, and we can now
deal with considerably long RRSs in comparison with the operating wavelength. The merit
of using DRTM is that we can treat very long RRSs compared with the wavelength without
much computer memory nor computation time. Thus, the DRTM has become one of the most
powerful tools in order to numerically analyze the long-distance propagation characteristics
of electromagnetic waves traveling along RRSs [Uchida,2008], [Uchida,2009], [Honda,2010].
In this chapter, we discuss the distance characteristics of electromagnetic waves propagating
along homogeneous RRSs which are described statistically in terms of the two parameters,
that is, height deviation h and correlation length c. The distance characteristics of propa-
gation are estimated by introducing an amplitude weighting factor for eld amplitude, an
13
order for an equivalent propagation distance, and a distance correction factor . The or-
der yields an equivalent distance indicating the distance to the -th power. The order was
introduced by Hata successfully as an empirical formula for the propagation characteristics in
the urban and suburban areas [Hata,1980]. In the present formulations, we determine these
parameters numerically by using the least square method. Once these parameters are deter-
mined for one type of RRSs, we can easily estimate the radio communication distance between
two sensors distributed on RRSs, provided the input power of a source antenna and the min-
imum detectable electric eld intensity of a receiver are specied.
The contents of the present chapter are described as follows. Section 1 is the introduction of
this chapter, and the background of this research is denoted. Section 2 discusses the statistical
properties of 1D RRSs and the convolution method is introduced for RRS generation. Sec-
tion 3 discusses DRTM for evaluation of electromagnetic waves propagating along RRSs. It is
shown that the DRTM is very effective to the eld evaluation especially in a complicated en-
vironment, since it discretizes not only the terrain prole but also the procedure for searching
rays, resulting in saving much computation time and computer memory. Section 4 discusses a
numerical method to estimate propagation or path loss characteristics along 1D RRSs. An es-
timation formula for the radio communication distance along the 1D RRSs is also introduced
in this Section. Section 5 is the conclusion of this chapter, and a few comments on the near
future problems are remarked.
2. Generation of 1D random rough surface
As mentioned in the introduction, the sensor network has attracted many researchers interest
recently in different technical elds like signal processing, antennas, wave propagation, low
power circuit design and so forth, just as the same as the case of radio frequency identication
(RFID) [Heidrich,2010]. The sensor devices are usually located on terrestrial surfaces such as
dessert, hilly terrain, forest, sea surface and so on. Since these surfaces are considered to be
statistically random, it is important to study statistics of the RRSs as well as the electromag-
netic wave propagation along them in order to construct reliable and efcient sensor network
systems [Honda,2009].
In this section, we describe the statistical properties of RRSs and we show three types of spec-
tral density functions, that is, Gaussian, n-th order of power-law and exponential spectra. We
also discuss the convolution method for RRS generation. The convolution method is exi-
ble and suitable for computer simulations to attack problems related to electromagnetic wave
scattering from RRSs and electromagnetic wave propagation along RRSs.
2.1 Spectral density function and auto-correlation function
In this study, we assume that 1D RRSs extend in x-direction and it is uniform in z-direction
with its height function as denoted by y = f (x). The spectral density function W(K) for a set
of RRSs is dened by using the height function and the spatial angular frequency K as follows:
W(K)dK = h
2
(1)
where h is the standard deviation of of the height function or height deviation, and
W(K) = lim
L
1
2
1
L
L/2
L/2
f (x)e
jKx
dx
(2)
where <> indicates the ensemble average of the RRS set. As is well-known, the auto-
correlation function is given by the Fourier transform of the spectral density function as fol-
lows:
(x) =
W(K)e
jKx
dK . (3)
Now we summarize three types of spectral density functions that are useful for numerical
simulations of the propagation characteristics of electromagnetic waves traveling along RRSs.
1. Gaussian Type of Spectrum:
The spectral density function of this type is dened by
W(K) =
clh
2
2
K
2
cl
2
4
(4)
where cl is the correlation length, and the auto-correlation function is given by
(x) = h
2
e
x
2
cl
2
.
(5)
2. N-th Order Power-Law Spectrum:
The spectral density function of this type is given by
W(K) =
clh
2
2
1 +

2
(N
1
2
)
2
(N)
K
2
cl
2
4
N
(6)
where (N) is the Gamma function with N > 1, and the auto-correlation function is
given by
(x) =
h
2
[1 +
x
2
Ncl
2
]
N
.
(7)
3. Exponential Spectrum:
W(K) =
clh
2
1 + K
2
cl
2
1
(8)
and the auto-correlation function is given by
(x) = h
2
e
|x|
cl
.
(9)
2.2 Convolution method for RRS generation
As is well-known, we should not use discrete Fourier transform (DFT) but fast Fourier trans-
form (FFT) for practical applications to save computation time. For simplicity of analyses,
however, we use DFT only for theoretical discussions. Now we consider a complex type of
1D array f corresponding to a discretized form of f (x) and its complex type of spectral array
F dened by
f = ( f
0
, f
1
, f
2
, , f
N2
, f
N1
) , F = (F
0
, F
1
, F
2
, , F
N2
, F
N1
) . (10)
order for an equivalent propagation distance, and a distance correction factor . The or-
der yields an equivalent distance indicating the distance to the -th power. The order was
introduced by Hata successfully as an empirical formula for the propagation characteristics in
the urban and suburban areas [Hata,1980]. In the present formulations, we determine these
parameters numerically by using the least square method. Once these parameters are deter-
mined for one type of RRSs, we can easily estimate the radio communication distance between
two sensors distributed on RRSs, provided the input power of a source antenna and the min-
imum detectable electric eld intensity of a receiver are specied.
The contents of the present chapter are described as follows. Section 1 is the introduction of
this chapter, and the background of this research is denoted. Section 2 discusses the statistical
properties of 1D RRSs and the convolution method is introduced for RRS generation. Sec-
tion 3 discusses DRTM for evaluation of electromagnetic waves propagating along RRSs. It is
shown that the DRTM is very effective to the eld evaluation especially in a complicated en-
vironment, since it discretizes not only the terrain prole but also the procedure for searching
rays, resulting in saving much computation time and computer memory. Section 4 discusses a
numerical method to estimate propagation or path loss characteristics along 1D RRSs. An es-
timation formula for the radio communication distance along the 1D RRSs is also introduced
in this Section. Section 5 is the conclusion of this chapter, and a few comments on the near
future problems are remarked.
2. Generation of 1D random rough surface
As mentioned in the introduction, the sensor network has attracted many researchers interest
recently in different technical elds like signal processing, antennas, wave propagation, low
power circuit design and so forth, just as the same as the case of radio frequency identication
(RFID) [Heidrich,2010]. The sensor devices are usually located on terrestrial surfaces such as
dessert, hilly terrain, forest, sea surface and so on. Since these surfaces are considered to be
statistically random, it is important to study statistics of the RRSs as well as the electromag-
netic wave propagation along them in order to construct reliable and efcient sensor network
systems [Honda,2009].
In this section, we describe the statistical properties of RRSs and we show three types of spec-
tral density functions, that is, Gaussian, n-th order of power-law and exponential spectra. We
also discuss the convolution method for RRS generation. The convolution method is exi-
ble and suitable for computer simulations to attack problems related to electromagnetic wave
scattering from RRSs and electromagnetic wave propagation along RRSs.
2.1 Spectral density function and auto-correlation function
In this study, we assume that 1D RRSs extend in x-direction and it is uniform in z-direction
with its height function as denoted by y = f (x). The spectral density function W(K) for a set
of RRSs is dened by using the height function and the spatial angular frequency K as follows:
W(K)dK = h
2
(1)
where h is the standard deviation of of the height function or height deviation, and
W(K) = lim
L
1
2
1
L
L/2
L/2
f (x)e
jKx
dx
(2)
where <> indicates the ensemble average of the RRS set. As is well-known, the auto-
correlation function is given by the Fourier transform of the spectral density function as fol-
lows:
(x) =
W(K)e
jKx
dK . (3)
Now we summarize three types of spectral density functions that are useful for numerical
simulations of the propagation characteristics of electromagnetic waves traveling along RRSs.
1. Gaussian Type of Spectrum:
The spectral density function of this type is dened by
W(K) =
clh
2
2
K
2
cl
2
4
(4)
where cl is the correlation length, and the auto-correlation function is given by
(x) = h
2
e
x
2
cl
2
.
(5)
2. N-th Order Power-Law Spectrum:
W(K) =
clh
2
2
1 +

2
(N
1
2
)
2
(N)
K
2
cl
2
4
N
(6)
where (N) is the Gamma function with N > 1, and the auto-correlation function is
given by
(x) =
h
2
[1 +
x
2
Ncl
2
]
N
.
(7)
3. Exponential Spectrum:
W(K) =
clh
2
1 + K
2
cl
2
1
(8)
and the auto-correlation function is given by
(x) = h
2
e
|x|
cl
.
(9)
2.2 Convolution method for RRS generation
As is well-known, we should not use discrete Fourier transform (DFT) but fast Fourier trans-
form (FFT) for practical applications to save computation time. For simplicity of analyses,
however, we use DFT only for theoretical discussions. Now we consider a complex type of
1D array f corresponding to a discretized form of f (x) and its complex type of spectral array
F dened by
f = ( f
0
, f
1
, f
2
, , f
N2
, f
N1
) , F = (F
0
, F
1
, F
2
, , F
N2
, F
N1
) . (10)
The complex type of spectral array F is the DFT of f. And the complex type of DFT is dened
as follows:
F = DFT(f) , F
=
N1
n=0
f
n
e
j2
n
N
( = 0, 1, , N 1) . (11)
Moreover, the inverse DFT is dened by
f = DFT
1
(F) , f
n
=
1
N
N1
=0
F
e
j2
n
N
(n = 0, 1, , N 1) . (12)
First, we discretize the spectral density function discussed in the preceding section by intro-
ducing the discretized spatial angular frequency K
n
as follows:
K
n
=
2n
N
1
c
(n = 0, 1, 2, , N 1) (13)
where N = N
1
N
2
. It is assumed that N
2
is the number of discretizedpoints per one correlation
length c and correlation beyond the distance N
1
c is negligibly small. Then we can obtain the
real type of 1D array w by using the spectral density function W(K) at the discretized spatial
angular frequencies as follows:
w = (w
0
, w
1
, , w
N1
) (14)
where the elements w
n
of the array are expressed as follows:
w
n
= 2W(K
n
)/N
1
c
n
n (0 n < N/2)
N n (N/2 n < N) .
(15)
It should be noted that the DFT of the above 1D array corresponds to the discretized auto-
correlation function of (x) as follows:
DFT(w) (x) . (16)
Thus we can utilize this relation to check the accuracy of the discretized numerical results for
the spectral density function of the RRSs we are dealing with.
Second, we introduce another real 1D array w by taking the square root of the former array as
follows:
w = (
w
0
/N,
w
1
/N, ,
w
N1
/N) . (17)
Performing the DFT of the above 1D array leads to a new weighting array dened by
W = (

W
0
,

W
1
, ,

W
N1
) = DFT( w) (18)
This weighting array includes all the information about the spectral properties of the RRSs,
and also it plays an important role as a weighting factor when we generate RRSs by the con-
volution method.
Third, we consider the random number generator necessary for computer simulations. C pro-
gramming language provides us the software rand(a) which produces a sequence of random
numbers ranging in [0, a] [Johnsonbaugh,1997]. Then we can generate another sequence of
random number x
i
in the following way:
u
1
= rand(2), u
2
= rand(1)
x
i
=
2 log(u
2
) cos(u
1
) (i = 0, 1, 2, ...).
(19)
It can be proved that the random numbers obtained by the above functions belong to the
normal distribution as follows:
(x
0
, x
1
, x
2
, ) N(0, 1) . (20)
As a result, we can generate a sequence of the discrete randomrough surface with arbitrary N
points by performing the discrete convolution between the sequence of the Gaussian random
number x
i
N(1, 0) given by Eq.19 and the weighting array

W
k
given by Eq.18. The nal
results are summarized as follows:
f
n
=
N1
k=0
W
k
x
n+k
(n = 0, 1, 2, 3, , N
1)
{x
i
} N(1, 0) (i = 0, 1, 2, , N + N
1) .
(21)
Eq.21 is the essential part of the convolution method, and it provides us any type of RRSs with
arbitrary length [Uchida,2007], [Uchida,2008].
It is worth noting that correlation of the generated RRSs is assumed to be negligibly small
outside the distance of N
1
c and the minimum discretized distance is c/N
2
. One of the ad-
vantages of the present convolution method is that we can generate continuous RRSs with an
arbitrary number of sample points N
> N, provided that the weighting array

W
k
in Eq.18 is
computed at the denite number of points N = N
1
N
2
. The other advantage is that the present
method is more exible and it saves more computation time than the conventional direct DFT
method [Thoros,1989], [Thoros,1990], [Phu,1994], [Tsang,1994], [Yoon,2000], [Yoon,2002].
3. Discrete ray tracing method (DRTM)
In this chapter, we apply DRTM to the investigation of propagation characteristics along ran-
dom rough surfaces whose height deviation h and correlation length c are much longer than
the wavelength, that is, h, c >> . In the past, we used the ray tracing method (RTM) to
analyze electromagnetic wave propagation along 1D RRSs. The RTM, however, requires lots
of computer memory and computation time, since its ray searching algorithm is based only
on the imaging method. The present DRTM, however, requires much less computer memory
and computation time than the RTM. This is the reason why we employ the DRTM for ray
searching and eld computing. First, we discretize the rough surface in term of piecewise-
linear lines, and second, we determine whether two lines are in the line of sight (LOS) or not
(NLOS), depending on whether the two representative points on the two lines can be seen
each other or not.
The eld analyses of DRTM are based on the well-known edge diffraction problem by a con-
ducting half plane which was rigorously solved by the Wiener-Hopf technique [Noble,1958].
The Wiener-Hopf solution cannot be rigorously applied to the diffraction problem by a plate
The complex type of spectral array F is the DFT of f. And the complex type of DFT is dened
as follows:
F = DFT(f) , F
=
N1
n=0
f
n
e
j2
n
N
( = 0, 1, , N 1) . (11)
Moreover, the inverse DFT is dened by
f = DFT
1
(F) , f
n
=
1
N
N1
=0
F
e
j2
n
N
(n = 0, 1, , N 1) . (12)
First, we discretize the spectral density function discussed in the preceding section by intro-
ducing the discretized spatial angular frequency K
n
as follows:
K
n
=
2n
N
1
c
(n = 0, 1, 2, , N 1) (13)
where N = N
1
N
2
. It is assumed that N
2
is the number of discretizedpoints per one correlation
length c and correlation beyond the distance N
1
c is negligibly small. Then we can obtain the
real type of 1D array w by using the spectral density function W(K) at the discretized spatial
angular frequencies as follows:
w = (w
0
, w
1
, , w
N1
) (14)
where the elements w
n
of the array are expressed as follows:
w
n
= 2W(K
n
)/N
1
c
n
n (0 n < N/2)
N n (N/2 n < N) .
(15)
It should be noted that the DFT of the above 1D array corresponds to the discretized auto-
correlation function of (x) as follows:
DFT(w) (x) . (16)
Thus we can utilize this relation to check the accuracy of the discretized numerical results for
the spectral density function of the RRSs we are dealing with.
Second, we introduce another real 1D array w by taking the square root of the former array as
follows:
w = (
w
0
/N,
w
1
/N, ,
w
N1
/N) . (17)
Performing the DFT of the above 1D array leads to a new weighting array dened by
W = (

W
0
,

W
1
, ,

W
N1
) = DFT( w) (18)
This weighting array includes all the information about the spectral properties of the RRSs,
and also it plays an important role as a weighting factor when we generate RRSs by the con-
volution method.
Third, we consider the random number generator necessary for computer simulations. C pro-
gramming language provides us the software rand(a) which produces a sequence of random
numbers ranging in [0, a] [Johnsonbaugh,1997]. Then we can generate another sequence of
random number x
i
in the following way:
u
1
= rand(2), u
2
= rand(1)
x
i
=
2 log(u
2
) cos(u
1
) (i = 0, 1, 2, ...).
(19)
It can be proved that the random numbers obtained by the above functions belong to the
normal distribution as follows:
(x
0
, x
1
, x
2
, ) N(0, 1) . (20)
As a result, we can generate a sequence of the discrete randomrough surface with arbitrary N
points by performing the discrete convolution between the sequence of the Gaussian random
number x
i
N(1, 0) given by Eq.19 and the weighting array

W
k
given by Eq.18. The nal
results are summarized as follows:
f
n
=
N1
k=0
W
k
x
n+k
(n = 0, 1, 2, 3, , N
1)
{x
i
} N(1, 0) (i = 0, 1, 2, , N + N
1) .
(21)
Eq.21 is the essential part of the convolution method, and it provides us any type of RRSs with
arbitrary length [Uchida,2007], [Uchida,2008].
It is worth noting that correlation of the generated RRSs is assumed to be negligibly small
outside the distance of N
1
c and the minimum discretized distance is c/N
2
. One of the ad-
vantages of the present convolution method is that we can generate continuous RRSs with an
arbitrary number of sample points N
> N, provided that the weighting array

W
k
in Eq.18 is
computed at the denite number of points N = N
1
N
2
. The other advantage is that the present
method is more exible and it saves more computation time than the conventional direct DFT
method [Thoros,1989], [Thoros,1990], [Phu,1994], [Tsang,1994], [Yoon,2000], [Yoon,2002].
3. Discrete ray tracing method (DRTM)
In this chapter, we apply DRTM to the investigation of propagation characteristics along ran-
dom rough surfaces whose height deviation h and correlation length c are much longer than
the wavelength, that is, h, c >> . In the past, we used the ray tracing method (RTM) to
analyze electromagnetic wave propagation along 1D RRSs. The RTM, however, requires lots
of computer memory and computation time, since its ray searching algorithm is based only
on the imaging method. The present DRTM, however, requires much less computer memory
and computation time than the RTM. This is the reason why we employ the DRTM for ray
searching and eld computing. First, we discretize the rough surface in term of piecewise-
linear lines, and second, we determine whether two lines are in the line of sight (LOS) or not
(NLOS), depending on whether the two representative points on the two lines can be seen
each other or not.
The eld analyses of DRTM are based on the well-known edge diffraction problem by a con-
ducting half plane which was rigorously solved by the Wiener-Hopf technique [Noble,1958].
The Wiener-Hopf solution cannot be rigorously applied to the diffraction problem by a plate
of nite width. When the distance between the two edges of the plate is much longer than
the wavelength, however, it can be approximately applied to this problem with an excellent
accuracy. This is the basic idea of the eld analyses based on DRTM. Numerical calculation
are carried out for the propagation characteristics of electromagnetic waves traveling along
RRSs with Gaussian, n-th order of power-law and exponential types of spectra.
3.1 RRS discretization in terms of piecewise-linear prole
A RRS of arbitrary length can be generated by the convolution method discussed in the pre-
ceding section. We treat here three types of spectral density functions for generating RRSs.
The rst is Gaussian, the second is n-th order of power-law and the third is exponential dis-
tribution, where the RRS parameters are correlation length cl and height deviation h. Fig.1
shows four examples of RRSs with Gaussian, rst and third order of power-lawand exponen-
tial spectra, and the parameters are selected as cl = 10.0 [m] and h = 1.0 [m]. It is shown that
the Gaussian spectrum exhibits the smoothest roughness.
Fig. 1. Examples of random rough surface.
The convolution method introduced in the preceding section provides us the data of position
vectors corresponding to the discretized RRS points as follows:
r
n
= (nx, f
n
) (n = 0, 1, 2, , N
1) (22)
where the minimum discretized distance is given by
x = c/N
2
. (23)
On the other hand, we can determine the normal vector of each straight line as follows:
n
n
= (u
z
a
n
)/|u
z
a
n
| (n = 0, 1, 2, , N
1) (24)
where u
z
is the unit vector in z-direction. Moreover, the vector corresponding to each straight
line is given by
a
i
= (r
i+1
r
i
). (25)
Thus all the informations regarding traced rays can be expressed in terms of the position
vectors r
n
in Eq.22 and the normal vectors n
n
in Eq.24, resulting in saving computer memory.
3.2 Algorithm for searching rays based on DRTM
Now we discuss the algorithm to trace discrete rays with respect to a discretized RRS. We
propose a procedure to approximately determine whether the two straight lines a
i
and a
j
(i = j) are in the line of sight (LOS) or not (NLOS) by checking whether the two representative
points on the two lines can be seen from each other or not. The representative point of a line
may be its center or one of its two edges, and in the following discussions, we employ the
central point as a representative point of a line. Thus the essence of nding rays is reduced
to checking whether the representative point of one straight line is in LOS or in NLOS of the
representative point of the other line.
One type of ray is determined by constructing the minimum distance between the two repre-
sentative points which are in NLOS, while the other type of ray is determined by connecting
the two representative points which are in LOS. The traced rays obtained in this way are ap-
proximate, but the algorithm is simple and thus we can save much computation time. More-
over, we can modify the discrete rays into more accurate ones by applying the principle of the
shortest path to the former case and the imaging method to the latter case. The former type is
shown in (a) of Fig.2, and the latter type is depicted in (b) of Fig.2. In these gures, S denotes
a source point and R indicates a receiver point.
(a) in case of NLOS (b) in case of LOS
Fig. 2. Examples of searching rays.
Let us explain the example of searched ray in (a) of Fig.2. First we nd a shortest path from(2)
to (4) which are in NLOS, and we also nd a straight line (4) to (8) which are in LOS. Moreover,
we add the straight line from S to (2) which are in LOS as well as the straight line from(8) to R
which are also in LOS. Thus we can drawan approximate discrete ray fromS to R through (2),
(3), (4) and (8). The discrete ray is shown by green lines. In order to construct a more accurate
ray, we modify the discrete ray so that the distance from S to (8) may be minimum, and we
apply the imaging method to the discrete ray from(4) to R through (8). The nal modied ray
is plotted in blue lines in (a) of Fig.2. The ray from S to (8) constitutes a diffraction. We call
it as a source diffraction, because it is associated with shadowing of the incident wave from
source S by the line (3).
Let us explain another example of searched ray shown in (b) of Fig.2. First, we nd the straight
line from (5) to (8) which are in LOS. Second, we add the lines from S to (5) and from (8) to R,
since S and (5) as well as (8) and R are in LOS. Thus we obtain an approximate discrete ray
from S to R through (5) and (8) as shown in green lines. In order to obtain more accurate ray,
we can modify the discrete ray based on the imaging method. The nal ray plotted in blue
lines shows that the ray emitted from S is rst reected fromthe line at (5) and next diffracted
at the right edge of the line at (8), and nally it reaches R. We call this type of diffraction as an
of nite width. When the distance between the two edges of the plate is much longer than
the wavelength, however, it can be approximately applied to this problem with an excellent
accuracy. This is the basic idea of the eld analyses based on DRTM. Numerical calculation
are carried out for the propagation characteristics of electromagnetic waves traveling along
RRSs with Gaussian, n-th order of power-law and exponential types of spectra.
3.1 RRS discretization in terms of piecewise-linear prole
A RRS of arbitrary length can be generated by the convolution method discussed in the pre-
ceding section. We treat here three types of spectral density functions for generating RRSs.
The rst is Gaussian, the second is n-th order of power-law and the third is exponential dis-
tribution, where the RRS parameters are correlation length cl and height deviation h. Fig.1
shows four examples of RRSs with Gaussian, rst and third order of power-lawand exponen-
tial spectra, and the parameters are selected as cl = 10.0 [m] and h = 1.0 [m]. It is shown that
the Gaussian spectrum exhibits the smoothest roughness.
Fig. 1. Examples of random rough surface.
The convolution method introduced in the preceding section provides us the data of position
vectors corresponding to the discretized RRS points as follows:
r
n
= (nx, f
n
) (n = 0, 1, 2, , N
1) (22)
where the minimum discretized distance is given by
x = c/N
2
. (23)
On the other hand, we can determine the normal vector of each straight line as follows:
n
n
= (u
z
a
n
)/|u
z
a
n
| (n = 0, 1, 2, , N
1) (24)
where u
z
is the unit vector in z-direction. Moreover, the vector corresponding to each straight
line is given by
a
i
= (r
i+1
r
i
). (25)
Thus all the informations regarding traced rays can be expressed in terms of the position
vectors r
n
in Eq.22 and the normal vectors n
n
in Eq.24, resulting in saving computer memory.
3.2 Algorithm for searching rays based on DRTM
Now we discuss the algorithm to trace discrete rays with respect to a discretized RRS. We
propose a procedure to approximately determine whether the two straight lines a
i
and a
j
(i = j) are in the line of sight (LOS) or not (NLOS) by checking whether the two representative
points on the two lines can be seen from each other or not. The representative point of a line
may be its center or one of its two edges, and in the following discussions, we employ the
central point as a representative point of a line. Thus the essence of nding rays is reduced
to checking whether the representative point of one straight line is in LOS or in NLOS of the
representative point of the other line.
One type of ray is determined by constructing the minimum distance between the two repre-
sentative points which are in NLOS, while the other type of ray is determined by connecting
the two representative points which are in LOS. The traced rays obtained in this way are ap-
proximate, but the algorithm is simple and thus we can save much computation time. More-
over, we can modify the discrete rays into more accurate ones by applying the principle of the
shortest path to the former case and the imaging method to the latter case. The former type is
shown in (a) of Fig.2, and the latter type is depicted in (b) of Fig.2. In these gures, S denotes
a source point and R indicates a receiver point.
(a) in case of NLOS (b) in case of LOS
Fig. 2. Examples of searching rays.
Let us explain the example of searched ray in (a) of Fig.2. First we nd a shortest path from(2)
to (4) which are in NLOS, and we also nd a straight line (4) to (8) which are in LOS. Moreover,
we add the straight line from S to (2) which are in LOS as well as the straight line from(8) to R
which are also in LOS. Thus we can drawan approximate discrete ray fromS to R through (2),
(3), (4) and (8). The discrete ray is shown by green lines. In order to construct a more accurate
ray, we modify the discrete ray so that the distance from S to (8) may be minimum, and we
apply the imaging method to the discrete ray from(4) to R through (8). The nal modied ray
is plotted in blue lines in (a) of Fig.2. The ray from S to (8) constitutes a diffraction. We call
it as a source diffraction, because it is associated with shadowing of the incident wave from
source S by the line (3).
Let us explain another example of searched ray shown in (b) of Fig.2. First, we nd the straight
line from (5) to (8) which are in LOS. Second, we add the lines from S to (5) and from (8) to R,
since S and (5) as well as (8) and R are in LOS. Thus we obtain an approximate discrete ray
from S to R through (5) and (8) as shown in green lines. In order to obtain more accurate ray,
we can modify the discrete ray based on the imaging method. The nal ray plotted in blue
lines shows that the ray emitted from S is rst reected fromthe line at (5) and next diffracted
at the right edge of the line at (8), and nally it reaches R. We call this type of diffraction as an
image diffraction, since it is associated with reection and the reection might be described as
an emission from the image point with respect to the related line.
3.3 Reection and diffraction coefcients
The purpose of this investigation is to evaluate the propagation characteristics of electromag-
netic waves traveling along RRSs from a source point S to a receiver point R. We assume that
the inuences of transmitted waves through RRSs on propagation are negligibly small. As a
result, the received electromagnetic waves at R are expressed in terms of incident, reected
and diffracted rays in LOS region, and they are denoted in terms of reected and diffracted
rays in NLOS region.
First we consider electromagnetic wave reection from a at ground plane composed of a
lossy dielectric. The lossy dielectric medium, for example, indicates a soil ground plane. Fig.3
shows a geometry of incidence and reection with source point S and receiver point R together
with the sources image point S
i
. In Fig.3, the polarization of the incident wave is assumed
such that electric eld is parallel to the ground plane (z-axis) or magnetic eld is parallel to
it. We call the former case as E-wave or horizontal polarization, and we call the latter case as
H-wave or vertical polarization [Mushuake,1985], [Collin,1985].
S
Si
R
Plane Reflector
r1
r2
ri
r0
n
Fig. 3. Incidence and reection.
The incident wave, which we also call a source eld, and the reected wave, which we also
call an image eld, are given by the following relations:
E
z
, H
z
= (r
0
) + R
e,h
()(r
1
+ r
2
) (26)
where E
z
and H
z
indicate E-wave (e) and H-wave (h), respectively. The distances r
0
, r
1
and r
2
are depicted in Fig.3, and the complex eld function expressing the amplitude and phase of a
eld is dened in terms of a propagation distance r as follows:
(r) =
e
jr
r
.
(27)
In the eld expressions, the time dependence e
jt
is assumed and suppressedthrough out this
chapter. The wavenumber in the free space is given by
=
0
(28)
where
0
and
0
denote permittivity and permeability of the free space, respectively. E
z
and
H
z
in Eq.26 correspond to R
e
() and R
h
(), respectively. As mentioned earlier, E or H-wave
indicates that electric or magnetic eld is parallel to z-axis, respectively.
The reection coefcients are expressed depending on the two different polarizations of the
incident wave as follows:
R
e
() =
cos
c
sin
2
cos +
c
sin
2
R
h
() =
c
cos
c
sin
2
c
cos +
c
sin
2
(29)
where is the incident angle as shown in Fig.3. Moreover, the complex permittivity of the
medium is given by
c
=
r
j

0
(30)
where
r
is dielectric constant and is conductivity of the medium.
S
R
r1
r2
r0
Region
Region
Edge
Fig. 4. Source diffraction from the edge of an half plane.
According to the rigorous solution for the plane wave diffraction by a half-plane [Noble,1958],
diffraction phenomenon can be classied into two types. One is related to incident wave or
eld emitted from a source, which we call source eld in short, as shown in Fig.4, and we call
this type of diffraction as a source diffraction. The other is related to reected wave or eld
emitted from an image, which we call image eld in short, as shown in (a) of Fig.5, and we call
this type of diffraction as an image diffraction. It should be noted that the rigorous solution
based no the Wiener-Hopf technique is applicable only to the geometry of a semi-innite half
plane, and its extension to nite plate results in an approximate solution. However, it exhibits
an excellent accuracy when the plate width is much longer than the wavelength. This is the
starting point of the eld analysis based on DRTM.
First we consider the source diffraction shown in Fig.4. In this case, we assume that the
diffracted wave is approximated by the Winner-Hopf (WH) solution [Noble,1958]. The to-
tal diffracted elds for different two polarizations, that is E and H-wave, are given by
E
z
, H
z
=
D
s
(r
0
) (Region I)
D
s
(r
1
+ r
2
) (Region II)
(31)
image diffraction, since it is associated with reection and the reection might be described as
an emission from the image point with respect to the related line.
3.3 Reection and diffraction coefcients
The purpose of this investigation is to evaluate the propagation characteristics of electromag-
netic waves traveling along RRSs from a source point S to a receiver point R. We assume that
the inuences of transmitted waves through RRSs on propagation are negligibly small. As a
result, the received electromagnetic waves at R are expressed in terms of incident, reected
and diffracted rays in LOS region, and they are denoted in terms of reected and diffracted
rays in NLOS region.
First we consider electromagnetic wave reection from a at ground plane composed of a
lossy dielectric. The lossy dielectric medium, for example, indicates a soil ground plane. Fig.3
shows a geometry of incidence and reection with source point S and receiver point R together
with the sources image point S
i
. In Fig.3, the polarization of the incident wave is assumed
such that electric eld is parallel to the ground plane (z-axis) or magnetic eld is parallel to
it. We call the former case as E-wave or horizontal polarization, and we call the latter case as
H-wave or vertical polarization [Mushuake,1985], [Collin,1985].
S
Si
R
Plane Reflector
r1
r2
ri
r0
n
Fig. 3. Incidence and reection.
The incident wave, which we also call a source eld, and the reected wave, which we also
call an image eld, are given by the following relations:
E
z
, H
z
= (r
0
) + R
e,h
()(r
1
+ r
2
) (26)
where E
z
and H
z
indicate E-wave (e) and H-wave (h), respectively. The distances r
0
, r
1
and r
2
are depicted in Fig.3, and the complex eld function expressing the amplitude and phase of a
eld is dened in terms of a propagation distance r as follows:
(r) =
e
jr
r
.
(27)
In the eld expressions, the time dependence e
jt
is assumed and suppressedthrough out this
chapter. The wavenumber in the free space is given by
=
0
(28)
where
0
and
0
denote permittivity and permeability of the free space, respectively. E
z
and
H
z
in Eq.26 correspond to R
e
() and R
h
(), respectively. As mentioned earlier, E or H-wave
indicates that electric or magnetic eld is parallel to z-axis, respectively.
The reection coefcients are expressed depending on the two different polarizations of the
incident wave as follows:
R
e
() =
cos
c
sin
2
cos +
c
sin
2
R
h
() =
c
cos
c
sin
2
c
cos +
c
sin
2
(29)
where is the incident angle as shown in Fig.3. Moreover, the complex permittivity of the
medium is given by
c
=
r
j

0
(30)
where
r
is dielectric constant and is conductivity of the medium.
S
R
r1
r2
r0
Region
Region
Edge
Fig. 4. Source diffraction from the edge of an half plane.
According to the rigorous solution for the plane wave diffraction by a half-plane [Noble,1958],
diffraction phenomenon can be classied into two types. One is related to incident wave or
eld emitted from a source, which we call source eld in short, as shown in Fig.4, and we call
this type of diffraction as a source diffraction. The other is related to reected wave or eld
emitted from an image, which we call image eld in short, as shown in (a) of Fig.5, and we call
this type of diffraction as an image diffraction. It should be noted that the rigorous solution
based no the Wiener-Hopf technique is applicable only to the geometry of a semi-innite half
plane, and its extension to nite plate results in an approximate solution. However, it exhibits
an excellent accuracy when the plate width is much longer than the wavelength. This is the
starting point of the eld analysis based on DRTM.
First we consider the source diffraction shown in Fig.4. In this case, we assume that the
diffracted wave is approximated by the Winner-Hopf (WH) solution [Noble,1958]. The to-
tal diffracted elds for different two polarizations, that is E and H-wave, are given by
E
z
, H
z
=
D
s
(r
0
) (Region I)
D
s
(r
1
+ r
2
) (Region II)
(31)
where the distances r
0
, r
1
and r
2
are depicted in Fig.4 and the complex eld function is dened
by Eq.27. The source diffraction coefcient is dened as follows:
D
s
=
1 D(r
1
, r
2
, )(r
1
+r
2
)/(r
0
) (Region I)
D(r
1
, r
2
, ) (Region II)
(32)
where the diffraction function used in the above equation is dened by
D(r
1
, r
2
, ) = e
jX
2
F(X) .
(33)
The complex type of Fresnel function for the diffraction function is dened by [Noble,1958]
F(X) =
e
4
j

X
e
ju
2
du (X > 0) (34)
where the argument is dened by
X =
(r
1
+ r
2
r
0
)
(35)
and the distances r
0
, r
1
and r
2
are shown in Fig.4.
Finally, we discuss the image diffraction as shown in (a) of Fig.5 where S is a source point, S
i
is its image point and R is a receiver point, respectively. Two edges of the line of the dicretized
rough surface are given by the position vectors dened by Eq.22 as follows:
r
i1
= (ix, f
i
), r
i2
= ((i + 1)x, f
i+1
) . (36)
Region
Region
Region
S
R
Si
2
r11
r21
r12
r22
Edge 2
Edge 1
r20 r10
ri2
ri0
ri1
(a) Diffraction angles
1
and
2
S
R
Edge 2 Edge 1
n
0
n n
(b) Reection angles
1
,
2
and
0
Fig. 5. Image diffraction.
Intersection between one straight line from r
i1
to r
i1
and the other straight line from the image
point S
i
to the receiver point R can be expressed as
r
i0
= r
i1
+ (r
i2
r
i1
) (37)
where is a constant to be determined. It is worth noting that the constant can be easily
determined in terms of the position vectors, and < 0 AC0 1 and 1 < correspond to
regions I, II and III in (a) of Fig.5, respectively.
By use of the image diffraction coefcients as well as the eld function in Eq.27, the image
diffraction elds corresponding to the three regions are summarized as follows:
E
z
, H
z
=
D
ie,h
(r
i1
+r
21
) (Rgion I)
D
ie,h
(r
i0
+r
20
) (Region II)
D
ie,h
(r
i2
+r
22
) (Region III) .
(38)
The image diffraction coefcients are expressed in terms of the reection coefcients in Eq.29,
the diffraction function in Eq.33 and the eld function in Eq.27 as follows:
D
ie,h
=
R
e,h
(
2
)D(r
i2
, r
22
,
2
)(r
i2
+r
22
)/(r
i1
+r
21
)
+R
e,h
(
1
)D(r
i1
, r
21
,
1
) (RgionI)
R
e,h
(
1
)D(r
i1
, r
21
,
1
)(r
i1
+r
21
)/(r
i0
+r
20
)
R
e,h
(
2
)D(r
i2
, r
22
,
2
)(r
i2
+r
22
)/(r
i0
+r
20
)
+R
e,h
(
0
) (RegionII)
R
e,h
(
1
)D(r
i1
, r
21
,
1
)(r
i1
+r
21
)/(r
i2
+r
22
)
+R
e,h
(
2
)D(r
i2
, r
22
,
2
) (RegionIII)
(39)
where the two diffraction angles
1
and
1
are shown in (a) of Fig.5 and the three reection
angles
0
,
1
and
1
are depicted in (b) of Fig.5.
The main feature of the image diffraction is described in the following way by using the traced
rays shown in (a) of Fig.5. In region I, we have two image diffraction rays (S r
i1
R)
and (S r
i2
R). In region II we have two image diffraction rays (S r
i1
R) and
(S r
i2
R) and a reection ray (S r
i0
R). In region III, we have two image diffraction
rays (S r
i1
R) and (S r
i2
R). It should be noted that all these elds are continuous
at the two boundaries of the three regions, that is, from I to II and from II to III.
3.4 Field evaluations based on DRTM
In the preceding sections, we have rstly proposed an algorithm to construct a discrete ray
starting from a source, repeating source or image diffractions successively, and terminating at
a receiver. Secondly, we have discussed the algorithm of DRTM to approximately evaluate the
electromagnetic elds in relation to source or image diffractions based on the modied rays
derived from the approximate discrete rays. As a result, electromagnetic elds along RRSs
can be calculated numerically by repeating the DRTM computations step by step.
We assume that the source antenna is a small dipole antenna with gain G = 1.5 [Mushi-
ake,1985] and input power P
i
[W]. The direction of the source antenna is denoted by a unit
vector p, and the E or H-wave corresponds to the antenna direction parallel to z or y-direction,
respectively. The elds of the small dipole antenna are classied into three types, that is, static,
induced and radiated elds. We call the rst two terms as the near elds and the last one as the
far eld [Mushiuake,1985],[Collin,1985]. Contrary to RFID where the near elds are mainly
used [Heidrich,2010], sensor networks mainly use the far elds which are predominant in the
region where the distance from the source is much longer than the wavelength (r >> ).
Thus neglecting the near elds, the electric eld radiating from a small dipole antenna is
expressed in the following form [Mushiake,1985], [Collin,1985]
E
0
=
30GP
i
[(u
r
p) u
r
](r) (40)
where r is a position vector from the source to a receiver point and r = |r|. The unit vec-
tor u
r
= r/r is in the direction from the source to the receiver point, and |u
r
p| = sin
where the distances r
0
, r
1
and r
2
are depicted in Fig.4 and the complex eld function is dened
by Eq.27. The source diffraction coefcient is dened as follows:
D
s
=
1 D(r
1
, r
2
, )(r
1
+r
2
)/(r
0
) (Region I)
D(r
1
, r
2
, ) (Region II)
(32)
where the diffraction function used in the above equation is dened by
D(r
1
, r
2
, ) = e
jX
2
F(X) .
(33)
The complex type of Fresnel function for the diffraction function is dened by [Noble,1958]
F(X) =
e
4
j

X
e
ju
2
du (X > 0) (34)
where the argument is dened by
X =
(r
1
+ r
2
r
0
)
(35)
and the distances r
0
, r
1
and r
2
are shown in Fig.4.
Finally, we discuss the image diffraction as shown in (a) of Fig.5 where S is a source point, S
i
is its image point and R is a receiver point, respectively. Two edges of the line of the dicretized
rough surface are given by the position vectors dened by Eq.22 as follows:
r
i1
= (ix, f
i
), r
i2
= ((i + 1)x, f
i+1
) . (36)
Region
Region
Region
S
R
Si
2
r11
r21
r12
r22
Edge 2
Edge 1
r20 r10
ri2
ri0
ri1
(a) Diffraction angles
1
and
2
S
R
Edge 2 Edge 1
n
0
n n
(b) Reection angles
1
,
2
and
0
Fig. 5. Image diffraction.
Intersection between one straight line from r
i1
to r
i1
and the other straight line from the image
point S
i
to the receiver point R can be expressed as
r
i0
= r
i1
+ (r
i2
r
i1
) (37)
where is a constant to be determined. It is worth noting that the constant can be easily
determined in terms of the position vectors, and < 0 AC0 1 and 1 < correspond to
regions I, II and III in (a) of Fig.5, respectively.
By use of the image diffraction coefcients as well as the eld function in Eq.27, the image
diffraction elds corresponding to the three regions are summarized as follows:
E
z
, H
z
=
D
ie,h
(r
i1
+r
21
) (Rgion I)
D
ie,h
(r
i0
+r
20
) (Region II)
D
ie,h
(r
i2
+r
22
) (Region III) .
(38)
The image diffraction coefcients are expressed in terms of the reection coefcients in Eq.29,
the diffraction function in Eq.33 and the eld function in Eq.27 as follows:
D
ie,h
=
R
e,h
(
2
)D(r
i2
, r
22
,
2
)(r
i2
+r
22
)/(r
i1
+r
21
)
+R
e,h
(
1
)D(r
i1
, r
21
,
1
) (RgionI)
R
e,h
(
1
)D(r
i1
, r
21
,
1
)(r
i1
+r
21
)/(r
i0
+r
20
)
R
e,h
(
2
)D(r
i2
, r
22
,
2
)(r
i2
+r
22
)/(r
i0
+r
20
)
+R
e,h
(
0
) (RegionII)
R
e,h
(
1
)D(r
i1
, r
21
,
1
)(r
i1
+r
21
)/(r
i2
+r
22
)
+R
e,h
(
2
)D(r
i2
, r
22
,
2
) (RegionIII)
(39)
where the two diffraction angles
1
and
1
are shown in (a) of Fig.5 and the three reection
angles
0
,
1
and
1
are depicted in (b) of Fig.5.
The main feature of the image diffraction is described in the following way by using the traced
rays shown in (a) of Fig.5. In region I, we have two image diffraction rays (S r
i1
R)
and (S r
i2
R). In region II we have two image diffraction rays (S r
i1
R) and
(S r
i2
R) and a reection ray (S r
i0
R). In region III, we have two image diffraction
rays (S r
i1
R) and (S r
i2
R). It should be noted that all these elds are continuous
at the two boundaries of the three regions, that is, from I to II and from II to III.
3.4 Field evaluations based on DRTM
In the preceding sections, we have rstly proposed an algorithm to construct a discrete ray
starting from a source, repeating source or image diffractions successively, and terminating at
a receiver. Secondly, we have discussed the algorithm of DRTM to approximately evaluate the
electromagnetic elds in relation to source or image diffractions based on the modied rays
derived from the approximate discrete rays. As a result, electromagnetic elds along RRSs
can be calculated numerically by repeating the DRTM computations step by step.
We assume that the source antenna is a small dipole antenna with gain G = 1.5 [Mushi-
ake,1985] and input power P
i
[W]. The direction of the source antenna is denoted by a unit
vector p, and the E or H-wave corresponds to the antenna direction parallel to z or y-direction,
respectively. The elds of the small dipole antenna are classied into three types, that is, static,
induced and radiated elds. We call the rst two terms as the near elds and the last one as the
far eld [Mushiuake,1985],[Collin,1985]. Contrary to RFID where the near elds are mainly
used [Heidrich,2010], sensor networks mainly use the far elds which are predominant in the
region where the distance from the source is much longer than the wavelength (r >> ).
Thus neglecting the near elds, the electric eld radiating from a small dipole antenna is
expressed in the following form [Mushiake,1985], [Collin,1985]
E
0
=
30GP
i
[(u
r
p) u
r
](r) (40)
where r is a position vector from the source to a receiver point and r = |r|. The unit vec-
tor u
r
= r/r is in the direction from the source to the receiver point, and |u
r
p| = sin
is the directivity of the small dipole antenna. The electric eld radiated from the source an-
tenna propagates along a RRS to a receiver point, decaying due to repeated source and image
diffractions as discussed in the preceding section. We have assumed that the propagation
model is 2D, which means that the RRSs are uniform in z-direction and the direction of prop-
agation is restricted only to the (x, y)-plane. This assumption indicates that the back and
forward diffractions are predominant and the side diffractions are negligibly small, and this
assumption might be valid as long as the isotropic 2D RRSs are concerned.
At a source diffraction point, the electric eld is subject to both amplitude and phase conver-
sions according to Eq.32, but this source diffraction gives rise to no conversion of polarization.
At an image diffraction point, however, not only amplitude and phase conversions but also
conversion of polarization occur. The latter conversion is described in such a way that E-wave
conversion occurs for electric eld component parallel to z-direction and H-wave conversion
does for electric eld components perpendicular to z-direction as shown in Eq.39. At a source
diffraction point, of course, the electric eld receives the same conversion both for E-wave and
H-wave as expressed in Eq.32.
Thus we can summarize the electric eld at a receiver point in the following dyadic form
E =
N
n=1
m=M
i
n
m=1
(D
i
nm
)
k=M
s
n
k=1
(D
s
nk
) E
0
(r
n
)
(41)
where E
0
is the electric eld vector of the n-th ray at the rst source or diffraction point. N is
the number of rays included for the eld computations, M
s
n
is the number of source diffrac-
tions of the n-th ray, and M
i
n
is the number of its image diffractions. Moreover, D
s
nk
is a dyadic
source diffraction coefcient at the k-th source diffraction point of the n-th ray, and the coef-
cient can be computed by using the source diffraction coefcient D
s
dened in Eq.32. On the
other hand, D
i
nm
is a dyadic image diffraction coefcient at the m-th image diffraction point
of the n-th ray, and the coefcient can be computed by using the image diffraction coefcients
D
ie
nm
and D
ih
nm
dened in Eq.39. It should be noted that the source diffraction coefcient is the
same both for E and H-waves, while the image diffraction coefcient is different depending
on the polarization of the elds at the image diffraction points. Moreover, the total distance of
the n-th ray is given by
r
n
=
m=M
s
n
+M
i
n
m=0
r
nm
(n = 1, 2, . . . , N).
(42)
where M
s
n
and M
i
n
are the number of source and image diffraction points of the n-th ray,
respectively.
It is worth noting that the specular reection froma plate is included automatically as a special
case of the image diffraction in the reection region II of (a) in Fig.5. In the present DRTM
computations, of course, the more the diffraction times increases, the more computation time
is required. However, we can neglect the higher order of diffractions because their effects are
small. In the present analyses, we include at most the three order of diffractions, resulting in
saving much computation time compared to the method of moments (MoM) [Yagbasan,2010].
4. Propagation characteristics of electromagnetic waves along RRSs
In the preceding sections, we have proposed the convolution method for generating RRSs
with two parameters, height deviation h and correlation length c. We have also proposed the
DRTM to compute electric elds which are rst radiated from a small dipole antenna, next
propagate along RRSs repeating source and image diffractions, and nally arrive at a receiver.
According to the DRTM process, the electromagnetic waves emitted from a source antenna
propagate along RRSs, repeating reection, diffraction and shadowing, and thus resulting in
a more attenuation than in the free space.
The eld distribution along one pattern of RRS exhibits one pattern of eld variation with
respect to propagating distance, and that of the other pattern of RRS shows another pattern of
eld variation. Accordingly every eld distribution is different depending on the seed of RRS
generation. However, as is evident from the theory of statistics, the ensemble average of the
eld distributions may show a denite propagation characteristics in a simple analytical form.
This situation was empirically conrmed by Hata in case of the propagation characteristics in
urban or suburban areas [Hata,1980].
4.1 Distance characteristics of averaged eld distribution
Now we show a numerical example to explain the statistical properties of electromagnetic
wave propagation along RRSs. In this numerical simulation, the source antenna is placed
at x=0 [m] and at 0.5 [m] high above RRSs, and the receiver point is movable along RRSs at
0.5 [m] high above them. The operating frequency is chosen as f=1 [GHz]. The RRS parameters
are selected as height deviation h=10 [m] and correlation length cl=50 [m], and the material
constants are chosen as dielectric constant
r
=5 and conductivity =0.0023 [S/m]. We assume
here that the terrestrial ground is composed of a dry soil [Sato,2002].
Fig. 6. Field distribution along one generated Gaussian RRS together with the ensemble aver-
age of 100 samples.
Fig.6 shows two electric eld distributions along RRSs; one curve in red is the eld distribu-
tion for one generated Gaussian RRS, and the other in blue is the ensemble average of the
eld distributions for 100 generated RRSs. In Fig.6, it is well demonstrated that one pattern of
the eld distributions is varying rapidly along the propagating distance, while the ensemble
average of them is expressed in terms of a smooth and monotonic curve. As a result, it is
concluded that we can approximate the ensemble average of the eld distributions in terms
of a simple analytic function.
Now we approximate the ensemble average of the electric eld intensity by an analytic func-
tion with three constants, , and , as follows:
E =
10

20
(r +)
[V/m] (43)
is the directivity of the small dipole antenna. The electric eld radiated from the source an-
tenna propagates along a RRS to a receiver point, decaying due to repeated source and image
diffractions as discussed in the preceding section. We have assumed that the propagation
model is 2D, which means that the RRSs are uniform in z-direction and the direction of prop-
agation is restricted only to the (x, y)-plane. This assumption indicates that the back and
forward diffractions are predominant and the side diffractions are negligibly small, and this
assumption might be valid as long as the isotropic 2D RRSs are concerned.
At a source diffraction point, the electric eld is subject to both amplitude and phase conver-
sions according to Eq.32, but this source diffraction gives rise to no conversion of polarization.
At an image diffraction point, however, not only amplitude and phase conversions but also
conversion of polarization occur. The latter conversion is described in such a way that E-wave
conversion occurs for electric eld component parallel to z-direction and H-wave conversion
does for electric eld components perpendicular to z-direction as shown in Eq.39. At a source
diffraction point, of course, the electric eld receives the same conversion both for E-wave and
H-wave as expressed in Eq.32.
Thus we can summarize the electric eld at a receiver point in the following dyadic form
E =
N
n=1
m=M
i
n
m=1
(D
i
nm
)
k=M
s
n
k=1
(D
s
nk
) E
0
(r
n
)
(41)
where E
0
is the electric eld vector of the n-th ray at the rst source or diffraction point. N is
the number of rays included for the eld computations, M
s
n
is the number of source diffrac-
tions of the n-th ray, and M
i
n
is the number of its image diffractions. Moreover, D
s
nk
is a dyadic
source diffraction coefcient at the k-th source diffraction point of the n-th ray, and the coef-
cient can be computed by using the source diffraction coefcient D
s
dened in Eq.32. On the
other hand, D
i
nm
is a dyadic image diffraction coefcient at the m-th image diffraction point
of the n-th ray, and the coefcient can be computed by using the image diffraction coefcients
D
ie
nm
and D
ih
nm
dened in Eq.39. It should be noted that the source diffraction coefcient is the
same both for E and H-waves, while the image diffraction coefcient is different depending
on the polarization of the elds at the image diffraction points. Moreover, the total distance of
the n-th ray is given by
r
n
=
m=M
s
n
+M
i
n
m=0
r
nm
(n = 1, 2, . . . , N).
(42)
where M
s
n
and M
i
n
are the number of source and image diffraction points of the n-th ray,
respectively.
It is worth noting that the specular reection froma plate is included automatically as a special
case of the image diffraction in the reection region II of (a) in Fig.5. In the present DRTM
computations, of course, the more the diffraction times increases, the more computation time
is required. However, we can neglect the higher order of diffractions because their effects are
small. In the present analyses, we include at most the three order of diffractions, resulting in
saving much computation time compared to the method of moments (MoM) [Yagbasan,2010].
4. Propagation characteristics of electromagnetic waves along RRSs
In the preceding sections, we have proposed the convolution method for generating RRSs
with two parameters, height deviation h and correlation length c. We have also proposed the
DRTM to compute electric elds which are rst radiated from a small dipole antenna, next
propagate along RRSs repeating source and image diffractions, and nally arrive at a receiver.
According to the DRTM process, the electromagnetic waves emitted from a source antenna
propagate along RRSs, repeating reection, diffraction and shadowing, and thus resulting in
a more attenuation than in the free space.
The eld distribution along one pattern of RRS exhibits one pattern of eld variation with
respect to propagating distance, and that of the other pattern of RRS shows another pattern of
eld variation. Accordingly every eld distribution is different depending on the seed of RRS
generation. However, as is evident from the theory of statistics, the ensemble average of the
eld distributions may show a denite propagation characteristics in a simple analytical form.
This situation was empirically conrmed by Hata in case of the propagation characteristics in
urban or suburban areas [Hata,1980].
4.1 Distance characteristics of averaged eld distribution
Now we show a numerical example to explain the statistical properties of electromagnetic
wave propagation along RRSs. In this numerical simulation, the source antenna is placed
at x=0 [m] and at 0.5 [m] high above RRSs, and the receiver point is movable along RRSs at
0.5 [m] high above them. The operating frequency is chosen as f=1 [GHz]. The RRS parameters
are selected as height deviation h=10 [m] and correlation length cl=50 [m], and the material
constants are chosen as dielectric constant
r
=5 and conductivity =0.0023 [S/m]. We assume
here that the terrestrial ground is composed of a dry soil [Sato,2002].
Fig. 6. Field distribution along one generated Gaussian RRS together with the ensemble aver-
age of 100 samples.
Fig.6 shows two electric eld distributions along RRSs; one curve in red is the eld distribu-
tion for one generated Gaussian RRS, and the other in blue is the ensemble average of the
eld distributions for 100 generated RRSs. In Fig.6, it is well demonstrated that one pattern of
the eld distributions is varying rapidly along the propagating distance, while the ensemble
average of them is expressed in terms of a smooth and monotonic curve. As a result, it is
concluded that we can approximate the ensemble average of the eld distributions in terms
of a simple analytic function.
Now we approximate the ensemble average of the electric eld intensity by an analytic func-
tion with three constants, , and , as follows:
E =
10

20
(r +)
[V/m] (43)
where the unit input power or P
i
=1 [W] in Eq.40 is assumed, and is an amplitude weighting
factor, is an order of propagation distance, and is a distance correction factor. Rewriting
the above relation in dB leads to the following equation [Hata,1980]:
E = 20 log
10
(r + ) [dB]. (44)
Next, we determine the three unknown constants , and based on the method of least
mean square (LMS) with the objective function dened by
(, , ) =
1
N
N
n=1
(E
d
n
E
n
)
2
min (45)
where N is the number of data. Moreover, E
d
n
is the averaged value of simulated or experi-
mental data at r = r
n
, and E
n
is the data computed by putting r = r
n
in Eq.43. Applying the
conjugate gradient method (CGM) [Press,1992] to Eq.45, we can numerically determine the
constants , and explicitly.
Fig. 7. Ensemble average of eld distribution compared with estimated eld distribution.
Fig.7 shows a comparison of the ensemble average of the eld distribution with the estimated
function based on the proposed algorithm. The estimated curve is given by Eq.43 with the
constants determined by CGM such as =22.6 [dB], =1.62 and =-6.85 [m]. It is worth noting
that the distance correction factor is small in comparison with the radio communication dis-
tance r, and thus we can neglect it as 0. Consequently, when we estimate the propagation
characteristics of electromagnetic waves traveling along RRSs, the most important parameters
are and in Eq.43.
4.2 Procedure for estimation of radio communication distance
In the preceding section, we have demonstrated that the ensemble average of the eld dis-
tributions of electromagnetic waves traveling along RRSs is well expressed in a simple and
analytic function with three parameters , and . In this section, we propose an algorithm
to estimate the radio communication distance when the two parameters for RRSs as well as
another two parameters for source and receiver are specied. The rst two parameters are the
height deviation h and the correlation length cl for RRSs, and the last two parameters are the
input power P
i
of a source antenna and the sensitivity of a receiver or the minimum detectable
electric intensity E
m
of the receiver.
Feasibility of the present method comes from the simplied form of the electric eld distri-
bution given by Eq.43. Once the three constants , and are determined based on the sim-
ulated or experimental data, we can analytically evaluate the radio communication distance
from a source with an arbitrary input power P
i
to a receiver with an arbitrary minimum de-
tectable electric intensity E
min
. In this case we have assumed that both the source and receiver
are on the homogeneous RRSs with the same parameters. Considering the dB expression such
as [dBm] for the input power P
i
and [dB V/m] for the minimum detectable electric intensity
E
min
, we can rewrite Eq.44 in the following form
E
min
120 = 20 log
10
(r + ) + P
i
30. (46)
Solving for r in the above equation, we can analytically determine the radio communication
distance r by the following relation:
r = 10
(+P
i
E
min
+90)/20
10
(+P
i
E
min
+90)/20
. (47)
Eq.47 is the estimation formula for the radio communication distance between the source
antenna and receiver placed on homogeneous RRSs. The parameters of the RRSs are the height
deviation h [m] and correlation length c [m]. Moreover, the input power of the source is
denoted by P
i
[dBm] and the minimum detectable electric intensity of th receiver is described
by E
min
[dB V/m]. As mentioned earlier, we can neglect the distance correction factor , and
we have the more simplied estimation formula in Eq.47.
We show some numerical examples of the radio communication distances along RRSs com-
puted by Eq.47. Fig.8 (a) shows radio communication distance r [m] versus the minimum de-
tectable electric eld intensity E
min
[dBV/m] of the receiver with the input power P
i
[dBm] of
the source antenna as a parameter. The height deviation of the RRSs is selected as h = 10 [m]
and the correlation length is chosen as cl = 50 [m]. The spectrum of the RRSs is assumed
to be Gaussian, and the material constants of the RRS are chosen such that dielectric con-
stant is
r
=5 and conductivity is = 0.023 [S/m]. The operating frequency is chosen
as f = 1 [GHz]. The minimum detectable electric intensity E
min
is varied from 40 [dB
V/m]to 80 [dBV/m], where the parameters for input power of source antenna are selected
as P
i
= 10, 8, , 8, 10 [dBm].
Fig.8 (b) shows radio communication distance r [m] versus minimum detectable electric eld
intensity E
min
[dBV/m] where the RRS height deviation is selected as h = 10 [m] and its
correlation length is chosen as c = 100 [m]. Other parameters are chosen as the same values
as in Fig.8 (a). In these two gures, it is demonstrated that the larger the input power of a
source antenna becomes, the longer is the radio communication distance. It is also shown that
the smaller the minimum detectable electric eld intensity of a receiver becomes, the longer is
the radio communication distance. Thus, based on the proposed procedure, we can estimate
the radio communication distances along RRSs.
where the unit input power or P
i
=1 [W] in Eq.40 is assumed, and is an amplitude weighting
factor, is an order of propagation distance, and is a distance correction factor. Rewriting
the above relation in dB leads to the following equation [Hata,1980]:
E = 20 log
10
(r + ) [dB]. (44)
Next, we determine the three unknown constants , and based on the method of least
mean square (LMS) with the objective function dened by
(, , ) =
1
N
N
n=1
(E
d
n
E
n
)
2
min (45)
where N is the number of data. Moreover, E
d
n
is the averaged value of simulated or experi-
mental data at r = r
n
, and E
n
is the data computed by putting r = r
n
in Eq.43. Applying the
conjugate gradient method (CGM) [Press,1992] to Eq.45, we can numerically determine the
constants , and explicitly.
Fig. 7. Ensemble average of eld distribution compared with estimated eld distribution.
Fig.7 shows a comparison of the ensemble average of the eld distribution with the estimated
function based on the proposed algorithm. The estimated curve is given by Eq.43 with the
constants determined by CGM such as =22.6 [dB], =1.62 and =-6.85 [m]. It is worth noting
that the distance correction factor is small in comparison with the radio communication dis-
tance r, and thus we can neglect it as 0. Consequently, when we estimate the propagation
characteristics of electromagnetic waves traveling along RRSs, the most important parameters
are and in Eq.43.
4.2 Procedure for estimation of radio communication distance
In the preceding section, we have demonstrated that the ensemble average of the eld dis-
tributions of electromagnetic waves traveling along RRSs is well expressed in a simple and
analytic function with three parameters , and . In this section, we propose an algorithm
to estimate the radio communication distance when the two parameters for RRSs as well as
another two parameters for source and receiver are specied. The rst two parameters are the
height deviation h and the correlation length cl for RRSs, and the last two parameters are the
input power P
i
of a source antenna and the sensitivity of a receiver or the minimum detectable
electric intensity E
m
of the receiver.
Feasibility of the present method comes from the simplied form of the electric eld distri-
bution given by Eq.43. Once the three constants , and are determined based on the sim-
ulated or experimental data, we can analytically evaluate the radio communication distance
from a source with an arbitrary input power P
i
to a receiver with an arbitrary minimum de-
tectable electric intensity E
min
. In this case we have assumed that both the source and receiver
are on the homogeneous RRSs with the same parameters. Considering the dB expression such
as [dBm] for the input power P
i
and [dB V/m] for the minimum detectable electric intensity
E
min
, we can rewrite Eq.44 in the following form
E
min
120 = 20 log
10
(r + ) + P
i
30. (46)
Solving for r in the above equation, we can analytically determine the radio communication
distance r by the following relation:
r = 10
(+P
i
E
min
+90)/20
10
(+P
i
E
min
+90)/20
. (47)
Eq.47 is the estimation formula for the radio communication distance between the source
antenna and receiver placed on homogeneous RRSs. The parameters of the RRSs are the height
deviation h [m] and correlation length c [m]. Moreover, the input power of the source is
denoted by P
i
[dBm] and the minimum detectable electric intensity of th receiver is described
by E
min
[dB V/m]. As mentioned earlier, we can neglect the distance correction factor , and
we have the more simplied estimation formula in Eq.47.
We show some numerical examples of the radio communication distances along RRSs com-
puted by Eq.47. Fig.8 (a) shows radio communication distance r [m] versus the minimum de-
tectable electric eld intensity E
min
[dBV/m] of the receiver with the input power P
i
[dBm] of
the source antenna as a parameter. The height deviation of the RRSs is selected as h = 10 [m]
and the correlation length is chosen as cl = 50 [m]. The spectrum of the RRSs is assumed
to be Gaussian, and the material constants of the RRS are chosen such that dielectric con-
stant is
r
=5 and conductivity is = 0.023 [S/m]. The operating frequency is chosen
as f = 1 [GHz]. The minimum detectable electric intensity E
min
is varied from 40 [dB
V/m]to 80 [dBV/m], where the parameters for input power of source antenna are selected
as P
i
= 10, 8, , 8, 10 [dBm].
Fig.8 (b) shows radio communication distance r [m] versus minimum detectable electric eld
intensity E
min
[dBV/m] where the RRS height deviation is selected as h = 10 [m] and its
correlation length is chosen as c = 100 [m]. Other parameters are chosen as the same values
as in Fig.8 (a). In these two gures, it is demonstrated that the larger the input power of a
source antenna becomes, the longer is the radio communication distance. It is also shown that
the smaller the minimum detectable electric eld intensity of a receiver becomes, the longer is
the radio communication distance. Thus, based on the proposed procedure, we can estimate
the radio communication distances along RRSs.
(a) h = 10 m and c = 50 m (b) h = 10 m and c = 100 m
Fig. 8. Radio communication distances along RRSs with Gaussian spectrum.
Fig.9 (a) and Fig.9 (b) show the radio communication distance versus minimum detectable
electric intensity of a receiver located above the RRSs with the 1st order of power-law spec-
trum. The RRS parameters are chosen as the height deviation is h = 10 [m] and the correlation
length is c = 50 [m] in (a), and h = 10 [m] and c = 100 [m] in (b), respectively. Other pa-
rameters are selected as the same as the former two Gaussian cases in Fig.8.
Fig. 9. Radio communication distances along RRSs with power-law spectrum.
electric intensity along the RRSs with exponential type of spectrum. The RRS parameters
are chosen as h = 10 [m] and c = 50 [m] in (a), and h = 10 [m] and c = 100 [m] in (b),
respectively. Other parameters are the same as the former examples in Figs.8 and 9.
It is evident from the above numerical examples that the radio communication distances vary
depending on the types of the RRS spectra, even though the RRS parameters h and c are the
same. It is shown that the radio communication distances along the RRSs with the Gaussian
type of spectrum are the longest of the three types of spectra and those with the exponential
type of spectrum are the shortest of the three.
Fig. 10. Radio communication distances along RRSs with exponential spectrum.
5. Conclusion
In this chapter, from a viewpoint of the application of radio communications to sensor net-
works, we have rst discussed the convolution method to generate random rough surfaces
(RRSs) to numerically simulate the propagation characteristics of electromagnetic waves trav-
eling along the RRSs. Second, we have introduced the discrete ray tracing method (DRTM)
to numerically evaluate the distributions of the electromagnetic waves along the RRSs. The
remarkable point of the present method is to discretize not only the RRSs prole but also the
ray tracing itself, resulting in saving much computation time. Third, we have proposed an
algorithm to estimate the ensemble average of the eld distributions in a simple and analytic
expressionby introducing the amplitude weighting factor , the order of propagation distance
and the distance correction factor . Finally, we have introduced a formula to estimate the
radio communication distance along RRSs, provided that the input power of a source antenna
and the minimum detectable electric eld intensity of a receiver are specied.
Numerical calculations were carried out for the dependence of radio communication distance
versus minimum detectable electric eld intensity of a receiver with the input power of a
source antenna as a parameter. The results of computer simulations have revealed us that the
longer the correlation length of the RRSs becomes, the longer are the radio communication
distances. It has also been found that the radio communication distances along the RRSs vary
depending on the types of spectra of the RRSs, and those of the Gaussian type of spectrum are
longer than those of any other types of spectra.
We have treated only the homogeneous 1D RRSs to estimate the radio communication dis-
tances. In a more realistic situation, however, it is required to deal with inhomogeneous 2D
RRSs [Uchida,2009]. This deserves as a near future investigation.
6. References
Aryanfar, F. & Sarabandi, K. (2007). Validation of Wireless Channel Models Using a Scaled
mm-Wave Measurement System, IEEE Antennas and Propagation Magazine, vol.49,
no.4, pp.124-134.
Collin, R.E. (1985). Antennas and Radio Wave Propagation, McGraw-Hill Inc., New York, pp.13-
86.
Hata, M. (1980). Empirical formula for propagation loss in land mobile radio services, IEEE
Trans. Veh. Technol., VT-29(3), pp.317-325.
Fig. 8. Radio communication distances along RRSs with Gaussian spectrum.
electric intensity of a receiver located above the RRSs with the 1st order of power-law spec-
trum. The RRS parameters are chosen as the height deviation is h = 10 [m] and the correlation
length is c = 50 [m] in (a), and h = 10 [m] and c = 100 [m] in (b), respectively. Other pa-
rameters are selected as the same as the former two Gaussian cases in Fig.8.
Fig. 9. Radio communication distances along RRSs with power-law spectrum.
electric intensity along the RRSs with exponential type of spectrum. The RRS parameters
are chosen as h = 10 [m] and c = 50 [m] in (a), and h = 10 [m] and c = 100 [m] in (b),
respectively. Other parameters are the same as the former examples in Figs.8 and 9.
It is evident from the above numerical examples that the radio communication distances vary
depending on the types of the RRS spectra, even though the RRS parameters h and c are the
same. It is shown that the radio communication distances along the RRSs with the Gaussian
type of spectrum are the longest of the three types of spectra and those with the exponential
type of spectrum are the shortest of the three.
Fig. 10. Radio communication distances along RRSs with exponential spectrum.
5. Conclusion
In this chapter, from a viewpoint of the application of radio communications to sensor net-
works, we have rst discussed the convolution method to generate random rough surfaces
(RRSs) to numerically simulate the propagation characteristics of electromagnetic waves trav-
eling along the RRSs. Second, we have introduced the discrete ray tracing method (DRTM)
to numerically evaluate the distributions of the electromagnetic waves along the RRSs. The
remarkable point of the present method is to discretize not only the RRSs prole but also the
ray tracing itself, resulting in saving much computation time. Third, we have proposed an
algorithm to estimate the ensemble average of the eld distributions in a simple and analytic
expressionby introducing the amplitude weighting factor , the order of propagation distance
and the distance correction factor . Finally, we have introduced a formula to estimate the
radio communication distance along RRSs, provided that the input power of a source antenna
and the minimum detectable electric eld intensity of a receiver are specied.
Numerical calculations were carried out for the dependence of radio communication distance
versus minimum detectable electric eld intensity of a receiver with the input power of a
source antenna as a parameter. The results of computer simulations have revealed us that the
longer the correlation length of the RRSs becomes, the longer are the radio communication
distances. It has also been found that the radio communication distances along the RRSs vary
depending on the types of spectra of the RRSs, and those of the Gaussian type of spectrum are
longer than those of any other types of spectra.
We have treated only the homogeneous 1D RRSs to estimate the radio communication dis-
tances. In a more realistic situation, however, it is required to deal with inhomogeneous 2D
RRSs [Uchida,2009]. This deserves as a near future investigation.
6. References
Aryanfar, F. & Sarabandi, K. (2007). Validation of Wireless Channel Models Using a Scaled
mm-Wave Measurement System, IEEE Antennas and Propagation Magazine, vol.49,
no.4, pp.124-134.
Collin, R.E. (1985). Antennas and Radio Wave Propagation, McGraw-Hill Inc., New York, pp.13-
86.
Hata, M. (1980). Empirical formula for propagation loss in land mobile radio services, IEEE
Trans. Veh. Technol., VT-29(3), pp.317-325.
Heidrich, J. et al. (2010). The Roots, Rules, and Rise of RFID, IEEE Microwave Magazine, vol.
11, no. 3 , pp.78-86.
Honda, J. et al. (2006). Effect of rough surface spectrum on propagation characteristics, IEEJ
Technical Reports, EMT-06-128, pp.65-70.
Honda, J. & Uchida, K. (2009). Discrete Ray-Tracing Method (DRTM) Analysis of Propagation
Characteristics along RandomRough Surface in Relation to Development of Wireless
Sensor Network, Proceedings of RWS 2009, TU2P-4, pp.248-251.
Honda, J., Uchida, K., Yoon, K.Y. (2010). Estimation of radio communication distance along
random rough surface, IEICE Trans. Electron., E93-C(1), pp.39-45.
Johnsonbaugh, R. & Kalin, M. (1997). C for Scientists and Engineers, Prentice-Hall, Inc., New
Jersey, pp.191-195.
Mushiake, Y. (1985). Antennas and Radio Propagation, Corona Publishing Co., Ltd., p.39.
Noble, B. (1958). Methods based on the Wiener-Hopf technique, Pergamon Press.
Phu, P., Ishimaru, A., Kuga Y. (1994). Co-polarized and cross-polarized enhanced backscatter-
ing from two-dimensional very rough surfaces at millimeter wave frequencies, Radio
Sci., 29(5), pp.1275-1291.
Press, W.H. et al. (1992). Numerical Recipes in Fortran 77: The Art of Scientic Computing,
Cambridge University Press, New York, pp.387-448.
Sato, M. (2002). Subsurface imaging by ground penetrating radar, IEICE Trans. Electron.,
vol.J85-C, no.7, pp.520-530.
Thoros, E.I. (1988). The validity of the Kirchhoff approximation for rough surface scattering
using a Gaussian roughness spectrum, J. Acoust. Soc. Am., 83(1), pp.78-92.
Thoros, E.I. (1990). Acoustic scattering from a Pierson-Moskowitz sea surface, J. Acoust. Soc.
Am., 88(1), pp.335-349.
Tsang, L., Chan, C.H., Pak, K. (1994). Backscattering enhancement of a two-dimensional ran-
dom rough surface (three-dimensional scattering) based on Monte Carlo simulation,
J. Opt. Soc. Am. A, 11(2), pp.711-715.
Uchida, K. et al. (2007). FVTD analysis of electromagnetic wave propagation along random
rough surface, IEICE Trans. Commun., J90-B(1), pp.48-55.
Uchida, K. et al. (2008). A@Analysis of electromagnetic wave propagation along rough surface
by using discrete ray tracing method, Proceedings of ISAP 2008, pp.939-942.
Uchida, K., Honda, J., Yoon, K.Y. (2009). Distance characteristics of propagation in relation to
inhomogeneity of random rough surface, Proceedings of ISMOT 2009, pp.1133-1136.
Yagbasan, A. et al. (2010). Characteristic Basis Function Method for solving Electromagnetic
scattering Problems Over Rough Terrain Proles, IEEE Trans. Antennas Propag., vol.
58, no. 5 , pp.1579-1589.
Yoon, K.Y., Tateiba, M., Uchida, K. (2000). FVTD simulation for random rough dielectric sur-
face scattering at low grazing angle, IEICE Trans. Electron., E83-C(12), 1836-1843.
Yoon, K.Y., Tateiba, M., Uchida, K. (2002). A numerical simulation of low-grazing-angle scat-
tering from ocean-like dielectric surfaces, IEICE Trans. Commun., E85-B(10), pp.2344-
2347.
Design of Radio-Frequency Transceivers for Wireless Sensor Networks 249
Design of Radio-Frequency Transceivers for Wireless Sensor Networks
0
Design of Radio-Frequency Transceivers
for Wireless Sensor Networks
Department of Electronic Engineering, TNLIST, Tsinghua University
Beijing, China
[email protected]
[email protected]
1. Introduction
The SoC (System-on-Chip) design for the WSN (Wireless Sensor Networks) nodes is the most
signicant technology of modern WSN design. There are a large amount of nodes in a WSN
system, the nodes are densely deployed either inside the environment or very close to it.
Each node is equipped with a sensor, an ADC (Analog-to-Digital Converter), a MCU (Mi-
cro Controller Unit), a storage unit, a power management unit, and a RF (Radio-Frequency)
transceiver, as shown in Fig. 1, so that it can sense, store, process, and communicate with other
sensors using multi-hop packet transmissions. The basic specications of WSN are reliability,
accuracy, exibility, expenses, the difculty of development and power consumption. Because
all the nodes are battery-powered, power consumption is the most important specication of
WSN.
The core part of a WSN node is the RF transceiver, which is used to realize the wireless com-
munication among the nodes. For a common used commercial chip, the distribution of power
consumption is shown in Fig. 2, where TX and RX represent the transmitting mode and re-
ceiving mode of the transceiver. We can see that the RF part consumes the most power. Be-
sides, modern RF design is composed of so many subjects that it requires IC designers to
have sufcient knowledge. As a result, the IC design of RF transceivers becomes the most
challenging research topic in the WSN eld.
As the applications of WSN become more and more widespread, many companies have de-
veloped highly-integrated chips for RF transceivers. The main specications of several com-
mercial chips are shown in Table. 1. The common characters of these chips can be summarized
into several aspects: 1)a low data rate, 2)low power consumption, 3)high sensitivity, 4)rela-
tively low output power, and 5)a simple modulation scheme.
There are several ways to achieve low power consumption in WSN: 1)reduce the radiated
power by using ad-hoc networks and multi-hop communication, 2)optimize the trade-off be-
tween communication and local computing, 3)design more power-efcient RF transceivers,
and 4)develop more energy-efcient protocols and routing algorithms. And the third one is
what we will talk about in this chapter.
14
Sensor ADC MCU
Transceiver
Power Management Unit
Storage Antenna
Fig. 1. The Wireless Sensor Networks.
C
u
r
r
e
n
t

C
o
n
s
u
m
p
t
i
o
n

(
m
A
)
Fig. 2. The power distribution in a commercial WSN chip.
2. The classication of WSN transceivers
In this section, we make a classication of WSN transceivers, which are sorted by both mod-
ulation schemes and system architectures. Although some complex schemes such as OFDM
(Orthogonal Frequency Division Multiplexing) can be adopted for the prospect of spectrum
effective usage, RF designers are still inclined to simple schemes such as OOK (On-Off Key-
ing), FSK (Frequency Shift Keying), UWB (Ultra-Wide Band), MSK (Minimum Shift Keying),
BPSK (Binary Phase Shift Keying), QPSK (Quadrature Phase Shift Keying), and so on for low
power consideration. From Table. 1, we can see that these simple modulation schemes are
often adopted in the common-used commercial chips. In this section, we will analyze the
advantages and disadvantages of the transceivers with these modulation schemes.
Chips FB
a
(MHz)DR
b
(kbps)Power
c
(mW) Sensitivity (dBm) OTP
d
(dBm) MS
e
TR1000 916.5 115.2 14.4/36 -98 -1.2 OOK/ASK
TRF6903 3001000 19.2 60/111 -103 -128 FSK/OOK
CC1000 3001000 76.8 30/87.8 -107 -2010 FSK
CC2420 2400 250 33.8/31.3 -95 0 O-QPSK
nRF905 433915 100 37.5/90 -100 -1010 GFSK
nRF2401 2400 01000 75/39 -80 -200 GFSK
a
FB: The working Frequency Band of the chips.
b
DR: The Data Rate of the transceivers.
c
Power: The Power consumption for receiving/transmitting mode.
d
OTP: The Output Transmitting Power.
e
MS: The Modulation Scheme of the transceiver.
Table 1. The main specications of several commercial RF chips common used in WSN.
2.1 The OOK transceiver
The OOK transceiver can often be realized by a simple architecture, since both the modulator
and demodulator are easy to implement. As a result, the power consumption can be reduced.
Another advantage is that a high data rate can be obtained by OOK. However, an AGC (Au-
tomatic Gain Control) with a wide dynamic range is often needed; a special coding is needed
to avoid the saturation caused by long series of 0 or 1 in the receiver; it is spectrally inef-
cient; the most serious defect of OOK is that it is strongly susceptible to interferers, then the
maximum communication distance of OOK transceivers is usually not long.
As a high data rate can be obtained by OOK transceivers, there are some works focused on
multi-gigabit short-range wireless communications (Jri et al., 2010). The transceivers work
at several GHz with low energy per bit. Besides, since the power consumption of OOK
transceivers can be made very small, some recently works design OOK receivers for wake-up
usages (Pletcher et al., 2009; Seungkee et al., 2010). The power consumption of these wake-up
receivers can be as low as a few tens of W.
One typical architecture of OOK transceivers is shown in Fig. 3. In the transmitter, the base-
band data is modulated by a mixer with a carrier generated by an oscillator or a PLL (Phase-
Locked Loop), and then the modulated signals are amplied by a PA (Power Amplier) and
emitted through an antenna. In the receiver, the signals received from an antenna is amplied
by a LNA (Low-Noise Amplier) rstly, and then detected by an envelop detector. At last, a
simple DEM (DEModulator) can be used to demodulate the signals into 1-0 series. Therefore,
in short-range and high-speed WSN applications, the OOK transceiver is a reasonable choice.
2.2 The FSK transceiver
As only the zero-crossing points of the signal contain useful information, FSK transceivers
can work without an AGC, and special coding is not necessary. The modulator and demod-
ulator are also easy to realize. As a result, the FSK transceiver has a simple architecture that
ensures low power consumption and low cost. The spectrum efcient and ability to avoid
interferences of FSK are both higher than that of OOK, so the communication distance of FSK
transceivers can be longer. Besides, frequency hopping can be realized for FSKschemes. How-
ever, the complexity of FSK transceivers is relatively higher than that of OOK transceivers.
Sensor ADC MCU
Transceiver
Power Management Unit
Storage Antenna
Fig. 1. The Wireless Sensor Networks.
C
u
r
r
e
n
t

C
o
n
s
u
m
p
t
i
o
n

(
m
A
)
Fig. 2. The power distribution in a commercial WSN chip.
2. The classication of WSN transceivers
In this section, we make a classication of WSN transceivers, which are sorted by both mod-
ulation schemes and system architectures. Although some complex schemes such as OFDM
(Orthogonal Frequency Division Multiplexing) can be adopted for the prospect of spectrum
effective usage, RF designers are still inclined to simple schemes such as OOK (On-Off Key-
ing), FSK (Frequency Shift Keying), UWB (Ultra-Wide Band), MSK (Minimum Shift Keying),
BPSK (Binary Phase Shift Keying), QPSK (Quadrature Phase Shift Keying), and so on for low
power consideration. From Table. 1, we can see that these simple modulation schemes are
often adopted in the common-used commercial chips. In this section, we will analyze the
advantages and disadvantages of the transceivers with these modulation schemes.
Chips FB
a
(MHz)DR
b
(kbps)Power
c
(mW) Sensitivity (dBm) OTP
d
(dBm) MS
e
TR1000 916.5 115.2 14.4/36 -98 -1.2 OOK/ASK
TRF6903 3001000 19.2 60/111 -103 -128 FSK/OOK
CC1000 3001000 76.8 30/87.8 -107 -2010 FSK
CC2420 2400 250 33.8/31.3 -95 0 O-QPSK
nRF905 433915 100 37.5/90 -100 -1010 GFSK
nRF2401 2400 01000 75/39 -80 -200 GFSK
a
FB: The working Frequency Band of the chips.
b
DR: The Data Rate of the transceivers.
c
Power: The Power consumption for receiving/transmitting mode.
d
OTP: The Output Transmitting Power.
e
MS: The Modulation Scheme of the transceiver.
Table 1. The main specications of several commercial RF chips common used in WSN.
2.1 The OOK transceiver
The OOK transceiver can often be realized by a simple architecture, since both the modulator
and demodulator are easy to implement. As a result, the power consumption can be reduced.
Another advantage is that a high data rate can be obtained by OOK. However, an AGC (Au-
tomatic Gain Control) with a wide dynamic range is often needed; a special coding is needed
to avoid the saturation caused by long series of 0 or 1 in the receiver; it is spectrally inef-
cient; the most serious defect of OOK is that it is strongly susceptible to interferers, then the
maximum communication distance of OOK transceivers is usually not long.
As a high data rate can be obtained by OOK transceivers, there are some works focused on
multi-gigabit short-range wireless communications (Jri et al., 2010). The transceivers work
at several GHz with low energy per bit. Besides, since the power consumption of OOK
transceivers can be made very small, some recently works design OOK receivers for wake-up
usages (Pletcher et al., 2009; Seungkee et al., 2010). The power consumption of these wake-up
receivers can be as low as a few tens of W.
One typical architecture of OOK transceivers is shown in Fig. 3. In the transmitter, the base-
band data is modulated by a mixer with a carrier generated by an oscillator or a PLL (Phase-
Locked Loop), and then the modulated signals are amplied by a PA (Power Amplier) and
emitted through an antenna. In the receiver, the signals received from an antenna is amplied
by a LNA (Low-Noise Amplier) rstly, and then detected by an envelop detector. At last, a
simple DEM (DEModulator) can be used to demodulate the signals into 1-0 series. Therefore,
in short-range and high-speed WSN applications, the OOK transceiver is a reasonable choice.
2.2 The FSK transceiver
As only the zero-crossing points of the signal contain useful information, FSK transceivers
can work without an AGC, and special coding is not necessary. The modulator and demod-
ulator are also easy to realize. As a result, the FSK transceiver has a simple architecture that
ensures low power consumption and low cost. The spectrum efcient and ability to avoid
interferences of FSK are both higher than that of OOK, so the communication distance of FSK
transceivers can be longer. Besides, frequency hopping can be realized for FSKschemes. How-
ever, the complexity of FSK transceivers is relatively higher than that of OOK transceivers.
Mixer
Oscillator
PA
Antenna
Envelop
Detector
DEM LNA
Antenna
Transmitter
Receiver
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d
Fig. 3. A typical OOK transceiver.
A typical FSK transceiver is shown in Fig. 4. For the transmitter, the PLL directly digital mod-
ulation can be adopted (Perrott et al., 1997). This technology will be described in detail in
section 4.4. The data stream from baseband is shaped rstly, and then input into a PLL to gen-
erate the FSK signals. Then the FSK signals are amplied by a PA and emitted by an antenna.
In the receiver, the received signals are amplied by a LNA, and down-converted by a mixer
that is followed by a lter to depress the interferences and the high-frequency component. The
FSK demodulator is easy to realize since only the zero-cross points contain data information,
and sometimes a zero-cross detector is enough. However, in many applications, the inuence
caused by frequency offset needs to be avoided. The power consumption of such transceiver
is relatively low because the modulation and demodulation of FSK signals is easy to realize.
As we analyzed above, OOK can realize a high data rate, but the communication distance
is short; FSK transceivers can be used for long-distance communication, but the data rate is
often not high. Therefore, a multi-mode transceiver in which OOKand FSKcan be compatible
with each other may be a good choice. The OOK modulation can be implemented by directly
modulating the output power of the PA in Fig. 4.
2.3 The UWB transceiver
UWB is dened as a signal that occupies a bandwidth wider than 500 MHz or a fractional
bandwidth larger than 20%. Although UWB is not a modulation scheme, we list the UWB
transceivers individually here because the advantages of UWB are remarkable: 1)a high data,
2)low cost and low power consumption, and 3)high accuracy in distance measuring due to
PLL
Shaping
Filter
Filter DEM
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d PA
LNA
Antenna
Antenna
Transmitter
Receiver
Mixer
Fig. 4. An typical FSK transceiver.
the narrow pulse width of a few nanoseconds. However, the disadvantage is also serious:
the spectrum is too wide and the emitting power is limited, so the single-hop transmission
distance is very short (usually less than 10 m).
The IR (Impulse-Radio) UWB transceiver can be implemented with ultra-low power con-
sumption. A typical example is shown in Fig. 5, where the architecture is similar to that of
the typical OOK transceiver. The transmitter has a very simple structure, which is composed
of an impulse generator and a PA. The impulse generator can be realized by a few logical
gates. The receiver includes a LNA, a mixer as a squarer, an integrator and a comparator.
In some applications, the architecture is even simpler than that of common OOK transceivers.
Therefore, ultra-lowpower can be obtained with UWB transceivers, which ts the short-range
WSN requirement very well.
2.4 Other Modulation Schemes
The spectrum efciency of MSK is higher than FSK, but the demodulation is more complex
and frequency hopping can not be realized for MSK. In 2003, IEEE released 802.15.4, which
supports BPSKand O-QPSK(Offset-QPSK). The ability of BPSKto avoid interferences is better
than that of FSK; and O-QPSK is more spectrally efcient. But the demodulation circuits
of both BPSK and O-QPSK are complex and ADCs are usually needed. For general WSN
applications that have a low data rate at several hundreds of Hz, the usage of these complex
modulations is not recommended. However, in a few applications that have a high data rate
and a strict spectrum limitation, MSK, BPSK, O-QPSK and even OFDM can be adopted.
2.5 The classication based on system architectures
In this section, we sort the transceivers according to the systemarchitecture, and the types can
be divided into superheterodyne, zero-IF (Intermediate Frequency), low-IF, slide-IF, super-
regenerative, amplier-sequenced hybrid architectures, and so on. These kinds of transceivers
are summarized into Table 2. In WSN applications, low-IF transceivers are often adopted for
high integration and low power consumption.
Mixer
Oscillator
PA
Antenna
Envelop
Detector
DEM LNA
Antenna
Transmitter
Receiver
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d
Fig. 3. A typical OOK transceiver.
A typical FSK transceiver is shown in Fig. 4. For the transmitter, the PLL directly digital mod-
ulation can be adopted (Perrott et al., 1997). This technology will be described in detail in
section 4.4. The data stream from baseband is shaped rstly, and then input into a PLL to gen-
erate the FSK signals. Then the FSK signals are amplied by a PA and emitted by an antenna.
In the receiver, the received signals are amplied by a LNA, and down-converted by a mixer
that is followed by a lter to depress the interferences and the high-frequency component. The
FSK demodulator is easy to realize since only the zero-cross points contain data information,
and sometimes a zero-cross detector is enough. However, in many applications, the inuence
caused by frequency offset needs to be avoided. The power consumption of such transceiver
is relatively low because the modulation and demodulation of FSK signals is easy to realize.
As we analyzed above, OOK can realize a high data rate, but the communication distance
is short; FSK transceivers can be used for long-distance communication, but the data rate is
often not high. Therefore, a multi-mode transceiver in which OOKand FSKcan be compatible
with each other may be a good choice. The OOK modulation can be implemented by directly
modulating the output power of the PA in Fig. 4.
2.3 The UWB transceiver
UWB is dened as a signal that occupies a bandwidth wider than 500 MHz or a fractional
bandwidth larger than 20%. Although UWB is not a modulation scheme, we list the UWB
transceivers individually here because the advantages of UWB are remarkable: 1)a high data,
2)low cost and low power consumption, and 3)high accuracy in distance measuring due to
PLL
Shaping
Filter
Filter DEM
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d PA
LNA
Antenna
Antenna
Transmitter
Receiver
Mixer
Fig. 4. An typical FSK transceiver.
the narrow pulse width of a few nanoseconds. However, the disadvantage is also serious:
the spectrum is too wide and the emitting power is limited, so the single-hop transmission
distance is very short (usually less than 10 m).
The IR (Impulse-Radio) UWB transceiver can be implemented with ultra-low power con-
sumption. A typical example is shown in Fig. 5, where the architecture is similar to that of
the typical OOK transceiver. The transmitter has a very simple structure, which is composed
of an impulse generator and a PA. The impulse generator can be realized by a few logical
gates. The receiver includes a LNA, a mixer as a squarer, an integrator and a comparator.
In some applications, the architecture is even simpler than that of common OOK transceivers.
Therefore, ultra-lowpower can be obtained with UWB transceivers, which ts the short-range
WSN requirement very well.
2.4 Other Modulation Schemes
The spectrum efciency of MSK is higher than FSK, but the demodulation is more complex
and frequency hopping can not be realized for MSK. In 2003, IEEE released 802.15.4, which
supports BPSKand O-QPSK(Offset-QPSK). The ability of BPSKto avoid interferences is better
than that of FSK; and O-QPSK is more spectrally efcient. But the demodulation circuits
of both BPSK and O-QPSK are complex and ADCs are usually needed. For general WSN
applications that have a low data rate at several hundreds of Hz, the usage of these complex
modulations is not recommended. However, in a few applications that have a high data rate
and a strict spectrum limitation, MSK, BPSK, O-QPSK and even OFDM can be adopted.
2.5 The classication based on system architectures
In this section, we sort the transceivers according to the systemarchitecture, and the types can
be divided into superheterodyne, zero-IF (Intermediate Frequency), low-IF, slide-IF, super-
regenerative, amplier-sequenced hybrid architectures, and so on. These kinds of transceivers
are summarized into Table 2. In WSN applications, low-IF transceivers are often adopted for
high integration and low power consumption.
Impulse Generator
Integrator Comparator LNA
PA
Antenna
Antenna
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d
Transmitter
Receiver
Fig. 5. A typical UWB transceiver.
A common transceiver architecture is shown in Fig. 6. In the transmitter, there is I/Q dual-
path chains to realize the vector-signal transmitting. The coding, spreading, shaping, and so
on are all nished in the digital domain. The data stream from digital baseband is converted
to analog signals by two DACs (Digital-to-Analog Converters), and then the interferences
and harmonics are ltered out by two lters. Two mixers are adopted for I/Q dual-path up-
conversion and modulation. The phase difference of the two LO (Local Oscillation) signals
input to the two mixers is 90
. At last, the I/Qsignals are summed by an adder, amplied by a

PA, and then emitted through an antenna. Assume the signal in the I path is a, the signal in the
Qpath is b, and then the transmission signal can be expressed as a +jb. Therefore, all the QAM
(Quadrature Amplitude Modulation) signals in the constellation diagramcan be generated by
this transmitter. In the receiver, the received signals are amplied by a LNA rstly, and then
down-converted by two mixers into I/Q dual-path signals. After being ltered by a lter,
the analog signals are converted to digital domain by two ADCs. At last, the demodulating,
de-spreading, synchronizing, decoding and so on are all done in digital domain. Such is the
common architecture of SDR (Software Dened Radio), all the modulation schemes can be
realized in this transceiver. Generally, the WSN usages have the characters of low cost, low
power and a low data rate, so such complex architecture of SDR is not necessary in many
WSN applications since simple analog circuits can be adequate. However, for specially WSN
applications that need a complex modulation scheme, the SDR transceiver is indispensable.
3. The system design of transceivers
In this section, we will describe how to make system-level plans for the transceivers. The
system-level specications of a receiver are composed of sensitivity, maximum received sig-
nal, co-channel interferences, ACS (Adjacent/Alternate Channel Selectivity), image rejection,
desensitization or blocking, and so on. And the specications of a transmitter include ACPR
(Adjacent Channel Power Ratio), EVM (Error Vector Magnitude), noise emission, and so
on. These system-level specications are decided by the protocols and network formation
of WSN, and are affected by every submodules performance, such as noise gure, gain, third
intercept point, and so on. The main task for the system designers is to assign the system-
Architecture Advantages Disadvantages
Superheterodyne
Simple structure Poor integration level
High sensitivity Large power consumption
High selectivity
Zero-IF
a
Easy to be integrated DC offset
Low power Flicker noise
Low-IF
Easy to be integrated Image rejection
Immune to DC offset Difcult baseband design
Slide-IF
Small image interference Difcult for LO design
Easy to be integrated
super-regenerative
High sensitivity Low data rate
Low power Seriously affected by PVT
b
Poor stability
Amplier-sequenced hybrid
Good stability Low data rate
Low power
High Sensitivity
a
IF: Intermediate Frequency
b
PVT: Process, Voltage, and Temperature
Table 2. The classication based on system architecture.
PA
Antenna
LNA
Antenna
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d
Filter
Filter
DAC
DAC
Filter
ADC
ADC
PLL
Mixer
Mixer
Mixer
Mixer
Transmitter
Receiver
Adder
Fig. 6. A common architecture of transceivers.
level specications to every individual submodules. Some of the conclusions in this section
are referenced to Gus book (Gu, 2005).
Impulse Generator
Integrator Comparator LNA
PA
Antenna
Antenna
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d
Transmitter
Receiver
Fig. 5. A typical UWB transceiver.
A common transceiver architecture is shown in Fig. 6. In the transmitter, there is I/Q dual-
path chains to realize the vector-signal transmitting. The coding, spreading, shaping, and so
on are all nished in the digital domain. The data stream from digital baseband is converted
to analog signals by two DACs (Digital-to-Analog Converters), and then the interferences
and harmonics are ltered out by two lters. Two mixers are adopted for I/Q dual-path up-
conversion and modulation. The phase difference of the two LO (Local Oscillation) signals
input to the two mixers is 90
. At last, the I/Qsignals are summed by an adder, amplied by a

PA, and then emitted through an antenna. Assume the signal in the I path is a, the signal in the
Qpath is b, and then the transmission signal can be expressed as a +jb. Therefore, all the QAM
(Quadrature Amplitude Modulation) signals in the constellation diagramcan be generated by
this transmitter. In the receiver, the received signals are amplied by a LNA rstly, and then
down-converted by two mixers into I/Q dual-path signals. After being ltered by a lter,
the analog signals are converted to digital domain by two ADCs. At last, the demodulating,
de-spreading, synchronizing, decoding and so on are all done in digital domain. Such is the
common architecture of SDR (Software Dened Radio), all the modulation schemes can be
realized in this transceiver. Generally, the WSN usages have the characters of low cost, low
power and a low data rate, so such complex architecture of SDR is not necessary in many
WSN applications since simple analog circuits can be adequate. However, for specially WSN
applications that need a complex modulation scheme, the SDR transceiver is indispensable.
3. The system design of transceivers
In this section, we will describe how to make system-level plans for the transceivers. The
system-level specications of a receiver are composed of sensitivity, maximum received sig-
nal, co-channel interferences, ACS (Adjacent/Alternate Channel Selectivity), image rejection,
desensitization or blocking, and so on. And the specications of a transmitter include ACPR
(Adjacent Channel Power Ratio), EVM (Error Vector Magnitude), noise emission, and so
on. These system-level specications are decided by the protocols and network formation
of WSN, and are affected by every submodules performance, such as noise gure, gain, third
intercept point, and so on. The main task for the system designers is to assign the system-
Architecture Advantages Disadvantages
Superheterodyne
Simple structure Poor integration level
High sensitivity Large power consumption
High selectivity
Zero-IF
a
Easy to be integrated DC offset
Low power Flicker noise
Low-IF
Easy to be integrated Image rejection
Immune to DC offset Difcult baseband design
Slide-IF
Small image interference Difcult for LO design
Easy to be integrated
super-regenerative
High sensitivity Low data rate
Low power Seriously affected by PVT
b
Poor stability
Amplier-sequenced hybrid
Good stability Low data rate
Low power
High Sensitivity
a
IF: Intermediate Frequency
b
PVT: Process, Voltage, and Temperature
Table 2. The classication based on system architecture.
PA
Antenna
LNA
Antenna
D
i
g
i
t
a
l

B
a
s
e
b
a
n
d
Filter
Filter
DAC
DAC
Filter
ADC
ADC
PLL
Mixer
Mixer
Mixer
Mixer
Transmitter
Receiver
Adder
Fig. 6. A common architecture of transceivers.
level specications to every individual submodules. Some of the conclusions in this section
are referenced to Gus book (Gu, 2005).
For the AWGN (Additive White Gaussian Noise) channel, the noise at the input of a receiver
chain can be equivalent to thermal noise, and the sensitivity of a receiver can be expressed as:
R
min
(dBm) = 174 +10log(BW) + NF + CNR
min
(1)
where BW is the receiver noise bandwidth in Hz, NF is the overall noise gure of the receiver
in dB, and CNR
min
is the minimum CNR (Carrier-to-Noise Ratio) required for obtaining the
required error rate.
As two different radio transmitters may use the same frequency in actual applications, RF re-
ceivers must have a certain ability to avoid the inuence of co-channel interferences. Another
important specication of receiver is the ACS, which measures the ability to receive a desired
signal at the operating channel frequency in the presence of adjacent/alternate channel sig-
nals at a frequency offset from the assigned channel. The blocking characteristic is similar to
the ACS, but it measures the anti-interference ability of an unwanted interferer at frequencies
other than those of adjacent channels. The adjacent/alternate channel interference signal is
usually modulated, while the blocking interferer is often dened as a continuous waveform
tone. Both the adjacent/alternate channel selectivity and the blocking characteristics of a re-
ceiver are determined mainly by the characters of the channel lter, and the phase noise and
spurs of the LOs in the adjacent/alternate channel bandwidth or around the unwanted inter-
ferer. For most wireless mobile systems, the desired signal level at receiver input for testing is
dened as 3 dB above the reference sensitivity level R
min
(Gu, 2005):
R
d,i
= R
min
+3dB (2)
For the single-conversion receiver, as shown in Fig. 6, it can be derived the ACS or the blocking
characteristic ACS
adj/alt/block
to have the following form:
ACS
adj/alt/block
= I
adj/alt/block
R
d,i
(3)
I
adj/alt/block
= 10lg
10
R
d,i
CNR
10
10
174+10lgBW+NF
10
10
PN+10lgBWR
IF
10
+10
SPR
IF
10
(4)
where CNR is the carrier-to-noise ratio for a given bit error rate, BW is the receiver noise
bandwidth, NF is the noise gure of the RF front end, PN and SP are the phase noise and
spurs of the LO signals, and R
IF
is the relative rejection of the IF channel lter to the adja-
cent/alternate channel signal or the blocking interferer.
The maximum received signal is decided by both the 1-dB compression point and the inter-
modulation behavior of the receiver. For a RF receiver chain with multi-stages, the input
signal strength needs to be smaller than the 1-dB compression point of the rst stage, and
the output signal strength of every previous stage must be smaller than the 1-dB composition
point of the following stage, otherwise the receiver will be saturated. The maximum received
signal is also limited by the intermodulation because the nonlinearity will become more se-
rious under larger signal strength. The intermodulation characteristics reect the linearity of
the receiver, and it is limited by the noise gure, the receiver noise bandwidth, the phase noise
and spurs of the LOs, and the cross-modulation. For the single-conversion receiver, as shown
in Fig. 6, the expression of the allowed maximum degradation of the desired signal caused by
noise and interferences at the receiver input is:
D
max
= R
d,i
CNR
min
(5)
The minimum third intercept point of a receiver can be expressed as:
I IP3
min
=
1
2
3I
min
10lg
10
Dmax
10
10
N
nf
10
P
pn
P
sp
10
Ncm
10
(6)
where I
min
is the minimum intermodulation interference tone level at receiver input. P
pn
and
P
sp
are the phase noise and spurs contributions of LO, and they can be expressed as:
P
pn
= 10
Npn+10lgBW+I
in
R
10
(7)
P
sp
= 10
Nsp+I
in
R
10
(8)
where N
pn
is the average phase noise over the receiver bandwidth at a frequency offset equal
to the space between the interference and the carrier, N
sp
is the magnitude of the spurs at a
frequency offset equal to or nearby the space between the interference and the carrier, I
in
is the
intermodulation interference tone, and R is the rejection to the interference tone of a lter.
N
nf
denotes the receiver inherent noise converted to its input port:
N
nf
= 174 + NF +10lgBW (9)
N
cm
is the cross-modulation product, which can be approximately expressed as:
N
cm
= I
in
2I IP3
lna
+2(P
Tx
+ IL
dRx
R
dTx
) + C (10)
where I IP3
lna
is the LNA input third intercept point, P
Tx
is the transmitter output power at
the antenna port in dBm, IL
dRx
is the receiver side insertion loss of the duplexer in dB, R
dTx
is the duplexer receiver side lter rejection to the transmission in dB, and C is a correction
factor associated with waveformmagnitude uctuation and interference tone offset frequency,
which is approximately -3.8 dB and -5.8 dB for the cellular and PCS band CDMA mobile
station receivers, respectively (Gu, 2005).
Then another specication can be introduced, that is the dynamic range, which is generally de-
ned as the ratio of the maximum input level to the minimum input level at which the circuit
satised the required bit error rate. A common used specication called SFDR (Spurious-Free
Dynamic Range) bases the denition of the upper end of the dynamic range on the intermod-
ulation behavior and the lower end on the sensitivity. The SFDR can be expressed as:
SFDR =
2(I IP3 NF)
3
SNR
min
(11)
where I IP3 is the input third intercept point of the RF front end, SNR
min
is the required
minimum SNR (Signal-to-Noise Ratio) of the IF circuits. The required I IP3 can be calculated
by equation (6), then the SFDR of a receiver can be obtained by equation (11).
Two strong in-band interference tones may cause a zero-IF or low-IF receiver to be completely
jammed if the second-order intercept point of the receiver is not high enough. The two strong
tones may generate in-channel interferences due to the second-order distortion of the receiver.
The maximum level of the two equal blocking interference tones can be expressed as (Gu,
2005):
I
block
=
1
2
10lg
10
R
d,i
CNR
min
10
10
N
nf
10
+ I IP2
(12)
where I IP2 is the input second intercept point of the receiver.
For the AWGN (Additive White Gaussian Noise) channel, the noise at the input of a receiver
chain can be equivalent to thermal noise, and the sensitivity of a receiver can be expressed as:
R
min
(dBm) = 174 +10log(BW) + NF + CNR
min
(1)
where BW is the receiver noise bandwidth in Hz, NF is the overall noise gure of the receiver
in dB, and CNR
min
is the minimum CNR (Carrier-to-Noise Ratio) required for obtaining the
required error rate.
As two different radio transmitters may use the same frequency in actual applications, RF re-
ceivers must have a certain ability to avoid the inuence of co-channel interferences. Another
important specication of receiver is the ACS, which measures the ability to receive a desired
signal at the operating channel frequency in the presence of adjacent/alternate channel sig-
nals at a frequency offset from the assigned channel. The blocking characteristic is similar to
the ACS, but it measures the anti-interference ability of an unwanted interferer at frequencies
other than those of adjacent channels. The adjacent/alternate channel interference signal is
usually modulated, while the blocking interferer is often dened as a continuous waveform
tone. Both the adjacent/alternate channel selectivity and the blocking characteristics of a re-
ceiver are determined mainly by the characters of the channel lter, and the phase noise and
spurs of the LOs in the adjacent/alternate channel bandwidth or around the unwanted inter-
ferer. For most wireless mobile systems, the desired signal level at receiver input for testing is
dened as 3 dB above the reference sensitivity level R
min
(Gu, 2005):
R
d,i
= R
min
+3dB (2)
For the single-conversion receiver, as shown in Fig. 6, it can be derived the ACS or the blocking
characteristic ACS
adj/alt/block
to have the following form:
ACS
adj/alt/block
= I
adj/alt/block
R
d,i
(3)
I
adj/alt/block
= 10lg
10
R
d,i
CNR
10
10
174+10lgBW+NF
10
10
PN+10lgBWR
IF
10
+10
SPR
IF
10
(4)
where CNR is the carrier-to-noise ratio for a given bit error rate, BW is the receiver noise
bandwidth, NF is the noise gure of the RF front end, PN and SP are the phase noise and
spurs of the LO signals, and R
IF
is the relative rejection of the IF channel lter to the adja-
cent/alternate channel signal or the blocking interferer.
The maximum received signal is decided by both the 1-dB compression point and the inter-
modulation behavior of the receiver. For a RF receiver chain with multi-stages, the input
signal strength needs to be smaller than the 1-dB compression point of the rst stage, and
the output signal strength of every previous stage must be smaller than the 1-dB composition
point of the following stage, otherwise the receiver will be saturated. The maximum received
signal is also limited by the intermodulation because the nonlinearity will become more se-
rious under larger signal strength. The intermodulation characteristics reect the linearity of
the receiver, and it is limited by the noise gure, the receiver noise bandwidth, the phase noise
and spurs of the LOs, and the cross-modulation. For the single-conversion receiver, as shown
in Fig. 6, the expression of the allowed maximum degradation of the desired signal caused by
noise and interferences at the receiver input is:
D
max
= R
d,i
CNR
min
(5)
The minimum third intercept point of a receiver can be expressed as:
I IP3
min
=
1
2
3I
min
10lg
10
Dmax
10
10
N
nf
10
P
pn
P
sp
10
Ncm
10
(6)
where I
min
is the minimum intermodulation interference tone level at receiver input. P
pn
and
P
sp
are the phase noise and spurs contributions of LO, and they can be expressed as:
P
pn
= 10
Npn+10lgBW+I
in
R
10
(7)
P
sp
= 10
Nsp+I
in
R
10
(8)
where N
pn
is the average phase noise over the receiver bandwidth at a frequency offset equal
to the space between the interference and the carrier, N
sp
is the magnitude of the spurs at a
frequency offset equal to or nearby the space between the interference and the carrier, I
in
is the
intermodulation interference tone, and R is the rejection to the interference tone of a lter.
N
nf
denotes the receiver inherent noise converted to its input port:
N
nf
= 174 + NF +10lgBW (9)
N
cm
is the cross-modulation product, which can be approximately expressed as:
N
cm
= I
in
2I IP3
lna
+2(P
Tx
+ IL
dRx
R
dTx
) + C (10)
where I IP3
lna
is the LNA input third intercept point, P
Tx
is the transmitter output power at
the antenna port in dBm, IL
dRx
is the receiver side insertion loss of the duplexer in dB, R
dTx
is the duplexer receiver side lter rejection to the transmission in dB, and C is a correction
factor associated with waveformmagnitude uctuation and interference tone offset frequency,
which is approximately -3.8 dB and -5.8 dB for the cellular and PCS band CDMA mobile
station receivers, respectively (Gu, 2005).
Then another specication can be introduced, that is the dynamic range, which is generally de-
ned as the ratio of the maximum input level to the minimum input level at which the circuit
satised the required bit error rate. A common used specication called SFDR (Spurious-Free
Dynamic Range) bases the denition of the upper end of the dynamic range on the intermod-
ulation behavior and the lower end on the sensitivity. The SFDR can be expressed as:
SFDR =
2(I IP3 NF)
3
SNR
min
(11)
where I IP3 is the input third intercept point of the RF front end, SNR
min
is the required
minimum SNR (Signal-to-Noise Ratio) of the IF circuits. The required I IP3 can be calculated
by equation (6), then the SFDR of a receiver can be obtained by equation (11).
Two strong in-band interference tones may cause a zero-IF or low-IF receiver to be completely
jammed if the second-order intercept point of the receiver is not high enough. The two strong
tones may generate in-channel interferences due to the second-order distortion of the receiver.
The maximum level of the two equal blocking interference tones can be expressed as (Gu,
2005):
I
block
=
1
2
10lg
10
R
d,i
CNR
min
10
10
N
nf
10
+ I IP2
(12)
where I IP2 is the input second intercept point of the receiver.
The main problem of the low-IF receiver architecture is the image rejection since the IF is so
low that it is difcult to separate the image from the desired signal by lters. The imbalance
between I and Q channel signals in the low-IF receiver determines the possible maximum
image rejection. The image rejection IR and the imbalances of the I/Q amplitude and the
phase have the following relationship:
IR = 10lg
1 + 2(1 +)cos + (1 +)
2
1 2(1 +)cos + (1 +)
2
(13)
where is the I and Q phase imbalance from the nominal 90
offset in degree, is the I and

Q amplitude imbalance, which is usually expressed in dB by using the formula 10lg(1 +).
For a full-duplex system, the receiver suffers from transmission leakage interferences particu-
larly in the LNA. If a strong interference tone appears near the desired signal of the receiver,
the amplitude modulation of the transmission leakage will cross-modulate the interference
tone in the LNA. The spectrum of the cross-modulated tone may partially spread into the
receiver channel bandwidth when the single-tone interferer is enough close to the desired sig-
nal. The receiver will be desensitized if the cross-modulation product getting into the receiver
channel band is high enough. The expression of the allowed single-tone interferer is (Gu,
2005):
I
st
= 10lg
_
10
Dmax
10
10
N
nf
10
10
[2(I IP3
lna
IL
dRx
)2(P
Tx
R
dTx
)C]
10
+ 10
Npn+10lgBW
10
+ 10
Nsp
10
_
(14)
where D
max
can be got by equation (5), N
nf
can be obtained by equation (9), P
Tx
is the trans-
mitter output power at the antenna in dBm, and C is the correction factor approximately equal
to (Gu, 2005):
C = M
A
+ 6 + 10lg
1.5 BW f
2 BW
(15)
where f is the space between the interference tone and the carrier frequency of the desired
receiving signal. M
A
can be calculated by the probability density function p(x) of the trans-
mission signals and the normalized low-frequency product of the second-order distortion:
M
A
= 10lg
_
_

_
(10
x
10
)
2
10
I M2
dc
10
2
p(x)dx
_
(16)
I M2
dc
= 10lg
_
_

_
10
x
10
_
2
p(x)dx
_
(17)
For the transmitter, the modulation accuracy is represented by EVM, which is dened as the
mean square error between the samples of the actual and the ideal signals, normalized by the
average power of the idea signal. The EVM of transmitters is inuenced by the inter-symbol
or inter-chip interference, the close-in phase noise of synthesized LO, the carrier leakage, the
I and Q imbalance, the nonlinearity, the in-channel bandwidth noise, the reverse modulation
of LO, and so on.
The inuence caused by inter-symbol or inter-chip interferences can be obtained by:
EVM
isi
=
_
+
k=
I
2
isi
(k) (18)
I
isi
(k) =
h
ir
(t
0
+ kT
s
)
h
ir
(t
0
)
=
_
t
0
+t
t
0
t
|h
ir
(t + kT
s
)|dt
_
t
0
+t
t
0
t
|h
ir
(t)|dt
(19)
where h
ir
(t) is the impulse response of the pulse-shaping lter, k is equal to 1, 2, 3, . . .,
Ts is the period of a symbol, and 2t is the during of a sampling pulse.
The inuence caused by close-in phase noise of synthesized LO can be obtained by:
EVM
pn
=
2 10
N
phase
10
BW
l f ,pll
(20)
where N
phase
is the average phase noise in dBc/Hz within the PLL loop bandwidth, and
BW
l f ,pll
is the bandwidth of the PLL loop lter in Hz.
The carrier leakage is mainly caused by the DC offset of the baseband, the LO-to-RF leakage
and the IF-to-RF leakage. The inuence on EVMcaused by carrier leakage can be obtained by:
EVM
cl
=
CL
o f f set
10
+
CL
lo
10
+
CL
i f
10
(21)
where CL
o f f set
, CL
lo
and CL
i f
represents the leakage results from the DC offset, the LO-to-RF
leakage and the IF-to-RF leakage, respectively.
The EVM caused by the I and Q imbalance can be expressed as:
EVM
iq
=
10
IR
10
(22)
where IR is the image suppression, which can be calculated by equation (13).
Its assumed that only the signal amplitude equal to and greater than the output 1-dB com-
pression of the power amplier P
1
will affect the modulation accuracy, then the EVM caused
by nonlinearity of the transmitter chain can be expressed as:
EVM
nonlin
=

0
P()
10
+1
20
1
d (23)
= P
Tx
P
1
(24)
where P
Tx
is the output power level, and P
is the amplitude probability density function of

the signal.
The EVM caused by in-channel bandwidth noise can be expressed by:
EVM
ibn
=
10
N
ibn
P
Tx
10
(25)
where N
ibn
is the integrated noise over the channel bandwidth, and P
Tx
is the transmission
power in dBm.
The transmission signals may be reected from the load of the modulator, then the reected
signals and their harmonics may modulate the LO if the frequency of the carrier or the har-
monics of the reected signals is equal to LO frequency. Then reverse modulation occurs. The
EVM caused by reverse modulation is:
EVM
rm
=
10
Nrm
10
(26)
where N
rm
is the integrated reverse modulation noise of the synthesized LO over the trans-
mission signal bandwidth below the LO level.
The main problem of the low-IF receiver architecture is the image rejection since the IF is so
low that it is difcult to separate the image from the desired signal by lters. The imbalance
between I and Q channel signals in the low-IF receiver determines the possible maximum
image rejection. The image rejection IR and the imbalances of the I/Q amplitude and the
phase have the following relationship:
IR = 10lg
1 + 2(1 +)cos + (1 +)
2
1 2(1 +)cos + (1 +)
2
(13)
where is the I and Q phase imbalance from the nominal 90
offset in degree, is the I and

Q amplitude imbalance, which is usually expressed in dB by using the formula 10lg(1 +).
For a full-duplex system, the receiver suffers from transmission leakage interferences particu-
larly in the LNA. If a strong interference tone appears near the desired signal of the receiver,
the amplitude modulation of the transmission leakage will cross-modulate the interference
tone in the LNA. The spectrum of the cross-modulated tone may partially spread into the
receiver channel bandwidth when the single-tone interferer is enough close to the desired sig-
nal. The receiver will be desensitized if the cross-modulation product getting into the receiver
channel band is high enough. The expression of the allowed single-tone interferer is (Gu,
2005):
I
st
= 10lg
_
10
Dmax
10
10
N
nf
10
10
[2(I IP3
lna
IL
dRx
)2(P
Tx
R
dTx
)C]
10
+ 10
Npn+10lgBW
10
+ 10
Nsp
10
_
(14)
where D
max
can be got by equation (5), N
nf
can be obtained by equation (9), P
Tx
is the trans-
mitter output power at the antenna in dBm, and C is the correction factor approximately equal
to (Gu, 2005):
C = M
A
+ 6 + 10lg
1.5 BW f
2 BW
(15)
where f is the space between the interference tone and the carrier frequency of the desired
receiving signal. M
A
can be calculated by the probability density function p(x) of the trans-
mission signals and the normalized low-frequency product of the second-order distortion:
M
A
= 10lg
_
_

_
(10
x
10
)
2
10
I M2
dc
10
2
p(x)dx
_
(16)
I M2
dc
= 10lg
_
_

_
10
x
10
_
2
p(x)dx
_
(17)
For the transmitter, the modulation accuracy is represented by EVM, which is dened as the
mean square error between the samples of the actual and the ideal signals, normalized by the
average power of the idea signal. The EVM of transmitters is inuenced by the inter-symbol
or inter-chip interference, the close-in phase noise of synthesized LO, the carrier leakage, the
I and Q imbalance, the nonlinearity, the in-channel bandwidth noise, the reverse modulation
of LO, and so on.
The inuence caused by inter-symbol or inter-chip interferences can be obtained by:
EVM
isi
=
_
+
k=
I
2
isi
(k) (18)
I
isi
(k) =
h
ir
(t
0
+ kT
s
)
h
ir
(t
0
)
=
_
t
0
+t
t
0
t
|h
ir
(t + kT
s
)|dt
_
t
0
+t
t
0
t
|h
ir
(t)|dt
(19)
where h
ir
(t) is the impulse response of the pulse-shaping lter, k is equal to 1, 2, 3, . . .,
Ts is the period of a symbol, and 2t is the during of a sampling pulse.
The inuence caused by close-in phase noise of synthesized LO can be obtained by:
EVM
pn
=
2 10
N
phase
10
BW
l f ,pll
(20)
where N
phase
is the average phase noise in dBc/Hz within the PLL loop bandwidth, and
BW
l f ,pll
is the bandwidth of the PLL loop lter in Hz.
The carrier leakage is mainly caused by the DC offset of the baseband, the LO-to-RF leakage
and the IF-to-RF leakage. The inuence on EVMcaused by carrier leakage can be obtained by:
EVM
cl
=
CL
o f f set
10
+
CL
lo
10
+
CL
i f
10
(21)
where CL
o f f set
, CL
lo
and CL
i f
represents the leakage results from the DC offset, the LO-to-RF
leakage and the IF-to-RF leakage, respectively.
The EVM caused by the I and Q imbalance can be expressed as:
EVM
iq
=
10
IR
10
(22)
where IR is the image suppression, which can be calculated by equation (13).
Its assumed that only the signal amplitude equal to and greater than the output 1-dB com-
pression of the power amplier P
1
will affect the modulation accuracy, then the EVM caused
by nonlinearity of the transmitter chain can be expressed as:
EVM
nonlin
=

0
P()
10
+1
20
1
d (23)
= P
Tx
P
1
(24)
where P
Tx
is the output power level, and P
is the amplitude probability density function of

the signal.
The EVM caused by in-channel bandwidth noise can be expressed by:
EVM
ibn
=
10
N
ibn
P
Tx
10
(25)
where N
ibn
is the integrated noise over the channel bandwidth, and P
Tx
is the transmission
power in dBm.
The transmission signals may be reected from the load of the modulator, then the reected
signals and their harmonics may modulate the LO if the frequency of the carrier or the har-
monics of the reected signals is equal to LO frequency. Then reverse modulation occurs. The
EVM caused by reverse modulation is:
EVM
rm
=
10
Nrm
10
(26)
where N
rm
is the integrated reverse modulation noise of the synthesized LO over the trans-
mission signal bandwidth below the LO level.
The overall EVM of the transmission signal can be expressed as:
EVM
total
= EVM
2
isi
+ EVM
2
pn
+ EVM
2
cl
+ EVM
2
iq
(27)
+ EVM
2
nonlin
+ EVM
2
ibn
+ EVM
2
rm
+ . . .
The ACPR specication is generally dened as ratio of the power integrated over an assigned
bandwidth in the adjacent/alternate channel to the total desired transmission power. The
ACPR can be expressed as:
ACPR =
f
a
+B
f
a
SPD( f )d f
f
o
+BW/2
f
o
BW/2
SPD( f )d f
(28)
where f
a
is the start frequency of the adjacent/alternate channel, B is the bandwidth of
measuring adjacent/alternate channel power, which varies with different mobile systems. The
general formula for the ACPR of a transmission signal at the output of the power amplier
with an output third intercept point OIP3 can be expressed as:
ACPR 2(P
Tx
OIP3) 9 +C
0
+10lg
B
BW
(29)
C
0
0.85 (PAR 3) (30)
where P
Tx
is the transmission signal power at the output of the power amplier, PAR is the
peak-to-average ratio of the random noise. In the mobile communication systems, the ad-
jacent/alternate power may be tested in a bandwidth B that is different from the desired
transmission signal bandwidth BW.
We only discuss the noise emissions that are those located outside of alternate channels here.
In general, we like to have lower gain power amplier for achieving low noise emission, but
this is completely opposite to the gain setting of the power amplier to obtain a good ACPR
performance. The noise emission in mW/Hz of the transmitter has an expression as:
P
nm
= G
Tx
P
n,in
+kT
0
G
Tx
(NF
Tx
1) (31)
where P
n,in
in mW/Hz is the noise at the transmitter input, G
Tx
is the overall transmitter gain,
NF
Tx
is the overall noise factor of the transmitter, and kT
0
= 10
174/10
mW/Hz.
4. The design of key modules in the transceiver
Acommon used transceiver is composed of a LNA, Mixers, lters, IF circuits, a PA, ADCs and
DACs, a PLL and so on. The individual performance and the matching among these modules
determine the performance of the whole transceiver system. A general description of these
modules will be given in this section.
4.1 Low-Noise Amplier
The specications of a LNA can be summarized into:
The working frequency;
The noise gure;
The third intercept point;
The voltage or power gain;
The reection coefcient at the input port and the isolation between the output port and the
input port;
The power consumption;
The most important specication of the LNA is the noise gure since the rst stage of a re-
ceiver chain decides the noise performance of the whole system. One of the common used
LNAs is the inductively source degenerated type, and a typical design example is shown in
Fig. 7(a). M1 is a common-source amplier transistor, L
S
is the source degenerated inductor,
L
G
is the gate inductor, V
I N
is the input Port, and V
OUT
is the output port; M2 is a common-
gate transistor that is used for isolation and gain enhancement; the load Z
L
can be a resistor,
a inductor or a inductor-capacitor tank. This structure has a large gain and a low noise gure,
but the input reection is a problem. There is a trade-off between the noise gure and the
input impedance matching.
V
B
V
IN
L
G
L
S
V
OUT
M1
M2
VDD
Z
L
V
B
V
OUT
VDD
Z
L
V
IN
(a) (b)
M3
Fig. 7. (a)An inductively source degenerated LNA. (b)The common-gate input type.
Another common used LNA is the common-gate input type, as shown in Fig. 7(b). If the
transconductance of the common-gate transistor M3 is g
m
, then the input resistance is equal
to 1/g
m
. Therefore, the input matching of common-gate LNA is easier to realize compared to
the source degenerated LNA. However, the noise performance is poor, since the common-gate
amplier has a low gain.
The WSN receivers usually have a high sensitivity, as shown in Table 1. According to equa-
tion (1), the sensitivity is proportional to the noise gure of the RF front end. Therefore, in
WSN applications, we often adopt the inductively source degenerated LNA assisted by some
low-noise technologies.
4.2 Mixer
The mixers in transceivers can be divided into two types: 1)the up-converting mixers and
2)the down-converting mixers. The up-converting mixers are used in the transmitter, while
the down-converting mixers are used for the receiver. The specications of a mixer can be
summarized into:
The working frequency including RF frequency, LO frequency, and IF frequency;
The noise gure;
The overall EVM of the transmission signal can be expressed as:
EVM
total
= EVM
2
isi
+ EVM
2
pn
+ EVM
2
cl
+ EVM
2
iq
(27)
+ EVM
2
nonlin
+ EVM
2
ibn
+ EVM
2
rm
+ . . .
The ACPR specication is generally dened as ratio of the power integrated over an assigned
bandwidth in the adjacent/alternate channel to the total desired transmission power. The
ACPR can be expressed as:
ACPR =
f
a
+B
f
a
SPD( f )d f
f
o
+BW/2
f
o
BW/2
SPD( f )d f
(28)
where f
a
is the start frequency of the adjacent/alternate channel, B is the bandwidth of
measuring adjacent/alternate channel power, which varies with different mobile systems. The
general formula for the ACPR of a transmission signal at the output of the power amplier
with an output third intercept point OIP3 can be expressed as:
ACPR 2(P
Tx
OIP3) 9 +C
0
+10lg
B
BW
(29)
C
0
0.85 (PAR 3) (30)
where P
Tx
is the transmission signal power at the output of the power amplier, PAR is the
peak-to-average ratio of the random noise. In the mobile communication systems, the ad-
jacent/alternate power may be tested in a bandwidth B that is different from the desired
transmission signal bandwidth BW.
We only discuss the noise emissions that are those located outside of alternate channels here.
In general, we like to have lower gain power amplier for achieving low noise emission, but
this is completely opposite to the gain setting of the power amplier to obtain a good ACPR
performance. The noise emission in mW/Hz of the transmitter has an expression as:
P
nm
= G
Tx
P
n,in
+kT
0
G
Tx
(NF
Tx
1) (31)
where P
n,in
in mW/Hz is the noise at the transmitter input, G
Tx
is the overall transmitter gain,
NF
Tx
is the overall noise factor of the transmitter, and kT
0
= 10
174/10
mW/Hz.
4. The design of key modules in the transceiver
Acommon used transceiver is composed of a LNA, Mixers, lters, IF circuits, a PA, ADCs and
DACs, a PLL and so on. The individual performance and the matching among these modules
determine the performance of the whole transceiver system. A general description of these
modules will be given in this section.
4.1 Low-Noise Amplier
The specications of a LNA can be summarized into:
The working frequency;
The noise gure;
The voltage or power gain;
The reection coefcient at the input port and the isolation between the output port and the
input port;
The most important specication of the LNA is the noise gure since the rst stage of a re-
ceiver chain decides the noise performance of the whole system. One of the common used
LNAs is the inductively source degenerated type, and a typical design example is shown in
Fig. 7(a). M1 is a common-source amplier transistor, L
S
is the source degenerated inductor,
L
G
is the gate inductor, V
I N
is the input Port, and V
OUT
is the output port; M2 is a common-
gate transistor that is used for isolation and gain enhancement; the load Z
L
can be a resistor,
a inductor or a inductor-capacitor tank. This structure has a large gain and a low noise gure,
but the input reection is a problem. There is a trade-off between the noise gure and the
input impedance matching.
V
B
V
IN
L
G
L
S
V
OUT
M1
M2
VDD
Z
L
V
B
V
OUT
VDD
Z
L
V
IN
(a) (b)
M3
Fig. 7. (a)An inductively source degenerated LNA. (b)The common-gate input type.
Another common used LNA is the common-gate input type, as shown in Fig. 7(b). If the
transconductance of the common-gate transistor M3 is g
m
, then the input resistance is equal
to 1/g
m
. Therefore, the input matching of common-gate LNA is easier to realize compared to
the source degenerated LNA. However, the noise performance is poor, since the common-gate
amplier has a low gain.
The WSN receivers usually have a high sensitivity, as shown in Table 1. According to equa-
tion (1), the sensitivity is proportional to the noise gure of the RF front end. Therefore, in
WSN applications, we often adopt the inductively source degenerated LNA assisted by some
low-noise technologies.
4.2 Mixer
The mixers in transceivers can be divided into two types: 1)the up-converting mixers and
2)the down-converting mixers. The up-converting mixers are used in the transmitter, while
the down-converting mixers are used for the receiver. The specications of a mixer can be
summarized into:
The working frequency including RF frequency, LO frequency, and IF frequency;
The noise gure;
The second intercept point (for zero-IF or low-IF receivers);
The voltage or power conversion gain;
The isolation between the RF port and the LO port, the RF port and the IF port, and the LO
port and the IF port;
The magnitude and phase imbalance between I and Q channel down converters (for the
receivers and transmitters that use I and Q dual-path converters);
A classical mixer is known as the Gilbert cell, as shown in Fig. 8. I
B
is a current source, RF+
and RF are the differential RF input ports, LO+and LO are the differential LOinput ports,
and the output differential currents are I
OUT
+ and I
OUT
; M1 and M2 convert the input RF
voltage into current, and M3-M6 are used as switches for mixing.
RF RF-
LO
LO-
I
OUT
I
OUT
-
M1 M2
M3 M4 M5 M6
I
B
Fig. 8. The Gilbert cell.
The Gilbert mixer is a typical example of active mixers, which have high gain, low noise, but
poor linearity. Another type of mixers are called passive mixers, which usually have low gain,
large noise and high linearity. The passive mixers can be divided into two types: 1)voltage-
mode mixers: the MOS switches are used for voltage switches, and loaded with high resis-
tance. Because of the nonlinearity of the switches, distortions will be enlarged when the am-
plitudes of RF and IF signals are increased and the switches are modulated. 2)Current-mode
mixers: the MOS switches are used for current switches, and loaded with low resistance. So
the amplitudes of RF and IF signals are relatively low in current-mode mixers, then the linear-
ity is improved. A typical example of current-mode passive mixers is shown in Fig. 9 (Valla
et al., 2005). The input I
I N
is a differential current signal, the output V
OUT
is a voltage sig-
nal. Transistors M1-M4 are used for switches that are controlled by the LO signals LO+ and
LO. An OTA (OperaTional Amplier) together with resistors R1, R2 and capacitors C1, C2
is adopted to amplify and lter the signals. Two additional capacitors C3 and C4 are used to
generate an extra pole to suppress the amplitude. The LNA and mixers are often designed
and tested together to realize an optimized trade-off among gain, noise gure and linearity
for different applications.
The distribution of WSN nodes is random, and the distance between two nodes may be very
short or very long, so the dynamic range of the transceivers is very important. According
to equation (11), in order to improve the SFDR performance, the noise gure needs to be
LO
LO
LO-
V
OUT
M1
M2
M3
M4
OTA
R1
R2
C1
C2
C3
C4
-
I
IN
Fig. 9. A typical current-mode passive mixer.
decreased and the IIP3 needs to be enhanced. Therefore, a combination of a low-noise induc-
tively source degenerated LNA and a current-mode passive mixer can be adopted for WSN
usages.
4.3 Active Filter
According to the pattern of implement, the active lters usually used for transceivers can be
summarized into three types: 1)switched-capacitor lters, in which the resistors are replaced
by switched capacitors; 2)active-RC lters, which is composed of OTAs and resistor-capacitor
networks; 3)gm-Clters, in which the resistors and inductors are replaced by transconductors.
For switched-capacitor lters, the advantages can be summarized here: 1)high precision with-
out tuning, 2)small chip area and low power, and 3)insensitive to parasitics. However, there
are several disadvantages: 1)affected by sampling, 2)requirement for extra clock generation
circuit, and 3)not suitable for high-frequency applications.
For active-RC lters, the advantages can be summarized into: 1)high precision with tuning,
2)easy to design with classical RC structures, 3)insensitive to parasitics, 4)no sampling effect,
and 5)large dynamic range. The disadvantages can be summarized into: 1)requirement for
tuning circuits and 2)limited working frequency caused by OTAs.
For gm-C lters, the advantages can be summarized into: 1)high precision with tuning, 2)able
to be realized based on simple open-loop OTAs, 3)lower power consumption than active-RC
lters, 4)no sampling effect, and 5)good frequency performance. The disadvantages can be
summarized into: 1)requirement for complex on-chip tuning circuits, 2)poor dynamic range,
and 3)sensitive to parasitics.
According to the transfer characters, the lters can also be divided into four types: 1)Butter-
worth lters, which has the maximum at amplitude in the pass band; 2)Chebyshev lters,
which has the minimumripples in the pass band; 3)Bessel lters, which has the maximumat
of group delay; 4)Ellipse lters, which has the minimumtransition band. The other characters
of these lters are summarized into Table 3.
The second intercept point (for zero-IF or low-IF receivers);
The voltage or power conversion gain;
The isolation between the RF port and the LO port, the RF port and the IF port, and the LO
port and the IF port;
The magnitude and phase imbalance between I and Q channel down converters (for the
receivers and transmitters that use I and Q dual-path converters);
A classical mixer is known as the Gilbert cell, as shown in Fig. 8. I
B
is a current source, RF+
and RF are the differential RF input ports, LO+and LO are the differential LOinput ports,
and the output differential currents are I
OUT
+ and I
OUT
; M1 and M2 convert the input RF
voltage into current, and M3-M6 are used as switches for mixing.
RF RF-
LO
LO-
I
OUT
I
OUT
-
M1 M2
M3 M4 M5 M6
I
B
Fig. 8. The Gilbert cell.
The Gilbert mixer is a typical example of active mixers, which have high gain, low noise, but
poor linearity. Another type of mixers are called passive mixers, which usually have low gain,
large noise and high linearity. The passive mixers can be divided into two types: 1)voltage-
mode mixers: the MOS switches are used for voltage switches, and loaded with high resis-
tance. Because of the nonlinearity of the switches, distortions will be enlarged when the am-
plitudes of RF and IF signals are increased and the switches are modulated. 2)Current-mode
mixers: the MOS switches are used for current switches, and loaded with low resistance. So
the amplitudes of RF and IF signals are relatively low in current-mode mixers, then the linear-
ity is improved. A typical example of current-mode passive mixers is shown in Fig. 9 (Valla
et al., 2005). The input I
I N
is a differential current signal, the output V
OUT
is a voltage sig-
nal. Transistors M1-M4 are used for switches that are controlled by the LO signals LO+ and
LO. An OTA (OperaTional Amplier) together with resistors R1, R2 and capacitors C1, C2
is adopted to amplify and lter the signals. Two additional capacitors C3 and C4 are used to
generate an extra pole to suppress the amplitude. The LNA and mixers are often designed
and tested together to realize an optimized trade-off among gain, noise gure and linearity
for different applications.
The distribution of WSN nodes is random, and the distance between two nodes may be very
short or very long, so the dynamic range of the transceivers is very important. According
to equation (11), in order to improve the SFDR performance, the noise gure needs to be
LO
LO
LO-
V
OUT
M1
M2
M3
M4
OTA
R1
R2
C1
C2
C3
C4
-
I
IN
Fig. 9. A typical current-mode passive mixer.
decreased and the IIP3 needs to be enhanced. Therefore, a combination of a low-noise induc-
tively source degenerated LNA and a current-mode passive mixer can be adopted for WSN
usages.
4.3 Active Filter
According to the pattern of implement, the active lters usually used for transceivers can be
summarized into three types: 1)switched-capacitor lters, in which the resistors are replaced
by switched capacitors; 2)active-RC lters, which is composed of OTAs and resistor-capacitor
networks; 3)gm-Clters, in which the resistors and inductors are replaced by transconductors.
For switched-capacitor lters, the advantages can be summarized here: 1)high precision with-
out tuning, 2)small chip area and low power, and 3)insensitive to parasitics. However, there
are several disadvantages: 1)affected by sampling, 2)requirement for extra clock generation
circuit, and 3)not suitable for high-frequency applications.
For active-RC lters, the advantages can be summarized into: 1)high precision with tuning,
2)easy to design with classical RC structures, 3)insensitive to parasitics, 4)no sampling effect,
and 5)large dynamic range. The disadvantages can be summarized into: 1)requirement for
tuning circuits and 2)limited working frequency caused by OTAs.
For gm-C lters, the advantages can be summarized into: 1)high precision with tuning, 2)able
to be realized based on simple open-loop OTAs, 3)lower power consumption than active-RC
lters, 4)no sampling effect, and 5)good frequency performance. The disadvantages can be
summarized into: 1)requirement for complex on-chip tuning circuits, 2)poor dynamic range,
and 3)sensitive to parasitics.
According to the transfer characters, the lters can also be divided into four types: 1)Butter-
worth lters, which has the maximum at amplitude in the pass band; 2)Chebyshev lters,
which has the minimumripples in the pass band; 3)Bessel lters, which has the maximumat
of group delay; 4)Ellipse lters, which has the minimumtransition band. The other characters
of these lters are summarized into Table 3.
Type
Amplitude-Frequency Characteristic Phase-Frequency
Pass Band Stop Band Transition Band characteristic
Butterworth Flat
Monotonic Gentle Monotonic
Moderate
Decreasing Decreasing
Chebyshev Fluctuant
Monotonic Steep Monotonic
Poor
Bessel Flat
Monotonic Slowly Monotonic
Excellent
Ellipse Fluctuant Fluctuant
Steep Monotonic
Poor
Decreasing
Table 3. The characteristics of different lters.
In wireless transceivers, there is a special kind of lter named complex lter, which is usu-
ally used in low-IF receivers for image rejection. A classical complex lter is designed in
1995 (Crols & Steyaert, 1995). Fig. 10 shows the block diagram, which has I/Q dual-path in-
puts and I/Q dual-path outputs. The input of the Q path Q
I N
is 90
delay of the input of the

I path I
I N
, and the output of Q path Q
OUT
is also 90
delay of the output of the I path I

OUT
.
That is Q
I N
= jI
I N
and Q
OUT
= jI
OUT
. The transfer function of this complex lter can be
A
o
/j
-1/A
c
/(A
o
)
-
c
/(A
o
)
-1/A
A
o
/j
I
IN
Q
IN
I
OUT
Q
OUT
Fig. 10. The block diagram of a rst-order complex lter.
expressed as:
H
c f
(j) =
A
1 + j(
c
)/
o
(32)
where
c
is the central frequency, and 2
o
is the double-sideband bandwidth. It is equivalent
to a low-pass lters pass band moved by
c
, and then the transfer curves of positive and
negative frequency become asymmetric. As a result, the image can be rejected.
As shown in Table 2, low-IF transceivers has the advantages of easy to be integrated and
immune to DC offset, so the low-IF SDR transceiver is adopted in many WSN applications.
Besides, the data rate of WSN is usually not very high, then the IF can be relatively low and
gm-C lters are not necessary. Therefore, an active-RC complex lter is suitable for such WSN
receivers because of their low power, large dynamic range and image rejection function.
4.4 Phase-Locked Loop
The PLL is the core part of a transceiver system, as its used for both the down-converting
in receiver and the up-converting in transmitter. A typical sigma-delta charge-pump PLL is
shown in Fig. 11 (Zhao et al., 2009), which is composed of a PFD(Phase-Frequency Detector), a
charge pump, a loop lter, a VCO (Voltage-Controlled Oscillator), a multi-modulus frequency
divider, and a sigma-delta modulator. Although the all-digital PLL has appeared in recent
years, the classical charge-pump PLL is still widely used in the industrial community.
PFD
VCO
Charge Pump Loop Filter
Sigma-Delta Modulator
Multi-Modulus Frequency Divider
Decimal Input
I
REF
I
OUT
Integer Input
Fig. 11. A typical sigma-delta charge-pump PLL (Zhao et al., 2009).
For WSN transceivers, we tend to design low-power, full-integrated and fast-settling PLL.
The power of the PLL is mainly limited by these modules: 1)VCO (Voltage-Controlled Os-
cillator), 2)prescaler, and 3)the buffer connected at the output of VCO. Therefore, the power
reduction of these modules is signicant to the low-power design of PLL.
For full-integrated design, the chip area needs to be decreased. In a typical PLL, the LF (Loop
Filter) and the inductors in LC-tank VCO take up the largest chip area. In order to decrease
the area of LF, some one has proposed a discrete-time architecture (Zhang et al., 2003). For the
applications in low-frequency bands, the inductors will be large if the resonant frequency of
the VCO is low, so the VCO is required to be designed at a high frequency with a frequency
divider connected after it.
The settling speed of PLL is decided by the loop bandwidth. Too Large bandwidth brings
not only fast settling, but also large in-band noise and spurs. As a result, there is a trade-off
between the settling speed, and the phase noise and the spur performance. How to set the
trade-off depends on the requirement of the transceiver system.
As we referenced in section 2.2, the PLL can be adopted for directly digital modulation, and
such method is proposed by Perrott in 1997 (Perrott et al., 1997). The architecture of such PLL
based transmitter is shown in Fig. 12. The data streamcan be shaped by a lter rstly, then the
shaped data are input into the sigma-delta modulator in order to change the dividing ratio. As
a result, the output frequency can be modulated by the variation of dividing ratio according
to the input data, and the FSK signals can be generated in this way. A PA is connected at the
output of the PLL so that the FSK signals can be emitted through an antenna. Generally, the
data rate can not be larger than the bandwidth of the PLL. Although some one has proposed
a compensation technology with a digital lter whose transfer function is the reciprocal of
PLLs (Perrott et al., 1997), mismatch and inaccuracy depress the performance in actual de-
signs. As a result, we would rather enlarge the bandwidth of PLL to obtain a relatively high
data rate.
Type
Amplitude-Frequency Characteristic Phase-Frequency
Pass Band Stop Band Transition Band characteristic
Butterworth Flat
Monotonic Gentle Monotonic
Moderate
Chebyshev Fluctuant
Monotonic Steep Monotonic
Poor
Bessel Flat
Monotonic Slowly Monotonic
Excellent
Ellipse Fluctuant Fluctuant
Steep Monotonic
Poor
Decreasing
Table 3. The characteristics of different lters.
In wireless transceivers, there is a special kind of lter named complex lter, which is usu-
ally used in low-IF receivers for image rejection. A classical complex lter is designed in
1995 (Crols & Steyaert, 1995). Fig. 10 shows the block diagram, which has I/Q dual-path in-
puts and I/Q dual-path outputs. The input of the Q path Q
I N
is 90
delay of the input of the

I path I
I N
, and the output of Q path Q
OUT
is also 90
delay of the output of the I path I

OUT
.
That is Q
I N
= jI
I N
and Q
OUT
= jI
OUT
. The transfer function of this complex lter can be
A
o
/j
-1/A
c
/(A
o
)
-
c
/(A
o
)
-1/A
A
o
/j
I
IN
Q
IN
I
OUT
Q
OUT
Fig. 10. The block diagram of a rst-order complex lter.
expressed as:
H
c f
(j) =
A
1 + j(
c
)/
o
(32)
where
c
is the central frequency, and 2
o
is the double-sideband bandwidth. It is equivalent
to a low-pass lters pass band moved by
c
, and then the transfer curves of positive and
negative frequency become asymmetric. As a result, the image can be rejected.
As shown in Table 2, low-IF transceivers has the advantages of easy to be integrated and
immune to DC offset, so the low-IF SDR transceiver is adopted in many WSN applications.
Besides, the data rate of WSN is usually not very high, then the IF can be relatively low and
gm-C lters are not necessary. Therefore, an active-RC complex lter is suitable for such WSN
receivers because of their low power, large dynamic range and image rejection function.
4.4 Phase-Locked Loop
The PLL is the core part of a transceiver system, as its used for both the down-converting
in receiver and the up-converting in transmitter. A typical sigma-delta charge-pump PLL is
shown in Fig. 11 (Zhao et al., 2009), which is composed of a PFD(Phase-Frequency Detector), a
charge pump, a loop lter, a VCO (Voltage-Controlled Oscillator), a multi-modulus frequency
divider, and a sigma-delta modulator. Although the all-digital PLL has appeared in recent
years, the classical charge-pump PLL is still widely used in the industrial community.
PFD
VCO
Decimal Input
I
REF
I
OUT
Integer Input
Fig. 11. A typical sigma-delta charge-pump PLL (Zhao et al., 2009).
For WSN transceivers, we tend to design low-power, full-integrated and fast-settling PLL.
The power of the PLL is mainly limited by these modules: 1)VCO (Voltage-Controlled Os-
cillator), 2)prescaler, and 3)the buffer connected at the output of VCO. Therefore, the power
reduction of these modules is signicant to the low-power design of PLL.
For full-integrated design, the chip area needs to be decreased. In a typical PLL, the LF (Loop
Filter) and the inductors in LC-tank VCO take up the largest chip area. In order to decrease
the area of LF, some one has proposed a discrete-time architecture (Zhang et al., 2003). For the
applications in low-frequency bands, the inductors will be large if the resonant frequency of
the VCO is low, so the VCO is required to be designed at a high frequency with a frequency
divider connected after it.
The settling speed of PLL is decided by the loop bandwidth. Too Large bandwidth brings
not only fast settling, but also large in-band noise and spurs. As a result, there is a trade-off
between the settling speed, and the phase noise and the spur performance. How to set the
trade-off depends on the requirement of the transceiver system.
As we referenced in section 2.2, the PLL can be adopted for directly digital modulation, and
such method is proposed by Perrott in 1997 (Perrott et al., 1997). The architecture of such PLL
based transmitter is shown in Fig. 12. The data streamcan be shaped by a lter rstly, then the
shaped data are input into the sigma-delta modulator in order to change the dividing ratio. As
a result, the output frequency can be modulated by the variation of dividing ratio according
to the input data, and the FSK signals can be generated in this way. A PA is connected at the
output of the PLL so that the FSK signals can be emitted through an antenna. Generally, the
data rate can not be larger than the bandwidth of the PLL. Although some one has proposed
a compensation technology with a digital lter whose transfer function is the reciprocal of
PLLs (Perrott et al., 1997), mismatch and inaccuracy depress the performance in actual de-
signs. As a result, we would rather enlarge the bandwidth of PLL to obtain a relatively high
data rate.
PFD
VCO
Shaping Filter
Data Stream
PA
I
reI
Antenna
Fig. 12. A transmitter based on PLL directly digital modulation.
For WSN usages, the data rate is often not high, so such PLL directly digital modulation is a
reasonable choice for common frequency and phase modulation schemes, such as FSK, MSK,
and so on. As a result, mixers can be moved away, the cost and power consumption of the
WSN transmitter can be saved a lot.
4.5 Power Amplier
Generally, PA is the most power hungry module of a transceiver. Therefore, the output power
of PA is usually relatively small for WSN usages, as shown in Table 1. The types of PA can be
divided into class A, B, C, AB, D, E, F and F
1
.
For class-A PAs, the amplier MOSFET is kept in the saturation region. The transistor always
dissipates power because the product of drain current and drain voltage is always positive.
It should be noticed that the maximum theoretical drain efciency of class-A PAs is just 50%.
However, drain efciencies of 3050% are common for practical class-A PA designs. The nor-
malized power output capability is about 1/8. The class-A amplier provides high linearity
at the cost of low efciency and relatively large device stresses.
In a class-B PA, the device is shut off in half of every cycle. It should be mentioned that
most practical class-B PAs are push-pull congurations of two MOSFETs. The peak drain
current and maximum output voltage are the same as for the class-A PAs. The maximum
drain efciency for a class-B PA is 78.5%. The normalized power capability of the class-B PAs
is 1/8, the same as for class-A PAs, since the output power, maximum drain voltage, and
maximum drain current are the same.
In a class-CPA, the transistor conducts less than half the time. As the conduction angle shrinks
toward zero, the efciency approaches 100%, but the gain and output power unfortunately
also tend toward zero at the same time. Furthermore, the normalized power capability of
class-C PAs approaches zero as the conduction angle approaches zero. In one word, the ef-
ciency can be large, but at the cost of normalized power capability, gain, and linearity.
The class-AB Pas conducts between 50% and 100% of a cycle. Both the conduction angle and
efciency of class-AB PA vary between that of class-A PA and class-B PA.
In a class-D PA, only one transistor is driven on at a given time, and one transistor handles
the positive half-cycles and the other handles the negative half-cycles, just as a push-pull
class-B PA. The difference between class-D and class-B is that the transistors are driven hard
enough to make them act like switches for class-D PA, rather than as linear ampliers. The
normalized power capability of class-D PAs is about 0.32, which is better than a class-B push-
pull and much better than a class-A PA. The MOS switches in class-D PAs function well only
at frequencies substantially below f
T
, which is the cut-off frequency. Usually, one transistor
fails to turn completely off before the other turns on, then the efcient is deteriorated.
The class-E PA uses a high-order reactive network that provides enough space to shape the
switch voltage to have both zero value and zero slope at switch turn-on, then the switch loss
is reduced. The efciency can approach theoretically 100% with idea switches. The normal-
ized power capability is about 0.098, which is worse than class-A PA. The class-E PA is more
demanding of its switch performance than even class-A PAs because of the poor power capa-
bility and the reduced efciency due to switch turn-off losses.
The termination of a class-F PA appears as an open circuit at odd harmonics of the carrier
beyond the fundamental and as a short circuit at even harmonics, while the class-F
1
employs
a termination that appears as an open circuit at even harmonics and as a short circuit at the
odd harmonics. The class-F PA is capable of 100% efciency in principle. The normalized
power capability of class-F PAs is about 0.16, which is half that of the class-D PAs.
In summary, there is a trade-off between the efciency and the linearity. For receivers with
constant-envelope modulation, such as FSK, high-efciency PAs can be adopted; for linear
operation such as ASK (Amplitude Shift Keying), or systems with high ACPR requirement,
high-linearity PAs can be adopted.
The high-efciency PAs such as class-E are usually used in WSN transceivers, as the power
consumption is the most signicant specications of WSN system.
4.6 IF circuits
The function of IF circuits includes demodulation, data decision, and clock recovery. There
are two main kinds of IF circuits: 1)the digital scheme. In common receivers, an ADC is
connected after the RF front end, and the frequency detecting, data decision and received
signal strength indicating are all realized in digital domain, then the performance of the circuit
can be easily improved in digital domain. Such is the general architecture for SDR as we
described in section 2.5. However, the ADC usually consumes a large amount of power, and a
high-linearity AGC circuit is required before the ADC. 2)The analog scheme. For low-power
applications such as WSN, the CMOS analog resolution is sometimes a reasonable choice since
the power consumption can be saved a lot.
4.7 ADC and DAC
There are two kinds of ADCs in a WSN node. One is connected after the sensor and used for
data sampling, as shown in Fig. 1; and another is used in the transceiver, as shown in Fig. 6.
The main parameters of an ADC can be summarized into several aspects:
Resolution: The minimum voltage level that can be discriminated by the ADC is V
re f
/2
N
for a N-bit ADC with an input range from 0 to V
re f
.
DNL (Differential Non-Linearity): The maximum deviation between the actual conversion
step and the idea conversion step.
INL (Integrated Non-Linearity): The maximum deviation of actual center of bin from its
idea location.
Offset: The non-zero voltage or current at the output of the ADC when the input is zero
since the OTAs or comparators have offset voltages and offset currents.
Gain Error: The deviation of actual input voltage from the idea value when the ADC out-
puts the full-scale bits.
PFD
VCO
Shaping Filter
Data Stream
PA
I
reI
Antenna
Fig. 12. A transmitter based on PLL directly digital modulation.
For WSN usages, the data rate is often not high, so such PLL directly digital modulation is a
reasonable choice for common frequency and phase modulation schemes, such as FSK, MSK,
and so on. As a result, mixers can be moved away, the cost and power consumption of the
WSN transmitter can be saved a lot.
4.5 Power Amplier
Generally, PA is the most power hungry module of a transceiver. Therefore, the output power
of PA is usually relatively small for WSN usages, as shown in Table 1. The types of PA can be
divided into class A, B, C, AB, D, E, F and F
1
.
For class-A PAs, the amplier MOSFET is kept in the saturation region. The transistor always
dissipates power because the product of drain current and drain voltage is always positive.
It should be noticed that the maximum theoretical drain efciency of class-A PAs is just 50%.
However, drain efciencies of 3050% are common for practical class-A PA designs. The nor-
malized power output capability is about 1/8. The class-A amplier provides high linearity
at the cost of low efciency and relatively large device stresses.
In a class-B PA, the device is shut off in half of every cycle. It should be mentioned that
most practical class-B PAs are push-pull congurations of two MOSFETs. The peak drain
current and maximum output voltage are the same as for the class-A PAs. The maximum
drain efciency for a class-B PA is 78.5%. The normalized power capability of the class-B PAs
is 1/8, the same as for class-A PAs, since the output power, maximum drain voltage, and
maximum drain current are the same.
In a class-CPA, the transistor conducts less than half the time. As the conduction angle shrinks
toward zero, the efciency approaches 100%, but the gain and output power unfortunately
also tend toward zero at the same time. Furthermore, the normalized power capability of
class-C PAs approaches zero as the conduction angle approaches zero. In one word, the ef-
ciency can be large, but at the cost of normalized power capability, gain, and linearity.
The class-AB Pas conducts between 50% and 100% of a cycle. Both the conduction angle and
efciency of class-AB PA vary between that of class-A PA and class-B PA.
In a class-D PA, only one transistor is driven on at a given time, and one transistor handles
the positive half-cycles and the other handles the negative half-cycles, just as a push-pull
class-B PA. The difference between class-D and class-B is that the transistors are driven hard
enough to make them act like switches for class-D PA, rather than as linear ampliers. The
normalized power capability of class-D PAs is about 0.32, which is better than a class-B push-
pull and much better than a class-A PA. The MOS switches in class-D PAs function well only
at frequencies substantially below f
T
, which is the cut-off frequency. Usually, one transistor
fails to turn completely off before the other turns on, then the efcient is deteriorated.
The class-E PA uses a high-order reactive network that provides enough space to shape the
switch voltage to have both zero value and zero slope at switch turn-on, then the switch loss
is reduced. The efciency can approach theoretically 100% with idea switches. The normal-
ized power capability is about 0.098, which is worse than class-A PA. The class-E PA is more
demanding of its switch performance than even class-A PAs because of the poor power capa-
bility and the reduced efciency due to switch turn-off losses.
The termination of a class-F PA appears as an open circuit at odd harmonics of the carrier
beyond the fundamental and as a short circuit at even harmonics, while the class-F
1
employs
a termination that appears as an open circuit at even harmonics and as a short circuit at the
odd harmonics. The class-F PA is capable of 100% efciency in principle. The normalized
power capability of class-F PAs is about 0.16, which is half that of the class-D PAs.
In summary, there is a trade-off between the efciency and the linearity. For receivers with
constant-envelope modulation, such as FSK, high-efciency PAs can be adopted; for linear
operation such as ASK (Amplitude Shift Keying), or systems with high ACPR requirement,
high-linearity PAs can be adopted.
The high-efciency PAs such as class-E are usually used in WSN transceivers, as the power
consumption is the most signicant specications of WSN system.
4.6 IF circuits
The function of IF circuits includes demodulation, data decision, and clock recovery. There
are two main kinds of IF circuits: 1)the digital scheme. In common receivers, an ADC is
connected after the RF front end, and the frequency detecting, data decision and received
signal strength indicating are all realized in digital domain, then the performance of the circuit
can be easily improved in digital domain. Such is the general architecture for SDR as we
described in section 2.5. However, the ADC usually consumes a large amount of power, and a
high-linearity AGC circuit is required before the ADC. 2)The analog scheme. For low-power
applications such as WSN, the CMOS analog resolution is sometimes a reasonable choice since
the power consumption can be saved a lot.
4.7 ADC and DAC
There are two kinds of ADCs in a WSN node. One is connected after the sensor and used for
data sampling, as shown in Fig. 1; and another is used in the transceiver, as shown in Fig. 6.
The main parameters of an ADC can be summarized into several aspects:
Resolution: The minimum voltage level that can be discriminated by the ADC is V
re f
/2
N
for a N-bit ADC with an input range from 0 to V
re f
.
DNL (Differential Non-Linearity): The maximum deviation between the actual conversion
step and the idea conversion step.
INL (Integrated Non-Linearity): The maximum deviation of actual center of bin from its
idea location.
Offset: The non-zero voltage or current at the output of the ADC when the input is zero
since the OTAs or comparators have offset voltages and offset currents.
Gain Error: The deviation of actual input voltage from the idea value when the ADC out-
puts the full-scale bits.
SNR (Signal-to-Noise Ratio): The theory equation of SNR for a N-bit ADC can be expressed
as:
SNR = 6.02N +1.76 (33)
SNDR (Signal-to-Noise and Distortion Ratio): The power of the noise and harmonics di-
vided by the power of signal.
SFDR (Spurious Free Dynamic Range): The ratio of the signals power to the maximum
harmonics power.
ENOB (Effective Number of Bits): The ENOB can be calculated by:
ENOB = (SNDR 1.76)/6.02 (34)
THD (Total Harmonic Distortion): The ratio of all the harmonics power to the signals
power.
Besides the specications above, another very important parameter is power consumption. In
the aspect of architecture, the ADCs can be divided into ash, SAR (Successive AppRoxima-
tion), folding, pipeline, sigma-delta, and so on.
For the ADC connected after the sensor in a WSN node, SAR ADCs may be the best choice
since the data collection is executed at most of the time. There are two reasons: 1)The only
power hungry module in a SAR ADC is the comparator, so the overall power consumption
is low; 2)the circuit structure of a SAR ADC is simple, so the cost of chip area is small. In the
WSN transceivers, the data rate is not high and the modulation scheme is simple, so a low
bandwidth and low precision ADC is usually enough. Therefore, a SAR ADC may also be a
good choice for WSN transceivers.
The specications of DAC also includes DNL, INL, SNR, SNDR, SFDR, power consumption,
and so on. The DACs are often used in transmitters, as shown in Fig. 6, the shaped digital
waves are converted to analog signals that are sent to up-converting mixers for modulation.
As a result, the performance of the DACs will affect the EVM and ACPR of the transmitters.
5. Conclusion
The goal of this chapter is to give a brief manual for WSN transceiver design. Section 1 gives
an introduction of WSN and the RF transceivers for WSN. The WSN transceivers are classi-
ed by both modulation schemes and architectures in section 2. How to calculate and assign
the system specications are described in section 3. The design of key modules is analyzed
briey in section 4. The readers is expected to master a top-to-down design method for WSN
transceivers through the chapter.
6. References
Crols, J. & Steyaert, M. (1995). An analog integrated polyphase lter for a high performance
low-IF receiver, Symposium on VLSI Circuits, Digest of Technical Papers, pp. 8788.
Gu, Q. (2005). RF System Design of Transceivers for Wireless Communications, Springer Sci-
ence+Business Media, LLC,.
Jri, L., Chen, Y. & Yenlin, H. (2010). A Low-Power Low-Cost Fully-Integrated 60-GHz
Transceiver System With OOK Modulation and On-Board Antenna Assembly, IEEE
Journal of Solid-State Circuits, DOI-10.1109/JSSC.2009.2034806, 45(2): 264275.
Perrott, M. H., Tewksbury, T. L. & Sodini, C. G. (1997). A 27-mW CMOS fractional-N syn-
thesizer using digital compensation for 2.5-Mb/s GFSK modulation, IEEE Journal of
Solid-State Circuits, DOI-10.1109/4.643663, 32(12): 20482060.
Pletcher, N. M., Gambini, S. & Rabaey, J. (2009). A 52 W Wake-Up Receiver With -72 dBm
Sensitivity Using an Uncertain-IF Architecture, IEEE Journal of Solid-State Circuits,
DOI-10.1109/JSSC.2008.2007438, 44(1): 269280.
Seungkee, M., Shashidharan, S., Stevens, M., Copani, T., Kiaei, S., Bakkaloglu, B. &
Chakraborty, S. (2010). A 2mW CMOS MICS-band BFSK transceiver with recon-
gurable antenna interface, IEEE Radio Frequency Integrated Circuits Symposium,,
pp. 289292.
Valla, M., Montagna, G., Castello, R., Tonietto, R. & Bietti, I. (2005). A 72-mW CMOS 802.11a
direct conversion front-end with 3.5-dB NF and 200-kHz 1/f noise corner, IEEE Jour-
nal of Solid-State Circuits, DOI-10.1109/JSSC.2004.842847, 40(4): 970 977.
Zhang, B., Allen, P. E. & Huard, J. M. (2003). A fast switching PLL frequency synthesizer
with an on-chip passive discrete-time loop lter in 0.25-m CMOS, IEEE Journal of
Solid-State Circuits, 38(6): 855 865.
Zhao, B., Mao, X., Yang, H. & Wang, H. (2009). A 1.41-1.72 GHz sigma-delta fractional-N
frequency synthesizer with a PVT insensitive VCO and a new prescaler, Analog Inte-
grated Circuits and Signal Processing, 59(3): 265273.
SNR (Signal-to-Noise Ratio): The theory equation of SNR for a N-bit ADC can be expressed
as:
SNR = 6.02N +1.76 (33)
SNDR (Signal-to-Noise and Distortion Ratio): The power of the noise and harmonics di-
vided by the power of signal.
SFDR (Spurious Free Dynamic Range): The ratio of the signals power to the maximum
harmonics power.
ENOB (Effective Number of Bits): The ENOB can be calculated by:
ENOB = (SNDR 1.76)/6.02 (34)
THD (Total Harmonic Distortion): The ratio of all the harmonics power to the signals
power.
Besides the specications above, another very important parameter is power consumption. In
the aspect of architecture, the ADCs can be divided into ash, SAR (Successive AppRoxima-
tion), folding, pipeline, sigma-delta, and so on.
For the ADC connected after the sensor in a WSN node, SAR ADCs may be the best choice
since the data collection is executed at most of the time. There are two reasons: 1)The only
power hungry module in a SAR ADC is the comparator, so the overall power consumption
is low; 2)the circuit structure of a SAR ADC is simple, so the cost of chip area is small. In the
WSN transceivers, the data rate is not high and the modulation scheme is simple, so a low
bandwidth and low precision ADC is usually enough. Therefore, a SAR ADC may also be a
good choice for WSN transceivers.
The specications of DAC also includes DNL, INL, SNR, SNDR, SFDR, power consumption,
and so on. The DACs are often used in transmitters, as shown in Fig. 6, the shaped digital
waves are converted to analog signals that are sent to up-converting mixers for modulation.
As a result, the performance of the DACs will affect the EVM and ACPR of the transmitters.
5. Conclusion
The goal of this chapter is to give a brief manual for WSN transceiver design. Section 1 gives
an introduction of WSN and the RF transceivers for WSN. The WSN transceivers are classi-
ed by both modulation schemes and architectures in section 2. How to calculate and assign
the system specications are described in section 3. The design of key modules is analyzed
briey in section 4. The readers is expected to master a top-to-down design method for WSN
transceivers through the chapter.
6. References
Crols, J. & Steyaert, M. (1995). An analog integrated polyphase lter for a high performance
low-IF receiver, Symposium on VLSI Circuits, Digest of Technical Papers, pp. 8788.
Gu, Q. (2005). RF System Design of Transceivers for Wireless Communications, Springer Sci-
ence+Business Media, LLC,.
Jri, L., Chen, Y. & Yenlin, H. (2010). A Low-Power Low-Cost Fully-Integrated 60-GHz
Transceiver System With OOK Modulation and On-Board Antenna Assembly, IEEE
Journal of Solid-State Circuits, DOI-10.1109/JSSC.2009.2034806, 45(2): 264275.
Perrott, M. H., Tewksbury, T. L. & Sodini, C. G. (1997). A 27-mW CMOS fractional-N syn-
thesizer using digital compensation for 2.5-Mb/s GFSK modulation, IEEE Journal of
Solid-State Circuits, DOI-10.1109/4.643663, 32(12): 20482060.
Pletcher, N. M., Gambini, S. & Rabaey, J. (2009). A 52 W Wake-Up Receiver With -72 dBm
Sensitivity Using an Uncertain-IF Architecture, IEEE Journal of Solid-State Circuits,
DOI-10.1109/JSSC.2008.2007438, 44(1): 269280.
Seungkee, M., Shashidharan, S., Stevens, M., Copani, T., Kiaei, S., Bakkaloglu, B. &
Chakraborty, S. (2010). A 2mW CMOS MICS-band BFSK transceiver with recon-
gurable antenna interface, IEEE Radio Frequency Integrated Circuits Symposium,,
pp. 289292.
Valla, M., Montagna, G., Castello, R., Tonietto, R. & Bietti, I. (2005). A 72-mW CMOS 802.11a
direct conversion front-end with 3.5-dB NF and 200-kHz 1/f noise corner, IEEE Jour-
nal of Solid-State Circuits, DOI-10.1109/JSSC.2004.842847, 40(4): 970 977.
Zhang, B., Allen, P. E. & Huard, J. M. (2003). A fast switching PLL frequency synthesizer
with an on-chip passive discrete-time loop lter in 0.25-m CMOS, IEEE Journal of
Solid-State Circuits, 38(6): 855 865.
Zhao, B., Mao, X., Yang, H. & Wang, H. (2009). A 1.41-1.72 GHz sigma-delta fractional-N
frequency synthesizer with a PVT insensitive VCO and a new prescaler, Analog Inte-
grated Circuits and Signal Processing, 59(3): 265273.
MAC & Mobility In Wireless Sensor Networks
Marwan Al-Jemeli, Vooi Voon Yap and Fawnizu Azmadi Bin Hussin
X

MAC & Mobility In Wireless Sensor Networks

Marwan Al-Jemeli
1
, Vooi Voon Yap
2
and Fawnizu Azmadi Bin Hussin
1
1
Universiti Teknologi PETRONAS
2
Universiti Tuanku Abdul Rahman
Malaysia

1. Introduction
The recent climate change has a significant impact on our planet environment. Therefore,
deploying sensor networks to monitor the environment is becoming important. With sensor
networks deployed in strategic location can provide the scientific communities useful data
to be analyzed and take action if necessary. Typical environmental applications of sensor
networks include, but not limited to, monitoring environmental conditions that affect crops
and livestock, biological, Earth, and environmental monitoring and many more. Monitoring
hazardous environment like volcanic activities is one of the important applications for
Wireless Sensor Network (WSN) (Sohraby et al, 2007). WSN communicate wirelessly to pass
and process information see Figure 1.

These sensor networks are deployed far away from the nearest permanent energy source
available which make them depending on their own energy source to provide the needed
information.

WSNs usually consist of a large number of low-cost, low-power, multifunctional (or uni-
functional) wireless devices deployed over a geographical area in an ad hoc fashion and
with or without careful planning (this depends on the application mainly whether it is
related to a real-time applications or non-real-time application). Individually, these devices
have limited resources and have limited processing and communication capabilities. The
cooperative operation behavior of these sensing devices gives a significant impact on a wide
range of applications in several fields, including science and engineering, military settings,
critical infrastructure protection, and environmental monitoring (Yu et al, 2006).
Networking distributed sensors are used in military and industrial applications and it dates
back at least to the 1970s. back then the systems were primarily wired and small in scale.
wireless technologies and low-power Very Large Scale of Integration (VLSI) design became
feasible and emerged in 1990 and after that researchers began envisioning and investigating
large-scale embedded wireless sensor networks for dense sensing applications
(Krishnamachari, 2005).
However, wireless sensor networks have a major problem, that is, network life time. Since
WSN uses batteries, it does them in terms of storage, and processing power. Limited
capabilities results in limited information efficiency. Current available technology on-shelf
15
allow us to produce sensors that consumes as little power as 100mW which means that the
sensors can remain operational efficiently (depending on the application and the deployed
nodes own capabilities) for about 10 months. Yet the life time of the network can be
extended for further than 10 months.

Fig. 1. Wireless Sensor Networks Example.

Some researchers proposed methods includes energy harvesting, solar energy and vibration
energy. But these methods can only provide a small amount of energy to power these
sensors, typically 20mw or less (Mainwaring et al,2002; Raghunathan et al, 2002).
Maintenance and recharging these sensors is not a good option, and it will increase the
expenses to keep the network alive and operational. Another alternative is to use energy
efficient information processing and transacting algorithms to manage the network
operation. We envisage that efficient routing and Medium Access Control (MAC) Protocols
can help resolve this problem.
Information processing and routing is a technique used widely when it comes to provide a
longer life time operation in wireless sensor networks however these techniques lacks the
integrity as it has to compensate between either providing an energy efficient operation
with the lack of high throughput or vice versa (Boukerche et al, 2005; Branyard et al, 2006).
Mobile
Client
Web
Client
Inst. Client
Wind
Speed > 5
One of the major levels of tweaking in networking systems is to manipulate the timing when
to deliver particular packets at a precise times to achieve efficient operation. From the
literature provided most of the available approaches consider the main purpose of
manipulating information processing technique is to achieve better energy consumption in
the nodes while sacrificing the system throughput quality and robustness (Branyard et al,
2006). the next section will discuss MAC theory and the related works that where done in
this area of research follows it in section three our own theory and methodology. Section
four and five will discuss the results that where obtained during our research . Section six
will review Mobility issues in brief and section seven reviews our proposed future goals in
this research.

2. MAC protocols effect in WSNs
MAC is the second layer after the physical layer in the Open System Interconnection (OSI)
model in networking systems, MAC protocols controls when to send and receive
distinguished packet between different nodes in a network. It controls the network interface
when to establish the connection or the transaction between two or more hosts.
Manipulating the operation of a MAC protocol can give its effect in terms of energy
consumption and message delay between nodes (Van Hoesel, Havigna, 2004).
Different MAC protocols were defined for WSN because of its application dependency.
MAC protocols have to compensate between providing energy efficient consumption with
the availability of decent throughput to make the system dependable (Van Hoesel, Havigna,
2004; Law et al, 2005) .
An essential characteristic of wireless communication is that it provides an inherently
shared medium. All MAC protocols for wireless networks manage the usage of the radio
interface to ensure efficient utilization of the shared bandwidth. MAC protocols designed
for wireless sensor networks have an additional goal of managing radio activity to conserve
energy. Thus, while traditional MAC protocols must balance throughput, delay, and
fairness concerns, WSN MAC protocols place an emphasis on energy efficiency as well
(Krishnamachari, 2005). MAC layer affects the energy efficiency mainly through the
adjustment of transmission scheduling and channel access. A common way to do that is via
sleep scheduling from a long time scale, or time-division multiple access (TDMA), from a
short time scale perspective. Similar to the shutdown technique of CPUs, sleep scheduling
also explores the energy vs. response time tradeoffs in wireless communication. From
previous studies, the response time is translated to network or application layer
transmission delay or throughput.
(Mathioudakis et al, 2008) presented the most energy wastage sources in MAC protocols for
WSNs:
The first source is caused by collisions, which occur when two or more nodes attempt
to transmit simultaneously. The need to re-transmit a packet that has been corrupted by
collision increases the energy consumption.
The second source of energy wastage is idle-listening, where a node listens for traffic
that it is not sent. In a sample fetching operation, a silent channel can be high in several
sensor applications.
The third source of waste is overhearing, which occurs when a sensor node receives
packets that are destined for other nodes.
allow us to produce sensors that consumes as little power as 100mW which means that the
sensors can remain operational efficiently (depending on the application and the deployed
nodes own capabilities) for about 10 months. Yet the life time of the network can be
extended for further than 10 months.

Fig. 1. Wireless Sensor Networks Example.

Some researchers proposed methods includes energy harvesting, solar energy and vibration
energy. But these methods can only provide a small amount of energy to power these
sensors, typically 20mw or less (Mainwaring et al,2002; Raghunathan et al, 2002).
Maintenance and recharging these sensors is not a good option, and it will increase the
expenses to keep the network alive and operational. Another alternative is to use energy
efficient information processing and transacting algorithms to manage the network
operation. We envisage that efficient routing and Medium Access Control (MAC) Protocols
can help resolve this problem.
Information processing and routing is a technique used widely when it comes to provide a
longer life time operation in wireless sensor networks however these techniques lacks the
integrity as it has to compensate between either providing an energy efficient operation
with the lack of high throughput or vice versa (Boukerche et al, 2005; Branyard et al, 2006).
Mobile
Client
Web
Client
Inst. Client
Wind
Speed > 5
One of the major levels of tweaking in networking systems is to manipulate the timing when
to deliver particular packets at a precise times to achieve efficient operation. From the
literature provided most of the available approaches consider the main purpose of
manipulating information processing technique is to achieve better energy consumption in
the nodes while sacrificing the system throughput quality and robustness (Branyard et al,
2006). the next section will discuss MAC theory and the related works that where done in
this area of research follows it in section three our own theory and methodology. Section
four and five will discuss the results that where obtained during our research . Section six
will review Mobility issues in brief and section seven reviews our proposed future goals in
this research.

2. MAC protocols effect in WSNs
MAC is the second layer after the physical layer in the Open System Interconnection (OSI)
model in networking systems, MAC protocols controls when to send and receive
distinguished packet between different nodes in a network. It controls the network interface
when to establish the connection or the transaction between two or more hosts.
Manipulating the operation of a MAC protocol can give its effect in terms of energy
consumption and message delay between nodes (Van Hoesel, Havigna, 2004).
Different MAC protocols were defined for WSN because of its application dependency.
MAC protocols have to compensate between providing energy efficient consumption with
the availability of decent throughput to make the system dependable (Van Hoesel, Havigna,
2004; Law et al, 2005) .
An essential characteristic of wireless communication is that it provides an inherently
shared medium. All MAC protocols for wireless networks manage the usage of the radio
interface to ensure efficient utilization of the shared bandwidth. MAC protocols designed
for wireless sensor networks have an additional goal of managing radio activity to conserve
energy. Thus, while traditional MAC protocols must balance throughput, delay, and
fairness concerns, WSN MAC protocols place an emphasis on energy efficiency as well
(Krishnamachari, 2005). MAC layer affects the energy efficiency mainly through the
adjustment of transmission scheduling and channel access. A common way to do that is via
sleep scheduling from a long time scale, or time-division multiple access (TDMA), from a
short time scale perspective. Similar to the shutdown technique of CPUs, sleep scheduling
also explores the energy vs. response time tradeoffs in wireless communication. From
previous studies, the response time is translated to network or application layer
transmission delay or throughput.
(Mathioudakis et al, 2008) presented the most energy wastage sources in MAC protocols for
WSNs:
The first source is caused by collisions, which occur when two or more nodes attempt
to transmit simultaneously. The need to re-transmit a packet that has been corrupted by
collision increases the energy consumption.
The second source of energy wastage is idle-listening, where a node listens for traffic
that it is not sent. In a sample fetching operation, a silent channel can be high in several
sensor applications.
The third source of waste is overhearing, which occurs when a sensor node receives
packets that are destined for other nodes.
The fourth is caused by control packet overheads, which are required to regulate access
to the transmission channel. Sending and receiving control packets consumes energy
too, and less useful data packets can be transmitted.
The fifth source is over-emitting where the destination node is not ready to receive
during the transmission procedure, and hence the packet is not correctly received.
Finally, the transition between different operation modes, such as sleep, idle, receive
and transmit, can result in significant energy consumption. Limiting the number of
transitions between sleep and active modes leads to a considerable energy saving.

2.1 Related Approaches
For wireless sensor networks the literature provided a lot of protocols and divided it into
two major categories (Shukur et al, 2009):
1. Contention Based MAC Protocols (CSMA carrier sense multiple access). The
wireless nodes here contend to enter the medium of connectivity (which is the
wireless medium in case of WSNs) and the winner node reserves the medium to
itself until it finishes its operation. Examples for this kind of protocols are: IEEE
802.11, S-MAC (Ye et al, 2001), T-MAC (Van Dam, Longendean, 2003), R-MAC (Du
et al, 2007) and others.
2. TDMA (time division multiple access) Based MAC Protocols. The medium here is
divided into time slots each node knows its time slot when to enter the medium
and do its operation. One popular TDMA based MAC protocol for WSNs is
ALOHA (PARK et al, 2006).
Contention based MAC protocols offers a more scalability approach over TDMA based
approaches because of the nature of TDMA approaches that requires slotting the time into
slots to each node which is improper when deploying a large number of nodes. To list dome
of the works in this area of research, follows are some approaches regarding CSMA based
MAC protocols:

A popular contention based MAC protocol for wireless networks is the IEEE 802.11 which is
the standard for WLAN applications. IEEE 802.11 performs well in terms of latency and
throughput but it is not efficient in terms of energy consumption because of the idle
listening problem. It has been shown that when the node is in idle listening state it
consumes energy equivalent to the receiving energy and that is why this protocol is not
suitable for WSNs applications (Ye et al, 2001).

Sensor-MAC, S-MAC is a contention based MAC protocol designed explicitly for wireless
sensor networks proposed by (Ye et al, 2001). While reducing energy consumption is the
primary goal of this design, the protocol also has good scalability and collision avoidance
capability. It achieves good scalability and collision avoidance by utilizing a combined
scheduling and contention scheme. It also achieves efficient energy consumption by using a
scheme of periodic listening and sleeping which reduces energy consumption. In addition, it
uses synchronization to form virtual clusters of nodes on the same sleep schedule. These
schedules coordinate nodes to minimize additional latency. The protocol also uses the same
mechanism to avoid the overhearing problem and hidden channel problem that is used in
IEEE 802.11. But the S-MAC has a problem of latency because of periodic listen and sleep
scheme which is dependent on the duty cycle.
WSNs applications have some unique operation characteristics, for example, low message
rate, insensitivity to latency. These characteristics can be exploited to reduce energy
consumption by introducing an active/sleep duty cycle. To handle load variations in time
and location, (Van Dam, Langendeon, 2003) proposed the Timeout MAC T-MAC protocol.
T-MAC can handle an adaptive duty cycle in a novel way: by dynamically ending the active
part of it. This reduces the amount of energy wasted on idle listening, in which nodes wait
for potentially incoming messages, while still maintaining a reasonable throughput. T-MAC
uses TA (time out) packet to end the active part when there is no data to send/receive on the
node. The protocol balances between energy efficient consumption and latency efficient
throughput due to the scheme of burst data sending more effective in terms of energy
consumption.

The concept of periodic listen and sleep approach was explored by (Suh, Ko, 2005). They
proposed a novel MAC scheme named as TEEM (Traffic aware, Energy Efficient MAC)
protocol. The proposed TEEM is based on the often cited contention-based MAC protocol S-
MAC. The protocol achieves energy efficient consumption by utilizing traffic information
of each sensor node.

Thus, Suh and Ko show that the listen time of nodes can be reduced by putting them into
sleep state earlier when they expect no data traffic to occur. In this method, they made two
important modifications to the S-MAC protocol: the first modification was to make all nodes
turn off the radio interface much earlier when no data packet transfer is expected to occur in
the networks, and secondly eliminating communication of a separate RTS control packet
even when data traffic is likely to occur. However, it lacks on latency efficiency to conserve
energy.

The cross-layer approach protocol was investigated by (Pack et al, 2006) . They proposed a
task aware MAC protocol for WSNs. The TA-MAC protocol determines the channel access
probability depending on a nodes and its neighbor nodes traffic loads through the
interaction with the data dissemination protocol. In this approach the TA-MAC protocol can
reduce energy consumption and improve the throughput by eliminating unnecessary
collisions. The TA-MAC protocol is feasible because it can be integrated with other energy
efficient MAC protocol example, SMAC. The TA-MAC protocol focuses on the
determination of channel access probability that is orthogonal to the previous MAC
protocols for WSNs.

Another work that explores the cross-layer approach was presented by (Du et al, 2007) . The
proposed scheme called Routing-enhanced MAC protocol (RMAC), exploits cross-layer
routing information in order provide delay guarantee without sacrificing energy efficiency.
Most importantly, RMAC can deliver a data packet multiple hops in a single operational
cycle. During the SLEEP period in RMAC, a relaying node for a data packet goes to sleep
and then wake up when its upstream node has the data packet ready to transmit to it. After
the data packet is received by this relaying node, it can also immediately forward the packet
to its next downstream node, as that node has just woken up and is ready to receive the data
packet. The mechanism is implemented using a packet called Pioneer. This packet travels to
The fourth is caused by control packet overheads, which are required to regulate access
to the transmission channel. Sending and receiving control packets consumes energy
too, and less useful data packets can be transmitted.
The fifth source is over-emitting where the destination node is not ready to receive
during the transmission procedure, and hence the packet is not correctly received.
Finally, the transition between different operation modes, such as sleep, idle, receive
and transmit, can result in significant energy consumption. Limiting the number of
transitions between sleep and active modes leads to a considerable energy saving.

2.1 Related Approaches
For wireless sensor networks the literature provided a lot of protocols and divided it into
two major categories (Shukur et al, 2009):
1. Contention Based MAC Protocols (CSMA carrier sense multiple access). The
wireless nodes here contend to enter the medium of connectivity (which is the
wireless medium in case of WSNs) and the winner node reserves the medium to
itself until it finishes its operation. Examples for this kind of protocols are: IEEE
802.11, S-MAC (Ye et al, 2001), T-MAC (Van Dam, Longendean, 2003), R-MAC (Du
et al, 2007) and others.
2. TDMA (time division multiple access) Based MAC Protocols. The medium here is
divided into time slots each node knows its time slot when to enter the medium
and do its operation. One popular TDMA based MAC protocol for WSNs is
ALOHA (PARK et al, 2006).
Contention based MAC protocols offers a more scalability approach over TDMA based
approaches because of the nature of TDMA approaches that requires slotting the time into
slots to each node which is improper when deploying a large number of nodes. To list dome
of the works in this area of research, follows are some approaches regarding CSMA based
MAC protocols:

A popular contention based MAC protocol for wireless networks is the IEEE 802.11 which is
the standard for WLAN applications. IEEE 802.11 performs well in terms of latency and
throughput but it is not efficient in terms of energy consumption because of the idle
listening problem. It has been shown that when the node is in idle listening state it
consumes energy equivalent to the receiving energy and that is why this protocol is not
suitable for WSNs applications (Ye et al, 2001).

Sensor-MAC, S-MAC is a contention based MAC protocol designed explicitly for wireless
sensor networks proposed by (Ye et al, 2001). While reducing energy consumption is the
primary goal of this design, the protocol also has good scalability and collision avoidance
capability. It achieves good scalability and collision avoidance by utilizing a combined
scheduling and contention scheme. It also achieves efficient energy consumption by using a
scheme of periodic listening and sleeping which reduces energy consumption. In addition, it
uses synchronization to form virtual clusters of nodes on the same sleep schedule. These
schedules coordinate nodes to minimize additional latency. The protocol also uses the same
mechanism to avoid the overhearing problem and hidden channel problem that is used in
IEEE 802.11. But the S-MAC has a problem of latency because of periodic listen and sleep
scheme which is dependent on the duty cycle.
WSNs applications have some unique operation characteristics, for example, low message
rate, insensitivity to latency. These characteristics can be exploited to reduce energy
consumption by introducing an active/sleep duty cycle. To handle load variations in time
and location, (Van Dam, Langendeon, 2003) proposed the Timeout MAC T-MAC protocol.
T-MAC can handle an adaptive duty cycle in a novel way: by dynamically ending the active
part of it. This reduces the amount of energy wasted on idle listening, in which nodes wait
for potentially incoming messages, while still maintaining a reasonable throughput. T-MAC
uses TA (time out) packet to end the active part when there is no data to send/receive on the
node. The protocol balances between energy efficient consumption and latency efficient
throughput due to the scheme of burst data sending more effective in terms of energy
consumption.

The concept of periodic listen and sleep approach was explored by (Suh, Ko, 2005). They
proposed a novel MAC scheme named as TEEM (Traffic aware, Energy Efficient MAC)
protocol. The proposed TEEM is based on the often cited contention-based MAC protocol S-
MAC. The protocol achieves energy efficient consumption by utilizing traffic information
of each sensor node.

Thus, Suh and Ko show that the listen time of nodes can be reduced by putting them into
sleep state earlier when they expect no data traffic to occur. In this method, they made two
important modifications to the S-MAC protocol: the first modification was to make all nodes
turn off the radio interface much earlier when no data packet transfer is expected to occur in
the networks, and secondly eliminating communication of a separate RTS control packet
even when data traffic is likely to occur. However, it lacks on latency efficiency to conserve
energy.

The cross-layer approach protocol was investigated by (Pack et al, 2006) . They proposed a
task aware MAC protocol for WSNs. The TA-MAC protocol determines the channel access
probability depending on a nodes and its neighbor nodes traffic loads through the
interaction with the data dissemination protocol. In this approach the TA-MAC protocol can
reduce energy consumption and improve the throughput by eliminating unnecessary
collisions. The TA-MAC protocol is feasible because it can be integrated with other energy
efficient MAC protocol example, SMAC. The TA-MAC protocol focuses on the
determination of channel access probability that is orthogonal to the previous MAC
protocols for WSNs.

Another work that explores the cross-layer approach was presented by (Du et al, 2007) . The
proposed scheme called Routing-enhanced MAC protocol (RMAC), exploits cross-layer
routing information in order provide delay guarantee without sacrificing energy efficiency.
Most importantly, RMAC can deliver a data packet multiple hops in a single operational
cycle. During the SLEEP period in RMAC, a relaying node for a data packet goes to sleep
and then wake up when its upstream node has the data packet ready to transmit to it. After
the data packet is received by this relaying node, it can also immediately forward the packet
to its next downstream node, as that node has just woken up and is ready to receive the data
packet. The mechanism is implemented using a packet called Pioneer. This packet travels to
all sensors in down-stream to synchronize the duty-cycles of the nodes to guarantee a multi-
hop packet delivery. In this way the protocol achieved latency efficient operation.

(Erazo, Qain, 2007) developed the S-MAC to SEA-MAC, a protocol which aims for energy
efficient operation for WSNs for environment monitoring. The protocol assumes only the
base station node has the time synchronization schedule. Sensor nodes are active only when
there is a sample to be taken from the environment which decreases the duty-cycle of the
node and preserves energy. The packet which is responsible for initiating important data
delivery in SEA-MAC is called TONE packet which is shorter in period than SYNC packet in
S-MAC.

The literature trawl has revealed that few protocols use TDMA-based scheduling because of
the overhead of time slot scheduling as sensor network deployment usually includes large
number of sensors. A protocol that uses TDMA-based scheduling is the Energy and Rate
(ER) proposed by (Kannan et al, 2003). The ER_MAC protocol has the ability of avoiding
extra energy wastage.
The main advantages of ER-MAC are:
packet loss due to collisions is absent because two nodes do not transmit in the
same slot. Although packet loss may occur due to other reasons like interference,
loss of signal strength etc.
no contention mechanism is required for a node to start sensing its packets since
the slots are pre-assigned to each node. No extra control overhead packets for
contention are required.
ER-MAC uses the concept of periodic listen and sleep. A sensor node switches off its radio
and goes into a sleep mode only when it is in its own time slot and does not have anything
to transmit. It has to keep the radio awake in the slots assigned to its neighbors in order to
receive packets from them even if the node with current slot has nothing to transmit.

Real-Time MAC (RT-MAC) proposed by (Sahoo, Baronia, 2007) is another TDMA-based
MAC protocol that can provide delay guarantee. TDMA based MAC protocols suffers from
latency caused by the assigning of time slots which takes up a lot of time because of the
number of sensor nodes deployed. RT-MAC overcomes this problem by reutilizing the
connection channel between two successive channel accesses of a sensor node. RT-MAC also
allows sensors to go to sleep which preserves energy. Although it provides delay guarantee,
the RT-MAC protocol requires a lot of computation that exhaust the sensor node itself in
some cases like clock drifting problem.
There are other works on design of MAC protocol based on TDMA scheme (Ganeriwal et al,
2003; Egea-L'opez et al, 2006); they all share the same complexity in time slot assigning.
To summarize the investigated literature, we devised a table that illustrates the categories of
MAC protocols proposed for WSNs showing their advantages and disadvantages. Refer to
Table 2:

MAC
Protocol
Category Main Advantage Main Disadvantage
IEEE 802.11
CSMA/C
A
The Highest system throughput
Inefficient energy
consumption
S-MAC
CSMA/C
A
Scalable, energy efficient due to
the sleep/listen scheme
Suffers from Latency issues
T-MAC
CSMA/C
A
Energy efficient, Reasonable
throughput
Requires extended control
packet to achieve efficient
operation
TEEM
CSMA/C
A
Energy efficient due to the
eliminating the use of RTS
packet
TA-MAC
CSMA/C
A
Cross-Layer approach Suffers from latency issues
R-MAC
CSMA/C
A
Enhanced throughput
Control Packet Delivery
overhead
SEA-MAC
CSMA/C
A
Energy efficient operation Suffers from Latency issues
ER-MAC TDMA Collision free environment Scalability and latency issues
RT-MAC TDMA
Increased the system
throughput
Excessive calculation and
clock drifting problems
Table 1. Summary of the related approaches for MAC Protocols.

3. Proposed solution and the methodology behind it
In this section we ill discuss our proposed solution and describe the operation on the
protocol. It also discusses how it manages control packets and data packets exchanges
between the network nodes. Energy consumption and packet exchange delay analysis are
also discussed. To prove the method proposed we devised simulation experiments using the
most common tool to simulate networking systems the Network Simulator 2 (NS2)
(Issariyakul, Hossain, 2008). The analysis equations were based on the theory of S-MAC.

3.1 The Network Simulator 2 (NS2)
NS2 is the most widely used tool in researches involved in general networking systems
analysis and wireless networking systems includes Mobile networking, Satellite networking,
Wireless Sensor Networks, LAN networks and other network technologies. NS2 is built
using C++ language and uses OTcl (Object Oriented Tcl) language as an interface with the
simulator. The network topology is built using OTcl and the packet operation protocol is
written in C++ (Issariyakul, Hossain, 2008).

3.1.2 Mobile Networking In NS2
The wireless model essentially consists of the MobileNode at the core, with additional
supporting features that allows simulations of multi-hop ad-hoc networks, wireless LANs
etc. A MobileNode thus is the basic Node object with added functionalities of a wireless and
mobile node like ability to move within a given topology, ability to receive and transmit
signals to and from a wireless channel.
all sensors in down-stream to synchronize the duty-cycles of the nodes to guarantee a multi-
hop packet delivery. In this way the protocol achieved latency efficient operation.

(Erazo, Qain, 2007) developed the S-MAC to SEA-MAC, a protocol which aims for energy
efficient operation for WSNs for environment monitoring. The protocol assumes only the
base station node has the time synchronization schedule. Sensor nodes are active only when
there is a sample to be taken from the environment which decreases the duty-cycle of the
node and preserves energy. The packet which is responsible for initiating important data
delivery in SEA-MAC is called TONE packet which is shorter in period than SYNC packet in
S-MAC.

The literature trawl has revealed that few protocols use TDMA-based scheduling because of
the overhead of time slot scheduling as sensor network deployment usually includes large
number of sensors. A protocol that uses TDMA-based scheduling is the Energy and Rate
(ER) proposed by (Kannan et al, 2003). The ER_MAC protocol has the ability of avoiding
extra energy wastage.
The main advantages of ER-MAC are:
packet loss due to collisions is absent because two nodes do not transmit in the
same slot. Although packet loss may occur due to other reasons like interference,
loss of signal strength etc.
no contention mechanism is required for a node to start sensing its packets since
the slots are pre-assigned to each node. No extra control overhead packets for
contention are required.
ER-MAC uses the concept of periodic listen and sleep. A sensor node switches off its radio
and goes into a sleep mode only when it is in its own time slot and does not have anything
to transmit. It has to keep the radio awake in the slots assigned to its neighbors in order to
receive packets from them even if the node with current slot has nothing to transmit.

Real-Time MAC (RT-MAC) proposed by (Sahoo, Baronia, 2007) is another TDMA-based
MAC protocol that can provide delay guarantee. TDMA based MAC protocols suffers from
latency caused by the assigning of time slots which takes up a lot of time because of the
number of sensor nodes deployed. RT-MAC overcomes this problem by reutilizing the
connection channel between two successive channel accesses of a sensor node. RT-MAC also
allows sensors to go to sleep which preserves energy. Although it provides delay guarantee,
the RT-MAC protocol requires a lot of computation that exhaust the sensor node itself in
some cases like clock drifting problem.
There are other works on design of MAC protocol based on TDMA scheme (Ganeriwal et al,
2003; Egea-L'opez et al, 2006); they all share the same complexity in time slot assigning.
To summarize the investigated literature, we devised a table that illustrates the categories of
MAC protocols proposed for WSNs showing their advantages and disadvantages. Refer to
Table 2:

MAC
Protocol
Category Main Advantage Main Disadvantage
IEEE 802.11
CSMA/C
A
The Highest system throughput
Inefficient energy
consumption
S-MAC
CSMA/C
A
Scalable, energy efficient due to
the sleep/listen scheme
T-MAC
CSMA/C
A
Energy efficient, Reasonable
throughput
Requires extended control
packet to achieve efficient
operation
TEEM
CSMA/C
A
Energy efficient due to the
eliminating the use of RTS
packet
TA-MAC
CSMA/C
A
Cross-Layer approach Suffers from latency issues
R-MAC
CSMA/C
A
Enhanced throughput
Control Packet Delivery
overhead
SEA-MAC
CSMA/C
A
Energy efficient operation Suffers from Latency issues
ER-MAC TDMA Collision free environment Scalability and latency issues
RT-MAC TDMA
Increased the system
throughput
Excessive calculation and
clock drifting problems
Table 1. Summary of the related approaches for MAC Protocols.

3. Proposed solution and the methodology behind it
In this section we ill discuss our proposed solution and describe the operation on the
protocol. It also discusses how it manages control packets and data packets exchanges
between the network nodes. Energy consumption and packet exchange delay analysis are
also discussed. To prove the method proposed we devised simulation experiments using the
most common tool to simulate networking systems the Network Simulator 2 (NS2)
(Issariyakul, Hossain, 2008). The analysis equations were based on the theory of S-MAC.

3.1 The Network Simulator 2 (NS2)
NS2 is the most widely used tool in researches involved in general networking systems
analysis and wireless networking systems includes Mobile networking, Satellite networking,
Wireless Sensor Networks, LAN networks and other network technologies. NS2 is built
using C++ language and uses OTcl (Object Oriented Tcl) language as an interface with the
simulator. The network topology is built using OTcl and the packet operation protocol is
written in C++ (Issariyakul, Hossain, 2008).

3.1.2 Mobile Networking In NS2
The wireless model essentially consists of the MobileNode at the core, with additional
supporting features that allows simulations of multi-hop ad-hoc networks, wireless LANs
etc. A MobileNode thus is the basic Node object with added functionalities of a wireless and
mobile node like ability to move within a given topology, ability to receive and transmit
signals to and from a wireless channel.
3.1.3 Routing and MAC protocols provided in NS2
Two MAC layer protocols are implemented for mobile networks, which are IEEE 802.11 and
TDMA, while S-MAC was added to NS2 as a Patch by (Ye et al,2001). The four different ad-
hoc routing protocols currently implemented for mobile networking in NS2 are dsdv, dsr,
aodv and tora.

3.2 The Proposed Scheme
The proposed scheme (Shukur, Yap, 2009, a; Shukur, Yap, 2009, b) considers the following:
1. Combining the functionality of SYNC packet with RTS packet will provide both
energy and latency efficient operation which will eliminate the need of sending
two packets and decrease control packet overhead. This packet from now on would
be referred to as SEEK.
2. To increase the throughput of the system (SEEK) packet will be sent all the way to
the down stream nodes before sending CTS packet to the upper stream node. This
will open the way to DATA packet to move through the stream of nodes until
DATA packet reaches the base station node. Figure 2 Describes the approach
mentioned above.

3.2.1 Energy Consumption analysis
The first step is to analyze the proposed approach energy consumption for three nodes
operation. The following assumptions are made for the analysis (using the scenario shown
in Figure 3-3 below:

1. All nodes in the way are by all means available for any packet transmission.
2. The packet delivery direction is from node 1 to node 3.
3. No collision happens between nodes (assuming that Carrier Sense is successful in
each transmission start).
4. SEEK packet follow this rule (SYNC<SEEK<SYNC+RTS).
5. DATA packet could be transmitted in one hop.
6. All control packets are fixed in size.
7. In a more realistic scenario upper-layer routing information provides the shortest
route to the destination.
8. DATA packet can be transferred in one hop.
9. If the next node in the way is in sleep mode (SEEK) works as the signal that wakes
up the node.

Fig. 2. Proposed Scheme operation for Synchronization in MAC layer Protocol.

The analysis scenario is described in Figure 3:

Fig. 3. Analysis Scenario.

Each Node operation is represented by a formula following the devised Synchronization
timeline which is described in figure 2 above. Starting with the operation of Node (1) form
F
l
o
w

d
i
r
e
c
t
i
o
n

Node 1
Node 2
Node 3
3.1.3 Routing and MAC protocols provided in NS2
Two MAC layer protocols are implemented for mobile networks, which are IEEE 802.11 and
TDMA, while S-MAC was added to NS2 as a Patch by (Ye et al,2001). The four different ad-
hoc routing protocols currently implemented for mobile networking in NS2 are dsdv, dsr,
aodv and tora.

3.2 The Proposed Scheme
The proposed scheme (Shukur, Yap, 2009, a; Shukur, Yap, 2009, b) considers the following:
1. Combining the functionality of SYNC packet with RTS packet will provide both
energy and latency efficient operation which will eliminate the need of sending
two packets and decrease control packet overhead. This packet from now on would
be referred to as SEEK.
2. To increase the throughput of the system (SEEK) packet will be sent all the way to
the down stream nodes before sending CTS packet to the upper stream node. This
will open the way to DATA packet to move through the stream of nodes until
DATA packet reaches the base station node. Figure 2 Describes the approach
mentioned above.

3.2.1 Energy Consumption analysis
The first step is to analyze the proposed approach energy consumption for three nodes
operation. The following assumptions are made for the analysis (using the scenario shown
in Figure 3-3 below:

1. All nodes in the way are by all means available for any packet transmission.
2. The packet delivery direction is from node 1 to node 3.
3. No collision happens between nodes (assuming that Carrier Sense is successful in
each transmission start).
4. SEEK packet follow this rule (SYNC<SEEK<SYNC+RTS).
5. DATA packet could be transmitted in one hop.
6. All control packets are fixed in size.
7. In a more realistic scenario upper-layer routing information provides the shortest
route to the destination.
8. DATA packet can be transferred in one hop.
9. If the next node in the way is in sleep mode (SEEK) works as the signal that wakes
up the node.

Fig. 2. Proposed Scheme operation for Synchronization in MAC layer Protocol.

The analysis scenario is described in Figure 3:

Fig. 3. Analysis Scenario.

Each Node operation is represented by a formula following the devised Synchronization
timeline which is described in figure 2 above. Starting with the operation of Node (1) form
F
l
o
w

d
i
r
e
c
t
i
o
n

Node 1
Node 2
Node 3
the Scenario above (figure 3) the formula of Energy consumption can be represented as
follow:
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
+

+ =
+ + =
d
T
) (t
rect P
c
T
)
1
m
3
(t
rect P
s
T
t
rect
t
P (t)
1
S

.....(1) .......... .......... .......... .......... .......... .......... .......... Xd(t)..... )
3
Y(t X(t) (t)
1
S

Where:
T
s
: SEEK packet time length.
T
c
: CTS packet time length.
T
d
: DATA packet time length.
: the delay in each state of transmitting SEEK packet and receiving CTS packet.
P
t
: Transmission Power.
P : Reception Power.
X(t): rectangular function of delay for SEEK packet.
Y(t): rectangular function of delay for CTS packet out from the exact node.
Z(t): rectangular function of delay for CTS packet received from the down stream node.
X
d
(t): rectangular function of delay for DATA packet.

Node (2) energy consumption is equal to the following equation:

(
(
|
|
.
|
\
|
|
|
.
|
\
|
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|

+
=
+ + + =
d
T
)
4
(t
rect P
c
T
)
4
(t
rect P
c
T
)
4
(t
rect P
c
T
)
1
m (t
rect
t
P
s
T
)
1
(t
rect
t
P
(t)
2
S
) ........(2 .......... .......... .......... .......... ).........
4
Xd(t )
4
Z(t Y(t) )
1
X(t (t)
2
S

Finally Node (3) energy consumption:

|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
+
=
+ + + =
d
T
)
4
(t
P
c
T
)
1
n (t
rect
t
P
s
T
)
2
(t
rect
t
P (t)
3
S
...(3) .......... .......... .......... .......... .......... ).........
4
(t
d
X Z(t) Y(t) )
2
X(t (t)
3
S

From Equation (1,2 and 3) we can compute the energy consumed by following equation (4):

E
s
= S
1
(t) + S
2
(t) + S
3
(t) (4)
Where (E
s
) represents the energy consumed by the proposed analysis system in Figure
(3-3). Substitute equations (1, 2 & 3) into (4) results in:
( )
(
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
(
(
|
|
.
|
\
|
|
|
.
|
\
|
(
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
(
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|

+
+

+
+

- +
- +
- +
-
+
- +

- + - =
d
T
)
4
(t
P
c
T
)
1
n (t
rect
t
P
s
T
)
2
(t
rect
t
P
d
T
)
4
(t
rect P
c
T
)
4
(t
rect P
d
T
4
l t
rect P
c
T
4
t
rect P
c
T
1
m t
rect
t
P
s
T
1
t
rect
t
P
d
T
l t
rect P
c
T
1
m
3
t
rect P
s
T
t
rect
t
P t
s
E

3.2.2 System Delay analysis
The proposed scheme deals with more that one node in a duty-cycle because of the
concurrent (SEEK) packet transmission so the packet delay will only be counted as (extra
SEEK packet) and (extra CTS packet) in the middle nodes, Below is the mathematical delay
approach of the proposed scheme following the same parameters and the same assumptions
made for energy consumption:

Node 1 delay:
D
1
(t) = T
s
+ T
c
+ T
d
..........................(5)

Node 2 delay:
D
2
(t) = + T
s
+ 2 * T
c
+ T
d
........... (6)

Node 3 delay:
D
3
(t) = T
s
+ T
c
+ T
d
.... (7)

From (5, 6 and 7) above a system delay equation can be derived:

D
s
(t) =
+
2
1
*
N
c s
T T o
(9)

N: the number of nodes in the system.

While for S-MAC (Ye et al, 2001), because each node have to go through the same operation
to send the data packet it is possible to describe S-MAC delay operation for the same system
as:

SYNC
t
: time length for SYNC packet.
RTS
t
: time length for RTS packet.
CTS
t
: time length for CTS packet.
DATA
t
: time length for DATA packet.
the Scenario above (figure 3) the formula of Energy consumption can be represented as
follow:
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
+

+ =
+ + =
d
T
) (t
rect P
c
T
)
1
m
3
(t
rect P
s
T
t
rect
t
P (t)
1
S

.....(1) .......... .......... .......... .......... .......... .......... .......... Xd(t)..... )
3
Y(t X(t) (t)
1
S

Where:
T
s
: SEEK packet time length.
T
c
: CTS packet time length.
T
d
: DATA packet time length.
: the delay in each state of transmitting SEEK packet and receiving CTS packet.
P
t
: Transmission Power.
P : Reception Power.
X(t): rectangular function of delay for SEEK packet.
Y(t): rectangular function of delay for CTS packet out from the exact node.
Z(t): rectangular function of delay for CTS packet received from the down stream node.
X
d
(t): rectangular function of delay for DATA packet.

Node (2) energy consumption is equal to the following equation:

(
(
|
|
.
|
\
|
|
|
.
|
\
|
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|

+
=
+ + + =
d
T
)
4
(t
rect P
c
T
)
4
(t
rect P
c
T
)
4
(t
rect P
c
T
)
1
m (t
rect
t
P
s
T
)
1
(t
rect
t
P
(t)
2
S
) ........(2 .......... .......... .......... .......... ).........
4
Xd(t )
4
Z(t Y(t) )
1
X(t (t)
2
S

Finally Node (3) energy consumption:

|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
+
=
+ + + =
d
T
)
4
(t
P
c
T
)
1
n (t
rect
t
P
s
T
)
2
(t
rect
t
P (t)
3
S
...(3) .......... .......... .......... .......... .......... ).........
4
(t
d
X Z(t) Y(t) )
2
X(t (t)
3
S

From Equation (1,2 and 3) we can compute the energy consumed by following equation (4):

E
s
= S
1
(t) + S
2
(t) + S
3
(t) (4)
Where (E
s
) represents the energy consumed by the proposed analysis system in Figure
(3-3). Substitute equations (1, 2 & 3) into (4) results in:
( )
(
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
(
(
|
|
.
|
\
|
|
|
.
|
\
|
(
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|
(
(
|
|
.
|
\
|
|
|
.
|
\
|
|
|
.
|
\
|

+
+

+
+

- +
- +
- +
-
+
- +

- + - =
d
T
)
4
(t
P
c
T
)
1
n (t
rect
t
P
s
T
)
2
(t
rect
t
P
d
T
)
4
(t
rect P
c
T
)
4
(t
rect P
d
T
4
l t
rect P
c
T
4
t
rect P
c
T
1
m t
rect
t
P
s
T
1
t
rect
t
P
d
T
l t
rect P
c
T
1
m
3
t
rect P
s
T
t
rect
t
P t
s
E

3.2.2 System Delay analysis
The proposed scheme deals with more that one node in a duty-cycle because of the
concurrent (SEEK) packet transmission so the packet delay will only be counted as (extra
SEEK packet) and (extra CTS packet) in the middle nodes, Below is the mathematical delay
approach of the proposed scheme following the same parameters and the same assumptions
made for energy consumption:

Node 1 delay:
D
1
(t) = T
s
+ T
c
+ T
d
..........................(5)

Node 2 delay:
D
2
(t) = + T
s
+ 2 * T
c
+ T
d
........... (6)

Node 3 delay:
D
3
(t) = T
s
+ T
c
+ T
d
.... (7)

From (5, 6 and 7) above a system delay equation can be derived:

D
s
(t) =
+
2
1
*
N
c s
T T o
(9)

N: the number of nodes in the system.

While for S-MAC (Ye et al, 2001), because each node have to go through the same operation
to send the data packet it is possible to describe S-MAC delay operation for the same system
as:

SYNC
t
: time length for SYNC packet.
RTS
t
: time length for RTS packet.
CTS
t
: time length for CTS packet.
DATA
t
: time length for DATA packet.
Node 1 delay (S-MAC):

D
1
(t) = SYNC
t
+ RTS
t
+ CTS
t
+ DATA
t
. ........... (10)


D
2
(t) = D
1
(t) + SYNC
t
+ RTS
t
+ CTS
t
+ DATA
t
. ........ (11)


D
3
(t) = D
2
(t) + SYNC
t
+ RTS
t
+ CTS
t
+ DATA
t
. ............ (12)

From (10, 11 and 12) we can reach to a system delay equation using S-MAC:
D
S-MAC
(t) =

N
1
t
DATA
t
CTS
t
RTS
t
SYNC 1)(t) D(N
........(13)

4. Implementation of the proposed solution
This section will discuss the implementaiton of the methodology and the simulation
parameters used. Two simulation scenarios are devised and simulation parameters with a
range of duty-cycles from (5% - 25%) for the first scenario and from (5%-40%) for the second
scenario in three steps to cover most of operation environment that can a WSN suffer. The
first simulation scenario (Figure 3) is represented by a straight line of nodes deployment .:

Fig. 3.a: A straight node deplyment used as the first simultion scene.

The simulation environment was built and made using NS2 version 2.33, the scenario
consists of five nodes in one row Starts from node 0 to node 4 considering node 4 as the
destination node in the simulation. Our main revals during where S-MAC protocol (as it is
considered the base protocol to propose the Sleep-Listen Theory) and SEA-MAC because it
is an inprovment over S-MAC in terms of the control packets handling. The proposed
approach will be referred as Proposed Protocol (PP-) before or after any protocol name.

below is a table of the parameters that where allocated for the scenarion above:

Parameter Amplitude
Simulation time 700 second
Duty-Cycle 5%, 10%, 25%
Routing Protocol None
Node Idle power 100 mW
Node Rx Power 100 mW
Node Tx Power 100 mW
Node Sleep Power 1 mW
Transition Power 20 mW
Transition time 5 ms
Energy model NS2 Energy model
Propogation model TwoRayGround
Initial Energy for each node 1000 mJ
Table 3. the simulation parameters for the first scene

The second simulation (Figure 4) scenario consist of ten nodes. nine nodes 0-8 formed a
square deployment and one node 9 was separated from the other as a base node. The
simulations are conducted on a wide range of duty cycles from 5% - 40% in three steps (5, 25
and 40).

Fig. 4. square shape node deployment to simulate the second scene.


D
1
(t) = SYNC
t
+ RTS
t
+ CTS
t
+ DATA
t
. ........... (10)


D
2
(t) = D
1
(t) + SYNC
t
+ RTS
t
+ CTS
t
+ DATA
t
. ........ (11)


D
3
(t) = D
2
(t) + SYNC
t
+ RTS
t
+ CTS
t
+ DATA
t
. ............ (12)

From (10, 11 and 12) we can reach to a system delay equation using S-MAC:
D
S-MAC
(t) =

N
1
t
DATA
t
CTS
t
RTS
t
SYNC 1)(t) D(N
........(13)

4. Implementation of the proposed solution
This section will discuss the implementaiton of the methodology and the simulation
parameters used. Two simulation scenarios are devised and simulation parameters with a
range of duty-cycles from (5% - 25%) for the first scenario and from (5%-40%) for the second
scenario in three steps to cover most of operation environment that can a WSN suffer. The
first simulation scenario (Figure 3) is represented by a straight line of nodes deployment .:

Fig. 3.a: A straight node deplyment used as the first simultion scene.

The simulation environment was built and made using NS2 version 2.33, the scenario
consists of five nodes in one row Starts from node 0 to node 4 considering node 4 as the
destination node in the simulation. Our main revals during where S-MAC protocol (as it is
considered the base protocol to propose the Sleep-Listen Theory) and SEA-MAC because it
is an inprovment over S-MAC in terms of the control packets handling. The proposed
approach will be referred as Proposed Protocol (PP-) before or after any protocol name.

below is a table of the parameters that where allocated for the scenarion above:

Parameter Amplitude
Simulation time 700 second
Duty-Cycle 5%, 10%, 25%
Routing Protocol None
Initial Energy for each node 1000 mJ
Table 3. the simulation parameters for the first scene

The second simulation (Figure 4) scenario consist of ten nodes. nine nodes 0-8 formed a
square deployment and one node 9 was separated from the other as a base node. The
simulations are conducted on a wide range of duty cycles from 5% - 40% in three steps (5, 25
and 40).

Fig. 4. square shape node deployment to simulate the second scene.

The Simulation parameters where allocated to this sceneario as in table 4 below:

Parameter Amplitude
Simulation time 7000 seconds
Duty-Cycle 5%, 25%, 40%
Routing Protocol DSR
Initial Energy for each node 100000mJ
Table 4. the simulation parameters for the second scene.

The next section will discuss the results of the simulation and the effect of the proposed
scheme on the operation of the MAC protocols and the Network itself.

4.1 First Scenario Results
To show the effect of the solution we propose we have to issue a comparison between our
rivals S-MAC and SEA-MAC. Our criteria of comparison are Energy consumption, the
number of collisions and the delay of packet transmission. Figure 5 shows the protocols
operation in a 5% duty-cycle:

Fig. 5. S-MAC vs. SEA-MAC at 5% Duty-Cycle.

840
850
860
870
880
890
900
910
920
930
100 200 300 400 500 600 700
Time (s)
S-MAC
SEA-MAC
940
R
e
s
i
d
u
a
l

E
n
e
r
g
y

(
m
J
)

S-MAC operation is more prone to lose energy than SEA-MAC as it uses much more for
(SYNC) packet than in SEA-MAC (TONE) packet. Figure 6-A shows the collisions occurance
for both protocols. Figure 6-B show the delay effecincy of each protocol:

0
0.5
1
1.5
2
2.5
0 100 200 300 400 500 600 700
Time(s)
N
o
.

o
f

C
o
l
l
i
s
i
o
n
s
S-MAC
SEA-MAC

Fig. 6-A: S-MAC vs. SEA-Mac in terms of collision occurance at 5% Duty-Cycle

0
0.5
1
1.5
2
2.5
3
3.5
4
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC
SEA-MAC

Fig. 6-B: S-MAC vs. SEA-MAC in terms of Message Delay at 5% Duty-Cycle

Overall, S-MAC offers better delay than SEA-MAC, if we increase the operation Duty-Cycle
to 25%, we will observe that S-MAC a better performance in terms of Delay and collisions
(Figure 7-A & B). But SEA-MAC offers better energy consumption over S-MAC because it
uses a shorter activating packet called (TONE). Figure 8 shows the energy consumption
effect.

The Simulation parameters where allocated to this sceneario as in table 4 below:

Parameter Amplitude
Simulation time 7000 seconds
Duty-Cycle 5%, 25%, 40%
Routing Protocol DSR
Initial Energy for each node 100000mJ
Table 4. the simulation parameters for the second scene.

The next section will discuss the results of the simulation and the effect of the proposed
scheme on the operation of the MAC protocols and the Network itself.

4.1 First Scenario Results
To show the effect of the solution we propose we have to issue a comparison between our
rivals S-MAC and SEA-MAC. Our criteria of comparison are Energy consumption, the
number of collisions and the delay of packet transmission. Figure 5 shows the protocols
operation in a 5% duty-cycle:

Fig. 5. S-MAC vs. SEA-MAC at 5% Duty-Cycle.

840
850
860
870
880
890
900
910
920
930
100 200 300 400 500 600 700
Time (s)
S-MAC
SEA-MAC
940
R
e
s
i
d
u
a
l

E
n
e
r
g
y

(
m
J
)

S-MAC operation is more prone to lose energy than SEA-MAC as it uses much more for
(SYNC) packet than in SEA-MAC (TONE) packet. Figure 6-A shows the collisions occurance
for both protocols. Figure 6-B show the delay effecincy of each protocol:

0
0.5
1
1.5
2
2.5
0 100 200 300 400 500 600 700
Time(s)
N
o
.

o
f

C
o
l
l
i
s
i
o
n
s
S-MAC
SEA-MAC

Fig. 6-A: S-MAC vs. SEA-Mac in terms of collision occurance at 5% Duty-Cycle

0
0.5
1
1.5
2
2.5
3
3.5
4
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC
SEA-MAC

Fig. 6-B: S-MAC vs. SEA-MAC in terms of Message Delay at 5% Duty-Cycle

Overall, S-MAC offers better delay than SEA-MAC, if we increase the operation Duty-Cycle
to 25%, we will observe that S-MAC a better performance in terms of Delay and collisions
(Figure 7-A & B). But SEA-MAC offers better energy consumption over S-MAC because it
uses a shorter activating packet called (TONE). Figure 8 shows the energy consumption
effect.

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC
SEA-MAC

Fig. 7-A: S-MAC vs. SEA-MAC Message Delay at 25% operation Duty-Cycle.

0
0.2
0.4
0.6
0.8
1
1.2
0 100 200 300 400 500 600 700
Time(s)
N
o
.

o
f

C
o
l
l
i
s
i
o
n
s
S-MAC
SEA-MAC

Fig. 7-B: S-MAC vs. SEA-MAC Collision Effect at 25% Duty-Cycle.

Fig. 8: S-MAC vs. SEA-MAC Energy consumption at 25% Duty-Cycle.

600
650
700
750
800
850
900
950
1000
100 200 300 400 500 600 700
Time (s)
S-MAC
SEA-MAC
R
e
s
i
d
u
a
l

E
n
e
r
g
y

(
m
J
)

After impl;ementing our theory on both protocols (indicated by adding (PP-) before or after
the protocols name), S-MAC started to perform better than SEA-MAC in terms of energy
consumption at low duty-cycle. While SEA-MAC kept the energy consumption better at
higher Duty-Cycles. And to mention that both protocols after improvments provided Zero-
collisions in the straight line simulation scenario. Figure 9-A,B,C&D shows the energy
consumption and delay efficency at 5% & 25% duty-cycles.

860
870
880
890
900
910
920
930
940
100 200 300 400 500 600 700
Time
R
e
s
i
d
u
a
l

E
n
e
r
g
y
PP+S-MAC
PP+SEA-MAC

Fig. 9-A: PP+S-MAC vs. PP+SEA-MAC energy consumption at 5% Duty-Cycle.

750
800
850
900
950
1000
100 200 300 400 500 600 700
Time
R
e
s
i
d
u
a
l

E
n
e
r
g
y
PP+S-MAC
PP+SEA-MAC

Fig. 9-B: PP+S-MAC vs. PP+SEA-MAC energy consumption at 25% Duty-Cycle.

(s)
(
m
J
)

(s)
(
m
J
)

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC
SEA-MAC

Fig. 7-A: S-MAC vs. SEA-MAC Message Delay at 25% operation Duty-Cycle.

0
0.2
0.4
0.6
0.8
1
1.2
0 100 200 300 400 500 600 700
Time(s)
N
o
.

o
f

C
o
l
l
i
s
i
o
n
s
S-MAC
SEA-MAC

Fig. 7-B: S-MAC vs. SEA-MAC Collision Effect at 25% Duty-Cycle.

Fig. 8: S-MAC vs. SEA-MAC Energy consumption at 25% Duty-Cycle.

600
650
700
750
800
850
900
950
1000
100 200 300 400 500 600 700
Time (s)
S-MAC
SEA-MAC
R
e
s
i
d
u
a
l

E
n
e
r
g
y

(
m
J
)

After impl;ementing our theory on both protocols (indicated by adding (PP-) before or after
the protocols name), S-MAC started to perform better than SEA-MAC in terms of energy
consumption at low duty-cycle. While SEA-MAC kept the energy consumption better at
higher Duty-Cycles. And to mention that both protocols after improvments provided Zero-
collisions in the straight line simulation scenario. Figure 9-A,B,C&D shows the energy
consumption and delay efficency at 5% & 25% duty-cycles.

860
870
880
890
900
910
920
930
940
100 200 300 400 500 600 700
Time
R
e
s
i
d
u
a
l

E
n
e
r
g
y
PP+S-MAC
PP+SEA-MAC


750
800
850
900
950
1000
100 200 300 400 500 600 700
Time
R
e
s
i
d
u
a
l

E
n
e
r
g
y
PP+S-MAC
PP+SEA-MAC

Fig. 9-B: PP+S-MAC vs. PP+SEA-MAC energy consumption at 25% Duty-Cycle.

(s)
(
m
J
)

(s)
(
m
J
)

0
0.5
1
1.5
2
2.5
3
3.5
4
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC-PP
SEA-MAC-PP

Fig. 9-C: PP+S-MAC vs. PP+SEA-MAC Delay effeciny at 5% Duty-Cycle.

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC-PP
SEA-MAC-PP

Fig. 9-D: PP+S-MAC vs. PP+SEA-MAC Delay comparison at 25% Duty-Cycle.

To summerize the result show above, The proposed scheme gave the effect on S-MAC and
made the consumption in terms of energy at low Duty-Cycle operation better than the
original scheme of S-MAC.
The proposed approach provided better operation in terms of energy consumption at high
Duty-Cycle operation than the original SEA-MAC scheme.
Both protocols provided better throughput for most of the scenarios after adding the
proposed scheme to the original scheme of the protocols.

4.2 The proposed Scheme effect for the second scenario
The second scenario has a new factor that gave an effect on the operation of both protocols
S-MAC and SEA-MAC (with or without the implementation of the proposed theory). This is
represented by the number of the deployed nodes. Increasing the number of the nodes can
give a positive effect on the network operation as it will help to conduct the inquiry
collection of the phenomena in a more fast paced operation. Figure 10-A,B &C shows the
energy consumption, Delay and collisions occurrences. This effect is observed in Figure 10-
A, where we can see the gap of consumption between SEA-MAC and S-MAC.

93000
94000
95000
96000
97000
98000
99000
100000
101000
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
E
n
e
r
g
y
(
m
J
)
S-MAC
SEA-MAC

Fig. 10-A: S-MAC vs. SEA-MAC energy consumption at 5% Duty-Cycle.

0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
(
s
)
S-MAC
SEA-MAC

Fig. 10-B: S-MAC vs. SEA-MAC Delay average at 5% Duty-Cycle.

0
0.5
1
1.5
2
2.5
3
3.5
4
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC-PP
SEA-MAC-PP

Fig. 9-C: PP+S-MAC vs. PP+SEA-MAC Delay effeciny at 5% Duty-Cycle.

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 100 200 300 400 500 600 700
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
S-MAC-PP
SEA-MAC-PP

Fig. 9-D: PP+S-MAC vs. PP+SEA-MAC Delay comparison at 25% Duty-Cycle.

To summerize the result show above, The proposed scheme gave the effect on S-MAC and
made the consumption in terms of energy at low Duty-Cycle operation better than the
original scheme of S-MAC.
The proposed approach provided better operation in terms of energy consumption at high
Duty-Cycle operation than the original SEA-MAC scheme.
Both protocols provided better throughput for most of the scenarios after adding the
proposed scheme to the original scheme of the protocols.

4.2 The proposed Scheme effect for the second scenario
The second scenario has a new factor that gave an effect on the operation of both protocols
S-MAC and SEA-MAC (with or without the implementation of the proposed theory). This is
represented by the number of the deployed nodes. Increasing the number of the nodes can
give a positive effect on the network operation as it will help to conduct the inquiry
collection of the phenomena in a more fast paced operation. Figure 10-A,B &C shows the
energy consumption, Delay and collisions occurrences. This effect is observed in Figure 10-
A, where we can see the gap of consumption between SEA-MAC and S-MAC.

93000
94000
95000
96000
97000
98000
99000
100000
101000
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
E
n
e
r
g
y
(
m
J
)
S-MAC
SEA-MAC

Fig. 10-A: S-MAC vs. SEA-MAC energy consumption at 5% Duty-Cycle.

0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
(
s
)
S-MAC
SEA-MAC

Fig. 10-B: S-MAC vs. SEA-MAC Delay average at 5% Duty-Cycle.

0
20
40
60
80
100
120
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
N
o
.

o
f

c
o
l
l
i
s
i
o
n
s
S-MAC
SEA-MAC

Fig. 10-C: S-MAC vs. SEA-MAC collisions ocurrances at 5% Duty-Cycle.

Adding the proposed approach to both protocols resulted in a different operation than the
original ones. Figure 11-A,B&C shows that, it is observed that S-MAC was improved over
SEA-MAC operation at low Duty-Cycle. This is due to the fact that S-MAC goes through
four stages of operation (SYNC+RTS+CST+ACK) while SEA-MAC has
(TONE+SYNC+RTS+CTS+ACK) which leads to a longer operation even with the
compression of two control packets (SYNC&RTS), SEA-MAC has longer operation time than
S-MAC at shorter duty-cycle.

98800
99000
99200
99400
99600
99800
100000
100200
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
E
n
e
r
g
y
(
m
J
)
S-MAC-PP
SEA-MAC-PP


0
0.5
1
1.5
2
2.5
3
3.5
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
(
s
)
S-MAC-PP
SEA-MAC-PP

Fig. 11-B: S-MAC-PP vs. SEA-MAC-PP average Delay at 5% Duty-Cycle.

0
10
20
30
40
50
60
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
N
o
.

o
f

c
o
l
l
i
s
i
o
n
s
S-MAC-PP
SEA-MAC-PP

Fig. 11-C: S-MAC-PP vs. SEA-MAC-PP collisions at 5% Duty-Cycles.

4.3 Pros and Cons of the proposed theory
Overall, the proposed approached satisfied the quest as it does improve the operation of
both protocols at different ranges of duty-cycle (we must note SEA-MAC with the proposed
approach offered better energy consumption and delay operation at higher duty-cycles than
S-MAC also implemented with the approach). Increasing the number of nodes result in
collision occurance rather than the situation with the straight line deployment. Overall
message delay is in favor of S-MAC at shorter duty-cycles and the advantage is to SEA-
MAC at longer duty-cycle.
In the next section we will discuss breifly the mobility issues in WSN as it is considered an
important part of this research area.

5. Mobility in WSN
Wireless sensor networks (WSN) offers a wide range of applications and it is also an intense
area of research. However, current research in wireless sensor networks focuses on
0
20
40
60
80
100
120
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
N
o
.

o
f

c
o
l
l
i
s
i
o
n
s
S-MAC
SEA-MAC

Fig. 10-C: S-MAC vs. SEA-MAC collisions ocurrances at 5% Duty-Cycle.

Adding the proposed approach to both protocols resulted in a different operation than the
original ones. Figure 11-A,B&C shows that, it is observed that S-MAC was improved over
SEA-MAC operation at low Duty-Cycle. This is due to the fact that S-MAC goes through
four stages of operation (SYNC+RTS+CST+ACK) while SEA-MAC has
(TONE+SYNC+RTS+CTS+ACK) which leads to a longer operation even with the
compression of two control packets (SYNC&RTS), SEA-MAC has longer operation time than
S-MAC at shorter duty-cycle.

98800
99000
99200
99400
99600
99800
100000
100200
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
E
n
e
r
g
y
(
m
J
)
S-MAC-PP
SEA-MAC-PP


0
0.5
1
1.5
2
2.5
3
3.5
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
A
v
g
.

M
e
s
s
a
g
e

D
e
l
a
y
(
s
)
S-MAC-PP
SEA-MAC-PP

Fig. 11-B: S-MAC-PP vs. SEA-MAC-PP average Delay at 5% Duty-Cycle.

0
10
20
30
40
50
60
0 1000 2000 3000 4000 5000 6000 7000
Time(s)
N
o
.

o
f

c
o
l
l
i
s
i
o
n
s
S-MAC-PP
SEA-MAC-PP

Fig. 11-C: S-MAC-PP vs. SEA-MAC-PP collisions at 5% Duty-Cycles.

4.3 Pros and Cons of the proposed theory
Overall, the proposed approached satisfied the quest as it does improve the operation of
both protocols at different ranges of duty-cycle (we must note SEA-MAC with the proposed
approach offered better energy consumption and delay operation at higher duty-cycles than
S-MAC also implemented with the approach). Increasing the number of nodes result in
collision occurance rather than the situation with the straight line deployment. Overall
message delay is in favor of S-MAC at shorter duty-cycles and the advantage is to SEA-
MAC at longer duty-cycle.
In the next section we will discuss breifly the mobility issues in WSN as it is considered an
important part of this research area.

5. Mobility in WSN
Wireless sensor networks (WSN) offers a wide range of applications and it is also an intense
area of research. However, current research in wireless sensor networks focuses on
stationary WSN where they are deployed in a stationary position providing the base station
with information about the subject under observation. However, a mobile sensor network is
a collection of WSN nodes. Each of these nodes is capable of sensing, communication and
moving around. It is the mobility capabilities that distinguish a mobile sensor network from
the conventional fixed WSN (Motari'c et al, 2002).
Mobile sensor networks offer many opportunities for research as these sensors involves: the
estimate location of the node in a movement scenario, an efficient DATA and information
processing schemes that can cope with the mobility measurements and requirements (this
includes the routing theory and the potential MAC Protocol Used).
Most of the discussed approaches interms of routing theory, MAC and also allocation the
location of the sensors are ment for stationary sensor nodes. Mobile sensor networks
requiers extra care when it comes to design and implementing a network related protocols
the conserns includes ad not exclusive to: energy consumption, message delay, location
estimation accuracy and scurity of information traveled between the nodes to the base
station.
To list some of the aspects that effects on designing an operapable Mobile sensor networks,
the next sections will give a brief explination about routing theory, MAC approaches and
Localaization scheme aimed for mobility applications.

5.1 Routing theory
Routing protocols are protocols aimed to offer transmitting the DATA through the network
by utilizing the best available routes (not always the shortest ones) to the destination. When
it comes to design routing protocols for mobile Sensor nodes, extra care should be taken in
terms of timing the transportation between the nodes. Most of the routing protocol that are
used and implemented for Wireless sensor networks (e.g. Ad hoc on demand Distance
Vector (AODV) and Dynamic Source Routing (DSR)) are originally designed and optimized
for ad hoc networks which utilizes devices like (Laptop computers and mobile phones)
which has much powerful energy sources than the ones available in sensor nodes. And to
the power issue mobility make the task even tougher.

5.2 MAC approaches
Even the approach discussed in this chapter does not satisfy the mobility issues in MAC
protocols aimed for mobile sensor networks. The results from the current work suggest that
the CSMA based MAC protocols has a better chance in overcoming this issue than TDMA
based MAC protocols because of the time slotting issues that comes along with TDMA
based systems. IEEE802.15.4 or best known as (Zigbee) is a MAC layer standard provided
by IEEE organization aimed for low power miniatures. Still, it cannot be considered yet as a
standard MAC protocols for mobile sensor networks as it is still in the development stages
for such applications.

5.3 Localization Issues
Locating the sensor is an important task in WSN as it provides information about the
phenomena monitored and what action should be taken at the occurrence of an action.
Proposed localization schemes are aimed manly for stationary networks and partially for
mobile networks. Some of the examples of localization techniques are (Boukerche et al,
2007):

RSSI: Received Signal Strength Indicator, which is the cheapest technique to establish a
node location as the medium used is wireless medium and most of the wireless adapter are
capable of capturing such information. The disadvantage of such approach is the accuracy
of the information calculated by such approach.

GPS: Geo- Positioning System, the most used approach mobile nodes application and in
some cases considered the easiest. The disadvantage of GPS systems is that it adds extra cost
to systems in terms of financial cost and energy consumption costs and also accuracy issues.

TOA: Time On Arrival systems, the most accurate approach to achieve the location of the
nodes. However there are some cons for this technique: first of all the cost is higher than
GPS systems. Second the accuracy issue is dependent on how violent the environment being
applied on as it requires a line-of-sight connection to capture the required information. And
the last issue, because it is a mounted platform so it will consume energy like the issue with
the GPS systems.

6. Future Research goals
The future research goal is to devise a template Network Model aimed for Mobile Wireless
Sensor Networks. The template will take in consideration the concerns discussed in section
five of this chapter. It is envisage that the proposed approach provided in this chapter can
assist to devise a MAC approach that can be applied for various applications in WSN. The
proposed template is designed for Habitat monitoring applications as they share some
similarities in terms of the configurations and crucial guarantees. Future work would to
utilize a Signal to noise Ratio estimator (Kamel, Jeoti, 2007) as a metric to define which
route is the best to chose and on which nodes signal can estimate the location of the node.
Cross-layer approach a definite approach and consideration that we aim utilize in our
template.

7. Acknowledgments
We are greateful to both Dr. Brahim Belhaouari Samir, department of Electrical and
Electronic Engineering in Universiti Teknologi PETRONAS, 31750 Tronoh, Perak for helping
us with proposed scheme mathmatical analysis. and Mr. Megual A. Erazo, Computer
science department in Florida international University for helping us to develop SEA-MAC
protocol. Our thanks goes also to Universiti Teknologi PETRONAS for funding this research
and achieve the aimed results.

stationary WSN where they are deployed in a stationary position providing the base station
with information about the subject under observation. However, a mobile sensor network is
a collection of WSN nodes. Each of these nodes is capable of sensing, communication and
moving around. It is the mobility capabilities that distinguish a mobile sensor network from
the conventional fixed WSN (Motari'c et al, 2002).
Mobile sensor networks offer many opportunities for research as these sensors involves: the
estimate location of the node in a movement scenario, an efficient DATA and information
processing schemes that can cope with the mobility measurements and requirements (this
includes the routing theory and the potential MAC Protocol Used).
Most of the discussed approaches interms of routing theory, MAC and also allocation the
location of the sensors are ment for stationary sensor nodes. Mobile sensor networks
requiers extra care when it comes to design and implementing a network related protocols
the conserns includes ad not exclusive to: energy consumption, message delay, location
estimation accuracy and scurity of information traveled between the nodes to the base
station.
To list some of the aspects that effects on designing an operapable Mobile sensor networks,
the next sections will give a brief explination about routing theory, MAC approaches and
Localaization scheme aimed for mobility applications.

5.1 Routing theory
Routing protocols are protocols aimed to offer transmitting the DATA through the network
by utilizing the best available routes (not always the shortest ones) to the destination. When
it comes to design routing protocols for mobile Sensor nodes, extra care should be taken in
terms of timing the transportation between the nodes. Most of the routing protocol that are
used and implemented for Wireless sensor networks (e.g. Ad hoc on demand Distance
Vector (AODV) and Dynamic Source Routing (DSR)) are originally designed and optimized
for ad hoc networks which utilizes devices like (Laptop computers and mobile phones)
which has much powerful energy sources than the ones available in sensor nodes. And to
the power issue mobility make the task even tougher.

5.2 MAC approaches
Even the approach discussed in this chapter does not satisfy the mobility issues in MAC
protocols aimed for mobile sensor networks. The results from the current work suggest that
the CSMA based MAC protocols has a better chance in overcoming this issue than TDMA
based MAC protocols because of the time slotting issues that comes along with TDMA
based systems. IEEE802.15.4 or best known as (Zigbee) is a MAC layer standard provided
by IEEE organization aimed for low power miniatures. Still, it cannot be considered yet as a
standard MAC protocols for mobile sensor networks as it is still in the development stages
for such applications.

5.3 Localization Issues
Locating the sensor is an important task in WSN as it provides information about the
phenomena monitored and what action should be taken at the occurrence of an action.
Proposed localization schemes are aimed manly for stationary networks and partially for
mobile networks. Some of the examples of localization techniques are (Boukerche et al,
2007):

RSSI: Received Signal Strength Indicator, which is the cheapest technique to establish a
node location as the medium used is wireless medium and most of the wireless adapter are
capable of capturing such information. The disadvantage of such approach is the accuracy
of the information calculated by such approach.

GPS: Geo- Positioning System, the most used approach mobile nodes application and in
some cases considered the easiest. The disadvantage of GPS systems is that it adds extra cost
to systems in terms of financial cost and energy consumption costs and also accuracy issues.

TOA: Time On Arrival systems, the most accurate approach to achieve the location of the
nodes. However there are some cons for this technique: first of all the cost is higher than
GPS systems. Second the accuracy issue is dependent on how violent the environment being
applied on as it requires a line-of-sight connection to capture the required information. And
the last issue, because it is a mounted platform so it will consume energy like the issue with
the GPS systems.

6. Future Research goals
The future research goal is to devise a template Network Model aimed for Mobile Wireless
Sensor Networks. The template will take in consideration the concerns discussed in section
five of this chapter. It is envisage that the proposed approach provided in this chapter can
assist to devise a MAC approach that can be applied for various applications in WSN. The
proposed template is designed for Habitat monitoring applications as they share some
similarities in terms of the configurations and crucial guarantees. Future work would to
utilize a Signal to noise Ratio estimator (Kamel, Jeoti, 2007) as a metric to define which
route is the best to chose and on which nodes signal can estimate the location of the node.
Cross-layer approach a definite approach and consideration that we aim utilize in our
template.

7. Acknowledgments
We are greateful to both Dr. Brahim Belhaouari Samir, department of Electrical and
Electronic Engineering in Universiti Teknologi PETRONAS, 31750 Tronoh, Perak for helping
us with proposed scheme mathmatical analysis. and Mr. Megual A. Erazo, Computer
science department in Florida international University for helping us to develop SEA-MAC
protocol. Our thanks goes also to Universiti Teknologi PETRONAS for funding this research
and achieve the aimed results.

8. References
Kazem Sohraby, Daniel Minoli and Taieb Znati WIRELESS SENSOR NETWORKS
Technology, Protocols, and Applications, 2007 by John Wiley & Sons, Inc.
Yang Yu, Viktor K Prasanna and Bhaskar Krishnamachari Information processing and
routing in wireless sensor networks, 2006 by World Scientific Publishing Co. Pte.
Ltd.
Bhaskar Krishnamachari Networking Wireless Sensors, Cambridge University Press 2005.
Alan Mainwaring, Joseph Polastre, Robert Szewczyk, David Culler and John Anderson
Wireless Sensor Networks for Habitat Monitoring, WSNA02, September 28, 2002,
Atlanta, Georgia, USA, ACM.
Vijay Raghunathan, Curt Schurgers, Sung Park, and Mani B. Srivastava Energy Aware
Wireless Sensor Networks, IEEE Signal Processing Magazine, 2002.
Azzedine Boukerche, Fernando H. S. Silva, Regina B. Araujo and Richard W. N. Pazzi A
Low Latency and Energy Aware Event Ordering Algorithm for Wireless Actor and
Sensor Networks, MSWiM05, October 1013, 2005, Montreal, Quebec, Canada,
ACM.
Rebecca Braynard, Adam Silberstein and Carla Ellis Extending Network Lifetime Using an
Automatically Tuned Energy-Aware MAC Protocol, Proceedings of the 2006
European Workshop on Wireless Sensor Networks, Zurich, Switzerland (2006).
Lodewijk van Hoesel and Paul J.M. Havinga MAC Protocol for WSNs, SenSys'04,
November 3-5, 2004, Baltimore, Maryland, USA, ACM.
Yee Wei Law, Lodewijk van Hoesel, Jeroen Doumen, Pieter Hartel and Paul Havinga
Energy Efficient Link Layer Jamming Attacks against Wireless Sensor Network
MAC Protocols, SASN05, November 7, 2005, Alexandria, Virginia, USA, ACM.
Ioannis Mathioudakis, Neil M.White, Nick R. Harris, Geoff V. Merrett, Wireless Sensor
Networks: A Case Study for Energy Efficient Environmental Monitoring,
Eurosensors Conference 2008, 7-11 September 2008, Dresden, Germany.
Marwan Ihsan Shukur, Lee Sheng Chyan and Vooi Voon Yap Wireless Sensor Networks:
Delay Guarentee and Energy Efficient MAC Protocols, Proceedings of World
Academy of Science, Engineering and Technology, WCSET 2009, 25-27 Feb. 2009,
Penang, Malaysia.
Wei Ye, John Heidemann and Deborah Estrin An Energy-Efficient MAC protocol for
Wireless Sensor Networks, USC/ISI Technical Report ISI-TR-543, September 2001.
Tijs Van Dam and Keon Langendoen An Adaptive Energy-Efficeint MAC Protocol for
Wireless Sensor Networks, SenSys03, November 5-7, 2003, ACM.
Shu Du, Amit Kumar Saha and David B. Johnson, RMAC: A Routing-Enhanced Duty-
Cycle MAC Protocol for Wireless Sensor Networks, INFOCOM 2007. 26th IEEE
International Conference on Computer Communications. IEEE.
Jin Kyung PARK, Woo Cheol Shin and Jun HA Energy-Aware Pure ALOHA for Wireless
Sensor Networks, IEIC Trans. Fundamentals, VOL.E89-A, No.6 June 2006.
Changsu Suh and Young-Bae Ko, A Traffic Aware, Energy Efficient MAC protocol for
Wireless Sensor Networks, Proceeding of the IEEE international symposium on
circuits and systems (IS CAS05), May. 2005.
Sangheon Pack, Jaeyoung Choi, Taekyoung Kwon and Yanghee Choi, TA-MAC: Task
Aware MAC Protocol for Wireless Sensor Networks, Vehicular Technology
Conference, 2006. VTC 2006-Spring. IEEE 63rd.
Miguel A. Erazo, Yi Qian, SEA-MAC: A Simple Energy Aware MAC Protocol for Wireless
Sensor Networks for Environmental Monitoring Applications, Wireless Pervasive
Computing, 2007. ISWPC '07. IEEE 2
nd
international symposium 2007.
Rajgopal Kannan, Ram Kalidini and S. S. Iyengar Energy and rate based MAC protocol for
Wireless Sensor Networks SIGMOD Record, Vol.32, No.4, December 2003.
Anirudha Sahoo and Prashant Baronia An Energy Efficient MAC in Wireless Sensor
Networks to Provide Delay Guarantee, Local & Metropolitan Area Networks,
2007. LANMAN 2007. 15th IEEE Workshop on.
Saurabh Ganeriwal, Ram Kumar and Mani B. Srivastava Timing-sync Protocol for Sensor
Networks, SenSys 03, November 5-7, 2003, Los Angeles, California, USA, ACM.
Esteban Egea-Lopez, Javier Vales-Alonso, Alejandro S. Martnez-Sala, Joan Garca-Haro,
Pablo Pavon-Marino, and M. Victoria Bueno-Delgado A Real-Time MAC
Protocol for Wireless Sensor Networks: Virtual TDMA for Sensors (VTS), ARCS
2006, LNCS 3894, pp. 382396, 2006, Springer-Verlag Berlin Heidelberg 2006.
Teerawat Issariyakul and Ekram Hossain Introduction to Network Simulator NS2,
SpringerLink publications-Springer US 2008.
Marwan Ihsan Shukur and Vooi Voon Yap An Approach for efficient energy consumption
and delay guarantee MAC Protocol for Wireless Sensor Networks, Proceedings of
International Conference on Computing and Informatics, ICOCI 2009, 24-25 June
2009, Kuala Lumpur, Malaysia, a.
Marwan Ihsan Shukur and Vooi Voon Yap Enhanced SEA-MAC: An Efficient MAC
Protocol for Wireless Sensor Networks for Environmental Monitoring
Applications, Conference on Innovative Technologies in Intelligent Systems and
Industrial Applications, IEEE CITISIA 2009, 25 July 2009, MONASH University
Sunway Campus, Malaysia, b.
Howard, A, Mataric, M.J., and Sukhatme, G.S., Mobile Sensor Network Deployment using
Potential Fields: A Distributed, Scalable Solution to the Area Coverage Problem,
Proceedings of the 6th International Symposium on Distributed Autonomous Robotics
Systems (DARS02) Fukuoka, Japan, June 25-27, 2002.
Azzedine Boukerche, Horacio A. B. F. Oliveira and Eduardo F. Nakamura Localization
Systems For Wireless Sensor Networks, IEEE Wireless Communications Mag. Dec.
2007.
Nidal S. Kamel and Varun Jeoti A Linear Prediction Based Estimation of Signal-to-Noise
Ratio in AWGN Channel, ETRI journal, Volume 29, Number 5, October 2007.

8. References
Kazem Sohraby, Daniel Minoli and Taieb Znati WIRELESS SENSOR NETWORKS
Technology, Protocols, and Applications, 2007 by John Wiley & Sons, Inc.
Yang Yu, Viktor K Prasanna and Bhaskar Krishnamachari Information processing and
routing in wireless sensor networks, 2006 by World Scientific Publishing Co. Pte.
Ltd.
Bhaskar Krishnamachari Networking Wireless Sensors, Cambridge University Press 2005.
Alan Mainwaring, Joseph Polastre, Robert Szewczyk, David Culler and John Anderson
Wireless Sensor Networks for Habitat Monitoring, WSNA02, September 28, 2002,
Atlanta, Georgia, USA, ACM.
Vijay Raghunathan, Curt Schurgers, Sung Park, and Mani B. Srivastava Energy Aware
Wireless Sensor Networks, IEEE Signal Processing Magazine, 2002.
Azzedine Boukerche, Fernando H. S. Silva, Regina B. Araujo and Richard W. N. Pazzi A
Low Latency and Energy Aware Event Ordering Algorithm for Wireless Actor and
Sensor Networks, MSWiM05, October 1013, 2005, Montreal, Quebec, Canada,
ACM.
Rebecca Braynard, Adam Silberstein and Carla Ellis Extending Network Lifetime Using an
Automatically Tuned Energy-Aware MAC Protocol, Proceedings of the 2006
European Workshop on Wireless Sensor Networks, Zurich, Switzerland (2006).
Lodewijk van Hoesel and Paul J.M. Havinga MAC Protocol for WSNs, SenSys'04,
November 3-5, 2004, Baltimore, Maryland, USA, ACM.
Yee Wei Law, Lodewijk van Hoesel, Jeroen Doumen, Pieter Hartel and Paul Havinga
Energy Efficient Link Layer Jamming Attacks against Wireless Sensor Network
MAC Protocols, SASN05, November 7, 2005, Alexandria, Virginia, USA, ACM.
Ioannis Mathioudakis, Neil M.White, Nick R. Harris, Geoff V. Merrett, Wireless Sensor
Networks: A Case Study for Energy Efficient Environmental Monitoring,
Eurosensors Conference 2008, 7-11 September 2008, Dresden, Germany.
Marwan Ihsan Shukur, Lee Sheng Chyan and Vooi Voon Yap Wireless Sensor Networks:
Delay Guarentee and Energy Efficient MAC Protocols, Proceedings of World
Academy of Science, Engineering and Technology, WCSET 2009, 25-27 Feb. 2009,
Penang, Malaysia.
Wei Ye, John Heidemann and Deborah Estrin An Energy-Efficient MAC protocol for
Wireless Sensor Networks, USC/ISI Technical Report ISI-TR-543, September 2001.
Tijs Van Dam and Keon Langendoen An Adaptive Energy-Efficeint MAC Protocol for
Wireless Sensor Networks, SenSys03, November 5-7, 2003, ACM.
Shu Du, Amit Kumar Saha and David B. Johnson, RMAC: A Routing-Enhanced Duty-
Cycle MAC Protocol for Wireless Sensor Networks, INFOCOM 2007. 26th IEEE
International Conference on Computer Communications. IEEE.
Jin Kyung PARK, Woo Cheol Shin and Jun HA Energy-Aware Pure ALOHA for Wireless
Sensor Networks, IEIC Trans. Fundamentals, VOL.E89-A, No.6 June 2006.
Changsu Suh and Young-Bae Ko, A Traffic Aware, Energy Efficient MAC protocol for
Wireless Sensor Networks, Proceeding of the IEEE international symposium on
circuits and systems (IS CAS05), May. 2005.
Sangheon Pack, Jaeyoung Choi, Taekyoung Kwon and Yanghee Choi, TA-MAC: Task
Aware MAC Protocol for Wireless Sensor Networks, Vehicular Technology
Conference, 2006. VTC 2006-Spring. IEEE 63rd.
Miguel A. Erazo, Yi Qian, SEA-MAC: A Simple Energy Aware MAC Protocol for Wireless
Sensor Networks for Environmental Monitoring Applications, Wireless Pervasive
Computing, 2007. ISWPC '07. IEEE 2
nd
international symposium 2007.
Rajgopal Kannan, Ram Kalidini and S. S. Iyengar Energy and rate based MAC protocol for
Wireless Sensor Networks SIGMOD Record, Vol.32, No.4, December 2003.
Anirudha Sahoo and Prashant Baronia An Energy Efficient MAC in Wireless Sensor
Networks to Provide Delay Guarantee, Local & Metropolitan Area Networks,
2007. LANMAN 2007. 15th IEEE Workshop on.
Saurabh Ganeriwal, Ram Kumar and Mani B. Srivastava Timing-sync Protocol for Sensor
Networks, SenSys 03, November 5-7, 2003, Los Angeles, California, USA, ACM.
Esteban Egea-Lopez, Javier Vales-Alonso, Alejandro S. Martnez-Sala, Joan Garca-Haro,
Pablo Pavon-Marino, and M. Victoria Bueno-Delgado A Real-Time MAC
Protocol for Wireless Sensor Networks: Virtual TDMA for Sensors (VTS), ARCS
2006, LNCS 3894, pp. 382396, 2006, Springer-Verlag Berlin Heidelberg 2006.
Teerawat Issariyakul and Ekram Hossain Introduction to Network Simulator NS2,
SpringerLink publications-Springer US 2008.
Marwan Ihsan Shukur and Vooi Voon Yap An Approach for efficient energy consumption
and delay guarantee MAC Protocol for Wireless Sensor Networks, Proceedings of
International Conference on Computing and Informatics, ICOCI 2009, 24-25 June
2009, Kuala Lumpur, Malaysia, a.
Marwan Ihsan Shukur and Vooi Voon Yap Enhanced SEA-MAC: An Efficient MAC
Protocol for Wireless Sensor Networks for Environmental Monitoring
Applications, Conference on Innovative Technologies in Intelligent Systems and
Industrial Applications, IEEE CITISIA 2009, 25 July 2009, MONASH University
Sunway Campus, Malaysia, b.
Howard, A, Mataric, M.J., and Sukhatme, G.S., Mobile Sensor Network Deployment using
Potential Fields: A Distributed, Scalable Solution to the Area Coverage Problem,
Proceedings of the 6th International Symposium on Distributed Autonomous Robotics
Systems (DARS02) Fukuoka, Japan, June 25-27, 2002.
Azzedine Boukerche, Horacio A. B. F. Oliveira and Eduardo F. Nakamura Localization
Systems For Wireless Sensor Networks, IEEE Wireless Communications Mag. Dec.
2007.
Nidal S. Kamel and Varun Jeoti A Linear Prediction Based Estimation of Signal-to-Noise
Ratio in AWGN Channel, ETRI journal, Volume 29, Number 5, October 2007.

X

Hybrid Optical and Wireless Sensor Networks

Southwest Jiaotong University
Chengdu, Sichuan, China

1. Introduction
Wireless sensor network (WSN) has attracted considerable attentions during the last few
years due to characteristics such as feasibility of rapid deployment, self-organization
(different from ad hoc networks though) and fault tolerance, as well as rapid development
of wireless communications and integrated electronics [1]. Such networks are constructed by
randomly but densely scattered tiny sensor nodes (Fig. 1). As sensor nodes are prone to
failures and the network topology changes very frequently, different protocols have been
proposed to save the overall energy dissipation in WSNs [2-5]. Among them, Low-Energy-
Adaptive-Clustering-Hierarchy (LEACH), first proposed by researchers from Massachusetts
Institute of Technology [5], is considered to be one of the most effective protocols in terms of
energy efficiency [6-7]. Another protocol, called Power-Efficient Gathering in Sensor
Information Systems (PEGASIS), is a near optimal chain-based protocol [8].

WSN
WSN Nodes
SINK
WSN
WSN Nodes
SINK

Fig. 1. Illustration of a wireless sensor network (WSN) with randomly scattered nodes (sink
node: no energy restriction; WSN nodes: with energy restriction);

On the other hand, distributed fiber sensors (DFS) have been intensively studied or even
deployed for analyzing loss, external pressure and temperature or birefringence distribution
along the fiber link, ranging from hundreds of meters to tens of kilometers [9-13].
Mechanisms include Rayleigh, Brillouin or Raman scattering or polarization effects, through
either time or frequency-domain analysis. Compared with conventional sensors including
Hybrid Optical and Wireless Sensor Networks
16

wireless ones, optical fiber sensors have intrinsic advantages such as high sensitivity, the
immunity to electromagnetic interference (EMI), superior endurance in harsh environments
and much longer lifetime.
Apparently it would be highly desirable to have integrated sensor networks that can take
advantages of both WSNs and fiber sensor networks (FSNs). Such hybrid sensor networks
can find major applications including monitoring inaccessible terrains (military, high-
voltage electricity facilities, etc.), long-term observation of earthquake activity and large area
environmental control with tunnels, and so on. So far hybrid sensor networks have been
studied as well [14-16], while optical sensors in these networks are generally point-like (e.g.
fiber-Bragg-grating based), and such nodes can be regarded as normal WSN nodes after
optical to wireless signal conversion.
In this chapter, we first review typical WSN protocols, mainly about LEACH and PEGASIS,
then evaluate the performance of LEACH protocols for different topologies, especially the
rectangle one. We propose an improved algorithm based on LEACH and PEGASIS for the
WSN, finally and most importantly, we propose an O-LEACH protocol for the hybrid
sensor network that is composed of a DFS link and two separated WSNs. Most analyses
about performance are done in terms of lifetime of the sensor networks.

2. Overview of WSN protocols
Wireless sensor networks (WSNs) generally are composed of small or tiny nodes with
sensing, computation, and communication capabilities. Various routing, power
management, and data dissemination protocols have been specifically designed for WSNs
where energy awareness is an essential design issue. Among them, routing protocols might
differ depending on the applications and network architectures. In general, routing
protocols for the wireless sensor networks can be divided into flat-based, hierarchical-based,
and location-based in terms of the underlying network structures [17-19]. As some protocols
may be discussed intensively in other chapters of this book, here we give a brief review
about major protocols.
(1) Flat-based routing protocol
Sensor nodes in flat-based routing protocols have the same role and collaborate together to
perform the sensing task and multi-hop communication. Since the flat routing is based on
flooding, it has several demerits, such as large routing overhead and high energy
dissipation. Flat-based routing protocol is used in the early stage of WSNs, such as Flooding,
Gossiping, SPIN, and Rumor.
(2) Hierarchical-based routing protocol
Hierarchical-based routing protocol is the main trend for WSNs routing protocols. In
hierarchical-based routing protocols, the network is divided into several logical groups
within a fixed area. The logical groups are called clusters. Sensor nodes collect the
information in a cluster and a head node aggregates the information. Each sensor node
delivers the sensing data to the head node in the cluster and the head node delivers the
aggregated data to the base station which is located outside of the sensor network. Contrary
to flat routing protocols, only a head node aggregates the collected information and sends it
to the base station. Due to these advantages, sensor nodes can remarkably save their own
s

energy. In general, a hierarchical routing technique is regarded as superior to flat routing
approaches. The classical Hierarchical-based routing protocols are LEACH, PEGASIS, H-
PEGASIS, TEEN, and APTEEN. We will discuss the LEACH and PEGASIS protocols in
more details later.
(3) Location-based routing protocol
Such protocol is based on the location information of senor nodes in WSNs. It assumes that
each node would know its own location and its neighbor sensor nodes location before
sensor nodes sensing and collecting the peripheral information. The distance between
neighbouring sensor nodes can be computed based on the incoming signal strength [17-18].

2.1 LEACH
Low Energy Adaptive Clustering Hierarchy (LEACH) was first introduced by Heinzelman,
et al. in [5, 20] with advantages such as energy efficiency, simplicity and load balancing
ability. LEACH is a cluster-based protocol, therefore the numbers of cluster heads and
cluster members generated by LEACH are important parameters for achieving better
performance.
In LEACH protocol, the sensor nodes in the network are divided into a number of clusters,
the nodes organize themselves into preferred local clusters, a sensor node is selected
randomly as the cluster head (CH) in each cluster and this role is rotated to evenly distribute
the energy load among nodes of the network. The CH nodes compress data arriving from
nodes that belong to the respective cluster, and send an aggregated packet to the BS in order
to further reduce the amount of information that must be transmitted to the BS, thus
reducing energy dissipation and enhancing system lifetime. After a given interval of time,
randomized rotation of the role of CH is conducted to maximize the uniformity of energy
dissipation of the network. Sensors elect themselves to be local cluster heads at any time
with a certain probability. Generally only ~ 5% of nodes need to act as CHs based on
simulation results. LEACH uses a TDMA/CDMA MAC to reduce intercluster and
intracluster collisions. As data collection is centralized and performed periodically, this
protocol is most appropriate when there is a need for constant monitoring by the sensor
network.
The operation of LEACH is broken up into rounds, where each round begins with a set-up
phase followed by a steady-state phase. In order to minimize overhead, the steady-state
phase takes longer time compared to the set-up phase. In the setup phase, the clusters are
organized and CHs are selected. In the steady state phase, the actual data transfer to the BS
takes place. During the setup phase, each node decides whether or not to become a cluster
head for the current round. A predetermined fraction of nodes, p, elect themselves as CHs.
A sensor node chooses a random number between 0 and 1. If this random number is less
than a threshold value T(n), , the node becomes a cluster head for the current round. The
threshold value is calculated based on Eq. (2-1):

1
1 *( mod )
( )
0
p
if n G
p r
T n
p
otherwise
(2-1)


wireless ones, optical fiber sensors have intrinsic advantages such as high sensitivity, the
immunity to electromagnetic interference (EMI), superior endurance in harsh environments
and much longer lifetime.
Apparently it would be highly desirable to have integrated sensor networks that can take
advantages of both WSNs and fiber sensor networks (FSNs). Such hybrid sensor networks
can find major applications including monitoring inaccessible terrains (military, high-
voltage electricity facilities, etc.), long-term observation of earthquake activity and large area
environmental control with tunnels, and so on. So far hybrid sensor networks have been
studied as well [14-16], while optical sensors in these networks are generally point-like (e.g.
fiber-Bragg-grating based), and such nodes can be regarded as normal WSN nodes after
optical to wireless signal conversion.
In this chapter, we first review typical WSN protocols, mainly about LEACH and PEGASIS,
then evaluate the performance of LEACH protocols for different topologies, especially the
rectangle one. We propose an improved algorithm based on LEACH and PEGASIS for the
WSN, finally and most importantly, we propose an O-LEACH protocol for the hybrid
sensor network that is composed of a DFS link and two separated WSNs. Most analyses
about performance are done in terms of lifetime of the sensor networks.

2. Overview of WSN protocols
Wireless sensor networks (WSNs) generally are composed of small or tiny nodes with
sensing, computation, and communication capabilities. Various routing, power
management, and data dissemination protocols have been specifically designed for WSNs
where energy awareness is an essential design issue. Among them, routing protocols might
differ depending on the applications and network architectures. In general, routing
protocols for the wireless sensor networks can be divided into flat-based, hierarchical-based,
and location-based in terms of the underlying network structures [17-19]. As some protocols
may be discussed intensively in other chapters of this book, here we give a brief review
about major protocols.
(1) Flat-based routing protocol
Sensor nodes in flat-based routing protocols have the same role and collaborate together to
perform the sensing task and multi-hop communication. Since the flat routing is based on
flooding, it has several demerits, such as large routing overhead and high energy
dissipation. Flat-based routing protocol is used in the early stage of WSNs, such as Flooding,
Gossiping, SPIN, and Rumor.
(2) Hierarchical-based routing protocol
Hierarchical-based routing protocol is the main trend for WSNs routing protocols. In
hierarchical-based routing protocols, the network is divided into several logical groups
within a fixed area. The logical groups are called clusters. Sensor nodes collect the
information in a cluster and a head node aggregates the information. Each sensor node
delivers the sensing data to the head node in the cluster and the head node delivers the
aggregated data to the base station which is located outside of the sensor network. Contrary
to flat routing protocols, only a head node aggregates the collected information and sends it
to the base station. Due to these advantages, sensor nodes can remarkably save their own
s

energy. In general, a hierarchical routing technique is regarded as superior to flat routing
approaches. The classical Hierarchical-based routing protocols are LEACH, PEGASIS, H-
PEGASIS, TEEN, and APTEEN. We will discuss the LEACH and PEGASIS protocols in
more details later.
(3) Location-based routing protocol
Such protocol is based on the location information of senor nodes in WSNs. It assumes that
each node would know its own location and its neighbor sensor nodes location before
sensor nodes sensing and collecting the peripheral information. The distance between
neighbouring sensor nodes can be computed based on the incoming signal strength [17-18].

2.1 LEACH
Low Energy Adaptive Clustering Hierarchy (LEACH) was first introduced by Heinzelman,
et al. in [5, 20] with advantages such as energy efficiency, simplicity and load balancing
ability. LEACH is a cluster-based protocol, therefore the numbers of cluster heads and
cluster members generated by LEACH are important parameters for achieving better
performance.
In LEACH protocol, the sensor nodes in the network are divided into a number of clusters,
the nodes organize themselves into preferred local clusters, a sensor node is selected
randomly as the cluster head (CH) in each cluster and this role is rotated to evenly distribute
the energy load among nodes of the network. The CH nodes compress data arriving from
nodes that belong to the respective cluster, and send an aggregated packet to the BS in order
to further reduce the amount of information that must be transmitted to the BS, thus
reducing energy dissipation and enhancing system lifetime. After a given interval of time,
randomized rotation of the role of CH is conducted to maximize the uniformity of energy
dissipation of the network. Sensors elect themselves to be local cluster heads at any time
with a certain probability. Generally only ~ 5% of nodes need to act as CHs based on
simulation results. LEACH uses a TDMA/CDMA MAC to reduce intercluster and
intracluster collisions. As data collection is centralized and performed periodically, this
protocol is most appropriate when there is a need for constant monitoring by the sensor
network.
The operation of LEACH is broken up into rounds, where each round begins with a set-up
phase followed by a steady-state phase. In order to minimize overhead, the steady-state
phase takes longer time compared to the set-up phase. In the setup phase, the clusters are
organized and CHs are selected. In the steady state phase, the actual data transfer to the BS
takes place. During the setup phase, each node decides whether or not to become a cluster
head for the current round. A predetermined fraction of nodes, p, elect themselves as CHs.
A sensor node chooses a random number between 0 and 1. If this random number is less
than a threshold value T(n), , the node becomes a cluster head for the current round. The
threshold value is calculated based on Eq. (2-1):

1
1 *( mod )
( )
0
p
if n G
p r
T n
p
otherwise
(2-1)


Where p is the desired percentage of the cluster heads (e.g. p=0.05), r is the current round,
and G is the set of nodes that have not been cluster heads in the last 1/p rounds. Using this
threshold, each node may be a cluster head sometime within 1/p rounds. All elected CHs
broadcast an advertisement message to the rest of nodes in the network that they are the
new CHs. After receiving the advertisement, all non-CH nodes decide on the cluster to
which they want to belong based on the signal strength of the advertisement. The non-CH
nodes then inform the appropriate CHs to be a member of the cluster. After receiving all the
messages from the nodes that would like to be included in the cluster and based on the
number of nodes in the cluster, the CH node creates a TDMA schedule and assigns each
node a time slot when it can transmit information. This schedule is broadcast to all the
nodes in the cluster. During the steady state phase, the sensor nodes can begin sensing and
transmitting data to the CHs. The CH node must keep its receiver on to receive all the data
from the nodes in the cluster. Each cluster communicates using different CDMA codes to
reduce interference from nodes belonging to other clusters. After receiving all the data, the
CH aggregates the data before sending it to the BS. This ends a typical round.
Advantages of LEACH include: (i) by using adaptive clusters and rotating cluster heads,
LEACH allows the energy requirements of the system to be distributed among all the
sensors; (ii) LEACH is able to perform local computation in each cluster to reduce the
amount of data that must be transmitted to the base station. On the other hand, there are
still some drawbacks about LEACH: (i) LEACH assumes that each node could communicate
with the sink and each node has computational power to support different MAC protocols,
which limits its application to networks deployed in large regions. (ii) LEACH does not
determine how to distribute the CHs uniformly through the network. Therefore, there is the
possibility that the elected CHs will be concentrated in one part of the network. (iii) LEACH
assumes that all nodes begin with the same amount of energy capacity in each election
round, assuming that being a CH consumes approximately the same amount of energy for
each node. Hence LEACH is not appropriate for non-uniform energy nodes [17, 20].

2.2 PEGASIS
In [8], an enhancement over the LEACH protocol called Power-Efficient Gathering in Sensor
Information Systems (PEGASIS) was proposed. The basic idea of the protocol is that nodes
only receive from and transmit to the closest neighbours, and they take turns being the
leader for communicating with the BS. This reduces the power required to transmit data per
round as the power draining is spread uniformly over all nodes. Hence, PEGASIS has two
main objectives: (i) to increase the lifetime of each node by using collaborative techniques; (ii)
to allow only local coordination between nodes that are close together so that the bandwidth
consumed in communication is reduced.
PEGASIS adopts a homogenous topology. In this topology, the BS lies far from sensors with
the fixed position. The data is collected and compressed before sent to the next node. Hence
the messages maintain ideally a fixed size when they are transmitted between sensors. To
locate the closest neighbour node in PEGASIS, each node uses the signal strength to
measure the distance to all neighbouring nodes and then adjusts the signal strength so that
only one node can be heard. The chain in PEGASIS consists of those nodes that are closest to
each other and form a path to the BS. The following describes the protocol briefly:
s

1) The chain starts with the furthest node from the BS to make sure that nodes father
from the BS have close neighbours. Based on the greedy algorithm, the neighbour
node joins into the chain with its distance increases gradually. When a node dies, the
chain is reconstructed in the same manner to bypass the dead node.
2) To gather data in each round, a token is generated by the BS to set the aggregating
direction after the token sent from the BS to an end node. Each node receives data
from one neighbour, fuses with its own data, and transmits to the other neighbour on
the chain.
3) Only one node transmits data to the BS in certain rounds, the leader is the node
whose number is (i mod N) where N represents the number of the nodes in round i.
PEGASIS is better than LEACH in terms of energy saving due to following facts: (i) During
the data localization, the distances that most of the nodes transmit information are much
shorter compared to that in LEACH. (ii) The amount of data for the leader to receive is much
less than LEACH. (iii) only one node transmits to the BS in each round.
Though PEGASIS has obvious advantages, it has some shortcomings. Firstly, though most
sensors are joined on a chain to form a basically homogenous structure, a sensor with too
much branches may perform many times of data receiving in a certain round thereby
resulting in unbalanced energy problem. Secondly, all the nodes must keep active before the
token arriving. This means there will be a large percentage of active nodes with nothing to
do from the beginning, meaning a waste of energy and time. Thirdly, once a sensor on the
chain was captured the whole net may be under the control by the attackers. The weak
security could be a great threat [17, 20].
LEACH that is a cluster-based protocol and PEGASIS that is a chain-based protocol are the
most classical Hierarchical-based routing protocols. They both have attracted intensive
attention, and lots of routing protocols are based on these two. Next we will investigate
some issues in details.

3. WSN Topologies
3.1 Shapes of different topologies
According to the shape of WSN monitoring area, application requirements and monitoring
of different targets, different topologies should be chosen for deploying the WSN: circular
topology is preferred for applications such as harbour, stadium etc. [21]; square topology is
suitable for irrigation in agriculture, nature reserve area etc.; rectangular topology can be
chosen for highway, railway, mine and other areas [22].
Here we study the life time of WSN in round, square, rectangular shapes of topology, and
the three topologies are shown in Figs. 3.1 (a-c). In Fig. 3.1(a), the circular area is 10,000m
2

(same as square, rectangular areas) with the radius R= 56.419m and the base station is
located at the center of the circle, i.e. (0, 0). In Fig. 3.1(b), the size of the square area is
100x100m
2
with the base station located on (0, 50) or (50, 175). In Fig.3.1(c), the size of the
rectangular area is 50*200m with the base station located on (0, 25) or (100, 150).


Where p is the desired percentage of the cluster heads (e.g. p=0.05), r is the current round,
and G is the set of nodes that have not been cluster heads in the last 1/p rounds. Using this
threshold, each node may be a cluster head sometime within 1/p rounds. All elected CHs
broadcast an advertisement message to the rest of nodes in the network that they are the
new CHs. After receiving the advertisement, all non-CH nodes decide on the cluster to
which they want to belong based on the signal strength of the advertisement. The non-CH
nodes then inform the appropriate CHs to be a member of the cluster. After receiving all the
messages from the nodes that would like to be included in the cluster and based on the
number of nodes in the cluster, the CH node creates a TDMA schedule and assigns each
node a time slot when it can transmit information. This schedule is broadcast to all the
nodes in the cluster. During the steady state phase, the sensor nodes can begin sensing and
transmitting data to the CHs. The CH node must keep its receiver on to receive all the data
from the nodes in the cluster. Each cluster communicates using different CDMA codes to
reduce interference from nodes belonging to other clusters. After receiving all the data, the
CH aggregates the data before sending it to the BS. This ends a typical round.
Advantages of LEACH include: (i) by using adaptive clusters and rotating cluster heads,
LEACH allows the energy requirements of the system to be distributed among all the
sensors; (ii) LEACH is able to perform local computation in each cluster to reduce the
amount of data that must be transmitted to the base station. On the other hand, there are
still some drawbacks about LEACH: (i) LEACH assumes that each node could communicate
with the sink and each node has computational power to support different MAC protocols,
which limits its application to networks deployed in large regions. (ii) LEACH does not
determine how to distribute the CHs uniformly through the network. Therefore, there is the
possibility that the elected CHs will be concentrated in one part of the network. (iii) LEACH
assumes that all nodes begin with the same amount of energy capacity in each election
round, assuming that being a CH consumes approximately the same amount of energy for
each node. Hence LEACH is not appropriate for non-uniform energy nodes [17, 20].

2.2 PEGASIS
In [8], an enhancement over the LEACH protocol called Power-Efficient Gathering in Sensor
Information Systems (PEGASIS) was proposed. The basic idea of the protocol is that nodes
only receive from and transmit to the closest neighbours, and they take turns being the
leader for communicating with the BS. This reduces the power required to transmit data per
round as the power draining is spread uniformly over all nodes. Hence, PEGASIS has two
main objectives: (i) to increase the lifetime of each node by using collaborative techniques; (ii)
to allow only local coordination between nodes that are close together so that the bandwidth
consumed in communication is reduced.
PEGASIS adopts a homogenous topology. In this topology, the BS lies far from sensors with
the fixed position. The data is collected and compressed before sent to the next node. Hence
the messages maintain ideally a fixed size when they are transmitted between sensors. To
locate the closest neighbour node in PEGASIS, each node uses the signal strength to
measure the distance to all neighbouring nodes and then adjusts the signal strength so that
only one node can be heard. The chain in PEGASIS consists of those nodes that are closest to
each other and form a path to the BS. The following describes the protocol briefly:
s

1) The chain starts with the furthest node from the BS to make sure that nodes father
from the BS have close neighbours. Based on the greedy algorithm, the neighbour
node joins into the chain with its distance increases gradually. When a node dies, the
chain is reconstructed in the same manner to bypass the dead node.
2) To gather data in each round, a token is generated by the BS to set the aggregating
direction after the token sent from the BS to an end node. Each node receives data
from one neighbour, fuses with its own data, and transmits to the other neighbour on
the chain.
3) Only one node transmits data to the BS in certain rounds, the leader is the node
whose number is (i mod N) where N represents the number of the nodes in round i.
PEGASIS is better than LEACH in terms of energy saving due to following facts: (i) During
the data localization, the distances that most of the nodes transmit information are much
shorter compared to that in LEACH. (ii) The amount of data for the leader to receive is much
less than LEACH. (iii) only one node transmits to the BS in each round.
Though PEGASIS has obvious advantages, it has some shortcomings. Firstly, though most
sensors are joined on a chain to form a basically homogenous structure, a sensor with too
much branches may perform many times of data receiving in a certain round thereby
resulting in unbalanced energy problem. Secondly, all the nodes must keep active before the
token arriving. This means there will be a large percentage of active nodes with nothing to
do from the beginning, meaning a waste of energy and time. Thirdly, once a sensor on the
chain was captured the whole net may be under the control by the attackers. The weak
security could be a great threat [17, 20].
LEACH that is a cluster-based protocol and PEGASIS that is a chain-based protocol are the
most classical Hierarchical-based routing protocols. They both have attracted intensive
attention, and lots of routing protocols are based on these two. Next we will investigate
some issues in details.

3. WSN Topologies
3.1 Shapes of different topologies
According to the shape of WSN monitoring area, application requirements and monitoring
of different targets, different topologies should be chosen for deploying the WSN: circular
topology is preferred for applications such as harbour, stadium etc. [21]; square topology is
suitable for irrigation in agriculture, nature reserve area etc.; rectangular topology can be
chosen for highway, railway, mine and other areas [22].
Here we study the life time of WSN in round, square, rectangular shapes of topology, and
the three topologies are shown in Figs. 3.1 (a-c). In Fig. 3.1(a), the circular area is 10,000m
2

(same as square, rectangular areas) with the radius R= 56.419m and the base station is
located at the center of the circle, i.e. (0, 0). In Fig. 3.1(b), the size of the square area is
100x100m
2
with the base station located on (0, 50) or (50, 175). In Fig.3.1(c), the size of the
rectangular area is 50*200m with the base station located on (0, 25) or (100, 150).


-60 -30 0 30 60
-60
-30
0
30
60
m
m

0 25 50 75 100
0
25
50
75
100
m
m

(a) Circular topology (b) Square topology

0 50 100 150 200
0
25
50
m
m
Zone 1 Zone 2
Zone 3
Zone 4

(c) Rectangle topology
Fig 3.1 Three different types of WSNs topology:
Sink (a):(0,0); (b):(0,50) or (50,175); (c):(0,25) or (100,150)

The probability of cluster head node in the LEACH protocol has a certain impact on the
WSNs lifetime. In our analysis, we divides the rectangle area into four smaller square areas,
with the communication distance of nodes keeping short (d
B
<d
0
, where d
B
is the
broadcasting distance of the cluster head). From the first-order radio model, we can get the
optimal cluster head probability formula as follows:

2
opt
toBS
N M
k
d

(3-1)

Where N is the number of sensor nodes. In the rectangular region 100 nodes are scattered
randomly, and the region is divided into four regions. In order to verify the difference of the
number of nodes distributed in different regions, we simulate 100 independent iterations of
the nodes number in each region, and get the averages in the four regions as zone1(25.32)
zone2(24.83)zone3(24.87) and zone4(24.98). It can be seen that the number of nodes in
each region are around 25, therere almost no difference in the average nodes numbers for
the four regions, so in the text, the nodes are uniform distributed in the four regions, i.e.
N=25. M is the side length of each small square region, here M=50. d
toBS
is the distance
between the sink and the node. As the distance of a node to the base station is different, we
can change the percentage of cluster heads in different regions to reach the optimal value so
that the lifetime of the whole WSN can be prolonged.
For the topology with a rectangular shape, we propose an improved LEACH algorithm. The
main idea is described as follows:
s

(1) Divide the rectangular area into several small square areas with the same size;
(2) Elect the cluster heads separately in each region, and the optimal probabilities of the
cluster heads for each region can be obtained from Eq. (3-1), i.e. the values are
p
1
=0.02p
2
=0.03, p
3
=0.03, p
4
=0.02
(3) After the cluster heads in each region are selected, the rest of the protocol is similar
to LEACH.
The improved algorithm elects cluster heads in each region according to its probability of
the cluster heads. In this way, it can make sure that there are clusters in every region and
ensure the clusters distributed more uniformly in every region, which reduces the energy
dissipation and improves the lifetime of the network.

3.2 Simulation results
We simulate the three shapes of topology that use (improved) LEACH as the routing
protocol. Parameters used in simulation are listed in table 3.1. There are 100 sensor nodes
randomly scattered with fixed position in each shape. We measure the round number when
the first node died, 20% of nodes died and 50% of nodes died respectively as the criterion to
estimate the lifetime of WSN.

Parameter Value
Number of nodes 100
Initial energy (J) 0.5
Data packet length (bit) 4000
Control packet length(bit) 200
Energy dissipation of one Tx (nJ) 50
Energy dissipation of one Rx (nJ) 50
Energy aggregation energy (nJ) 5
Energy loss-free space (pJ/bit/m
2
) 10
Energy loss-multipass fading (pJ/bit/m
4
) 0.0013
Table 3.1 Parameters used in simulation

Table 3.2 shows the round numbers (the lifetime) of WSN for different BS locations and
different percentage of dead nodes in circle, square and rectangular shapes of topology. As
the sensor nodes distributed randomly in WSN that may statistically vary, we simulate
every case for 100 iterations to get more accurate results. The percentage of the cluster heads
is set to 5% in three shapes of topology. It can be seen from Table 3.2 that the longest lifetime
of WSN is the circle shape of topology. The BS located in the center of the circle that is
symmetric, so that the energy dissipation of nodes are more even and the lifetime of the
network is prolonged. For the rectangle shape, the lifetime is different as the position of BS
changes. Simulation results indicate that the lifetime for the BS in (0, 50) is longer that in
(50,175). As the BS in (0, 50) is nearer to the sensor area, the energy for transmitting data to
the BS is reduced.


-60 -30 0 30 60
-60
-30
0
30
60
m
m

0 25 50 75 100
0
25
50
75
100
m
m

(a) Circular topology (b) Square topology

0 50 100 150 200
0
25
50
m
m
Zone 1 Zone 2
Zone 3
Zone 4

(c) Rectangle topology
Fig 3.1 Three different types of WSNs topology:
Sink (a):(0,0); (b):(0,50) or (50,175); (c):(0,25) or (100,150)

The probability of cluster head node in the LEACH protocol has a certain impact on the
WSNs lifetime. In our analysis, we divides the rectangle area into four smaller square areas,
with the communication distance of nodes keeping short (d
B
<d
0
, where d
B
is the
broadcasting distance of the cluster head). From the first-order radio model, we can get the
optimal cluster head probability formula as follows:

2
opt
toBS
N M
k
d

(3-1)

Where N is the number of sensor nodes. In the rectangular region 100 nodes are scattered
randomly, and the region is divided into four regions. In order to verify the difference of the
number of nodes distributed in different regions, we simulate 100 independent iterations of
the nodes number in each region, and get the averages in the four regions as zone1(25.32)
zone2(24.83)zone3(24.87) and zone4(24.98). It can be seen that the number of nodes in
each region are around 25, therere almost no difference in the average nodes numbers for
the four regions, so in the text, the nodes are uniform distributed in the four regions, i.e.
N=25. M is the side length of each small square region, here M=50. d
toBS
is the distance
between the sink and the node. As the distance of a node to the base station is different, we
can change the percentage of cluster heads in different regions to reach the optimal value so
that the lifetime of the whole WSN can be prolonged.
For the topology with a rectangular shape, we propose an improved LEACH algorithm. The
main idea is described as follows:
s

(1) Divide the rectangular area into several small square areas with the same size;
(2) Elect the cluster heads separately in each region, and the optimal probabilities of the
cluster heads for each region can be obtained from Eq. (3-1), i.e. the values are
p
1
=0.02p
2
=0.03, p
3
=0.03, p
4
=0.02
(3) After the cluster heads in each region are selected, the rest of the protocol is similar
to LEACH.
The improved algorithm elects cluster heads in each region according to its probability of
the cluster heads. In this way, it can make sure that there are clusters in every region and
ensure the clusters distributed more uniformly in every region, which reduces the energy
dissipation and improves the lifetime of the network.

We simulate the three shapes of topology that use (improved) LEACH as the routing
protocol. Parameters used in simulation are listed in table 3.1. There are 100 sensor nodes
randomly scattered with fixed position in each shape. We measure the round number when
the first node died, 20% of nodes died and 50% of nodes died respectively as the criterion to
estimate the lifetime of WSN.

Parameter Value
Number of nodes 100
Initial energy (J) 0.5
Data packet length (bit) 4000
Control packet length(bit) 200
Energy dissipation of one Tx (nJ) 50
Energy dissipation of one Rx (nJ) 50
Energy aggregation energy (nJ) 5
Energy loss-free space (pJ/bit/m
2
) 10
Energy loss-multipass fading (pJ/bit/m
4
) 0.0013

Table 3.2 shows the round numbers (the lifetime) of WSN for different BS locations and
different percentage of dead nodes in circle, square and rectangular shapes of topology. As
the sensor nodes distributed randomly in WSN that may statistically vary, we simulate
every case for 100 iterations to get more accurate results. The percentage of the cluster heads
is set to 5% in three shapes of topology. It can be seen from Table 3.2 that the longest lifetime
of WSN is the circle shape of topology. The BS located in the center of the circle that is
symmetric, so that the energy dissipation of nodes are more even and the lifetime of the
network is prolonged. For the rectangle shape, the lifetime is different as the position of BS
changes. Simulation results indicate that the lifetime for the BS in (0, 50) is longer that in
(50,175). As the BS in (0, 50) is nearer to the sensor area, the energy for transmitting data to
the BS is reduced.


Topology Circle Square Rectangle
Sink (0, 0) (0,50) (50, 175) (0, 25) (100,150)
1% 757 740 639 470 555
20% 854 846 715 644 652
50% 923 919 788 785 717
Table 3.2 Lifetime comparisons of different shapes of topology

It can be seen form Table 3.2 that the lifetimes of the rectangular are poor for the BS both
near and far away from the sensor area using conventional LEACH protocol. Then we use
the improved LEACH algorithm that divides the rectangle region into four equal square
regions. The cluster heads are elected separately in each region to make sure the cluster
heads distribute uniformly in four regions.

1 10 20 30 40 50 60 70 80 90
550
650
750
850
950
1000
Perecentage of dead nodes(%)
R
o
u
n
d
Proposed
LEACH

Fig 3.2 The relationship between percentage of dead nodes and number of rounds

Fig. 3.2 shows the improvement using the modified LEACH algorithm in terms of the
lifetime of the network. It can be seen from the figure that the surviving round number of
improved algorithm is increased under the same percentage of dead nodes, corresponding
to the lifetime improvement of 19% (for the case of the first died node) compared with the
conventional LEACH algorithm. This is due to the fact that the improved algorithm reduces
the energy dissipation by dividing the rectangle region into sub-regions.
From the above analysis, when sensor nodes within a monitoring area can be manually
deployed, one can select the circle topology of the WSN to maximize the lifetime. When the
topology area is square and the base station location is variable, one can make the base
station close to the WSN area to save energy and extend the network lifetime. When the
monitoring area is a rectangular one, one can use the idea of partition to extend the
lifetime.

s

4. LEACH & PEGASIS
4.1 Introduction of LEACH & PEGASIS algorithm
Brief introductions about LEACH and PEGASIS have been given in section 2. As pointed,
there are three shortcomings for the LEACH protocol:
(1) The number of cluster heads is uncertain. If the number of cluster heads is large,
cluster heads that need communicate directly with the base station will consume
more energy. If the number of cluster heads is small, common nodes that need
communicate with the remote cluster heads will consume more energy.
(2) When cluster heads are selected, the remaining energy of cluster heads is not
considered. After a node is elected as the cluster head, maybe the remaining energy
is not enough for the next round of communication, therefore may lead to failure of
the entire cluster, and member nodes of the cluster will lose data. A blind spot will
appear within the monitoring area.
(3) Many cluster heads communicate directly with the base station. Especially the cluster
heads are far away from the base station. Transmission of data consumes a lot of
energy. As the cluster head dies prematurely, the total energy of network consumes
excessively.
To overcome the shortcomings of LEACH, we take the advantages of PEGASIS to construct
a chain using greedy algorithm which only uses a node as the cluster head to communicate
with the base station. Studies show that the approach of data fusion and multi-hop based on
cluster can save the energy of nodes even better [23]. Here we propose an improved routing
protocol called LEACH-P.
(1) The optimal number of cluster head is defined by Eq. (4-1).

2
2
fs
mp toBS
N M
m
d
(4-1)

Where m is the optimal number of cluster heads; N is the number of nodes;
fs
is signal
amplification factor in free space;
amp
is signal amplification factor of the multipath
fading channel; M is the side length; d
toBS
is the distance between the cluster head and
the base station.
(2) Cluster heads are decided by Eq. (2-1) in the LEACH algorithm without taking into
account the residual energy of nodes. The new algorithm detects the residual energy
of the cluster head to meet the required energy threshold E(r), i.e. the minimum
energy to complete one round communication, which is sum of the energy for
broadcasting information, receiving data packets and confirming messages from
cluster members to, as well as communicating with its neighboring cluster heads.
In our simulation (parameters and definitions of acronyms are listed in Table. 4.1. We
assume that the coverage of broadcasting is half of the diagonal of the field, which is
less than d
0
. Energy consumed through broadcasting is:

2
* * *
fs
ETX cPL cPL DB
(4-2)


Topology Circle Square Rectangle
Sink (0, 0) (0,50) (50, 175) (0, 25) (100,150)
1% 757 740 639 470 555
20% 854 846 715 644 652
50% 923 919 788 785 717
Table 3.2 Lifetime comparisons of different shapes of topology

It can be seen form Table 3.2 that the lifetimes of the rectangular are poor for the BS both
near and far away from the sensor area using conventional LEACH protocol. Then we use
the improved LEACH algorithm that divides the rectangle region into four equal square
regions. The cluster heads are elected separately in each region to make sure the cluster
heads distribute uniformly in four regions.

1 10 20 30 40 50 60 70 80 90
550
650
750
850
950
1000
Perecentage of dead nodes(%)
R
o
u
n
d
Proposed
LEACH

Fig 3.2 The relationship between percentage of dead nodes and number of rounds

Fig. 3.2 shows the improvement using the modified LEACH algorithm in terms of the
lifetime of the network. It can be seen from the figure that the surviving round number of
improved algorithm is increased under the same percentage of dead nodes, corresponding
to the lifetime improvement of 19% (for the case of the first died node) compared with the
conventional LEACH algorithm. This is due to the fact that the improved algorithm reduces
the energy dissipation by dividing the rectangle region into sub-regions.
From the above analysis, when sensor nodes within a monitoring area can be manually
deployed, one can select the circle topology of the WSN to maximize the lifetime. When the
topology area is square and the base station location is variable, one can make the base
station close to the WSN area to save energy and extend the network lifetime. When the
monitoring area is a rectangular one, one can use the idea of partition to extend the
lifetime.

s

4. LEACH & PEGASIS
4.1 Introduction of LEACH & PEGASIS algorithm
Brief introductions about LEACH and PEGASIS have been given in section 2. As pointed,
there are three shortcomings for the LEACH protocol:
(1) The number of cluster heads is uncertain. If the number of cluster heads is large,
cluster heads that need communicate directly with the base station will consume
more energy. If the number of cluster heads is small, common nodes that need
communicate with the remote cluster heads will consume more energy.
(2) When cluster heads are selected, the remaining energy of cluster heads is not
considered. After a node is elected as the cluster head, maybe the remaining energy
is not enough for the next round of communication, therefore may lead to failure of
the entire cluster, and member nodes of the cluster will lose data. A blind spot will
appear within the monitoring area.
(3) Many cluster heads communicate directly with the base station. Especially the cluster
heads are far away from the base station. Transmission of data consumes a lot of
energy. As the cluster head dies prematurely, the total energy of network consumes
excessively.
To overcome the shortcomings of LEACH, we take the advantages of PEGASIS to construct
a chain using greedy algorithm which only uses a node as the cluster head to communicate
with the base station. Studies show that the approach of data fusion and multi-hop based on
cluster can save the energy of nodes even better [23]. Here we propose an improved routing
protocol called LEACH-P.
(1) The optimal number of cluster head is defined by Eq. (4-1).

2
2
fs
mp toBS
N M
m
d
(4-1)

Where m is the optimal number of cluster heads; N is the number of nodes;
fs
is signal
amplification factor in free space;
amp
is signal amplification factor of the multipath
fading channel; M is the side length; d
toBS
is the distance between the cluster head and
the base station.
(2) Cluster heads are decided by Eq. (2-1) in the LEACH algorithm without taking into
account the residual energy of nodes. The new algorithm detects the residual energy
of the cluster head to meet the required energy threshold E(r), i.e. the minimum
energy to complete one round communication, which is sum of the energy for
broadcasting information, receiving data packets and confirming messages from
cluster members to, as well as communicating with its neighboring cluster heads.
In our simulation (parameters and definitions of acronyms are listed in Table. 4.1. We
assume that the coverage of broadcasting is half of the diagonal of the field, which is
less than d
0
. Energy consumed through broadcasting is:

2
* * *
fs
ETX cPL cPL DB
(4-2)


The energy that each cluster head needs to receive confirmation and data packets
from its cluster members is:

19*(( ) * *
DA
ERX E PL ERX cPL (4-3)

In each cluster, the average number of nodes that the cluster head needs to receive
the information is 19.
The energy that the cluster head needs to send information to its neighboring cluster
head is:
2
( ) * * *
DA fs
ETX E PL PL DB (4-4)

The distance between cluster heads is random, and we assume it as the broadcasting
distance between nodes.
From above, we can obtain that the energy threshold value is ~0.0048J.
(3) The new algorithm (LEACH-P) randomly selects five cluster heads linked as a chain.
The one with maximum residual energy is chosen to transfer information to the sink.
Other cluster heads reduce energy dissipation through data fusion. LEACH-P uses
the wireless communication model described in [24]. The energy dissipation that
transmits K-bit data to the receiver over a certain distance d is:

2
0
4
0
( , )
elec fs
elec mp
kE k d
d d
t d d
kE k d
E k d

(4-5)

Where E
elec
is the energy dissipation of the transmitter;
fs
and
mp
are the energy
dissipations of the power amplifiers; d
0
is the constant. The energy that a node needs to
receive K-bit data is:
( )
r elec
E k kE (4-6)

The energy to fuse a number (L) of K-bit packets is:

( , )
f DA
E L k LkE (4-7)

Where E
DA
is the energy dissipation of fusing 1-bit data.

The advantage of the PEGASIS link performance has been described and proved in [25]. The
topology of the improved algorithm is shown in Fig 4.1. In the figure, the nodes within a
cluster have the same symbol. There are five cluster heads with 5 symbols, one of them is
elected as the leading cluster head that is responsible for the information exchange with the
base station.

s

Parameter Value
Sink location (50, 175)
Sensing region 100x100
The number of nodes(N) 100
Initial energy(E
0
) (J) 0.5
Data packet length(PL) (bit) 4000
Control packet length(cPL) (bit) 200
Energy dissipation of one (ETX) (nJ) 50
Energy dissipation of one (ERX) (nJ) 50
Energy aggregation energy(EDA) (nJ) 5
Energy loss-free space(Efs) (pJ/bit/m
2
) 10
Energy loss-multipass fading (Emp) (pJ/bit/m
4
) 0.0013

0 20 40 60 80 100
0
20
40
60
80
100
120
140
160
180
Length/m
W
i
d
t
h
/
m
Sink(50,175)
Cluter head
Leading cluster head
Data transmission direction

Fig. 4.1 Topology of the improved algorithm

The new algorithm not only solves the problem in the LEACH protocol that multiple cluster
heads communicate directly with the base station to introduce more energy dissipation, but
also overcomes the shortcoming of the long delay in the PEGASIS protocol. The flowchart of
the algorithm is shown in Fig 4.2. In each round of communication, the improved algorithm
is still divided into two stagesthe setup stage and the stable stage.
In the setup stage, the election of cluster head depends on whether there is any dead node
and the residual energy of cluster heads must be greater than E(r). After the election of
cluster heads, the cluster heads broadcast information (advertisement message, ADV). Non-
cluster head node chooses the cluster according to the signal strength after receiving the


The energy that each cluster head needs to receive confirmation and data packets
from its cluster members is:

19*(( ) * *
DA
ERX E PL ERX cPL (4-3)

In each cluster, the average number of nodes that the cluster head needs to receive
the information is 19.
The energy that the cluster head needs to send information to its neighboring cluster
head is:
2
( ) * * *
DA fs
ETX E PL PL DB (4-4)

The distance between cluster heads is random, and we assume it as the broadcasting
distance between nodes.
From above, we can obtain that the energy threshold value is ~0.0048J.
(3) The new algorithm (LEACH-P) randomly selects five cluster heads linked as a chain.
The one with maximum residual energy is chosen to transfer information to the sink.
Other cluster heads reduce energy dissipation through data fusion. LEACH-P uses
the wireless communication model described in [24]. The energy dissipation that
transmits K-bit data to the receiver over a certain distance d is:

2
0
4
0
( , )
elec fs
elec mp
kE k d
d d
t d d
kE k d
E k d

(4-5)

Where E
elec
is the energy dissipation of the transmitter;
fs
and
mp
are the energy
dissipations of the power amplifiers; d
0
is the constant. The energy that a node needs to
receive K-bit data is:
( )
r elec
E k kE (4-6)

The energy to fuse a number (L) of K-bit packets is:

( , )
f DA
E L k LkE (4-7)

Where E
DA
is the energy dissipation of fusing 1-bit data.

The advantage of the PEGASIS link performance has been described and proved in [25]. The
topology of the improved algorithm is shown in Fig 4.1. In the figure, the nodes within a
cluster have the same symbol. There are five cluster heads with 5 symbols, one of them is
elected as the leading cluster head that is responsible for the information exchange with the
base station.

s

Parameter Value
Sink location (50, 175)
Sensing region 100x100
The number of nodes(N) 100
Initial energy(E
0
) (J) 0.5
Data packet length(PL) (bit) 4000
Control packet length(cPL) (bit) 200
Energy dissipation of one (ETX) (nJ) 50
Energy dissipation of one (ERX) (nJ) 50
Energy aggregation energy(EDA) (nJ) 5
Energy loss-free space(Efs) (pJ/bit/m
2
) 10
Energy loss-multipass fading (Emp) (pJ/bit/m
4
) 0.0013

0 20 40 60 80 100
0
20
40
60
80
100
120
140
160
180
Length/m
W
i
d
t
h
/
m
Sink(50,175)
Cluter head
Leading cluster head
Data transmission direction

Fig. 4.1 Topology of the improved algorithm

The new algorithm not only solves the problem in the LEACH protocol that multiple cluster
heads communicate directly with the base station to introduce more energy dissipation, but
also overcomes the shortcoming of the long delay in the PEGASIS protocol. The flowchart of
the algorithm is shown in Fig 4.2. In each round of communication, the improved algorithm
is still divided into two stagesthe setup stage and the stable stage.
In the setup stage, the election of cluster head depends on whether there is any dead node
and the residual energy of cluster heads must be greater than E(r). After the election of
cluster heads, the cluster heads broadcast information (advertisement message, ADV). Non-
cluster head node chooses the cluster according to the signal strength after receiving the


information, and sends a request that includes the cluster ID, its own ID, as well as its
remaining energy state to join the cluster. Cluster heads are connected into a chain after the
establishment of clusters, and the cluster head with maximum residual energy is assigned as
the leading cluster head to communicate with the BS directly.

In the stable stage, common nodes of the cluster send information to the cluster head
according to the TDMA time slot table. The cluster head receives data and integrates data
into a packet. The packets are transmitted along the chain to the leading cluster head
according to the Token. The leading cluster head receives and fuses data packets, then sends
to the base station.

Fig. 4.2 Flow chart of proposed LEACH-P algorithm
s

4.2 Simulation results and analysis
The performance of improved algorithm (LEACH-P) is evaluated in terms of the lifetime of
the network and data transmission delay. 100 sensor nodes are randomly distributed in the
sensing area of 100x100m
2
. Most simulation parameters are the same as those of the LEACH
algorithm, with specific parameters listed in Table 4.1.
The lifetime in terms of rounds corresponding to 1%, 20%, 50%of nodes died are simulated
for both the LEACH and improved algorithm. The results are compared in Table 4.2. The
lifetime of the new algorithm improves by 17% for the case with 1% dead nodes. The
numbers of rounds when 20% and 50% of nodes died are also improved compared with the
LEACH protocol. Similar to previous simulation, we take 100 iterations for each case to
reduce the statistical fluctuation.

Percentage of dead nodes LEACH LEACH-P
1% 639 751
20% 715 848
50% 788 919
100% 1296 1276
Table 4.2 Lifetime comparisons of different algorithms

Fig 4.3 also shows the lifetime comparison for the LEACH and improved algorithm. It can
be seen that the round number corresponding to the first dead node and all dead nodes is
~ 639 and 1240 for the LEACH protocol, respectively. These two numbers are improved to
~751 and 1300 using the new algorithm. The Lower part of Fig 4.3 shows the variation of the
cluster head number in the new algorithm with a fixed number as 5 until ~ 800 rounds
(dead nodes appear). Compared with LEACH, the proposed algorithm can prolong the
network lifetime and balance the energy dissipation of network nodes as well.

0 200 400 600 800 1000 1200 1400
0
50
100
Round
N
u
m
b
e
r

o
f

s
u
r
v
i
v
a
l

n
o
d
e
s
0 200 400 600 800 1000 1200 1400
0
5
10
Round
N
u
m
b
e
r

o
f

c
l
u
s
t
e
r

h
e
a
d
s
LEACH-P
LEACH-P
LEACH

Fig. 4.3 Relationship between rounds and number of survival nodes


information, and sends a request that includes the cluster ID, its own ID, as well as its
remaining energy state to join the cluster. Cluster heads are connected into a chain after the
establishment of clusters, and the cluster head with maximum residual energy is assigned as
the leading cluster head to communicate with the BS directly.

In the stable stage, common nodes of the cluster send information to the cluster head
according to the TDMA time slot table. The cluster head receives data and integrates data
into a packet. The packets are transmitted along the chain to the leading cluster head
according to the Token. The leading cluster head receives and fuses data packets, then sends
to the base station.

Fig. 4.2 Flow chart of proposed LEACH-P algorithm
s

4.2 Simulation results and analysis
The performance of improved algorithm (LEACH-P) is evaluated in terms of the lifetime of
the network and data transmission delay. 100 sensor nodes are randomly distributed in the
sensing area of 100x100m
2
. Most simulation parameters are the same as those of the LEACH
algorithm, with specific parameters listed in Table 4.1.
The lifetime in terms of rounds corresponding to 1%, 20%, 50%of nodes died are simulated
for both the LEACH and improved algorithm. The results are compared in Table 4.2. The
lifetime of the new algorithm improves by 17% for the case with 1% dead nodes. The
numbers of rounds when 20% and 50% of nodes died are also improved compared with the
LEACH protocol. Similar to previous simulation, we take 100 iterations for each case to
reduce the statistical fluctuation.

Percentage of dead nodes LEACH LEACH-P
1% 639 751
20% 715 848
50% 788 919
100% 1296 1276
Table 4.2 Lifetime comparisons of different algorithms

Fig 4.3 also shows the lifetime comparison for the LEACH and improved algorithm. It can
be seen that the round number corresponding to the first dead node and all dead nodes is
~ 639 and 1240 for the LEACH protocol, respectively. These two numbers are improved to
~751 and 1300 using the new algorithm. The Lower part of Fig 4.3 shows the variation of the
cluster head number in the new algorithm with a fixed number as 5 until ~ 800 rounds
(dead nodes appear). Compared with LEACH, the proposed algorithm can prolong the
network lifetime and balance the energy dissipation of network nodes as well.

0 200 400 600 800 1000 1200 1400
0
50
100
Round
N
u
m
b
e
r

o
f

s
u
r
v
i
v
a
l

n
o
d
e
s
0 200 400 600 800 1000 1200 1400
0
5
10
Round
N
u
m
b
e
r

o
f

c
l
u
s
t
e
r

h
e
a
d
s
LEACH-P
LEACH-P
LEACH

Fig. 4.3 Relationship between rounds and number of survival nodes


WSNs are generally deployed in harsh environments, where the base station is far away
from the sensor field. The location of the base station has a great effect on the lifetime of the
sensor network. When the distance between the base station and the monitored region is too
large, some cluster heads will lead to excessive energy dissipation in the LEACH, shortening
the lifetime of the network. On the other hand, the leading cluster head transmits data to the
base station gathered from other cluster heads using multi-hop in the new algorithm.
Therefore, the increasing distance between the base station and the monitored regions has
less effect on the network lifetime. In Fig. 4.4, the network lifetime of LEACH and proposed
algorithm are calculated as we vary the distance between the BS and sensor field. It can be
seen from Fig. 4.4 that:
(1) When the base station location changes from 100 to 250, the lifetime of WSN for the
LEACH-P algorithm keeps almost constant (~760). When using the LEACH
algorithm, the lifetime is reduced from 737 to 328.
(2) When the base station location changes from 250 to 400, the lifetime of LEACH-P
algorithm and LEACH algorithm are both reduced, but LEACH-P algorithm is still
better than the conventional LEACH.

100 150 200 250 300 350 400
0
100
200
300
400
500
600
700
800
Vertical coordinate of Sink
R
o
u
n
d

o
f

f
i
r
s
t

d
e
a
d

n
o
d
e
LEACH
LEACH-P

Fig. 4.4 Relationship between the sink location and the round of first dead node

Furthermore, Fig. 4.5 shows the numbers of live nodes using LEACH-P and LEACH
protocols with the position of sink fixed at (50,300). Now the number of the first dead node
for LEACH is only 187, but it is improved to 658 for the LEACH-P algorithm. Compared
with LEACH, LEACH-P can prolong the network lifetime by 351%. Figs. 4.4 and 4.5 further
indicate that LEACH-P is superior to LEACH with more evenly energy distribution among
nodes even the sink location is far away from the sensor field.

s

0 200 400 600 800 1000 1200 1400
0
20
40
60
80
100
Round
N
u
m
b
e
r

o
f

s
u
r
v
i
v
a
l

n
o
d
e
s
LEACH
LEACH-P

Fig. 4.5 Comparison of two algorithms for the sink location at (50,300)

Another significant advantage of LEACH-P is that it not only overcomes the issue of long delay
of PEGASIS, but also inherits the idea of greedy into a chain. In order to verify the improvement
of the transmission delay of the WSN, we calculate and compare the maximum distance that the
data should be transmitted for every round in LEACH, PEGASIS and LEACH-P.
(1) When data is transmitted according to LEACH, all common nodes of the cluster
send data to the cluster head in accordance with the TDMA time slots. The cluster
head will fuse the data and send it to the sink. The longest distance of transmitting
data in each round corresponds to the maximum distance of the sum of both
common nodes to the cluster head and the cluster head to the sink.
(2) When data is transmitted using the token mechanism in PEGASIS, there is only one
token which goes through the whole chain. The longest distance of transmitting
data in each round corresponds to the sum of the distances of both the length of the
entire chain and the leading cluster head to the sink.
(3) When data is transmitted using the TDMA and token mechanism in LEACH-P, in
each cluster, the cluster head allocates TDMA time slot to common nodes. Common
nodes send the data to the cluster head. Between cluster heads, the cluster heads are
linked into chains and then transfer the data to the leading cluster head according to
the token mechanism. The leading cluster head fuses the data and sends it to the sink.
The longest distance of transmitting data in each round corresponds to the sum of the
maximum distance of common nodes to cluster heads in two ends of the chain, the
length of the chain and the distance between the leading cluster head and the sink.
Again, we simulate all cases with 100 iterations to get the statistical average value. The
performance (network delay) is evaluated in terms of the average distance before the first
dead node appearing in wireless sensor network.


WSNs are generally deployed in harsh environments, where the base station is far away
from the sensor field. The location of the base station has a great effect on the lifetime of the
sensor network. When the distance between the base station and the monitored region is too
large, some cluster heads will lead to excessive energy dissipation in the LEACH, shortening
the lifetime of the network. On the other hand, the leading cluster head transmits data to the
base station gathered from other cluster heads using multi-hop in the new algorithm.
Therefore, the increasing distance between the base station and the monitored regions has
less effect on the network lifetime. In Fig. 4.4, the network lifetime of LEACH and proposed
algorithm are calculated as we vary the distance between the BS and sensor field. It can be
seen from Fig. 4.4 that:
(1) When the base station location changes from 100 to 250, the lifetime of WSN for the
LEACH-P algorithm keeps almost constant (~760). When using the LEACH
algorithm, the lifetime is reduced from 737 to 328.
(2) When the base station location changes from 250 to 400, the lifetime of LEACH-P
algorithm and LEACH algorithm are both reduced, but LEACH-P algorithm is still
better than the conventional LEACH.

100 150 200 250 300 350 400
0
100
200
300
400
500
600
700
800
Vertical coordinate of Sink
R
o
u
n
d

o
f

f
i
r
s
t

d
e
a
d

n
o
d
e
LEACH
LEACH-P

Fig. 4.4 Relationship between the sink location and the round of first dead node

Furthermore, Fig. 4.5 shows the numbers of live nodes using LEACH-P and LEACH
protocols with the position of sink fixed at (50,300). Now the number of the first dead node
for LEACH is only 187, but it is improved to 658 for the LEACH-P algorithm. Compared
with LEACH, LEACH-P can prolong the network lifetime by 351%. Figs. 4.4 and 4.5 further
indicate that LEACH-P is superior to LEACH with more evenly energy distribution among
nodes even the sink location is far away from the sensor field.

s

0 200 400 600 800 1000 1200 1400
0
20
40
60
80
100
Round
N
u
m
b
e
r

o
f

s
u
r
v
i
v
a
l

n
o
d
e
s
LEACH
LEACH-P

Fig. 4.5 Comparison of two algorithms for the sink location at (50,300)

Another significant advantage of LEACH-P is that it not only overcomes the issue of long delay
of PEGASIS, but also inherits the idea of greedy into a chain. In order to verify the improvement
of the transmission delay of the WSN, we calculate and compare the maximum distance that the
data should be transmitted for every round in LEACH, PEGASIS and LEACH-P.
(1) When data is transmitted according to LEACH, all common nodes of the cluster
send data to the cluster head in accordance with the TDMA time slots. The cluster
head will fuse the data and send it to the sink. The longest distance of transmitting
data in each round corresponds to the maximum distance of the sum of both
common nodes to the cluster head and the cluster head to the sink.
(2) When data is transmitted using the token mechanism in PEGASIS, there is only one
token which goes through the whole chain. The longest distance of transmitting
data in each round corresponds to the sum of the distances of both the length of the
entire chain and the leading cluster head to the sink.
(3) When data is transmitted using the TDMA and token mechanism in LEACH-P, in
each cluster, the cluster head allocates TDMA time slot to common nodes. Common
nodes send the data to the cluster head. Between cluster heads, the cluster heads are
linked into chains and then transfer the data to the leading cluster head according to
the token mechanism. The leading cluster head fuses the data and sends it to the sink.
The longest distance of transmitting data in each round corresponds to the sum of the
maximum distance of common nodes to cluster heads in two ends of the chain, the
length of the chain and the distance between the leading cluster head and the sink.
Again, we simulate all cases with 100 iterations to get the statistical average value. The
performance (network delay) is evaluated in terms of the average distance before the first
dead node appearing in wireless sensor network.


0 20 40 60 80 100
200
400
600
800
1000
1200
Number of Rounds
A
v
e
r
a
g
e

d
i
s
t
a
n
c
e
/
R
o
u
n
d
LEACH
LEACH-P
PEGASIS

Fig. 4.6 Comparison of transmission delay in terms of maximum distance for three protocols

Fig. 4.6 illustrates the simulation results. In PEGASIS algorithm, the distribution of nodes
and the chain length are quite different from round to round, so the average value is still
fluctuating greatly. The average longest distances of transmitting data in each round for
PEGASIS, LEACH and LEACH-P are 1026.7m, 209.9m and 357.33m, respectively. The real-
time of LEACH-P is slightly lower than LEACH. However, compared with PEGASIS,
LEACH-P increases by a dramatic value, i.e. ~290%.
Therefore, we can conclude for this section that LEACH-P combines the advantages of both
LEACH and PEGASIS. It can not only reduce the energy dissipation of cluster heads
compared to the LEACH algorithm in large-scale sensor networks, but also overcome the
issue of poor real-time in the PEGASIS algorithm.

5. O-LEACH
Above sections are mainly about randomly scattered WSN nodes with different topologies.
However, in some particular areas that are difficult to place wireless sensor nodes, we can
lay distributed fiber sensor (DFS) along. DFS can achieve measurements such as
temperature, strain /stress and so on, which associates with wireless sensor nodes to
construct a new hybrid optical wireless sensor network. Incorporating distributed optical
fiber sensor in rectangular topological region makes the WSN more suitable to work in
harsh and large-scale regions. Meanwhile, the reliability and security of system and data are
further protected.

5.1 O-LEACH algorithm description
We investigate an infrastructure of hybrid sensor network which is composed of a DFS link
and two separated WSNs, as shown in Fig 5.1. The DFS link is located at the center of the
whole sensor field and can cover a certain area. The two WSN fields are filled with
s

randomly scattered nodes as usual. These nodes can communicate with each others. Unlike
simple WSNs, since the DFS has to be powered on for data processing, we use one end of
the DFS as the sink or the base station for all WSN nodes. We specially propose a new
energy efficient communication protocol, optical LEACH (O-LEACH), based on the WSN
LEACH protocol.
WSN
Field I
WSN
Field II
Distributed Fiber Sensor Link
DFS
Coverage
Area
WSN Nodes
SINK
Width
D
WSN
Field I
WSN
Field II
DFS
Coverage
Area
WSN Nodes
SINK
Width
D

Fig. 5.1 Sensor field consisting of a distributed fiber sensor (DFS) link and two WSN fields
(I and II), the sink node is located at one end of the DFS and the width of the DFS coverage
area is D.

As a more general topology, Fig. 5.2 shows such hybrid sensor networks that have potential
to cover much more broad areas under certain guidelines: (1) cascade of multiple
rectangular regions in which the base station location of sensor node is (100,150), (2) the DFS
is located in the monitor area with massive volume of data, harsh environment and poor
security (located in the middle of the rectangular region for this paper) to link the
rectangular region and the location of fibers base station is (0, 25). The DFS can cover a
certain area, for example 10m (vertical axis), to monitor the pressure or temperature
information within the coverage area, and give the data back to the fiber base station.

0 50 100 150 200
0
10
20
30
40
50
Rectangular topology N Length/m
W
i
d
t
h
/
m
Sink N(100,150)
0 50 100 150 200
0
10
20
30
40
50
Rectangular topology1 Length/m
W
i
d
t
h
/
m
Sink 1(100,150)
Fiber's
sink
Distributed optical
fiber sensor
Distributed optical
fiber sensor
...........

Fig. 5.2 The topology incorporating distributed optical fiber sensors

As nodes of two WSNs are power limited, the protocol is mainly dealing with these nodes.
The flowchart of the O-LEACH protocol is shown in Fig. 5.3. As the operation of the
standard LEACH protocol is separated into the setup phase and the steady phase, we also


0 20 40 60 80 100
200
400
600
800
1000
1200
Number of Rounds
A
v
e
r
a
g
e

d
i
s
t
a
n
c
e
/
R
o
u
n
d
LEACH
LEACH-P
PEGASIS

Fig. 4.6 Comparison of transmission delay in terms of maximum distance for three protocols

Fig. 4.6 illustrates the simulation results. In PEGASIS algorithm, the distribution of nodes
and the chain length are quite different from round to round, so the average value is still
fluctuating greatly. The average longest distances of transmitting data in each round for
PEGASIS, LEACH and LEACH-P are 1026.7m, 209.9m and 357.33m, respectively. The real-
time of LEACH-P is slightly lower than LEACH. However, compared with PEGASIS,
LEACH-P increases by a dramatic value, i.e. ~290%.
Therefore, we can conclude for this section that LEACH-P combines the advantages of both
LEACH and PEGASIS. It can not only reduce the energy dissipation of cluster heads
compared to the LEACH algorithm in large-scale sensor networks, but also overcome the
issue of poor real-time in the PEGASIS algorithm.

5. O-LEACH
Above sections are mainly about randomly scattered WSN nodes with different topologies.
However, in some particular areas that are difficult to place wireless sensor nodes, we can
lay distributed fiber sensor (DFS) along. DFS can achieve measurements such as
temperature, strain /stress and so on, which associates with wireless sensor nodes to
construct a new hybrid optical wireless sensor network. Incorporating distributed optical
fiber sensor in rectangular topological region makes the WSN more suitable to work in
harsh and large-scale regions. Meanwhile, the reliability and security of system and data are
further protected.

5.1 O-LEACH algorithm description
We investigate an infrastructure of hybrid sensor network which is composed of a DFS link
and two separated WSNs, as shown in Fig 5.1. The DFS link is located at the center of the
whole sensor field and can cover a certain area. The two WSN fields are filled with
s

randomly scattered nodes as usual. These nodes can communicate with each others. Unlike
simple WSNs, since the DFS has to be powered on for data processing, we use one end of
the DFS as the sink or the base station for all WSN nodes. We specially propose a new
energy efficient communication protocol, optical LEACH (O-LEACH), based on the WSN
LEACH protocol.
WSN
Field I
WSN
Field II
DFS
Coverage
Area
WSN Nodes
SINK
Width
D
WSN
Field I
WSN
Field II
DFS
Coverage
Area
WSN Nodes
SINK
Width
D

Fig. 5.1 Sensor field consisting of a distributed fiber sensor (DFS) link and two WSN fields
(I and II), the sink node is located at one end of the DFS and the width of the DFS coverage
area is D.

As a more general topology, Fig. 5.2 shows such hybrid sensor networks that have potential
to cover much more broad areas under certain guidelines: (1) cascade of multiple
rectangular regions in which the base station location of sensor node is (100,150), (2) the DFS
is located in the monitor area with massive volume of data, harsh environment and poor
security (located in the middle of the rectangular region for this paper) to link the
rectangular region and the location of fibers base station is (0, 25). The DFS can cover a
certain area, for example 10m (vertical axis), to monitor the pressure or temperature
information within the coverage area, and give the data back to the fiber base station.

0 50 100 150 200
0
10
20
30
40
50
Rectangular topology N Length/m
W
i
d
t
h
/
m
Sink N(100,150)
0 50 100 150 200
0
10
20
30
40
50
Rectangular topology1 Length/m
W
i
d
t
h
/
m
Sink 1(100,150)
Fiber's
sink
Distributed optical
fiber sensor
Distributed optical
fiber sensor
...........

Fig. 5.2 The topology incorporating distributed optical fiber sensors

As nodes of two WSNs are power limited, the protocol is mainly dealing with these nodes.
The flowchart of the O-LEACH protocol is shown in Fig. 5.3. As the operation of the
standard LEACH protocol is separated into the setup phase and the steady phase, we also


separate the O-LEACH operation into two phases. The steady phase is as same as the
LEACH one. During the setup phase, there are two major differences between O-LEACH
and LEACH: (1) nodes of WSNs cannot be deployed in the DFS coverage area; (2) the cluster
head and the node should be within the same WSN field if two WSNs cannot communicate
with each other (i.e. checking if Position
Cluster_head
=Position
Node_I
in the flowchart). For most
applications, it would be better to assume that two WSN fields are isolated due to the
following reasons: (1) save information transfer energy, as transferring data over the DFS
terrain would waste more energy; (2) wireless communication over the DFS area is not
allowed for some applications. However, we simulate the case that nodes inside different
WSN fields can communicate with each other as well for reference.

Node I belongs
to fiber sensor coverage
area?
NO
Regenerate a
round node
YES
Node I is
cluster-head?
NO
YES
Announce
cluster-head
status
Wait for
join-request
message
Create TDMA
schedule and send
to cluster members
(t=0)
Wait for
cluster-head
announcement
Send join-request
message to chosen
cluster-head
Position
Cluster-head
=Position
Node_I
?
NO
YES
Wait for
schedule from
cluster-head
(t=0)
Steady-state operation
for t=Tround
Steady Phase
(similar to LEACH)
Setup Phase
Node I belongs
area?
NO
Regenerate a
round node
YES
Node I is
cluster-head?
NO
YES
Announce
cluster-head
status
Wait for
join-request
message
Create TDMA
schedule and send
to cluster members
(t=0)
Wait for
cluster-head
announcement
Send join-request
message to chosen
cluster-head
Position
Cluster-head
=Position
Node_I
?
NO
YES
Wait for
schedule from
cluster-head
(t=0)
for t=Tround
Steady Phase
(similar to LEACH)
Setup Phase

Fig. 5.3 Flowchart of the O-LEACH protocol

s

Based on the proposed O-LEACH protocol, we simulate the network performance in terms
of the node lifetime. In our simulation model: (1) most of parameters (e.g. probability of a
node to become a cluster head, data packet length, control packet length, etc.) are as the
same as other LEACH-based simulation models (listed in Table 4.1); (2) the position of the
sink in the LEACH model can be put in different places, while in our LEACH and O-
LEACH ones, we put the sink of all the cases at the same position, i.e. in the middle of one
edge of the sensor field; (3) as the network energy dissipation is a totally statistical behavior
due to the random distribution of WSN nodes, we simulate every case for 1000 independent
iterations (over days).
First we compare the network performance in terms of lifetime for the simple topology
shown in Fig. 5.1. Three cases are simulated: original LEACH without DFS; O-LEACH with
varying width of DFS coverage area (D) that two WSNs can either communicate with each
other or not. As people use different parameters to evaluate the lifetime, i.e. the round
number corresponding to the appearance of the first dead node, half of the dead nodes or
the last survival node (refered as first-dead, half-dead and fully-dead in the following
part of the paper), we obtain all the three parameters and find that the network
improvement may end up with quite different conclusions through these parameters.
For the LEACH case, we obtain that the average round number corresponding to first-
dead, half-dead and fully-dead are 731, 915, and 1741, respectively. Figs 5.4 (a) and (b)
show the average lifetime evaluated by three -dead parameters for situations that the two
WSNs can communicate with each other or not as we vary the value of D (from 5 to 50). We
can see that: (1) the network performance in terms of lifetime keeps almost constant
regardless the width of DFS coverage. (2) In the case that two WSNs cannot communicate
with each other, nodes can save energy on broadcasting over smaller area (i.e. shorter
distance), therefore, the average lifetime corresponding to either first-dead or half-dead
is improved ~20% compared to the case that two WSNs are connected, while the last nodes
lifetime (fully-dead) is more than doubled. (3) Compare to the conventional LEACH
protocol, if two WSNs can communicate with each other, the improvement of O-LEACH is
very limited (~2% and ~ 16% in terms of first-dead and fully-dead, respectively).
Therefore, it is expected and mostly required for such hybrid sensor networks to employ O-
LEACH with two isolated WSNs. Furthermore, typical lifetime evolutions are compared as
well in Fig. 5.4 (c) where D equals 20. These curves are specially chosen from thousands of
simulated iterations with performance close to average ones. Results of LEACH protocol are
also included.


separate the O-LEACH operation into two phases. The steady phase is as same as the
LEACH one. During the setup phase, there are two major differences between O-LEACH
and LEACH: (1) nodes of WSNs cannot be deployed in the DFS coverage area; (2) the cluster
head and the node should be within the same WSN field if two WSNs cannot communicate
with each other (i.e. checking if Position
Cluster_head
=Position
Node_I
in the flowchart). For most
applications, it would be better to assume that two WSN fields are isolated due to the
following reasons: (1) save information transfer energy, as transferring data over the DFS
terrain would waste more energy; (2) wireless communication over the DFS area is not
allowed for some applications. However, we simulate the case that nodes inside different
WSN fields can communicate with each other as well for reference.

Node I belongs
area?
NO
Regenerate a
round node
YES
Node I is
cluster-head?
NO
YES
Announce
cluster-head
status
Wait for
join-request
message
Create TDMA
schedule and send
to cluster members
(t=0)
Wait for
cluster-head
announcement
Send join-request
message to chosen
cluster-head
Position
Cluster-head
=Position
Node_I
?
NO
YES
Wait for
schedule from
cluster-head
(t=0)
for t=Tround
Steady Phase
(similar to LEACH)
Setup Phase
Node I belongs
area?
NO
Regenerate a
round node
YES
Node I is
cluster-head?
NO
YES
Announce
cluster-head
status
Wait for
join-request
message
Create TDMA
schedule and send
to cluster members
(t=0)
Wait for
cluster-head
announcement
Send join-request
message to chosen
cluster-head
Position
Cluster-head
=Position
Node_I
?
NO
YES
Wait for
schedule from
cluster-head
(t=0)
for t=Tround
Steady Phase
(similar to LEACH)
Setup Phase

Fig. 5.3 Flowchart of the O-LEACH protocol

s

Based on the proposed O-LEACH protocol, we simulate the network performance in terms
of the node lifetime. In our simulation model: (1) most of parameters (e.g. probability of a
node to become a cluster head, data packet length, control packet length, etc.) are as the
same as other LEACH-based simulation models (listed in Table 4.1); (2) the position of the
sink in the LEACH model can be put in different places, while in our LEACH and O-
LEACH ones, we put the sink of all the cases at the same position, i.e. in the middle of one
edge of the sensor field; (3) as the network energy dissipation is a totally statistical behavior
due to the random distribution of WSN nodes, we simulate every case for 1000 independent
iterations (over days).
First we compare the network performance in terms of lifetime for the simple topology
shown in Fig. 5.1. Three cases are simulated: original LEACH without DFS; O-LEACH with
varying width of DFS coverage area (D) that two WSNs can either communicate with each
other or not. As people use different parameters to evaluate the lifetime, i.e. the round
number corresponding to the appearance of the first dead node, half of the dead nodes or
the last survival node (refered as first-dead, half-dead and fully-dead in the following
part of the paper), we obtain all the three parameters and find that the network
improvement may end up with quite different conclusions through these parameters.
For the LEACH case, we obtain that the average round number corresponding to first-
dead, half-dead and fully-dead are 731, 915, and 1741, respectively. Figs 5.4 (a) and (b)
show the average lifetime evaluated by three -dead parameters for situations that the two
WSNs can communicate with each other or not as we vary the value of D (from 5 to 50). We
can see that: (1) the network performance in terms of lifetime keeps almost constant
regardless the width of DFS coverage. (2) In the case that two WSNs cannot communicate
with each other, nodes can save energy on broadcasting over smaller area (i.e. shorter
distance), therefore, the average lifetime corresponding to either first-dead or half-dead
is improved ~20% compared to the case that two WSNs are connected, while the last nodes
lifetime (fully-dead) is more than doubled. (3) Compare to the conventional LEACH
protocol, if two WSNs can communicate with each other, the improvement of O-LEACH is
very limited (~2% and ~ 16% in terms of first-dead and fully-dead, respectively).
Therefore, it is expected and mostly required for such hybrid sensor networks to employ O-
LEACH with two isolated WSNs. Furthermore, typical lifetime evolutions are compared as
well in Fig. 5.4 (c) where D equals 20. These curves are specially chosen from thousands of
simulated iterations with performance close to average ones. Results of LEACH protocol are
also included.


0
1000
2000
3000
4000
0 20 40 60
DFS Coverage Width (%)
R
o
u
n
d
First_Dead
Half_Dead
Fully_Dead
0
1000
2000
3000
4000
5000
0 20 40 60
R
o
u
n
d
First_Dead
Half_Dead
Fully_Dead

(a) (b)

0
20
40
60
80
100
0 1000 2000 3000 4000 5000
Round Number
N
u
m
b
e
r

o
f

L
i
v
e

N
o
d
e
s
O-LEACH (Isolated)
O-LEACH (Connected)
LEACH

(c)
Fig. 5.4 Simulation results of network performance in terms of lifetime (round number)
using O-LEACH protocol: (a) two WSNs can communicate with each other; (b) two WSNs
are isolated; (c) typical lifetime evolution curves

A legitimate question for above network model is how close to the reality of the hybrid
sensor network in terms of the coverage of DFS and WSN. As pointed in the introduction
section, the distance of typical DFS link can vary from hundreds of meters to tens of
kilometers, while the coverage diameter of WSN node is tens of meters. Therefore, it would
be interesting to look into the case that the length of DFS link increases and the number of
WSN nodes increases proportionally (to keep the approximate density).
To keep straightforward but simple, we fix the width of the whole sensor field to 100 and
the coverage percentage of DFS to 20% (i.e. D equals to 20). Also we only consider the case
that two WSNs are isolated. Fig. 5.5 illustrates the trend of normalized lifetime performance
with increasing link length of DFS. The normalization is done using the ratio of the -dead
parameter to the total node number. It is obvious that 100 is the optimum number for WSN
nodes with parameters listed in Table I. As the length of the DFS link increases, the lifetime
reduces dramatically, especially the first-dead parameter. More wireless sinks are
required for longer DFS links, and the performance evaluation of various optical wireless
sensor network topologies are of great interests for further investigation. On the other hand,
our protocol and the simulation model can be adapted into networks with different
parameters so that we can find the optimized network design.
s

1
10
100
1000
10000
0 100 200 300 400 500 600
Length of DFS Link (a.u.)
N
o
r
m
a
l
i
z
e
d

L
i
f
e
t
i
m
e

(
a
.
u
.
)
First_Dead
Half_Dead
Fully_Dead

Fig. 5.5 Normalized lifetime as we increase the length of DFS link (number of WSN nodes
are increased proportionally, the widths of the DFS coverage area and the whole sensor filed
are fixed)

For the general topology shown in Fig.5.2, we simulate a 50*200m rectangular WSN region
that is divided into four small regions. In the total area, we randomly scattered 100 nodes,
the number of nodes in each region is: 212923 and 27. The DFS is located in the middle
of the total rectangular region. We assume that the nodes in the upper and the lower part of
the optical region can communicate with each other, and the simulation results are shown in
Fig 5.6.
0 200 400 600 800 1000
0
20
40
60
80
100
Round Number
N
u
m
b
e
r

o
f

L
i
v
e

N
o
d
e
s
Zone 1
Zone 2
Zone 3
Zone 4
Total

Fig. 5.6 Performance (# of live nodes vs. round) after introducing the DFS into the
rectangular sensor area

We can see from Fig. 5.6 that the numbers of the first dead node in the four regions are 663,
805, 779 and 698, respectively. The first 20% nodes die slowly, but the remaining ones die


0
1000
2000
3000
4000
0 20 40 60
R
o
u
n
d
First_Dead
Half_Dead
Fully_Dead
0
1000
2000
3000
4000
5000
0 20 40 60
R
o
u
n
d
First_Dead
Half_Dead
Fully_Dead

(a) (b)

0
20
40
60
80
100
0 1000 2000 3000 4000 5000
Round Number
N
u
m
b
e
r

o
f

L
i
v
e

N
o
d
e
s
O-LEACH (Isolated)
O-LEACH (Connected)
LEACH

(c)
Fig. 5.4 Simulation results of network performance in terms of lifetime (round number)
using O-LEACH protocol: (a) two WSNs can communicate with each other; (b) two WSNs
are isolated; (c) typical lifetime evolution curves

A legitimate question for above network model is how close to the reality of the hybrid
sensor network in terms of the coverage of DFS and WSN. As pointed in the introduction
section, the distance of typical DFS link can vary from hundreds of meters to tens of
kilometers, while the coverage diameter of WSN node is tens of meters. Therefore, it would
be interesting to look into the case that the length of DFS link increases and the number of
WSN nodes increases proportionally (to keep the approximate density).
To keep straightforward but simple, we fix the width of the whole sensor field to 100 and
the coverage percentage of DFS to 20% (i.e. D equals to 20). Also we only consider the case
that two WSNs are isolated. Fig. 5.5 illustrates the trend of normalized lifetime performance
with increasing link length of DFS. The normalization is done using the ratio of the -dead
parameter to the total node number. It is obvious that 100 is the optimum number for WSN
nodes with parameters listed in Table I. As the length of the DFS link increases, the lifetime
reduces dramatically, especially the first-dead parameter. More wireless sinks are
required for longer DFS links, and the performance evaluation of various optical wireless
sensor network topologies are of great interests for further investigation. On the other hand,
our protocol and the simulation model can be adapted into networks with different
parameters so that we can find the optimized network design.
s

1
10
100
1000
10000
0 100 200 300 400 500 600
Length of DFS Link (a.u.)
N
o
r
m
a
l
i
z
e
d

L
i
f
e
t
i
m
e

(
a
.
u
.
)
First_Dead
Half_Dead
Fully_Dead

Fig. 5.5 Normalized lifetime as we increase the length of DFS link (number of WSN nodes
are increased proportionally, the widths of the DFS coverage area and the whole sensor filed
are fixed)

For the general topology shown in Fig.5.2, we simulate a 50*200m rectangular WSN region
that is divided into four small regions. In the total area, we randomly scattered 100 nodes,
the number of nodes in each region is: 212923 and 27. The DFS is located in the middle
of the total rectangular region. We assume that the nodes in the upper and the lower part of
the optical region can communicate with each other, and the simulation results are shown in
Fig 5.6.
0 200 400 600 800 1000
0
20
40
60
80
100
Round Number
N
u
m
b
e
r

o
f

L
i
v
e

N
o
d
e
s
Zone 1
Zone 2
Zone 3
Zone 4
Total

Fig. 5.6 Performance (# of live nodes vs. round) after introducing the DFS into the
rectangular sensor area

We can see from Fig. 5.6 that the numbers of the first dead node in the four regions are 663,
805, 779 and 698, respectively. The first 20% nodes die slowly, but the remaining ones die


rapidly in the total region. The results further demonstrate that the hybrid sensor network
incorporating DFS with the O-LEACH protocol can evenly distribute the energy load among
nodes, therefore prolong the overall lifetime of the network.

6. Conclusion
We discussed several improved algorithms (protocols) that can be used for WSNs or hybrid
sensor networks with distributed fiber sensors involved. As sensor networks are much more
complicated in real applications, more thorough and careful optimization of routing
algorithms are required to meet specific requirements, such as real-time, long lifetime,
security, and so on.

7. References
[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci (2002). A survey on sensor
networks, IEEE Communication Magazine, vol. 40, no.8, pp.102-114
[2] J. M. Kahn, R. H. Katz, and K. S. J. Pister (1999). Next century challenges: mobile
networking for smart dust, Proc. ACM MobiCom 99, Washington DC, pp. 27178
[3] V. Rodoplu and T. H. Meng (1999). Minimum energy mobile wireless networks, IEEE
JSAC, vol. 17, no. 8, pp.13331344
[4] K. Sohrabi et al. (2000). Protocols for self-organization of a wireless sensor network, IEEE
Pers. Commun., pp.1627
[5] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan (2000). Energy-efficient
communication protocol for wireless microsensor networks, IEEE Proc. Hawaii Intl.
Conf. Sys. Sci., pp. 110
[6] X. Fan, Y. Song (2007). Improvement on LEACH protocol of wireless sensor network,
IEEE SENSORCOMM, pp.260-264
[7] H. Jeong, C.-S. Nam, Y.-S. Jeong, D.-R. Shin (2008). A mobile agent based LEACH in
wireless sensor network, Conf. on Advanced Comm. Technol. (ICACT), pp. 75-78
[8] Stephanie Lindsey and Cauligi S. Raghavendra (2002). PEGASIS: Power-Efficient
Gathering in Sensor Information System, 2002 IEEE Aerospace Conference, vol. 3,
pp.1125-1130
[9] X. Bao, D. J. Webb, and D. A. Jackson (1993). 32-km distributed temperature sensor using
Brillouin loss in optical fiber, Opt. Lett., vol. 18, pp.15611563.
[10] D. Garus, T. Gogolla, K. Krebber, F. Schliep (1997). Brillouin optical-fiber frequency-
domain analysis for distributed temperature and strain measurements, J. Lightwave
Technol., vol.15, no.4, pp.654-662
[11] S.M. Maughan, H. H. Kee, T. P. Newson (2001). A calibrated 27-km distributed fiber
temperature sensor based on microwave heterodyne detection of spontaneous
Brillouin scattered power, IEEE Photon. Technol. Lett., vol. 13, no 5, pp. 511-513
[12] J. C. Juarez, E. W. Maier, K. N. Choi, H. F. Taylor (2005). Distributed fiber-optic
intrusion sensor system, J. Lightwave Technol. vol.23, no.6, pp.2081-2087
[13] D. Iida, F. Ito (2008). Detection sensitivity of Brillouin scattering near Fresnel reflection
in BOTDR measurement, J. Lightwave Technol., vol. 26, no.4, pp.417-424
[14] D. Kedar and S. Arnon (2003). Laser Firefly Clustering; a New Concept in
Atmospheric Probing, IEEE Photon. Tech. Lett., vol.15, no.1 pp. 16721624
s

[15] S. Teramoto, and T. Ohtsuki (2004). Optical wireless sensor network system using
corner cube retroreflectors (CCRs), IEEE Globecom04, pp.1035-1039
[16] D. Kedar, S. Arnon (2005). Second generation laser firefly clusters: an improved
scheme for distributed sensing in the atmosphere, Appl. Opt., vol. 44, no.6, pp.984-
992
[17] Jamal N. AL-Karaki, Ahmed E. Kamal (2004). Routing Techniques in Wireless Sensor
Networks: A Survey, IEEE Wireless Communications, Dec.
[18] W. Heinzelman, J. Kulik, and H. Balakrishnan (1999). Adaptive Protocols for
Information Dissemination in Wireless Sensor Networks, Proc. 5th ACM/IEEE
Mobicom, Seattle, WA, pp. 17485.
[19] J. Kulik, W. R. Heinzelman, and H. Balakrishnan (2002). Negotiation-Based Protocols
for Disseminating Information in Wireless Sensor Networks, Wireless Networks, vol.
8, pp. 16985.
[20] Wendi Beth Heinzelman (2000). Application-Specific Protocol Architectures for
Wireless Networks (PhD), Boston: Massachusetts Institute of Technology
[21] Vivek Mhatre, Catherine Rosenberg (2004). Design guidelines for wireless sensor
networks: Communication, clustering and aggregation, Ad Hoc Networks, vol.2, no.1,
pp. 45-63
[22] Ning Xu, Sumit Rangwala, Krishna Kant Chintalapudi, Deepak Ganesan, Alan Broad,
Ramesh Govindan, Deborah Estrin (2004). A Wireless sensor network for structural
monitoring, Proc. 2nd international conference on Embedded networked sensor systems,
Baltimore, MD, USA, pp.13-24.
[23] Katayoun Sohrabi, Jay Gao, Vishal Ailawadhi , Gregory J.Pottie (2000). Protocols for
Self-organization of a Wireless Sensor Network, IEEE Personal Communications,
vol.7, no.5, pp.16-27
[24] ISO.16484-5, Building automation and control systems part 5 data communication
protocol, 2003.
[25] Stephanie Lindsey, Cauligi Raghavendra, Krishna M. Sivalingam (2002). Data
Gathering Algorithms in Sensor Networks Using Energy Metrics, IEEE Transactions
on Parallel and Distributed Systems, vol.13, no.9, pp.924-935


rapidly in the total region. The results further demonstrate that the hybrid sensor network
incorporating DFS with the O-LEACH protocol can evenly distribute the energy load among
nodes, therefore prolong the overall lifetime of the network.

6. Conclusion
We discussed several improved algorithms (protocols) that can be used for WSNs or hybrid
sensor networks with distributed fiber sensors involved. As sensor networks are much more
complicated in real applications, more thorough and careful optimization of routing
algorithms are required to meet specific requirements, such as real-time, long lifetime,
security, and so on.

7. References
[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci (2002). A survey on sensor
networks, IEEE Communication Magazine, vol. 40, no.8, pp.102-114
[2] J. M. Kahn, R. H. Katz, and K. S. J. Pister (1999). Next century challenges: mobile
networking for smart dust, Proc. ACM MobiCom 99, Washington DC, pp. 27178
[3] V. Rodoplu and T. H. Meng (1999). Minimum energy mobile wireless networks, IEEE
JSAC, vol. 17, no. 8, pp.13331344
[4] K. Sohrabi et al. (2000). Protocols for self-organization of a wireless sensor network, IEEE
Pers. Commun., pp.1627
[5] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan (2000). Energy-efficient
communication protocol for wireless microsensor networks, IEEE Proc. Hawaii Intl.
Conf. Sys. Sci., pp. 110
[6] X. Fan, Y. Song (2007). Improvement on LEACH protocol of wireless sensor network,
IEEE SENSORCOMM, pp.260-264
[7] H. Jeong, C.-S. Nam, Y.-S. Jeong, D.-R. Shin (2008). A mobile agent based LEACH in
wireless sensor network, Conf. on Advanced Comm. Technol. (ICACT), pp. 75-78
[8] Stephanie Lindsey and Cauligi S. Raghavendra (2002). PEGASIS: Power-Efficient
Gathering in Sensor Information System, 2002 IEEE Aerospace Conference, vol. 3,
pp.1125-1130
[9] X. Bao, D. J. Webb, and D. A. Jackson (1993). 32-km distributed temperature sensor using
Brillouin loss in optical fiber, Opt. Lett., vol. 18, pp.15611563.
[10] D. Garus, T. Gogolla, K. Krebber, F. Schliep (1997). Brillouin optical-fiber frequency-
domain analysis for distributed temperature and strain measurements, J. Lightwave
Technol., vol.15, no.4, pp.654-662
[11] S.M. Maughan, H. H. Kee, T. P. Newson (2001). A calibrated 27-km distributed fiber
temperature sensor based on microwave heterodyne detection of spontaneous
Brillouin scattered power, IEEE Photon. Technol. Lett., vol. 13, no 5, pp. 511-513
[12] J. C. Juarez, E. W. Maier, K. N. Choi, H. F. Taylor (2005). Distributed fiber-optic
intrusion sensor system, J. Lightwave Technol. vol.23, no.6, pp.2081-2087
[13] D. Iida, F. Ito (2008). Detection sensitivity of Brillouin scattering near Fresnel reflection
in BOTDR measurement, J. Lightwave Technol., vol. 26, no.4, pp.417-424
[14] D. Kedar and S. Arnon (2003). Laser Firefly Clustering; a New Concept in
Atmospheric Probing, IEEE Photon. Tech. Lett., vol.15, no.1 pp. 16721624
s

[15] S. Teramoto, and T. Ohtsuki (2004). Optical wireless sensor network system using
corner cube retroreflectors (CCRs), IEEE Globecom04, pp.1035-1039
[16] D. Kedar, S. Arnon (2005). Second generation laser firefly clusters: an improved
scheme for distributed sensing in the atmosphere, Appl. Opt., vol. 44, no.6, pp.984-
992
[17] Jamal N. AL-Karaki, Ahmed E. Kamal (2004). Routing Techniques in Wireless Sensor
Networks: A Survey, IEEE Wireless Communications, Dec.
[18] W. Heinzelman, J. Kulik, and H. Balakrishnan (1999). Adaptive Protocols for
Information Dissemination in Wireless Sensor Networks, Proc. 5th ACM/IEEE
Mobicom, Seattle, WA, pp. 17485.
[19] J. Kulik, W. R. Heinzelman, and H. Balakrishnan (2002). Negotiation-Based Protocols
for Disseminating Information in Wireless Sensor Networks, Wireless Networks, vol.
8, pp. 16985.
[20] Wendi Beth Heinzelman (2000). Application-Specific Protocol Architectures for
Wireless Networks (PhD), Boston: Massachusetts Institute of Technology
[21] Vivek Mhatre, Catherine Rosenberg (2004). Design guidelines for wireless sensor
networks: Communication, clustering and aggregation, Ad Hoc Networks, vol.2, no.1,
pp. 45-63
[22] Ning Xu, Sumit Rangwala, Krishna Kant Chintalapudi, Deepak Ganesan, Alan Broad,
Ramesh Govindan, Deborah Estrin (2004). A Wireless sensor network for structural
monitoring, Proc. 2nd international conference on Embedded networked sensor systems,
Baltimore, MD, USA, pp.13-24.
[23] Katayoun Sohrabi, Jay Gao, Vishal Ailawadhi , Gregory J.Pottie (2000). Protocols for
Self-organization of a Wireless Sensor Network, IEEE Personal Communications,
vol.7, no.5, pp.16-27
[24] ISO.16484-5, Building automation and control systems part 5 data communication
protocol, 2003.
[25] Stephanie Lindsey, Cauligi Raghavendra, Krishna M. Sivalingam (2002). Data
Gathering Algorithms in Sensor Networks Using Energy Metrics, IEEE Transactions
on Parallel and Distributed Systems, vol.13, no.9, pp.924-935

Range-free Area Localization Scheme for Wireless Sensor Networks 321
Range-free Area Localization Scheme for Wireless Sensor Networks
Vijay R. Chandrasekhar, Winston K.G. Seah, Zhi Ang Eu and Arumugam P. Venkatesh
X

Range-free Area Localization Scheme
for Wireless Sensor Networks

Vijay R. Chandrasekhar
1
, Winston K.G. Seah
2
,
Zhi Ang Eu
3
and Arumugam P. Venkatesh
4

1
Stanford University, USA
*

2
Victoria University of Wellington, New Zealand
*

3
National University of Singapore
4
National University of Singapore

Abstract
For large wireless sensor networks, identifying the exact location of every sensor may not be
feasible and the cost may be very high. A coarse estimate of the sensors locations is usually
sufficient for many applications. In this chapter, we describe an efficient Area Localization
Scheme (ALS) for wireless sensor networks. ALS is a range-free scheme that tries to estimate
the position of every sensor within a certain area rather than its exact location. Furthermore,
the powerful sinks instead of the sensors handle all complex calculations. This reduces the
energy consumed by the sensors and helps extend the lifetime of the network. The
granularity of the areas estimated for each node can be easily adjusted by varying some
system parameters, thus making the scheme very flexible. We first study ALS under ideal
two-ray physical layer conditions (as a benchmark) before proceeding to test the scheme in
more realistic non-ideal conditions modelled by the two-ray physical layer model, Rayleigh
fading and lognormal shadowing. We compare the performance of ALS to range-free
localization schemes like APIT (Approximate Point In Triangle) and DV (Distance Vector)
Hop, and observe that the ALS outperforms them. We also implement ALS on an
experimental testbed and, show that at least 80% of nodes lie within a one-hop region of
their estimated areas. Both simulation and experimental results have verified that ALS is a
promising technique for range-free localization in large sensor networks.

Keywords: Localization, Wireless Sensor Network, Positioning, Range-free

1. Introduction
Deployment of low cost wireless sensors is envisioned to be a promising technique for
applications ranging from early warning systems for natural disasters (like tsunamis and

*
This work done by these authors in the Institute for Infocomm Research, Singapore.
17
wildfires), ecosystem monitoring, real-time health monitoring, and military surveillance.
The deployment and management of large scale wireless sensor networks is a challenge
because of the limited processing capability and power constraints on each sensor. Research
issues pertaining to wireless sensor networks, from the physical layer to the application
layer, as well as cross-layer issues like power management and topology management, have
been addressed[1]. Sensor network data is typically interpreted with reference to a sensors
location, e.g. reporting the occurrence of an event, tracking of a moving object or monitoring
the physical conditions of a region. Localization, the process of determining the location of a
sensor node in a wireless sensor network, is a challenging problem as reliance on technology
like GPS [2] is infeasible due to cost and energy constraints, and also physical constraints
like indoor environments.
In very large and dense wireless sensor networks, it may not be feasible to accurately
measure the exact location of every sensor and furthermore, a coarse estimate of the sensors
location may suffice for most applications. A preliminary design of the Area Localization
Scheme (ALS) [3] has been proposed, which can only function in an (unrealistic) ideal
channel and definitely not in a real environment with fading, shadowing and other forms of
interference. In this chapter, we describe algorithms and techniques that will enable the
Area Localization Scheme (ALS) to be deployable in a real environment. ALS is a centralized
range-free scheme that provides an estimation of a sensors location within a certain area,
rather than the exact coordinates of the sensor. The granularity of the location estimate is
determined by the size of areas which a sensor node falls within and this can be easily
adjusted by varying the system parameters. The advantage of this scheme lies in its
simplicity, as no measurements need to be made by the sensors. Since ALS is a range-free
scheme, we compare its performance to other range-free schemes like APIT (Approximate
Point In Triangle) [4], DV-Hop[5] and DHL (Density-aware Hopcount-based Localization)
[6]. To validate our schemes, we first use simulations developed in Qualnet[31] to evaluate
the performance of ALS and show that it outperforms other range-free localization schemes.
We then follow with an implementation of ALS on a wireless sensor network test bed and
conduct tests in both indoor and outdoor environments. We observe that at least 90% of
nodes lie within a 1-hop region of their estimated areas, i.e. within their individual
transmission radius.
The rest of the paper is organized as follows. Section 2 provides a survey of related work on
wireless sensor network localization. Section 3 then describes the key aspects of the basic
Area Localization Scheme. Section 4 describes the simulation environment and evaluates the
performance of the ALS and compares it to other range-free schemes. Section 5 discusses the
performance of the ALS evaluated on a wireless sensor network test bed for both indoor and
outdoor environments. This section also discusses how the ALS scheme is extended to a
generic physical layer model from the two-ray model used in the simulation studies. Section
6 presents our conclusions and plans for future work.

2. Related Work
A number of localization schemes have been proposed to date. The localization schemes
take into account a number of factors like the network topology, device capabilities, signal
propagation models and energy requirements. Most localization schemes require the
location of some nodes in the network to be known. Nodes whose locations are known are
referred to as anchor nodes or reference nodes in the literature. The localization schemes that
use reference nodes can be broadly classified into three categories: range-based schemes,
range-free schemes and schemes that use signal processing or probabilistic techniques
(hereafter referred to as probabilistic schemes). There also exist schemes that do not require
such reference locations in the network.

A. Range-based Schemes
In range-based schemes, the distance or angle measurements from a fixed set of reference
points are known. Multilateration, which encompass atomic, iterative and collaborative
multilateration techniques, are then used to estimate the location of each sensor node.
Range-based schemes use ToA (Time of Arrival), TDoA (Time Difference of Arrival), AOA
(Angle of Arrival) or RSSI (Received Signal Strength Indicator) to estimate their distances to
anchor nodes. UWB based localization schemes [7][8], GPS [2], Cricket [9] and other
schemes [11][12][13] use ToA or TDoA of acoustic or RF signals from multiple anchor nodes
for localization. However, the fast propagation of RF signals implies that a small error in
measurement could lead to large errors. Clock synchronization between multiple reference
nodes or between the sender and the receiver is also an extremely critical issue in schemes
that use ToA or TDoA. AOA allows sensor nodes to calculate the relative angles between
neighbouring nodes [14][15]. However, schemes that use AOA entail sensors and reference
nodes to be equipped with special antenna configurations which may not be feasible to
embed on each sensor. Complex non-linear equations also need to be solved[15]. Schemes
that use RSSI [16][17][18] have to deal with problems caused by large variances in reading,
multi-path fading, background interference and irregular signal propagation.

B. Range-free Schemes
Range-free localization schemes usually do not make use of any of the techniques
mentioned above to estimate distances to reference nodes, e.g. centroid scheme [19] and
APIT [4]. Range quantization methods like DV-Hop [5] and DHL [6] associate each 1-hop
connection with an estimated distance, while others apply RSSI quantization [20]. These
schemes also use multilateration techniques but rely on measures like hop count to estimate
distances to anchor nodes. Range-free schemes offer a less precise estimate of location
compared to range-based schemes.

C. Probabilistic Schemes
The third class of schemes use signal processing techniques or probabilistic schemes to do
localization. The fingerprinting scheme [21], which uses complex signal processing, is an
example of such a scheme. The major drawback of fingerprinting schemes is the substantial
effort required for generating a signal signature database, before localization can be
performed. Hence, it is not suitable for adhoc deployment scenarios in consideration.

D. Schemes without Anchor/Reference Points
The fourth class of schemes is different from the first three in that it does not require anchor
nodes or beacon signals. In [22], a central server models the network as a series of equations
representing proximity constraints between nodes, and then uses sophisticated optimization
wildfires), ecosystem monitoring, real-time health monitoring, and military surveillance.
The deployment and management of large scale wireless sensor networks is a challenge
because of the limited processing capability and power constraints on each sensor. Research
issues pertaining to wireless sensor networks, from the physical layer to the application
layer, as well as cross-layer issues like power management and topology management, have
been addressed[1]. Sensor network data is typically interpreted with reference to a sensors
location, e.g. reporting the occurrence of an event, tracking of a moving object or monitoring
the physical conditions of a region. Localization, the process of determining the location of a
sensor node in a wireless sensor network, is a challenging problem as reliance on technology
like GPS [2] is infeasible due to cost and energy constraints, and also physical constraints
like indoor environments.
In very large and dense wireless sensor networks, it may not be feasible to accurately
measure the exact location of every sensor and furthermore, a coarse estimate of the sensors
location may suffice for most applications. A preliminary design of the Area Localization
Scheme (ALS) [3] has been proposed, which can only function in an (unrealistic) ideal
channel and definitely not in a real environment with fading, shadowing and other forms of
interference. In this chapter, we describe algorithms and techniques that will enable the
Area Localization Scheme (ALS) to be deployable in a real environment. ALS is a centralized
range-free scheme that provides an estimation of a sensors location within a certain area,
rather than the exact coordinates of the sensor. The granularity of the location estimate is
determined by the size of areas which a sensor node falls within and this can be easily
adjusted by varying the system parameters. The advantage of this scheme lies in its
simplicity, as no measurements need to be made by the sensors. Since ALS is a range-free
scheme, we compare its performance to other range-free schemes like APIT (Approximate
Point In Triangle) [4], DV-Hop[5] and DHL (Density-aware Hopcount-based Localization)
[6]. To validate our schemes, we first use simulations developed in Qualnet[31] to evaluate
the performance of ALS and show that it outperforms other range-free localization schemes.
We then follow with an implementation of ALS on a wireless sensor network test bed and
conduct tests in both indoor and outdoor environments. We observe that at least 90% of
nodes lie within a 1-hop region of their estimated areas, i.e. within their individual
transmission radius.
The rest of the paper is organized as follows. Section 2 provides a survey of related work on
wireless sensor network localization. Section 3 then describes the key aspects of the basic
Area Localization Scheme. Section 4 describes the simulation environment and evaluates the
performance of the ALS and compares it to other range-free schemes. Section 5 discusses the
performance of the ALS evaluated on a wireless sensor network test bed for both indoor and
outdoor environments. This section also discusses how the ALS scheme is extended to a
generic physical layer model from the two-ray model used in the simulation studies. Section
6 presents our conclusions and plans for future work.

2. Related Work
A number of localization schemes have been proposed to date. The localization schemes
take into account a number of factors like the network topology, device capabilities, signal
propagation models and energy requirements. Most localization schemes require the
location of some nodes in the network to be known. Nodes whose locations are known are
referred to as anchor nodes or reference nodes in the literature. The localization schemes that
use reference nodes can be broadly classified into three categories: range-based schemes,
range-free schemes and schemes that use signal processing or probabilistic techniques
(hereafter referred to as probabilistic schemes). There also exist schemes that do not require
such reference locations in the network.

A. Range-based Schemes
In range-based schemes, the distance or angle measurements from a fixed set of reference
points are known. Multilateration, which encompass atomic, iterative and collaborative
multilateration techniques, are then used to estimate the location of each sensor node.
Range-based schemes use ToA (Time of Arrival), TDoA (Time Difference of Arrival), AOA
(Angle of Arrival) or RSSI (Received Signal Strength Indicator) to estimate their distances to
anchor nodes. UWB based localization schemes [7][8], GPS [2], Cricket [9] and other
schemes [11][12][13] use ToA or TDoA of acoustic or RF signals from multiple anchor nodes
for localization. However, the fast propagation of RF signals implies that a small error in
measurement could lead to large errors. Clock synchronization between multiple reference
nodes or between the sender and the receiver is also an extremely critical issue in schemes
that use ToA or TDoA. AOA allows sensor nodes to calculate the relative angles between
neighbouring nodes [14][15]. However, schemes that use AOA entail sensors and reference
nodes to be equipped with special antenna configurations which may not be feasible to
embed on each sensor. Complex non-linear equations also need to be solved[15]. Schemes
that use RSSI [16][17][18] have to deal with problems caused by large variances in reading,
multi-path fading, background interference and irregular signal propagation.

B. Range-free Schemes
Range-free localization schemes usually do not make use of any of the techniques
mentioned above to estimate distances to reference nodes, e.g. centroid scheme [19] and
APIT [4]. Range quantization methods like DV-Hop [5] and DHL [6] associate each 1-hop
connection with an estimated distance, while others apply RSSI quantization [20]. These
schemes also use multilateration techniques but rely on measures like hop count to estimate
distances to anchor nodes. Range-free schemes offer a less precise estimate of location
compared to range-based schemes.

C. Probabilistic Schemes
The third class of schemes use signal processing techniques or probabilistic schemes to do
localization. The fingerprinting scheme [21], which uses complex signal processing, is an
example of such a scheme. The major drawback of fingerprinting schemes is the substantial
effort required for generating a signal signature database, before localization can be
performed. Hence, it is not suitable for adhoc deployment scenarios in consideration.

D. Schemes without Anchor/Reference Points
The fourth class of schemes is different from the first three in that it does not require anchor
nodes or beacon signals. In [22], a central server models the network as a series of equations
representing proximity constraints between nodes, and then uses sophisticated optimization
techniques to estimate the location of every node in the network. In [23], Capkun et al.
propose an infrastructure-less GPS-free positioning algorithm.

E. Area-based Localization
Most of the localization schemes mentioned above calculate a sensor nodes exact position,
except for [4], which uses an area-based approach. In [4], anchor nodes send out beacon
packets at the highest power level that they can. A theoretical method, based on RSSI
measurements, called Approximate Point in Triangle (APIT), is defined to determine
whether a point lies inside a triangle formed by connecting three anchor nodes. A sensor
node uses the APIT test with different combinations of three audible anchor nodes (audible
anchors are anchor nodes from which beacon packets are received) until all combinations
are exhausted. Each APIT test determines whether or not the node lies inside a distinct
triangular region. The intersection of all the triangular regions is then considered to estimate
the area in which the sensor is located. The APIT algorithm performs well when the average
number of audible anchors is high (for example, more than 20). As a result, a major
drawback of the algorithm is that it is highly computationally intensive. An average of 20
audible anchors would imply that the intersection of
20
C
3
= 1140 areas need to be considered.
Furthermore, the algorithm performs well only when the anchor nodes are randomly
distributed throughout the network, which is not always feasible in a real deployment
scenario.

3. Area Localization Scheme Fundamentals
In ALS, the nodes in the wireless sensor network are divided into three categories according
to their different functions: reference nodes, sensor nodes and sinks.

A. Reference/Anchor nodes
The main responsibility of the reference/anchor (both terms will be used interchangeably)
nodes is to send out beacon signals to help sensor nodes locate themselves. Reference nodes
are either equipped with GPS to provide accurate location information or placed in pre-
determined locations. In addition, the reference nodes can send out radio signals at varying
power levels as required. For an Ideal Isotropic Antenna, the received power at a distance d
from the transmitter is given by:

2
4
|
.
|
\
|
=
d
G G P P
r t t r
t

(1)

while the two-ray ground reflection model considers both the direct path and a ground
reflection path, and the received power at a distance d is given by:

4
2 2
d
P G G h h
P
t t r t r
r
=
for
t
r t
h h
d
4
>
(2)

where P
r
is the received power, P
t
is the transmitted power, d is the distance between the
transmitter and receiver, is the wavelength and, h
t
and h
r
are the heights of the transmitter
and receiver respectively. G
t
and G
r
represent the gains of the transmitter and receiver
respectively in equations (1) and (2).
From the above equations, it can be clearly seen that if the received power is fixed at a
certain value, the radio signal with a higher transmitted power reaches a greater distance.
Using one of the physical layer models described above and the threshold power that each
sensor can receive, the reference node can calculate the power required to reach different
distances. Each reference node then devises a set of increasing power levels such that the
highest power level can cover the entire area in consideration. The reference nodes then
broadcast several rounds of radio signals. The beacon packet contains the ID of the reference
node and the power level at which the signal is transmitted (which can be simply
represented by an integer value, as explained below.)
Let PS denote the set of increasing power levels of beacon signals sent out by a reference
node. For now, let us assume that all the reference nodes in the system send out the same set
PS of beacon signals. In the ALS scheme, a sensor node simply listens and records the power
levels of beacon signals it receives from each reference node. In real environments, small
scale fading and shadowing can cause the power levels received by the sensor nodes to vary
significantly from the expected power levels calculated by the path loss models in equations
(1) and (2). Sending out beacon signals in the set PS only once might lead to inaccurate
beacon reception by sensor nodes. As a result, the reference nodes send out the beacon
signals in set PS multiple times. The sensor nodes can then calculate the statistical average
(mode or mean) of the received power levels from each reference node.
Let the number of power levels in set PS be denoted by N
p
and the N
p
power levels in set PS
be represented by P
1
,P
2
, P
3
,,P
Np
. The power levels P
1
, P
2
, P
3
,,P
Np
can be represented by
simple integers, e.g. increasing values corresponding to increasing power levels; therefore
sensor nodes only need to take note of these integer values that are contained in the beacon
packets and the hardware design can be kept simple as there is no need for accurate
measurement of the received power level. Let the number of times that the same set of
beacon signals PS are sent out be denoted by N
r
, also referred to as the number of rounds.
The power MP in dB required to cover the entire area is calculated from equation (1) or (2),
based on the physical layer model in consideration. The power LP in dB required to cover a
small distance (say 10 m) is also calculated. The values P
1
,P
2
, P
3
,,P
Np
are then set to be
N
p
uniformly distributed values in the range [LP, MP] in the dB scale. The simple procedure
followed by the reference nodes is shown below:

1 for i = 1: Nr
2 for j=1: N
p

3 Send beacon signal at power levelP
j

4 end for
5 end for

The transmissions by the different reference nodes do not need to be synchronized.
However, the reference nodes schedule the beacon signal transmissions so to avoid
collisions. The transmitted set of power levels PS need not be the same for all the reference
nodes, and can be configured by the network administrator. Also, the set of power levels PS
need not be uniformly distributed too. It is also not necessary for the reference nodes to
techniques to estimate the location of every node in the network. In [23], Capkun et al.
propose an infrastructure-less GPS-free positioning algorithm.

E. Area-based Localization
Most of the localization schemes mentioned above calculate a sensor nodes exact position,
except for [4], which uses an area-based approach. In [4], anchor nodes send out beacon
packets at the highest power level that they can. A theoretical method, based on RSSI
measurements, called Approximate Point in Triangle (APIT), is defined to determine
whether a point lies inside a triangle formed by connecting three anchor nodes. A sensor
node uses the APIT test with different combinations of three audible anchor nodes (audible
anchors are anchor nodes from which beacon packets are received) until all combinations
are exhausted. Each APIT test determines whether or not the node lies inside a distinct
triangular region. The intersection of all the triangular regions is then considered to estimate
the area in which the sensor is located. The APIT algorithm performs well when the average
number of audible anchors is high (for example, more than 20). As a result, a major
drawback of the algorithm is that it is highly computationally intensive. An average of 20
audible anchors would imply that the intersection of
20
C
3
= 1140 areas need to be considered.
Furthermore, the algorithm performs well only when the anchor nodes are randomly
distributed throughout the network, which is not always feasible in a real deployment
scenario.

3. Area Localization Scheme Fundamentals
In ALS, the nodes in the wireless sensor network are divided into three categories according
to their different functions: reference nodes, sensor nodes and sinks.

A. Reference/Anchor nodes
The main responsibility of the reference/anchor (both terms will be used interchangeably)
nodes is to send out beacon signals to help sensor nodes locate themselves. Reference nodes
are either equipped with GPS to provide accurate location information or placed in pre-
determined locations. In addition, the reference nodes can send out radio signals at varying
power levels as required. For an Ideal Isotropic Antenna, the received power at a distance d
from the transmitter is given by:

2
4
|
.
|
\
|
=
d
G G P P
r t t r
t

(1)

while the two-ray ground reflection model considers both the direct path and a ground
reflection path, and the received power at a distance d is given by:

4
2 2
d
P G G h h
P
t t r t r
r
=
for
t
r t
h h
d
4
>
(2)

where P
r
is the received power, P
t
is the transmitted power, d is the distance between the
transmitter and receiver, is the wavelength and, h
t
and h
r
are the heights of the transmitter
and receiver respectively. G
t
and G
r
represent the gains of the transmitter and receiver
respectively in equations (1) and (2).
From the above equations, it can be clearly seen that if the received power is fixed at a
certain value, the radio signal with a higher transmitted power reaches a greater distance.
Using one of the physical layer models described above and the threshold power that each
sensor can receive, the reference node can calculate the power required to reach different
distances. Each reference node then devises a set of increasing power levels such that the
highest power level can cover the entire area in consideration. The reference nodes then
broadcast several rounds of radio signals. The beacon packet contains the ID of the reference
node and the power level at which the signal is transmitted (which can be simply
represented by an integer value, as explained below.)
Let PS denote the set of increasing power levels of beacon signals sent out by a reference
node. For now, let us assume that all the reference nodes in the system send out the same set
PS of beacon signals. In the ALS scheme, a sensor node simply listens and records the power
levels of beacon signals it receives from each reference node. In real environments, small
scale fading and shadowing can cause the power levels received by the sensor nodes to vary
significantly from the expected power levels calculated by the path loss models in equations
(1) and (2). Sending out beacon signals in the set PS only once might lead to inaccurate
beacon reception by sensor nodes. As a result, the reference nodes send out the beacon
signals in set PS multiple times. The sensor nodes can then calculate the statistical average
(mode or mean) of the received power levels from each reference node.
Let the number of power levels in set PS be denoted by N
p
and the N
p
power levels in set PS
be represented by P
1
,P
2
, P
3
,,P
Np
. The power levels P
1
, P
2
, P
3
,,P
Np
can be represented by
simple integers, e.g. increasing values corresponding to increasing power levels; therefore
sensor nodes only need to take note of these integer values that are contained in the beacon
packets and the hardware design can be kept simple as there is no need for accurate
measurement of the received power level. Let the number of times that the same set of
beacon signals PS are sent out be denoted by N
r
, also referred to as the number of rounds.
The power MP in dB required to cover the entire area is calculated from equation (1) or (2),
based on the physical layer model in consideration. The power LP in dB required to cover a
small distance (say 10 m) is also calculated. The values P
1
,P
2
, P
3
,,P
Np
are then set to be
N
p
uniformly distributed values in the range [LP, MP] in the dB scale. The simple procedure
followed by the reference nodes is shown below:

1 for i = 1: Nr
2 for j=1: N
p

3 Send beacon signal at power levelP
j

4 end for
5 end for

The transmissions by the different reference nodes do not need to be synchronized.
However, the reference nodes schedule the beacon signal transmissions so to avoid
collisions. The transmitted set of power levels PS need not be the same for all the reference
nodes, and can be configured by the network administrator. Also, the set of power levels PS
need not be uniformly distributed too. It is also not necessary for the reference nodes to
know each others position and levels of transmitted power, but there should be at least one
sink or a central agent that stores the location information of all the reference nodes.

B. Sensor node
A sensor node is a unit device that monitors the environment. Sensors typically have limited
computing capability, storage capacity, communications range and battery power. Due to
power constraints, it is not desirable forsensor nodes to make complex calculations and send
out information frequently.

1) Signal Coordinate Representation:
In the ALS scheme, the sensors save a list of reference nodes and their respective transmitted
power levels and forward the information to the nearest sink when requested or appended
to sensed data. The sinks use this information to identify the area in which the sensors
reside in. However, if the number of reference nodes is large, the packets containing location
information may be long, which might result in more traffic in the network. A naming
scheme is hence designed.
The sensor nodes use a signal coordinate representation to indicate their location
information to the sinks. Power contour lines can be drawn on an area based on the set of
beacon signal power levels PS transmitted by each reference node, and their corresponding
distances travelled. The power contour lines divide the region in consideration into many
sub-regions (which we refer to as areas) as shown in Figure 1 below. Each area in the region
can be represented by a unique set of n coordinates, hereafter referred to as the signal
coordinate.
Suppose there are n reference nodes, which are referred to as R
1
, R
2, ,
and R
n.
For a sensor in
an area, let the lowest transmitted power levels it receives from the n reference nodes be S
1
,
S
2, ,
and S
n
respectively. S
1
, S
2, ,
and S
n
are simple integer numbers indicating the different
power levels rather than the actual signal strengths. The mappings between integer levels
and the actual power values are saved at the reference nodes and sinks. The signal
coordinate is defined as the representation < S
1
, S
2, ,
S
n
> such that each S
i
, the i
th
element, is
the lowest power level received from R
i
.
For example, consider a square region with reference nodes at the four corners, as shown in
Figure 1. In this case, the set of power levels PS is the same for all the four reference nodes
and there are three power levels in the set PS. The smallest power level in the power set PS
is represented by the integer 1 while the highest power level is represented by the integer 3.
For each node, the contour lines drawn on the region represent the farthest distances that
the beacon signals at each power level can travel. Contour lines for beacon power levels 1
and 2 are drawn. The power level 3 for each corner reference node can reach beyond the
corner that is diagonally opposite to it and so, its corresponding contour line is not seen on
the square region. Thus, for each reference node, the two contour lines corresponding to
power levels 1 and 2 divide the region into three (arc) areas.
Fig

Fo
fro
sig
the
rep
squ
sta
no
coo
Th
tra
<S
ne
coo
the

2)
In
rec
g. 1. Example of A
r a sensor node i
om reference nod
gnals at power lev
e sensor from re
presented by the
uare region can b
ated in the signa
ode i forms the
ordinate to identi
hus, if all the sens
ansmitted by ea
S
1
,S
2
,,S
n
> to ind
ed to get inform
ordinate in its req
eir own to see if t
Algorithm
the ALS scheme
cords the inform
ALS under ideal i
in the shaded are
des 1, 2 and 3 is
vels 2 and 3 from
eference node 4 i
e unique signal c
be represented by
l coordinate defi
i
th
element of t
ify the area in wh
sors and sinks ag
ch reference no
dicate their area l
mation from sen
quest and the sen
they lie in the rele
e, the sensor nod
ation that it rece
isotropic conditio
ea(lower right) in
3.The sensor no
m reference node 4
is 2. As a result,
coordinate <3,3,3
y a unique signal
inition, the lowe
the signal coord
hich they are loca
gree in advance o
ode, the sensor
location informat
nsors specific to
nsors simply com
evant area.
e simply listens
eives from them.
ons; shaded region
n Fig. 1, the lowes
ode in the shaded
4. So, the lowest p
, the shaded are
3,2>. Similarly, e
l coordinate, as s
st power level re
dinate. Sensors u
ted.
on the set(s) of b
nodes can use
tion to the sinks.
a certain area,
mpare the incomi
to signals from a
A sensor node a

n is <3, 3, 3, 2>
st power level re
d area receives b
power level receiv
ea in the figure c
every other area
shown in the figu
eceived from ref
use this unique
beacon power lev
the signal coor
Similarly, when
it includes the
ng signal coordin
all reference node
at a particular lo
eceived
beacon
ved by
can be
in the
ure. As
ference
signal
vels PS
rdinate
a sink
signal
nate to
es and
ocation
know each others position and levels of transmitted power, but there should be at least one
sink or a central agent that stores the location information of all the reference nodes.

B. Sensor node
A sensor node is a unit device that monitors the environment. Sensors typically have limited
computing capability, storage capacity, communications range and battery power. Due to
power constraints, it is not desirable forsensor nodes to make complex calculations and send
out information frequently.

1) Signal Coordinate Representation:
In the ALS scheme, the sensors save a list of reference nodes and their respective transmitted
power levels and forward the information to the nearest sink when requested or appended
to sensed data. The sinks use this information to identify the area in which the sensors
reside in. However, if the number of reference nodes is large, the packets containing location
information may be long, which might result in more traffic in the network. A naming
scheme is hence designed.
The sensor nodes use a signal coordinate representation to indicate their location
information to the sinks. Power contour lines can be drawn on an area based on the set of
beacon signal power levels PS transmitted by each reference node, and their corresponding
distances travelled. The power contour lines divide the region in consideration into many
sub-regions (which we refer to as areas) as shown in Figure 1 below. Each area in the region
can be represented by a unique set of n coordinates, hereafter referred to as the signal
coordinate.
Suppose there are n reference nodes, which are referred to as R
1
, R
2, ,
and R
n.
For a sensor in
an area, let the lowest transmitted power levels it receives from the n reference nodes be S
1
,
S
2, ,
and S
n
respectively. S
1
, S
2, ,
and S
n
are simple integer numbers indicating the different
power levels rather than the actual signal strengths. The mappings between integer levels
and the actual power values are saved at the reference nodes and sinks. The signal
coordinate is defined as the representation < S
1
, S
2, ,
S
n
> such that each S
i
, the i
th
element, is
the lowest power level received from R
i
.
For example, consider a square region with reference nodes at the four corners, as shown in
Figure 1. In this case, the set of power levels PS is the same for all the four reference nodes
and there are three power levels in the set PS. The smallest power level in the power set PS
is represented by the integer 1 while the highest power level is represented by the integer 3.
For each node, the contour lines drawn on the region represent the farthest distances that
the beacon signals at each power level can travel. Contour lines for beacon power levels 1
and 2 are drawn. The power level 3 for each corner reference node can reach beyond the
corner that is diagonally opposite to it and so, its corresponding contour line is not seen on
the square region. Thus, for each reference node, the two contour lines corresponding to
power levels 1 and 2 divide the region into three (arc) areas.
Fig

Fo
fro
sig
the
rep
squ
sta
no
coo
Th
tra
<S
ne
coo
the

2)
In
rec
g. 1. Example of A
r a sensor node i
om reference nod
gnals at power lev
e sensor from re
presented by the
uare region can b
ated in the signa
ode i forms the
ordinate to identi
hus, if all the sens
ansmitted by ea
S
1
,S
2
,,S
n
> to ind
ed to get inform
ordinate in its req
eir own to see if t
Algorithm
the ALS scheme
cords the inform
ALS under ideal i
in the shaded are
des 1, 2 and 3 is
vels 2 and 3 from
eference node 4 i
e unique signal c
be represented by
l coordinate defi
i
th
element of t
ify the area in wh
sors and sinks ag
ch reference no
dicate their area l
mation from sen
quest and the sen
they lie in the rele
e, the sensor nod
ation that it rece
isotropic conditio
ea(lower right) in
3.The sensor no
m reference node 4
is 2. As a result,
coordinate <3,3,3
y a unique signal
inition, the lowe
the signal coord
hich they are loca
gree in advance o
ode, the sensor
location informat
nsors specific to
nsors simply com
evant area.
e simply listens
eives from them.
ons; shaded region
n Fig. 1, the lowes
ode in the shaded
4. So, the lowest p
, the shaded are
3,2>. Similarly, e
l coordinate, as s
st power level re
dinate. Sensors u
ted.
on the set(s) of b
nodes can use
tion to the sinks.
a certain area,
mpare the incomi
to signals from a
A sensor node a

n is <3, 3, 3, 2>
st power level re
d area receives b
power level receiv
ea in the figure c
every other area
shown in the figu
eceived from ref
use this unique
beacon power lev
the signal coor
Similarly, when
it includes the
ng signal coordin
all reference node
at a particular lo
eceived
beacon
ved by
can be
in the
ure. As
ference
signal
vels PS
rdinate
a sink
signal
nate to
es and
ocation
may receive localization signals (beacon messages)at different power levels from the same
reference node, as explained above. The sensor records its signal coordinate and forwards
the information to the sink(s) using the existing data delivery scheme, as and when
requested.
Let the signal coordinate of a node be denoted <S
1
, S
2
,,S
n
> where n is the number of
reference nodes. A sensor node uses variables L
11
, L
12,
,L
1Nr
to represent the lowest power
levels received by the sensor from reference node 1 during rounds 1 to N
r.
Similarly, let
L
i1
, L
i2
,,L
iNr
represent the lowest power levels received by the sensor from reference node i
during rounds 1 to N
r.
Let the number of reference nodes be n. Initially, all the values
L
11
, L
12,
, L
1Nr
, L
21
, L
22
, , L
2Nr
, , L
n1
, L
n2
, , L
nNr
are set to zero. The zeros imply that the
sensor nodes have received no signals from the reference nodes.
The pseudo-code running on each sensor node is shown below. After initialization, the
sensor nodes start an infinite loop to receive beacon messages from reference nodes and
follow the algorithm shown below. Since a reference node sends out several rounds of
beacon signals, the sensor node may hear multiple rounds of beacon signals from the same
reference node. If the sensor receives a signal from reference node i for the first time during
round j, it sets L
ij
to be the lowest received power level for that round; otherwise, if the
received power level from reference node i in round j is lower than the current value in
L
ij
, L
ij
is set to the latest received power level. After all the reference nodes have completed
sending out beacon messages, the power levels L
i1
to L
iNr
on each sensor represent the lowest
power levels received from reference node i during rounds 1 to N
r
respectively.

Initialization:
1 for i=1 to n
2 for j = 1 to Nr
3 L
ij
= 0
4 end for
5 end for
Loop:
1 Receive a message
2 if (the message is from reference nodei during round j)
3 if (L
ij
= 0 || received power level <L
ij
) ; received power level integer
representation
4 L
ij
= received power level
5 end if
6 end if

Each reference node sends out beacon signals at all the power levels in the set PS N
r
times
(N
r
rounds). In real conditions, fading and shadowing can cause the power levels to vary
erratically about the expected signal strength predicted by the large scale fading model.
Hence, the lowest signal power level received by a sensor from a reference node need not be
the same for all the rounds 1 to N
r
, i.e. all the values L
i1
to L
iNr
need not be the same. One is
then faced with the problem of deciding which value L
ix
to pick as S
i,
the i
th
element of the
signal coordinate.
Hence, a threshold value CONFIDENCE_LEVEL is defined. This parameter represents the
confidence level with which the values S
1
, S
2
, , S
n
can be estimated, and is an operational
va
to
fre
i
th
fre
{L
i
fur

Fig
lue that end user
80% of N
r
in ou
equency greater t
element in the n
equency greater t
1
..L
iNr
} are consid
rther refinement m
g. 2. Illustration o
rs can specify to
ur performance
than CONFIDENC
nodes signal co
than CONFIDEN
dered possible ca
may be necessary
(a) Bla
(Black
(b) Blac
of Signal Coordin
suit their require
studies. If there
CE_LEVEL in the
oordinate, i.e. S
i
NCE_LEVEL, then
andidates of the i
y.
ack region <{2,3}
Red) regions <{
ck region <{2,3},
ate Representatio
ements. For exam
is a power leve
e set {L
i
, , L
iNr
},
= L
ix
. If there is
n all the distinct p
i
th
element of the
}, 3, 3, 3>.
1,2,3}, 3,3,3>
, 3, 3,{2,3}>
on
mple, this value w
el L
ix
that occur
then L
ix
is set to
s no power leve
power levels in t
signal coordinat

was set
s with
be the
el with
the set
te, and
may receive localization signals (beacon messages)at different power levels from the same
reference node, as explained above. The sensor records its signal coordinate and forwards
the information to the sink(s) using the existing data delivery scheme, as and when
requested.
Let the signal coordinate of a node be denoted <S
1
, S
2
,,S
n
> where n is the number of
reference nodes. A sensor node uses variables L
11
, L
12,
,L
1Nr
to represent the lowest power
levels received by the sensor from reference node 1 during rounds 1 to N
r.
Similarly, let
L
i1
, L
i2
,,L
iNr
represent the lowest power levels received by the sensor from reference node i
during rounds 1 to N
r.
Let the number of reference nodes be n. Initially, all the values
L
11
, L
12,
, L
1Nr
, L
21
, L
22
, , L
2Nr
, , L
n1
, L
n2
, , L
nNr
are set to zero. The zeros imply that the
sensor nodes have received no signals from the reference nodes.
The pseudo-code running on each sensor node is shown below. After initialization, the
sensor nodes start an infinite loop to receive beacon messages from reference nodes and
follow the algorithm shown below. Since a reference node sends out several rounds of
beacon signals, the sensor node may hear multiple rounds of beacon signals from the same
reference node. If the sensor receives a signal from reference node i for the first time during
round j, it sets L
ij
to be the lowest received power level for that round; otherwise, if the
received power level from reference node i in round j is lower than the current value in
L
ij
, L
ij
is set to the latest received power level. After all the reference nodes have completed
sending out beacon messages, the power levels L
i1
to L
iNr
on each sensor represent the lowest
power levels received from reference node i during rounds 1 to N
r
respectively.

Initialization:
1 for i=1 to n
2 for j = 1 to Nr
3 L
ij
= 0
4 end for
5 end for
Loop:
1 Receive a message
2 if (the message is from reference nodei during round j)
3 if (L
ij
= 0 || received power level <L
ij
) ; received power level integer
representation
4 L
ij
= received power level
5 end if
6 end if

Each reference node sends out beacon signals at all the power levels in the set PS N
r
times
(N
r
rounds). In real conditions, fading and shadowing can cause the power levels to vary
erratically about the expected signal strength predicted by the large scale fading model.
Hence, the lowest signal power level received by a sensor from a reference node need not be
the same for all the rounds 1 to N
r
, i.e. all the values L
i1
to L
iNr
need not be the same. One is
then faced with the problem of deciding which value L
ix
to pick as S
i,
the i
th
element of the
signal coordinate.
Hence, a threshold value CONFIDENCE_LEVEL is defined. This parameter represents the
confidence level with which the values S
1
, S
2
, , S
n
can be estimated, and is an operational
va
to
fre
i
th
fre
{L
i
fur

Fig
lue that end user
80% of N
r
in ou
equency greater t
element in the n
equency greater t
1
..L
iNr
} are consid
rther refinement m
g. 2. Illustration o
rs can specify to
ur performance
than CONFIDENC
nodes signal co
than CONFIDEN
dered possible ca
may be necessary
(a) Bla
(Black
(b) Blac
of Signal Coordin
suit their require
studies. If there
CE_LEVEL in the
oordinate, i.e. S
i
NCE_LEVEL, then
andidates of the i
y.
ack region <{2,3}
Red) regions <{
ck region <{2,3},
ate Representatio
ements. For exam
is a power leve
e set {L
i
, , L
iNr
},
= L
ix
. If there is
n all the distinct p
i
th
element of the
}, 3, 3, 3>.
1,2,3}, 3,3,3>
, 3, 3,{2,3}>
on
mple, this value w
el L
ix
that occur
then L
ix
is set to
s no power leve
power levels in t
signal coordinat

was set
s with
be the
el with
the set
te, and
This concept is further illustrated by a couple of examples and we assume the same scenario
as in Fig. 1. In Fig. 1, we have assumed ideal isotropic channel conditions and each element
in the signal coordinate has been ascertained with a high confidence level.

Fig. 2 illustrates scenarios of non-ideal channel conditions where beacons messages may be lost.
Fig. 2(a) shows the case <{2,3},3,3,3>, where the first element of the signal coordinate is either
1 or 2. This happens when the lowest power level received from reference 1 during the N
r

rounds of beacon messages oscillates between 1 and 2. Both values (1 and 2) can be
considered as possible candidates for S
1,
if no power level L
1x
occurs with frequency greater
than CONFIDENCE_LEVEL in the set {L
11
, ..., L
1Nr
}. The union of the black and red regions in
Fig. 2(a) represents the region <0, 3, 3, 3>, where the value of 0 implies that there is no
information available on the first element. This could happen in the case when no beacon
packets are received from reference node 1, and the signal coordinate region <{1,2,3},3,3,3>
is considered as a result. Thus, every element S
i
in the set <S
1
, S
2
, ,S
n
> need not be a
unique value, but could be a set of values as shown in Fig. 2(b). While more than one
element of a signal coordinate may have multiple values, we consider a signal coordinate to
be valid only if at least half of its values have been determined with a high confidence level.
From the above description, it can be clearly seen that the sensor nodes do not perform any
complicated calculations to estimate their location. Neither do they need to exchange
information with their neighbours.

C. Sink
In wireless sensor networks, data from sensor nodes are forwarded to a sink for processing.
From a hardware point of view, a sink usually has much higher computing and data
processing capabilities than a sensor node. In ALS, a sensor node sends its signal coordinate
(location information) to a sink according to the data delivery scheme in use. The sensor
itself does not know the exact location of the area in which it resides nor does it know what
its signal coordinate represents. It is up to the sink(s) to determine the sensors location
based on the signal coordinate information obtained from the sensor. One assumption of the
ALS scheme is that the sink knows the positions of all the reference nodes and their
respective transmitted power levels, whether by directly communicating with the reference
nodes, or from a central server, which contains this information. Therefore, with the
knowledge of the physical layer model and signal propagation algorithms, the sink is able to
derive the map of areas based on the information of the transmitted signals from the
reference nodes. With the map and the signal coordinate information, the sink can then
determine which area a sensor is in from the received data, tagged with the signal
coordinate.
In the ALS scheme, the choosing of the signal propagation model plays an important part in
the estimation accuracy. For different networks, different signal propagation models can be
used to draw out the signal map according to the physical layer conditions. An irregular signal
model could divide the whole region into many differently shaped areas, as shown in Fig. 3.
Any adjustments made to the underlying physical layer model will have no impact on the
sensor nodes, which just need to measure their signal coordinates and forward them to the
sink. An immediate observation is the diverse area granularity, which affects the accuracy of
the location estimation. The granularity issue will be discussed in the next section.
A key advantage of ALS is its simplicity for the sensors with all the complex calculations
done by the sink. Thus, the localization process consumes little power at the sensor nodes,
helps to extend the life of the whole network. Furthermore, it has a covert feature whereby
anyone eavesdropping on the transmission will not be able to infer the location of sensors
from the signal coordinates contained in the packets.

Fig. 3. Irregular contour lines arising from a non-ideal signal model

4. Performance Evaluation of ALS
We evaluate ALS using simulations as well as field experimentation using commercially
available wireless sensor nodes.

A. Performance metrics for ALS
The metrics, accuracy and granularity, are used to evaluate the performance of the scheme.
High levels of accuracy and granularity are desired; however, accuracy begins to suffer as
granularity increases, since the probability of estimating the location of a node correctly in a
smaller area decreases. Hence, in order to have a fair evaluation of ALS, we normalize the
accuracy with respect to the granularity or average area estimate, that is, normalized accuracy
= accuracy / average area estimate.
Another metric, average error, is defined to compare the performance of ALS to other range
free schemes. The Center of Gravity (COG) or centroid of the final area estimate is assumed
to be location of the node. Average error is then defined to be the average of the Euclidian
distances between the original and estimated locations for all the nodes in the network.

B. Simulation scenario and parameters
The QUALNET 3.8 simulation environment is used to evaluate the performance of ALS. The
system parameters used in our simulations are described below.
This concept is further illustrated by a couple of examples and we assume the same scenario
as in Fig. 1. In Fig. 1, we have assumed ideal isotropic channel conditions and each element
in the signal coordinate has been ascertained with a high confidence level.

Fig. 2 illustrates scenarios of non-ideal channel conditions where beacons messages may be lost.
Fig. 2(a) shows the case <{2,3},3,3,3>, where the first element of the signal coordinate is either
1 or 2. This happens when the lowest power level received from reference 1 during the N
r

rounds of beacon messages oscillates between 1 and 2. Both values (1 and 2) can be
considered as possible candidates for S
1,
if no power level L
1x
occurs with frequency greater
than CONFIDENCE_LEVEL in the set {L
11
, ..., L
1Nr
}. The union of the black and red regions in
Fig. 2(a) represents the region <0, 3, 3, 3>, where the value of 0 implies that there is no
information available on the first element. This could happen in the case when no beacon
packets are received from reference node 1, and the signal coordinate region <{1,2,3},3,3,3>
is considered as a result. Thus, every element S
i
in the set <S
1
, S
2
, ,S
n
> need not be a
unique value, but could be a set of values as shown in Fig. 2(b). While more than one
element of a signal coordinate may have multiple values, we consider a signal coordinate to
be valid only if at least half of its values have been determined with a high confidence level.
From the above description, it can be clearly seen that the sensor nodes do not perform any
complicated calculations to estimate their location. Neither do they need to exchange
information with their neighbours.

C. Sink
In wireless sensor networks, data from sensor nodes are forwarded to a sink for processing.
From a hardware point of view, a sink usually has much higher computing and data
processing capabilities than a sensor node. In ALS, a sensor node sends its signal coordinate
(location information) to a sink according to the data delivery scheme in use. The sensor
itself does not know the exact location of the area in which it resides nor does it know what
its signal coordinate represents. It is up to the sink(s) to determine the sensors location
based on the signal coordinate information obtained from the sensor. One assumption of the
ALS scheme is that the sink knows the positions of all the reference nodes and their
respective transmitted power levels, whether by directly communicating with the reference
nodes, or from a central server, which contains this information. Therefore, with the
knowledge of the physical layer model and signal propagation algorithms, the sink is able to
derive the map of areas based on the information of the transmitted signals from the
reference nodes. With the map and the signal coordinate information, the sink can then
determine which area a sensor is in from the received data, tagged with the signal
coordinate.
In the ALS scheme, the choosing of the signal propagation model plays an important part in
the estimation accuracy. For different networks, different signal propagation models can be
used to draw out the signal map according to the physical layer conditions. An irregular signal
model could divide the whole region into many differently shaped areas, as shown in Fig. 3.
Any adjustments made to the underlying physical layer model will have no impact on the
sensor nodes, which just need to measure their signal coordinates and forward them to the
sink. An immediate observation is the diverse area granularity, which affects the accuracy of
the location estimation. The granularity issue will be discussed in the next section.
A key advantage of ALS is its simplicity for the sensors with all the complex calculations
done by the sink. Thus, the localization process consumes little power at the sensor nodes,
helps to extend the life of the whole network. Furthermore, it has a covert feature whereby
anyone eavesdropping on the transmission will not be able to infer the location of sensors
from the signal coordinates contained in the packets.

Fig. 3. Irregular contour lines arising from a non-ideal signal model

4. Performance Evaluation of ALS
We evaluate ALS using simulations as well as field experimentation using commercially
available wireless sensor nodes.

A. Performance metrics for ALS
The metrics, accuracy and granularity, are used to evaluate the performance of the scheme.
High levels of accuracy and granularity are desired; however, accuracy begins to suffer as
granularity increases, since the probability of estimating the location of a node correctly in a
smaller area decreases. Hence, in order to have a fair evaluation of ALS, we normalize the
accuracy with respect to the granularity or average area estimate, that is, normalized accuracy
= accuracy / average area estimate.
Another metric, average error, is defined to compare the performance of ALS to other range
free schemes. The Center of Gravity (COG) or centroid of the final area estimate is assumed
to be location of the node. Average error is then defined to be the average of the Euclidian
distances between the original and estimated locations for all the nodes in the network.

B. Simulation scenario and parameters
The QUALNET 3.8 simulation environment is used to evaluate the performance of ALS. The
system parameters used in our simulations are described below.
Region of deployment: Square of size 500m 500m.
Physical layer: For the ideal case, it is modelled by the two-ray model given in equation
(2). In the non-ideal case, Rayleigh fading and lognormal shadowing are also factored
into the two-ray model.
Node placement: A wireless sensor network with 500 nodes (eight of which are reference
nodes) is used. The sensors are placed randomly throughout the region, and the eight
reference nodes are positioned at the four corners and the four mid points of the sides of
the square region. Although there are eight reference nodes, only four transmit beacon
signals during each round of ALS. The sensor nodes in the network are assumed to be
static, and the maximum velocity of objects in the surrounding is set to 1 m/s.
Reference-to-Node Range ratio (RNR): This parameter refers to the average distance a
reference beacon signal travels divided by the average distance a regular node signal
travels. The radio range of sensors is set to 50 m, while the radio range of reference nodes
is set to 1000 m, which is large enough for the beacon signals to cover the entire area.
Therefore, the RNR value is 20.
Node Density (ND): The node density refers to the average number of nodes within a
nodes radio transmission area. This value is close to 13 for the network scenario in
consideration.
Reference Node Percentage (RNP): The reference node percentage refers to the number of
reference nodes divided by the total number of nodes. In our case, the system has a low
RNP of 1.6% (8/500).
Receiver Threshold Power: The receiver threshold power refers to the lowest signal
strength of a packet that a node can receive. The value is set to -85 dBm.
N
r
: Number of times each beacon signal is sent out by a reference node. This parameter is
set to 20.
CONFIDENCE_LEVEL: 80%.

C. Simulation studyof ALS under ideal conditions
LP is set to -13 dBm and MP is set to 17 dBm. The number of power levels is then increased
from 3 to 7 and the performance of the scheme is observed. All the sensors lie in their
estimated areas as the experiment is carried out under ideal conditions. On the other hand,
the granularity increases as the average area estimate decreases (Table 1), and as a result, the
normalized accuracy metric improves, shown in Fig 4.

Ideal conditions
Iteration
No.
No. of power
levels
LP
(dBm)
MP(dB
m)
Avg. Area Est. as
% of area size
% nodes that lie in
their estimated area
1 3 -13 17 58.5 100
2 4 -13 17 17.4 100
3 5 -13 17 8.3 100
4 6 -13 17 5.8 100
5 7 -13 17 4.6 100
Table 1. Ideal case granularity increases as the number of power levels increases.

Fig. 4. Ideal case: Normalized Accuracy (accuracy/granularity) vs. Number of power levels.

D. Simulation study of ALS under non-ideal conditions
We first demonstrate the impact of decreasing the difference in adjacent power levels on the
signal coordinates measured by the sensors. A signal coordinate <S
1
, S
2
, S
3
, S
4
> is considered
to be valid only if at least two of the four elements S
i
can be measured with a confidence
level of 80%. The measured signal coordinate is considered wrong if any valid element, S
i
,
differs from the actual value.
LP is set to -13dBm, while MP is set to 17dBm, as in the ideal case, and the number of power
levels is increased from 3 to 7. The difference in adjacent power levels is (MP-LP)/(N
p
-1). For
example, when N
p
is set to 3, the three power levels are -13 dBm, 2 dBm and 17 dBm, and the
difference in adjacent power levels is 15 dBm.
It is observed that the percentage of nodes that measure their signal coordinate correctly
decreases from 96% to 28% as the number of power levels increases from 3 to 7. Fading and
shadowing can cause the received signal strength to vary by as much as +10 dBm to -30
dBm of the expected value. The variance in measured signal coordinate increases, as the
fading effect causes the received signal strength to vary by much more than the difference in
adjacent power levels. As a result, fewer signal coordinates are measured correctly with a
high confidence level (Fig. 5.). Nodes that were close to the edges of regions in the area were
more prone to error than the nodes that are in the centre a region.
Number of Power Levels
N
o
r
m
a
l
i
z
e
d

A
c
c
u
r
a
c
y

Region of deployment: Square of size 500m 500m.
Physical layer: For the ideal case, it is modelled by the two-ray model given in equation
(2). In the non-ideal case, Rayleigh fading and lognormal shadowing are also factored
into the two-ray model.
Node placement: A wireless sensor network with 500 nodes (eight of which are reference
nodes) is used. The sensors are placed randomly throughout the region, and the eight
reference nodes are positioned at the four corners and the four mid points of the sides of
the square region. Although there are eight reference nodes, only four transmit beacon
signals during each round of ALS. The sensor nodes in the network are assumed to be
static, and the maximum velocity of objects in the surrounding is set to 1 m/s.
Reference-to-Node Range ratio (RNR): This parameter refers to the average distance a
reference beacon signal travels divided by the average distance a regular node signal
travels. The radio range of sensors is set to 50 m, while the radio range of reference nodes
is set to 1000 m, which is large enough for the beacon signals to cover the entire area.
Therefore, the RNR value is 20.
Node Density (ND): The node density refers to the average number of nodes within a
nodes radio transmission area. This value is close to 13 for the network scenario in
consideration.
Reference Node Percentage (RNP): The reference node percentage refers to the number of
reference nodes divided by the total number of nodes. In our case, the system has a low
RNP of 1.6% (8/500).
Receiver Threshold Power: The receiver threshold power refers to the lowest signal
strength of a packet that a node can receive. The value is set to -85 dBm.
N
r
: Number of times each beacon signal is sent out by a reference node. This parameter is
set to 20.
CONFIDENCE_LEVEL: 80%.

C. Simulation studyof ALS under ideal conditions
LP is set to -13 dBm and MP is set to 17 dBm. The number of power levels is then increased
from 3 to 7 and the performance of the scheme is observed. All the sensors lie in their
estimated areas as the experiment is carried out under ideal conditions. On the other hand,
the granularity increases as the average area estimate decreases (Table 1), and as a result, the
normalized accuracy metric improves, shown in Fig 4.

Ideal conditions
Iteration
No.
No. of power
levels
LP
(dBm)
MP(dB
m)
Avg. Area Est. as
% of area size
% nodes that lie in
their estimated area
1 3 -13 17 58.5 100
2 4 -13 17 17.4 100
3 5 -13 17 8.3 100
4 6 -13 17 5.8 100
5 7 -13 17 4.6 100
Table 1. Ideal case granularity increases as the number of power levels increases.

Fig. 4. Ideal case: Normalized Accuracy (accuracy/granularity) vs. Number of power levels.

D. Simulation study of ALS under non-ideal conditions
We first demonstrate the impact of decreasing the difference in adjacent power levels on the
signal coordinates measured by the sensors. A signal coordinate <S
1
, S
2
, S
3
, S
4
> is considered
to be valid only if at least two of the four elements S
i
can be measured with a confidence
level of 80%. The measured signal coordinate is considered wrong if any valid element, S
i
,
differs from the actual value.
LP is set to -13dBm, while MP is set to 17dBm, as in the ideal case, and the number of power
levels is increased from 3 to 7. The difference in adjacent power levels is (MP-LP)/(N
p
-1). For
example, when N
p
is set to 3, the three power levels are -13 dBm, 2 dBm and 17 dBm, and the
difference in adjacent power levels is 15 dBm.
It is observed that the percentage of nodes that measure their signal coordinate correctly
decreases from 96% to 28% as the number of power levels increases from 3 to 7. Fading and
shadowing can cause the received signal strength to vary by as much as +10 dBm to -30
dBm of the expected value. The variance in measured signal coordinate increases, as the
fading effect causes the received signal strength to vary by much more than the difference in
adjacent power levels. As a result, fewer signal coordinates are measured correctly with a
high confidence level (Fig. 5.). Nodes that were close to the edges of regions in the area were
more prone to error than the nodes that are in the centre a region.
Number of Power Levels
N
o
r
m
a
l
i
z
e
d

A
c
c
u
r
a
c
y


Fig
lev

Fro
be
AL
set
Ta
no
at
reg
dis
the
the
the
com
car
as
Th
g. 5. Percentage o
vels
om the above dis
set as large as p
LS, with the differ
t PS having thre
able 2. We consid
odes at the four co
the mid-points o
gion after the refe
stinct set of powe
e areas obtained
e rounds complet
e area estimate. T
mbination of man
rried out under b
a benchmark to c
he results obtained
P
e
r
c
e
n
t
a
g
e

o
f

N
o
d
e
s

(
%
)

of nodes that me
scussion, it is evi
ossible. We there
rence between ad
ee distinct power
der 10 rounds in
orners send out b
of the four sides
erence nodes sen
er contour lines.
from each round
ted do not interse
Thus, the final ar
ny small regions
both ideal and no
compare how we
d are shown in Ta
easure signal coo
dent that the diff
efore use a differe
djacent power lev
r levels. The LP
our simulations.
beacon signals, an
send out beacon
d out six sets of b
The final area es
d of the beaconing
ect, the largest int
ea estimate of ea
in the final area
on-ideal condition
ell the ALS schem
able 2 and Fig. 7.
Number of Power L
ordinate correctly
ference in adjacen
ent set of power
vels being set to 1
and MP for eac
For the first five
nd for the next fi
n signals. For exa
beacon signals. E
stimate of a senso
g process. If the a
tersecting area ob
ach sensor node i
shown in Fig. 6.
ns. The ideal con
me performs unde

Levels
y vs Number of
nt power levels s
levels in each rou
15 to 20 dBm, wit
h round are sho
e rounds, the ref
ve, the reference
ample, Fig. 6 show
ach colour repres
or is the intersect
areas obtained fr
btained is conside
s one small regio
The experiment i
nditions scenario
er non-ideal cond

power
should
und of
th each
own in
ference
nodes
ws the
sents a
tion of
rom all
ered as
on or a
is then
serves
ditions.

Fig. 6. Region after six rounds of ALS; each color represents a set of power levels.

For the non-ideal case, as the number of rounds increases from 1 to 10, the accuracy decreases
from 98.47% to 69.9% (Table 2). The accuracy drops because a wrong signal coordinate
measured in any one round of ALS would result in the final area being estimated incorrectly,
as the intersection of areas from all rounds is considered in the final area estimate. On the
other hand, granularity increases as the average area estimate decreases from 59.99% of the
area size to 1.33% of the area size (Table 2) and this is a consequence of the intersection area for
each sensor becoming smaller and smaller as the number of rounds increases.

Ideal two-ray
conditions
Non-ideal two-ray conditions
Number
of rounds
finished
No. of
power
levels
LP
(dBm)
MP
(dBm)
Avg. Area
Est. as % of
area size
%nodes
correctly
localized
Avg. Area
Est. as % of
area size
% nodes
correctly
localized
% of nodes
that lie 1-
hop away
1 3 -16 30 49.76 100 59.99 98.47 1.53
2 3 -14 30 30.06 100 42.17 94.39 5.61
3 3 -13 30 19.26 100 31.18 89.29 10.71
4 3 -11 30 5.10 100 11.30 84.18 15.82
5 3 -9 30 2.80 100 5.58 82.65 9.69
6 3 -16 30 1.55 100 3.97 81.63 10.71
7 3 -14 30 1.17 100 3.09 77.55 10.71
8 3 -13 30 0.96 100 2.49 75.51 12.76
9 3 -11 30 0.71 100 1.72 72.96 18.37
10 3 -9 30 0.60 100 1.33 69.90 21.9
Table 2. Data and results for the non-ideal case

Fig
lev

Fro
be
AL
set
Ta
no
at
reg
dis
the
the
the
com
car
as
Th
g. 5. Percentage o
vels
om the above dis
set as large as p
LS, with the differ
t PS having thre
able 2. We consid
odes at the four co
the mid-points o
gion after the refe
stinct set of powe
e areas obtained
e rounds complet
e area estimate. T
mbination of man
rried out under b
a benchmark to c
he results obtained
P
e
r
c
e
n
t
a
g
e

o
f

N
o
d
e
s

(
%
)

of nodes that me
scussion, it is evi
ossible. We there
rence between ad
ee distinct power
der 10 rounds in
orners send out b
of the four sides
erence nodes sen
er contour lines.
from each round
ted do not interse
Thus, the final ar
ny small regions
both ideal and no
compare how we
d are shown in Ta
easure signal coo
dent that the diff
efore use a differe
djacent power lev
r levels. The LP
our simulations.
beacon signals, an
send out beacon
d out six sets of b
The final area es
d of the beaconing
ect, the largest int
ea estimate of ea
in the final area
on-ideal condition
ell the ALS schem
able 2 and Fig. 7.
Number of Power L
ordinate correctly
ference in adjacen
ent set of power
vels being set to 1
and MP for eac
For the first five
nd for the next fi
n signals. For exa
beacon signals. E
stimate of a senso
g process. If the a
tersecting area ob
ach sensor node i
shown in Fig. 6.
ns. The ideal con
me performs unde

Levels
y vs Number of
nt power levels s
levels in each rou
15 to 20 dBm, wit
h round are sho
e rounds, the ref
ve, the reference
ample, Fig. 6 show
ach colour repres
or is the intersect
areas obtained fr
btained is conside
s one small regio
The experiment i
nditions scenario
er non-ideal cond

power
should
und of
th each
own in
ference
nodes
ws the
sents a
tion of
rom all
ered as
on or a
is then
serves
ditions.

Fig. 6. Region after six rounds of ALS; each color represents a set of power levels.

For the non-ideal case, as the number of rounds increases from 1 to 10, the accuracy decreases
from 98.47% to 69.9% (Table 2). The accuracy drops because a wrong signal coordinate
measured in any one round of ALS would result in the final area being estimated incorrectly,
as the intersection of areas from all rounds is considered in the final area estimate. On the
other hand, granularity increases as the average area estimate decreases from 59.99% of the
area size to 1.33% of the area size (Table 2) and this is a consequence of the intersection area for
each sensor becoming smaller and smaller as the number of rounds increases.

Ideal two-ray
conditions
Non-ideal two-ray conditions
Number
of rounds
finished
No. of
power
levels
LP
(dBm)
MP
(dBm)
Avg. Area
Est. as % of
area size
%nodes
correctly
localized
Avg. Area
Est. as % of
area size
% nodes
correctly
localized
% of nodes
that lie 1-
hop away
1 3 -16 30 49.76 100 59.99 98.47 1.53
2 3 -14 30 30.06 100 42.17 94.39 5.61
3 3 -13 30 19.26 100 31.18 89.29 10.71
4 3 -11 30 5.10 100 11.30 84.18 15.82
5 3 -9 30 2.80 100 5.58 82.65 9.69
6 3 -16 30 1.55 100 3.97 81.63 10.71
7 3 -14 30 1.17 100 3.09 77.55 10.71
8 3 -13 30 0.96 100 2.49 75.51 12.76
9 3 -11 30 0.71 100 1.72 72.96 18.37
10 3 -9 30 0.60 100 1.33 69.90 21.9
Table 2. Data and results for the non-ideal case

(a) Normalized accuracy starts to flatten out as the number of rounds increases

(b) Average error decreases as the number of rounds increases.
Fig. 7. ALS performance after multiple rounds

A
v
e
r
a
g
e

E
r
r
o
r

(
R
)

Number of rounds
Number of rounds
N
o
r
m
a
l
i
z
e
d

A
c
c
u
r
a
c
y

For the non-ideal case, the normalized accuracy metric improves, and starts to flatten out as
the number of rounds increases. The performance metric increases as the decrease in
average area estimate is greater than the decrease in accuracy after each additional round of
ALS. The performance flattens out because of the quantization of power levels, and the
constraint of maintaining a significant difference between adjacent power levels. ALS can be
stopped once desired accuracy levels and granularity are obtained. The desired average area
estimate and accuracy level, as well as the computational complexity of performing an extra
round, with the increased overhead in beacon messages should all be taken into account
before an additional round is executed. After 10 rounds, the average error drops below 0.5*R
(where R is the Radio Range of a sensor) for both the ideal and non-ideal conditions
(Fig. 7(b)).

E. One-hop Neighbourhood
Nodes that are closer to contour line boundaries are more prone to have their signal
coordinates measured wrongly. An analysis was carried out to investigate the error patterns
of nodes that did not lie in their estimated areas. It was observed that nodes, whose
locations were estimated incorrectly, very often lie in an adjacent area to their actual location
area.
Let the average area estimate of the nodes in the sensor network be denoted by A (for
example, A = 1.33% of region size at the end of 10 rounds in our simulation). The area
estimate of each node can then be approximated by a circle of area A (of radius
(A/)).
Circles with radius
(A/) and 2
(A/) are drawn from the estimated location of the node.

The circular ring between radii
(A/) and 2
(A/) is defined as the one-hop

neighbourhood region of the node. This concept of one-hop neighbourhood is illustrated
with an example in Fig. 8. Referring to Table 2, we observe that the average area estimate is
large for the first four rounds. As a result, all nodes lie within their estimated area or in the
one-hop neighbourhood. As more rounds of ALS are executed, the accuracy decreases and
the number of nodes that fall in the one-hop neighbourhood increases from 9.69% to 21.9%.
It can be seen that, when A = 1.33%, more than 90% of nodes either lie in their estimated
areas or in an area one-hop away.
The significance of the one-hop neighbourhood lies in various application scenarios that
ALS can be applied to. Consider an application scenario where a particular sensor in the
network detects an event and an unmanned vehicle is sent to the area (estimated by ALS) to
investigate. If the vehicle fails to find the sensor in the estimated area, it would then expand
its search in the surrounding areas that are one-hop away, two-hop away and so on. The
chance of finding the sensor within a one-hop range of the estimated area is very high (>
90%), as evident from Table 2.

F. Comparison with other range-free schemes
In this section, we compare ALS against other range-free localization schemes proposed for
wireless sensor networks. The range-free area localization schemes and range-free distance
vector based localization schemes chosen for comparison with ALS are the following: PIT
(Point in Triangle) and APIT (Approximate Point in Triangle) schemes [4], DV-Hop [5], and
DHL [6]. For a fair comparison, the chosen algorithms share a common set of system
parameters described in Section 4.B. The results obtained after ten rounds of ALS are
compared to the other two categories of range-free schemes.

(a) Normalized accuracy starts to flatten out as the number of rounds increases

(b) Average error decreases as the number of rounds increases.
Fig. 7. ALS performance after multiple rounds

A
v
e
r
a
g
e

E
r
r
o
r

(
R
)

Number of rounds
Number of rounds
N
o
r
m
a
l
i
z
e
d

A
c
c
u
r
a
c
y

For the non-ideal case, the normalized accuracy metric improves, and starts to flatten out as
the number of rounds increases. The performance metric increases as the decrease in
average area estimate is greater than the decrease in accuracy after each additional round of
ALS. The performance flattens out because of the quantization of power levels, and the
constraint of maintaining a significant difference between adjacent power levels. ALS can be
stopped once desired accuracy levels and granularity are obtained. The desired average area
estimate and accuracy level, as well as the computational complexity of performing an extra
round, with the increased overhead in beacon messages should all be taken into account
before an additional round is executed. After 10 rounds, the average error drops below 0.5*R
(where R is the Radio Range of a sensor) for both the ideal and non-ideal conditions
(Fig. 7(b)).

E. One-hop Neighbourhood
Nodes that are closer to contour line boundaries are more prone to have their signal
coordinates measured wrongly. An analysis was carried out to investigate the error patterns
of nodes that did not lie in their estimated areas. It was observed that nodes, whose
locations were estimated incorrectly, very often lie in an adjacent area to their actual location
area.
Let the average area estimate of the nodes in the sensor network be denoted by A (for
example, A = 1.33% of region size at the end of 10 rounds in our simulation). The area
estimate of each node can then be approximated by a circle of area A (of radius
(A/)).
Circles with radius
(A/) and 2
(A/) are drawn from the estimated location of the node.

The circular ring between radii
(A/) and 2
(A/) is defined as the one-hop

neighbourhood region of the node. This concept of one-hop neighbourhood is illustrated
with an example in Fig. 8. Referring to Table 2, we observe that the average area estimate is
large for the first four rounds. As a result, all nodes lie within their estimated area or in the
one-hop neighbourhood. As more rounds of ALS are executed, the accuracy decreases and
the number of nodes that fall in the one-hop neighbourhood increases from 9.69% to 21.9%.
It can be seen that, when A = 1.33%, more than 90% of nodes either lie in their estimated
areas or in an area one-hop away.
The significance of the one-hop neighbourhood lies in various application scenarios that
ALS can be applied to. Consider an application scenario where a particular sensor in the
network detects an event and an unmanned vehicle is sent to the area (estimated by ALS) to
investigate. If the vehicle fails to find the sensor in the estimated area, it would then expand
its search in the surrounding areas that are one-hop away, two-hop away and so on. The
chance of finding the sensor within a one-hop range of the estimated area is very high (>
90%), as evident from Table 2.

F. Comparison with other range-free schemes
In this section, we compare ALS against other range-free localization schemes proposed for
wireless sensor networks. The range-free area localization schemes and range-free distance
vector based localization schemes chosen for comparison with ALS are the following: PIT
(Point in Triangle) and APIT (Approximate Point in Triangle) schemes [4], DV-Hop [5], and
DHL [6]. For a fair comparison, the chosen algorithms share a common set of system
parameters described in Section 4.B. The results obtained after ten rounds of ALS are
compared to the other two categories of range-free schemes.

Fig. 8. One-hop neighbourhood: green area represents the estimated area of a node in the
final area, while the surrounding red area represents the corresponding one-hop
neighbourhood

1) Comparison with area based scheme:APIT (Approximate Point in Triangle)
In the PIT and APIT schemes[4], a node chooses three reference nodes from all audible
reference nodes (reference nodes from which a beacon was received) and tests whether it is
inside the triangle formed by connecting these three reference nodes. The theoretical method
used to determine whether a point is inside a triangle or not is called the Point-In-Triangle
(PIT) test. The PIT test can be carried out only under ideal physical layer conditions, when
every node in the network is mobile can move around its own position. Due to the
infeasibility of conducting such a test, an APIT (Approximate Point in Triangle) test is
proposed. The APIT uses RSSI information of beacon signals to determine whether it is
inside or outside a given triangle. The PIT or APIT tests are carried out with different
audible reference node combinations until all combinations are exhausted. The information
is then processed by a central server to narrow down the possible area that a target node
resides in. An area scan aggregation algorithm is used to determine the intersection of the
areas and determine the final area estimate of the node.
Fig. 9 shows all the possible triangles for the given configuration of the eight reference
nodes. There are 52 triangles in total (
8
C
3
4). The sensor nodes determine whether they are
in or out of each of the 52 triangles, and the final area estimate computed is a small region or
combination of regions on the area. Since PIT and APIT are area localization schemes, their
performance are compared with ALS using the normalized accuracy metric. The following
five cases are compared and results shown in Fig. 10:

i) ALS under ideal physical layer conditions after six rounds
ii) PIT under ideal physical layer conditions
iii) APIT under ideal physical layer conditions
iv) ALS under non-ideal physical layer conditions after six rounds
v) APIT under non-ideal physical layer conditions

Fig. 9. All possible triangles for PIT and APIT schemes with 8 reference nodes: 4 at the
corners and 4 at the mid-points of sides. There are 52 triangles in total.

The PIT and APIT schemes are carried out under ideal conditions to establish the
performance limits that can be achieved with the APIT algorithm under non-ideal conditions.
For the given scenario, it is observed (as shown in Fig. 7(a)) that ALS under ideal conditions
outperforms both PIT and APIT after just six rounds.
Not all APIT tests yield correct results, even under ideal physical layer conditions. As a
result, the performance of APIT under ideal conditions is slightly lower than PIT, due to
lower accuracy levels. Under non-ideal conditions, it is observed that ALS performs much
better than APIT. This is primarily because fluctuating RSSI values causes a number of APIT
tests to be incorrect. It is also observed that only around 60% of the 52 APIT tests are correct
for each sensor. This results in large area estimates on the network area. Thus, lower
accuracy levels and higher area estimates cause the performance of the APIT scheme to
suffer. ALS, on the other hand, is more resilient to fading and shadowing due to the
significant difference in adjacent beacon power levels.

Fig. 8. One-hop neighbourhood: green area represents the estimated area of a node in the
final area, while the surrounding red area represents the corresponding one-hop
neighbourhood

1) Comparison with area based scheme:APIT (Approximate Point in Triangle)
In the PIT and APIT schemes[4], a node chooses three reference nodes from all audible
reference nodes (reference nodes from which a beacon was received) and tests whether it is
inside the triangle formed by connecting these three reference nodes. The theoretical method
used to determine whether a point is inside a triangle or not is called the Point-In-Triangle
(PIT) test. The PIT test can be carried out only under ideal physical layer conditions, when
every node in the network is mobile can move around its own position. Due to the
infeasibility of conducting such a test, an APIT (Approximate Point in Triangle) test is
proposed. The APIT uses RSSI information of beacon signals to determine whether it is
inside or outside a given triangle. The PIT or APIT tests are carried out with different
audible reference node combinations until all combinations are exhausted. The information
is then processed by a central server to narrow down the possible area that a target node
resides in. An area scan aggregation algorithm is used to determine the intersection of the
areas and determine the final area estimate of the node.
Fig. 9 shows all the possible triangles for the given configuration of the eight reference
nodes. There are 52 triangles in total (
8
C
3
4). The sensor nodes determine whether they are
in or out of each of the 52 triangles, and the final area estimate computed is a small region or
combination of regions on the area. Since PIT and APIT are area localization schemes, their
performance are compared with ALS using the normalized accuracy metric. The following
five cases are compared and results shown in Fig. 10:

i) ALS under ideal physical layer conditions after six rounds
ii) PIT under ideal physical layer conditions
iii) APIT under ideal physical layer conditions
iv) ALS under non-ideal physical layer conditions after six rounds
v) APIT under non-ideal physical layer conditions

Fig. 9. All possible triangles for PIT and APIT schemes with 8 reference nodes: 4 at the
corners and 4 at the mid-points of sides. There are 52 triangles in total.

The PIT and APIT schemes are carried out under ideal conditions to establish the
performance limits that can be achieved with the APIT algorithm under non-ideal conditions.
For the given scenario, it is observed (as shown in Fig. 7(a)) that ALS under ideal conditions
outperforms both PIT and APIT after just six rounds.
Not all APIT tests yield correct results, even under ideal physical layer conditions. As a
result, the performance of APIT under ideal conditions is slightly lower than PIT, due to
lower accuracy levels. Under non-ideal conditions, it is observed that ALS performs much
better than APIT. This is primarily because fluctuating RSSI values causes a number of APIT
tests to be incorrect. It is also observed that only around 60% of the 52 APIT tests are correct
for each sensor. This results in large area estimates on the network area. Thus, lower
accuracy levels and higher area estimates cause the performance of the APIT scheme to
suffer. ALS, on the other hand, is more resilient to fading and shadowing due to the
significant difference in adjacent beacon power levels.

Fig. 10. ALS outperforms PIT and APIT under ideal and non-ideal scenarios respectively

The ALS scheme is much more computationally efficient than APIT. For the scenario in
consideration, the area estimate obtained from the intersection of just 10 regions for ALS,
one from each round, results in a better performance than APIT, which considers the
intersection of 52 regions. Thus, ALS achieves the desired performance level as APIT at a
much lower computational cost. The computational complexity in number of areas is given
by O(N
r
) for ALS and O(
N
C
3
) for APIT.

2) Comparison with distance vector based schemes: DV-Hop and DHL
Distance Vector based localization schemes estimate the point location of a node. The
location estimation error is then defined to be the Euclidian distance between the actual
position and the estimated position of the node. The average of the location estimation
errors of all the nodes in the network is used to compare the performance of the three
localization schemes. The location errors are normalized with respect to the transmission
range of the node. For ALS and APIT, the Center of Gravity (COG) of the final predicted
region is used as the estimated position of the node. Again, localization using PIT, APIT and
ALS schemes are carried out under ideal conditions to establish the performance limits that
can be achieved by the algorithms under non-ideal conditions.
DV-Hop localization uses a mechanism that is similar to classical distance vector routing.
Each reference node broadcasts a beacon, which contains its location information and a hop-
count parameter initialized to one. The beacon is flooded throughout the network. Each
sensor node maintains the minimum counter value per reference node of all beacons it
receives and ignores those beacons with higher hop-count values. Beacons are flooded
outward with hop-count values incremented at every intermediate hop. Through this
mechanism, all nodes in the network (including other reference nodes) get the shortest
distance, in hops, to every reference node. In order to convert hop-count into physical
distance, the system estimates the average distance per hop without range-based techniques.
Once a node can calculate the distance estimation to more than three reference nodes in the
plane, it uses triangulation (or multilateration) to estimate its position. The DV-Hop scheme
performs well in networks with uniform node density, as the size of each hop is assumed to
be constant. The DHL scheme is an enhancement to the DV-Hop scheme for networks with
non-uniform node density. In the DHL scheme, the size of each hop is not assumed to be
constant. Instead, the size of each hop depends on the density of nodes in the
neighbourhood.
In the simulations, the RNR parameter is set to 1 for DV-Hop and DHL, i.e. the radio range
for both the reference and sensor nodes is set to 50m. From the results shown in Fig. 11, it
can be observed that even under non-ideal physical layer conditions, ALS performs better
than the other range-free schemes.

Fig. 11. Average estimation error for different algorithms. Under ideal and non-ideal
conditions, ALS (after 10 rounds) outperforms all the other schemes.

The average estimation error of DV-Hop and APIT under non-ideal conditions is greater
than R (radio range of sensor node.) The APIT scheme under non-ideal conditions is
severely affected by the fluctuations in RSSI of the beacon packets. APIT would perform
better if there were more reference nodes, but the improved performance would be at the
expense of a much higher computational cost. For example, it is observed from the APIT
simulations in [4] that achieving an average estimation error of close to 0.5R, for a similar set
of system parameters used here, would require more than 15 audible reference nodes. This
would entail computing the intersection of more than 455 (
15
C
3
) areas and hence, is highly
computation intensive. The performance of DV-Hop is contingent on the density
distribution of nodes in the network and the estimate used for the average distance of a
single hop. The performance of DV-Hop suffers if the distribution of nodes in the network is
non-uniform. For both schemes, viz. DV-Hop and DHL, we observe that the localization
error of the nodes along the sides is higher than nodes at the centre of the region [24].


Fig. 10. ALS outperforms PIT and APIT under ideal and non-ideal scenarios respectively

The ALS scheme is much more computationally efficient than APIT. For the scenario in
consideration, the area estimate obtained from the intersection of just 10 regions for ALS,
one from each round, results in a better performance than APIT, which considers the
intersection of 52 regions. Thus, ALS achieves the desired performance level as APIT at a
much lower computational cost. The computational complexity in number of areas is given
by O(N
r
) for ALS and O(
N
C
3
) for APIT.

2) Comparison with distance vector based schemes: DV-Hop and DHL
Distance Vector based localization schemes estimate the point location of a node. The
location estimation error is then defined to be the Euclidian distance between the actual
position and the estimated position of the node. The average of the location estimation
errors of all the nodes in the network is used to compare the performance of the three
localization schemes. The location errors are normalized with respect to the transmission
range of the node. For ALS and APIT, the Center of Gravity (COG) of the final predicted
region is used as the estimated position of the node. Again, localization using PIT, APIT and
ALS schemes are carried out under ideal conditions to establish the performance limits that
can be achieved by the algorithms under non-ideal conditions.
DV-Hop localization uses a mechanism that is similar to classical distance vector routing.
Each reference node broadcasts a beacon, which contains its location information and a hop-
count parameter initialized to one. The beacon is flooded throughout the network. Each
sensor node maintains the minimum counter value per reference node of all beacons it
receives and ignores those beacons with higher hop-count values. Beacons are flooded
outward with hop-count values incremented at every intermediate hop. Through this
mechanism, all nodes in the network (including other reference nodes) get the shortest
distance, in hops, to every reference node. In order to convert hop-count into physical
distance, the system estimates the average distance per hop without range-based techniques.
Once a node can calculate the distance estimation to more than three reference nodes in the
plane, it uses triangulation (or multilateration) to estimate its position. The DV-Hop scheme
performs well in networks with uniform node density, as the size of each hop is assumed to
be constant. The DHL scheme is an enhancement to the DV-Hop scheme for networks with
non-uniform node density. In the DHL scheme, the size of each hop is not assumed to be
constant. Instead, the size of each hop depends on the density of nodes in the
neighbourhood.
In the simulations, the RNR parameter is set to 1 for DV-Hop and DHL, i.e. the radio range
for both the reference and sensor nodes is set to 50m. From the results shown in Fig. 11, it
can be observed that even under non-ideal physical layer conditions, ALS performs better
than the other range-free schemes.

Fig. 11. Average estimation error for different algorithms. Under ideal and non-ideal
conditions, ALS (after 10 rounds) outperforms all the other schemes.

The average estimation error of DV-Hop and APIT under non-ideal conditions is greater
than R (radio range of sensor node.) The APIT scheme under non-ideal conditions is
severely affected by the fluctuations in RSSI of the beacon packets. APIT would perform
better if there were more reference nodes, but the improved performance would be at the
expense of a much higher computational cost. For example, it is observed from the APIT
simulations in [4] that achieving an average estimation error of close to 0.5R, for a similar set
of system parameters used here, would require more than 15 audible reference nodes. This
would entail computing the intersection of more than 455 (
15
C
3
) areas and hence, is highly
computation intensive. The performance of DV-Hop is contingent on the density
distribution of nodes in the network and the estimate used for the average distance of a
single hop. The performance of DV-Hop suffers if the distribution of nodes in the network is
non-uniform. For both schemes, viz. DV-Hop and DHL, we observe that the localization
error of the nodes along the sides is higher than nodes at the centre of the region [24].

5. Performance of ALS on a WSN Test bed
The performance of the ALS depends on the model used for the physical layer. The radio
environment can be modelled by the empirical log-distance path loss model, as shown
below:

) ( ) log( 10 ) ( ] )[ (
0
0
dB X
d
d
n d PL dB d PL

(3)

where n is the path loss exponent which indicates the rate at which the path loss increases
with distance, d
0
is the reference distance (typically set to 1 m), d is the distance between the
transmitter and receiver, PL(d
0
) is the power received at distance d
0
, and X
is a zero-mean
Gaussian distributed random variable (in dB) with standard deviation . X

describes the
random shadowing effects over a large number of measurements for the same transmitter-
receiver separation.
To extend ALS to any generic physical layer model, we implement a testing phase where the
parameters n and X
are estimated, which can be achieved by the reference nodes mutually

measuring the received signal strength from one anothers beacons. The model in equation
(3) can then be used to determine the different transmit power levels and to draw out the
signal map.
We implemented the ALS on a wireless sensor network test bed [25]. MicaZ motes by
Crossbow Technology Inc. [32] are used as both reference and sensor nodes. The MicaZ
motes allow transmission of signals at only 8 power levels: -25, -15, -10, -7, -5, -3, -1 and 0
dBm, corresponding to MicaZ transmission power settings of 3, 7, 11, 15, 19, 23, 27 and 31
respectively. As a result, we were constrained to use only these eight power levels for the
reference nodes. However, for a real deployment, we anticipate the anchor nodes to be more
sophisticated devices with the ability to finetune the power at which they transmit beacon
signals.

Fig. 12. ALS experiment in an obstacle free environment the experiment was carried out in
a 30m30m region in a soccer field. There were no obstacles present inside the region.

Fig. 13. ALS experiments in an obstacle ridden environment the experiment was carried
out in a 30m30m region in a park with several trees in the region. Several of the sensors
were placed behind trees, as seen in the picture.

The experiments were carried out in both indoor and outdoor environments. For the
outdoor scenario, the experiment was first carried out in an environment with no obstacles
(Fig. 12), and subsequently in an obstacle-ridden environment (trees, park benches, etc) (Fig.
13). RSSI measurements were made to estimate the path loss exponent of radio signals in
each environment. The path loss exponent is calculated using regression analysis on the
RSSI measurements, and was determined to be 2.92 for the indoor environment and 2.96 for
the outdoor environment. The path loss exponents were then used to estimate the ranges of
the beacon signals sent out at different power levels by reference nodes. The measured and
estimated range measurements for the indoor and outdoor environments are shown in
Table 3, and we observed that the estimated range values tally with the measured range
values for different power levels.

Power
Level
Indoor (Reference
Node Height = 12 cm)
Estimated (m)
Indoor (Reference
Measured (m)
Outdoor(Reference
Estimated (m)
Outdoor
(Reference Node
Height = 190 cm)
Measured (m)
3 2.2 2.5 2.5 2.5
7 4.9 5.5 13.0 13.5
11 7.3 8.0 19.2 17.5
15 9.3 9.0 24.3 24.5
19 10.8 13.0 28.3 30.0
23 12.0 13.0 33.1 33.5
27 14.0 15.0 38.7 37.0
31 15.0 15.0 50.0 50.0
Table 3. Estimated and Measured Range measurements for different MicaZ power levels for
indoor and outdoor environments. The slightly greater differences in indoor estimated and
measured values are due multipath effects.
5. Performance of ALS on a WSN Test bed
The performance of the ALS depends on the model used for the physical layer. The radio
environment can be modelled by the empirical log-distance path loss model, as shown
below:

) ( ) log( 10 ) ( ] )[ (
0
0
dB X
d
d
n d PL dB d PL

(3)

where n is the path loss exponent which indicates the rate at which the path loss increases
with distance, d
0
is the reference distance (typically set to 1 m), d is the distance between the
transmitter and receiver, PL(d
0
) is the power received at distance d
0
, and X
is a zero-mean
Gaussian distributed random variable (in dB) with standard deviation . X

describes the
random shadowing effects over a large number of measurements for the same transmitter-
receiver separation.
To extend ALS to any generic physical layer model, we implement a testing phase where the
parameters n and X
are estimated, which can be achieved by the reference nodes mutually

measuring the received signal strength from one anothers beacons. The model in equation
(3) can then be used to determine the different transmit power levels and to draw out the
signal map.
We implemented the ALS on a wireless sensor network test bed [25]. MicaZ motes by
Crossbow Technology Inc. [32] are used as both reference and sensor nodes. The MicaZ
motes allow transmission of signals at only 8 power levels: -25, -15, -10, -7, -5, -3, -1 and 0
dBm, corresponding to MicaZ transmission power settings of 3, 7, 11, 15, 19, 23, 27 and 31
respectively. As a result, we were constrained to use only these eight power levels for the
reference nodes. However, for a real deployment, we anticipate the anchor nodes to be more
sophisticated devices with the ability to finetune the power at which they transmit beacon
signals.

Fig. 12. ALS experiment in an obstacle free environment the experiment was carried out in
a 30m30m region in a soccer field. There were no obstacles present inside the region.

Fig. 13. ALS experiments in an obstacle ridden environment the experiment was carried
out in a 30m30m region in a park with several trees in the region. Several of the sensors
were placed behind trees, as seen in the picture.

The experiments were carried out in both indoor and outdoor environments. For the
outdoor scenario, the experiment was first carried out in an environment with no obstacles
(Fig. 12), and subsequently in an obstacle-ridden environment (trees, park benches, etc) (Fig.
13). RSSI measurements were made to estimate the path loss exponent of radio signals in
each environment. The path loss exponent is calculated using regression analysis on the
RSSI measurements, and was determined to be 2.92 for the indoor environment and 2.96 for
the outdoor environment. The path loss exponents were then used to estimate the ranges of
the beacon signals sent out at different power levels by reference nodes. The measured and
estimated range measurements for the indoor and outdoor environments are shown in
Table 3, and we observed that the estimated range values tally with the measured range
values for different power levels.

Power
Level
Indoor (Reference
Estimated (m)
Indoor (Reference
Measured (m)
Outdoor(Reference
Estimated (m)
Outdoor
(Reference Node
Height = 190 cm)
Measured (m)
3 2.2 2.5 2.5 2.5
7 4.9 5.5 13.0 13.5
11 7.3 8.0 19.2 17.5
15 9.3 9.0 24.3 24.5
19 10.8 13.0 28.3 30.0
23 12.0 13.0 33.1 33.5
27 14.0 15.0 38.7 37.0
31 15.0 15.0 50.0 50.0
Table 3. Estimated and Measured Range measurements for different MicaZ power levels for
indoor and outdoor environments. The slightly greater differences in indoor estimated and
measured values are due multipath effects.
We observed that the range measurements vary with the height at which the reference
nodes are placed, as also noted in [18]. In particular, the communication ranges of the nodes
increased when the reference nodes were raised above the ground. We, therefore, raised the
height of the reference nodes for our experiments. For the indoor environment, the reference
nodes are placed on plastic pots at a height of 12 cm above ground level. Similarly, for the
outdoor environment, the reference nodes are mounted on wooden easels at a height of 190
cm above ground level, while the sensors are placed on plastic pots, 12 cm above ground
level (as shown in Fig. 14.)
We also observed that the range measurements depend on the relative orientation between
the transmitter and receiver antennas, and the radio patterns of the MicaZ antennas are not
circular. Both these observations are confirmed by the results obtained by Lymberopoulos et
al. [26] and Tan et al. [27]. Therefore, we run each ALS experiment four times with four
different relative orientations between the transmitter and receiver antennas. For example,
all sensors are placed facing a certain direction initially (North). The ALS experiment is then
carried out four times, with reference nodes facing North, South, East and West directions
respectively. The results are then combined to obtain the final area estimate of each sensor.
This helps alleviate the problems caused by the radio irregularity of the sensor antennas.

Fig. 14. Reference nodes mounted 190 cm above ground; sensor nodes are mounted on
plastic pots 12 cm above ground level.

The experimental set up is similar to the set up described in the simulation scenario. An area
size of 10m 10m is chosen for the indoor environment while the area size of 30m 30m is
chosen for the outdoor environment. Between 30 and 35 sensors are deployed randomly
throughout the area, with eight reference nodes positioned at the four corners and the four
mid-points of the sides of the square region. Since packets are often lost due to varying
channel conditions, the CONFIDENCE_LEVEL parameter is set to a relatively low value of
30%. The results of the experiments are summarized in the Table 4 below.

Table 4. Summary of Experimental Results

We also plot the actual versus point estimated locations for the sensors in the different
environments, namely, Fig. 15 for indoor, Fig. 16 for outdoor (open field/no obstacles) and
Fig. 17 for outdoor (park/with obstacles). The crosses are the actual locations of the sensors,
the circles are the predicted point estimated location of the sensors (the approximate centre
of the region) and the squares are the locations of the reference nodes. A shorter line
connecting a cross-circle pair denotes higher accuracy.

Fig. 15. Actual versus Estimated Locations of Sensors in Indoor Environment

The set of ideal results refers to the case where every sensor node measures its signal
coordinate correctly and hence the area in which it is located, correctly. The average error
increases from 2.04 m to 3.37 m as we move from an obstacle-free outdoor environment to
an obstacle-ridden environment. Moving from an obstacle-free to an obstacle-ridden
environment outdoors, we notice that the accuracy drops but the number of nodes present
in the one-hop neighbourhood increases. For all the scenarios, more than 80% of sensor
nodes lie within their predicted area or within a one-hop region of the predicted areas with
the average area size estimate of less than 3.5%. This means that in real deployments,
sensors can be located quickly once the predicted region is calculated. The performance of

Ideal results Experimental results
Environment Description
No. of
power
levels
Avg.
Error
Accuracy
Avg. Area
Estimate
Accuracy /
Avg. Area
Estimate
Avg.
Error
Accuracy
One hop
accuracy
Avg. Area
Estimate
Accuracy /
Avg. Area
Estimate
Indoor 10m10m 3 0.71
35/35
(100%)
4.28% 23.36 1.09
21/35
(60%)
9/35
(25.71%)
3.48% 17.24
Outdoor (No
obstacles)
30m30m 4 1.45
30/30
(100%)
1.70% 58.82 2.04
17/30
(56.67%)
13/30
(43.33%)
2.76% 20.53
Outdoor (With
obstacles)
30m30m 4 1.45
30/30
(100%)
1.70% 58.82 3.37
16/30
(53.33%)
9/30
(30.00%)
2.58% 20.67
We observed that the range measurements vary with the height at which the reference
nodes are placed, as also noted in [18]. In particular, the communication ranges of the nodes
increased when the reference nodes were raised above the ground. We, therefore, raised the
height of the reference nodes for our experiments. For the indoor environment, the reference
nodes are placed on plastic pots at a height of 12 cm above ground level. Similarly, for the
outdoor environment, the reference nodes are mounted on wooden easels at a height of 190
cm above ground level, while the sensors are placed on plastic pots, 12 cm above ground
level (as shown in Fig. 14.)
We also observed that the range measurements depend on the relative orientation between
the transmitter and receiver antennas, and the radio patterns of the MicaZ antennas are not
circular. Both these observations are confirmed by the results obtained by Lymberopoulos et
al. [26] and Tan et al. [27]. Therefore, we run each ALS experiment four times with four
different relative orientations between the transmitter and receiver antennas. For example,
all sensors are placed facing a certain direction initially (North). The ALS experiment is then
carried out four times, with reference nodes facing North, South, East and West directions
respectively. The results are then combined to obtain the final area estimate of each sensor.
This helps alleviate the problems caused by the radio irregularity of the sensor antennas.

Fig. 14. Reference nodes mounted 190 cm above ground; sensor nodes are mounted on
plastic pots 12 cm above ground level.

The experimental set up is similar to the set up described in the simulation scenario. An area
size of 10m 10m is chosen for the indoor environment while the area size of 30m 30m is
chosen for the outdoor environment. Between 30 and 35 sensors are deployed randomly
throughout the area, with eight reference nodes positioned at the four corners and the four
mid-points of the sides of the square region. Since packets are often lost due to varying
channel conditions, the CONFIDENCE_LEVEL parameter is set to a relatively low value of
30%. The results of the experiments are summarized in the Table 4 below.

Table 4. Summary of Experimental Results

We also plot the actual versus point estimated locations for the sensors in the different
environments, namely, Fig. 15 for indoor, Fig. 16 for outdoor (open field/no obstacles) and
Fig. 17 for outdoor (park/with obstacles). The crosses are the actual locations of the sensors,
the circles are the predicted point estimated location of the sensors (the approximate centre
of the region) and the squares are the locations of the reference nodes. A shorter line
connecting a cross-circle pair denotes higher accuracy.

Fig. 15. Actual versus Estimated Locations of Sensors in Indoor Environment

The set of ideal results refers to the case where every sensor node measures its signal
coordinate correctly and hence the area in which it is located, correctly. The average error
increases from 2.04 m to 3.37 m as we move from an obstacle-free outdoor environment to
an obstacle-ridden environment. Moving from an obstacle-free to an obstacle-ridden
environment outdoors, we notice that the accuracy drops but the number of nodes present
in the one-hop neighbourhood increases. For all the scenarios, more than 80% of sensor
nodes lie within their predicted area or within a one-hop region of the predicted areas with
the average area size estimate of less than 3.5%. This means that in real deployments,
sensors can be located quickly once the predicted region is calculated. The performance of

Ideal results Experimental results
Environment Description
No. of
power
levels
Avg.
Error
Accuracy
Avg. Area
Estimate
Accuracy /
Avg. Area
Estimate
Avg.
Error
Accuracy
One hop
accuracy
Avg. Area
Estimate
Accuracy /
Avg. Area
Estimate
Indoor 10m10m 3 0.71
35/35
(100%)
4.28% 23.36 1.09
21/35
(60%)
9/35
(25.71%)
3.48% 17.24
Outdoor (No
obstacles)
30m30m 4 1.45
30/30
(100%)
1.70% 58.82 2.04
17/30
(56.67%)
13/30
(43.33%)
2.76% 20.53
Outdoor (With
obstacles)
30m30m 4 1.45
30/30
(100%)
1.70% 58.82 3.37
16/30
(53.33%)
9/30
(30.00%)
2.58% 20.67
ALS becomes worse as we move from the open field to the park because multipath effects
from obstacles cause fluctuations in signal strength resulting in sensor nodes receiving
incorrect signal coordinates. The accuracy of localization also depends on many other
factors, such as, type of hardware, environment, number of reference nodes, and the size of
the deployment area.

Fig. 16. Actual versus Estimated Locations of Sensors in Outdoor Environment (no obstacles)

Fig. 17. Actual versus Estimated Locations of Sensors in Outdoor Environment (with obstacles)
In addition, we also compared ALS with other localization schemes that utilize radio
signals, which have also been implemented and tested experimentally. From Table 5, show
the results of ALS with such schemes, and it is clear that the performance of ALS in the field
trials is comparable or better than these other localization schemes.

Table 5. Comparisons of Field Test results against other localization schemes

6. Conclusion
The ALS is a range free localization scheme that provides a coarse estimation of the location
of a sensor within a certain area. While the sensors simply record the signal levels received
from reference nodes, the sinks carry out most of the complicated computations. The
granularity of the area estimates can be increased easily by modifying certain system
parameters. The simulations results in Qualnet show that ALS is a promising scheme as
more than 90% of nodes are located in their estimated areas or in a region one-hop away.
We implemented the ALS algorithm on a wireless sensor network testbed and tested it in
both indoor and outdoor scenarios. We observed that 100% of nodes lie within a one-hop
region in an outdoor, obstacle-free environment, and 83% (or more) of nodes of lie within a
one-hop region in an indoor or outdoor, obstacle-ridden environment. With high quality
transmission device and antennas implemented at the reference nodes (as compared to the
motes used in the experiment), it is highly foreseeable that the accuracy of ALS will increase.
As part of our ongoing and future work, we have first addressed the issue of non-uniform
areas (e.g. as shown in Fig. 6) by aggregating areas of different sizes to create more
uniformity [33]. Subsequently, we will be improving the reference nodes and also develop
Scheme
Hardware
Platform
Environment
No of
reference
nodes
Region Size Error
ALS micaZ Indoor 8 10m x 10m 1.09m
ALS micaZ
Outdoor (no
obstacles)
8 30m x 30m 2.04m
ALS micaZ
Outdoor
(with
obstacles)
8 30m x 30m 3.37m
Ranging-
based [18]
mica2dot
Outdoor
(open field)
(not
mentioned;
49-node
network)
50m x 50m 4.91m
Mote Track
[28]
mica2 Indoor 20 1742 m
2

50
th
percentile: 2m
90
th
percentile: 3m
Ecolocation
[29]
mica2 Outdoor 10 12m x 12m around 1m
Probability
Grid [30]
mica2 Outdoor 3
5x5 grid;
approx 12m
apart
79% of radio range
where radio range is
about 15m
ALS becomes worse as we move from the open field to the park because multipath effects
from obstacles cause fluctuations in signal strength resulting in sensor nodes receiving
incorrect signal coordinates. The accuracy of localization also depends on many other
factors, such as, type of hardware, environment, number of reference nodes, and the size of
the deployment area.

Fig. 16. Actual versus Estimated Locations of Sensors in Outdoor Environment (no obstacles)

Fig. 17. Actual versus Estimated Locations of Sensors in Outdoor Environment (with obstacles)
In addition, we also compared ALS with other localization schemes that utilize radio
signals, which have also been implemented and tested experimentally. From Table 5, show
the results of ALS with such schemes, and it is clear that the performance of ALS in the field
trials is comparable or better than these other localization schemes.

Table 5. Comparisons of Field Test results against other localization schemes

6. Conclusion
The ALS is a range free localization scheme that provides a coarse estimation of the location
of a sensor within a certain area. While the sensors simply record the signal levels received
from reference nodes, the sinks carry out most of the complicated computations. The
granularity of the area estimates can be increased easily by modifying certain system
parameters. The simulations results in Qualnet show that ALS is a promising scheme as
more than 90% of nodes are located in their estimated areas or in a region one-hop away.
We implemented the ALS algorithm on a wireless sensor network testbed and tested it in
both indoor and outdoor scenarios. We observed that 100% of nodes lie within a one-hop
region in an outdoor, obstacle-free environment, and 83% (or more) of nodes of lie within a
one-hop region in an indoor or outdoor, obstacle-ridden environment. With high quality
transmission device and antennas implemented at the reference nodes (as compared to the
motes used in the experiment), it is highly foreseeable that the accuracy of ALS will increase.
As part of our ongoing and future work, we have first addressed the issue of non-uniform
areas (e.g. as shown in Fig. 6) by aggregating areas of different sizes to create more
uniformity [33]. Subsequently, we will be improving the reference nodes and also develop
Scheme
Hardware
Platform
Environment
No of
reference
nodes
Region Size Error
ALS micaZ Indoor 8 10m x 10m 1.09m
ALS micaZ
Outdoor (no
obstacles)
8 30m x 30m 2.04m
ALS micaZ
Outdoor
(with
obstacles)
8 30m x 30m 3.37m
Ranging-
based [18]
mica2dot
Outdoor
(open field)
(not
mentioned;
49-node
network)
50m x 50m 4.91m
Mote Track
[28]
mica2 Indoor 20 1742 m
2

50
th
percentile: 2m
90
th
percentile: 3m
Ecolocation
[29]
mica2 Outdoor 10 12m x 12m around 1m
Probability
Grid [30]
mica2 Outdoor 3
5x5 grid;
approx 12m
apart
79% of radio range
where radio range is
about 15m
routing protocols that is able to utilize the location information provided by the ALS
algorithm. A sensor can therefore estimate whether it is nearer or further away from the
destination, compared to its previous hop, based on the signal coordinate information of its
neighbour, the destination and itself, and this information can be used for developing fast
and efficient routing protocols. Another benefit is the covert nature of the scheme, which
can be exploited to meet privacy needs.

7. References
[1] I. Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci, A Survey on Sensor
Networks, IEEE Communications Magazine, Vol. 40, No. 8, pp 102-114, Aug2002.
[2] Global Positioning System standard Positioning Service Specification, 2
nd
Edition, June
2, 1995.
[3] Q. Yao, S. K. Tan, Y. Ge, B.S. Yeo, and Q. Yin, An Area Localization Scheme for Large
Wireless Sensor Networks,Proceedings of the IEEE 61st Semiannual Vehicular
Technology Conference (VTC2005-Spring), May 30 - Jun 1, 2005, Stockholm, Sweden.
[4] T. He, C. Huang, B. Blum, J. Stankovic and T. Abdelzaher, Range-Free Localization
Schemes for Large Scale Sensor Networks, Proceedings of the 9
th
ACM International
Conference on Mobile Computing and Networking (Mobicom 2003), Sep 14-19 2003, San
Diego, CA, USA.
[5] D. Niculescu and B. Nath, DV Based Positioning in Ad Hoc Networks,
Telecommunication Systems, Vol. 22, No. 1-4, pp 268-280, 2003.
[6] S.Y. Wong, J.G. Lim, S.V. Rao and Winston K.G. Seah, Density-aware Hop-count
Localization (DHL) in wireless sensor networks with variable density, Proceedings
of the IEEE Wireless Communications and Networking Conference (WCNC 2005), 13-17
Mar 2005, New Orleans, L.A.,USA.
[7] S. Gezici, Z. Tian, G. Giannakis, H. Kobayashi, A. Molisch, V.Poor and Z. Sahinoglu,
Localization via Ultra Wide Band Radios, IEEE Signal Processing Magazine, Vol. 22,
No. 4,Jul 2005, pp. 70-84.
[8] Y. Xu, J. Shi and X. Wu, A UWB-based localization scheme in wireless sensor
networks, Proceedings of the IET Conference on Wireless, Mobile and Sensor Networks
2007 (CCWMSN07), Dec 12-14, 2007, Shanghai, China.
[9] N. B. Priyantha, A. Chakraborty and H. Balakrishnan, The Cricket Location-Support
system, Proceedings of the 6th ACM International Conference on Mobile Computing and
Networking (Mobicom 2000), Aug 6-11, 2000, Boston, MA, USA.
[10] Y. Kwon, K. Mechitov, S. Sundresh, W. Kim and G. Agha,"Resilient Localization for
Sensor Networks in Outdoor Environments", Proceedings of 25th IEEE International
Conference on Distributed Computing Systems (ICDCS 2005), Jun 6-10, 2005,
Columbus, Ohio, USA.
[11] P. Bahl and V. Padmanabhan, RADAR: an in-building RF-based user location and
tracking system, Proceedings of the 19
th
Annual Joint Conference of the IEEE Computer
and Communications Societies (INFOCOM 2000),Mar 26-30, 2000, Tel Aviv, Israel.
[12] X. Cheng, A. Thaeler, G. Xue and D. Chen, TPS: A Time-Based Positioning Scheme for
Outdoor Sensor Networks, Proceedings of the 23
rd
Annual Joint Conference of the IEEE
Computer and Communications Societies (INFOCOM 2004), Mar 7-11, 2004, Hong
Kong.
[13] A. Savvides, C. C. Han and M. B. Srivastava, Dynamic Fine-grained Localization in
Ad-Hoc networks of Sensors,Proceedings of the 7
th
ACM International Conference on
Mobile Computing and Networking (Mobicom 2001), Jul 16-21, 2001, Rome, Italy.
[14] D. Niculescu and B. Nath, Ad Hoc Positioning System (APS) Using AOA, Proceedings
of the 22
nd
Annual Joint Conference of the IEEE Computer and Communications Societies
(INFOCOM 2003), Mar 30-Apr 3, 2003, San Francisco, CA, USA.
[15] N. Malhotra, M. Krasniewski, C. Yang, S. Bagchi, and W. Chappell, Location
Estimation in Ad-hoc networks with Directional Antennas,Proceedings of 25
th
IEEE
International Conference on Distributed Computing Systems (ICDCS 2005), Jun 6-10,
2005, Columbus, Ohio, USA.
[16] L. Girod and D. Estrin, Robust Range Estimation Using Acoustic and Multimodal
Sensing, Proceedings of the International Conference on Intelligent Robots and Systems
(IROS 2001), Oct 29-Nov 3, 2001, Maui, HI, USA.
[17] L.Evers, S. Dulman and P. Havinga, A Distributed Precision Based Localization
Algorithm for Ad-Hoc Networks, Proceedings of the 2
nd
International Conference on
Pervasive Computing (PERVASIVE 2004), Apr 21-23, 2004, Linz, Vienna, Austria.
[18] K. Whitehouse, C. Karlof and D. Culler, A practical evaluation of radio signal strength for
ranging-based localization, ACM SIGMOBILE Mobile Computing and Communications
Review, Special Issue on Localization, Vol. 11 , No. 1, pp. 41-52, Jan 2007.
[19] N. Bulusu, J. Heidemann and D. Estrin, GPS-less Low Cost Outdoor Localization for
Very Small Devices, IEEE Personal Communications Magazine,Vol. 7, No. 5, pp. 28-
34, Oct 2000.
[20] X. Li, H. Shi and Y. Shang, Sensor network localisation based on sorted RSSI
quantisation, International Journal of Ad Hoc and Ubiquitous Computing, Vol. 1, No.
4, pp. 222-229, 2006.
[21] R. Battiti, M. Brunato, and A. Villani, "Statistical learning theory for location
fingerprinting in wireless LANs" Tech. Rep. DIT-02-0086, Dipartimento di
Informatica e Telecomunicazioni, Universita di Trento, 2002.
[22] L. Doherty, K. Pister, and L. Ghaoui, Convex Position Estimation in Wireless Sensor
Networks, Proceedings of the 20
th
Annual Joint Conference of the IEEE Computer and
Communications Societies (INFOCOM 2001), Apr 22-26, 2001, Anchorage, AK, USA.
[23] S. Capkun, M. Hamdi and J. Hubaux, GPS-free positioning in mobile ad-hoc
networks, Proceedings of the 34
th
Annual Hawaii International conference on System
Sciences, Jan 3-6, 2001, Hawaii, USA.
[24] Jeffrey Tay, Vijay R. Chandrasekhar and Winston K.G. Seah, Selective Iterative
Multilateration for Hop Count Based Localization in Wireless Sensor Networks.
Proceedings of the 7th International Conference on Mobile Data Management (MDM06),
May 13-16, Nara, Japan, 2006.
[25] Vijay R. Chandrasekhar, Z.A. Eu, Winston K.G. Seah and Arumugam P. Venkatesh,
Experimental Analysis of Area Localization for Wireless Sensor Networks,
Proceedings of the IEEE Wireless Communications and Networking Conference
(WCNC2007), Mar 11-15, 2007, Hong Kong.
[26] D. Lymberopoulos, Q. Lindsey and A. Savvides, An Empirical Analysis of Radio
Signal Strength Variability in IEEE 802.15.4 Networks using Monopole Antennas,
Proceedings of the Second European Workshop on Sensor Networks (EWSN 2006), Feb 13-
15, 2006, ETH, Zurich, Switzerland.
routing protocols that is able to utilize the location information provided by the ALS
algorithm. A sensor can therefore estimate whether it is nearer or further away from the
destination, compared to its previous hop, based on the signal coordinate information of its
neighbour, the destination and itself, and this information can be used for developing fast
and efficient routing protocols. Another benefit is the covert nature of the scheme, which
can be exploited to meet privacy needs.

7. References
[1] I. Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci, A Survey on Sensor
Networks, IEEE Communications Magazine, Vol. 40, No. 8, pp 102-114, Aug2002.
[2] Global Positioning System standard Positioning Service Specification, 2
nd
Edition, June
2, 1995.
[3] Q. Yao, S. K. Tan, Y. Ge, B.S. Yeo, and Q. Yin, An Area Localization Scheme for Large
Wireless Sensor Networks,Proceedings of the IEEE 61st Semiannual Vehicular
Technology Conference (VTC2005-Spring), May 30 - Jun 1, 2005, Stockholm, Sweden.
[4] T. He, C. Huang, B. Blum, J. Stankovic and T. Abdelzaher, Range-Free Localization
Schemes for Large Scale Sensor Networks, Proceedings of the 9
th
ACM International
Conference on Mobile Computing and Networking (Mobicom 2003), Sep 14-19 2003, San
Diego, CA, USA.
[5] D. Niculescu and B. Nath, DV Based Positioning in Ad Hoc Networks,
Telecommunication Systems, Vol. 22, No. 1-4, pp 268-280, 2003.
[6] S.Y. Wong, J.G. Lim, S.V. Rao and Winston K.G. Seah, Density-aware Hop-count
Localization (DHL) in wireless sensor networks with variable density, Proceedings
of the IEEE Wireless Communications and Networking Conference (WCNC 2005), 13-17
Mar 2005, New Orleans, L.A.,USA.
[7] S. Gezici, Z. Tian, G. Giannakis, H. Kobayashi, A. Molisch, V.Poor and Z. Sahinoglu,
Localization via Ultra Wide Band Radios, IEEE Signal Processing Magazine, Vol. 22,
No. 4,Jul 2005, pp. 70-84.
[8] Y. Xu, J. Shi and X. Wu, A UWB-based localization scheme in wireless sensor
networks, Proceedings of the IET Conference on Wireless, Mobile and Sensor Networks
2007 (CCWMSN07), Dec 12-14, 2007, Shanghai, China.
[9] N. B. Priyantha, A. Chakraborty and H. Balakrishnan, The Cricket Location-Support
system, Proceedings of the 6th ACM International Conference on Mobile Computing and
Networking (Mobicom 2000), Aug 6-11, 2000, Boston, MA, USA.
[10] Y. Kwon, K. Mechitov, S. Sundresh, W. Kim and G. Agha,"Resilient Localization for
Sensor Networks in Outdoor Environments", Proceedings of 25th IEEE International
Conference on Distributed Computing Systems (ICDCS 2005), Jun 6-10, 2005,
Columbus, Ohio, USA.
[11] P. Bahl and V. Padmanabhan, RADAR: an in-building RF-based user location and
tracking system, Proceedings of the 19
th
Annual Joint Conference of the IEEE Computer
and Communications Societies (INFOCOM 2000),Mar 26-30, 2000, Tel Aviv, Israel.
[12] X. Cheng, A. Thaeler, G. Xue and D. Chen, TPS: A Time-Based Positioning Scheme for
Outdoor Sensor Networks, Proceedings of the 23
rd
Annual Joint Conference of the IEEE
Computer and Communications Societies (INFOCOM 2004), Mar 7-11, 2004, Hong
Kong.
[13] A. Savvides, C. C. Han and M. B. Srivastava, Dynamic Fine-grained Localization in
Ad-Hoc networks of Sensors,Proceedings of the 7
th
ACM International Conference on
Mobile Computing and Networking (Mobicom 2001), Jul 16-21, 2001, Rome, Italy.
[14] D. Niculescu and B. Nath, Ad Hoc Positioning System (APS) Using AOA, Proceedings
of the 22
nd
Annual Joint Conference of the IEEE Computer and Communications Societies
(INFOCOM 2003), Mar 30-Apr 3, 2003, San Francisco, CA, USA.
[15] N. Malhotra, M. Krasniewski, C. Yang, S. Bagchi, and W. Chappell, Location
Estimation in Ad-hoc networks with Directional Antennas,Proceedings of 25
th
IEEE
International Conference on Distributed Computing Systems (ICDCS 2005), Jun 6-10,
2005, Columbus, Ohio, USA.
[16] L. Girod and D. Estrin, Robust Range Estimation Using Acoustic and Multimodal
Sensing, Proceedings of the International Conference on Intelligent Robots and Systems
(IROS 2001), Oct 29-Nov 3, 2001, Maui, HI, USA.
[17] L.Evers, S. Dulman and P. Havinga, A Distributed Precision Based Localization
Algorithm for Ad-Hoc Networks, Proceedings of the 2
nd
International Conference on
Pervasive Computing (PERVASIVE 2004), Apr 21-23, 2004, Linz, Vienna, Austria.
[18] K. Whitehouse, C. Karlof and D. Culler, A practical evaluation of radio signal strength for
ranging-based localization, ACM SIGMOBILE Mobile Computing and Communications
Review, Special Issue on Localization, Vol. 11 , No. 1, pp. 41-52, Jan 2007.
[19] N. Bulusu, J. Heidemann and D. Estrin, GPS-less Low Cost Outdoor Localization for
Very Small Devices, IEEE Personal Communications Magazine,Vol. 7, No. 5, pp. 28-
34, Oct 2000.
[20] X. Li, H. Shi and Y. Shang, Sensor network localisation based on sorted RSSI
quantisation, International Journal of Ad Hoc and Ubiquitous Computing, Vol. 1, No.
4, pp. 222-229, 2006.
[21] R. Battiti, M. Brunato, and A. Villani, "Statistical learning theory for location
fingerprinting in wireless LANs" Tech. Rep. DIT-02-0086, Dipartimento di
Informatica e Telecomunicazioni, Universita di Trento, 2002.
[22] L. Doherty, K. Pister, and L. Ghaoui, Convex Position Estimation in Wireless Sensor
Networks, Proceedings of the 20
th
Annual Joint Conference of the IEEE Computer and
Communications Societies (INFOCOM 2001), Apr 22-26, 2001, Anchorage, AK, USA.
[23] S. Capkun, M. Hamdi and J. Hubaux, GPS-free positioning in mobile ad-hoc
th
Annual Hawaii International conference on System
Sciences, Jan 3-6, 2001, Hawaii, USA.
[24] Jeffrey Tay, Vijay R. Chandrasekhar and Winston K.G. Seah, Selective Iterative
Multilateration for Hop Count Based Localization in Wireless Sensor Networks.
Proceedings of the 7th International Conference on Mobile Data Management (MDM06),
May 13-16, Nara, Japan, 2006.
[25] Vijay R. Chandrasekhar, Z.A. Eu, Winston K.G. Seah and Arumugam P. Venkatesh,
Experimental Analysis of Area Localization for Wireless Sensor Networks,
Proceedings of the IEEE Wireless Communications and Networking Conference
(WCNC2007), Mar 11-15, 2007, Hong Kong.
[26] D. Lymberopoulos, Q. Lindsey and A. Savvides, An Empirical Analysis of Radio
Signal Strength Variability in IEEE 802.15.4 Networks using Monopole Antennas,
Proceedings of the Second European Workshop on Sensor Networks (EWSN 2006), Feb 13-
15, 2006, ETH, Zurich, Switzerland.
[27] Eddie B.S. Tan, J.G. Lim, Winston K.G. Seah and S.V. Rao, On the Practical Issues in
Hop Count Localization of Sensors in a Multihop Network, Proceedings of the 63rd
IEEE Vehicular Technology Conference (VTC2006-Spring), May 8-10, 2006, Melbourne,
Victoria, Australia.
[28] K. Lorincz and M. Welsh, Motetrack: A Robust, Decentralized Approach to RF-Based
Location Tracking, Proceedings of the International Workshop on Location- and
Context-Awareness (LoCA2005), May 12-13, 2005, Munich, Germany.
[29] K. Yedavalli, B. Krishnamachari, S. Ravula and B. Srinivasan, Ecolocation: A Sequence
Based Technique for RF Localization in Wireless Sensor Networks, Proceedings of
Information Processing in Sensor Networks (IPSN2005), Apr 25-27, 2005, Los Angeles,
CA, USA.
[30] R. Stoleru and J. A. Stankovic, Probability Grid: A Location Estimation Scheme for
Wireless Sensor Networks, Proceedings of Sensor and Ad Hoc Communications and
Networks Conference (SECON2004), Oct 4-7, 2004, Santa Clara, CA, USA.
[31] Scalable Networks Inc., QualNet Simulator, available from: http://www.scalable-
networks.com/.
[32] Crossbow Technology Inc., homepage: http://www.xbow.com.
[33] V.A. Pillai, Winston K.G. Seah and Y.H. Chew, "Improved Area Estimates for
Localization in Wireless Sensor Networks", Proceedings of the 16th Asia-Pacific
Conference on Communications (APCC), Auckland, New Zealand, Nov 1-3, 2010.

Information and Data Processing Technologies
Part 3
Information and Data Processing Technologies
Data Fusion Approach for Error Correction in Wireless Sensor Networks 353
Data Fusion Approach for Error Correction in Wireless Sensor Networks
Maen Takruri and Subhash Challa
0
Data Fusion Approach for Error Correction
in Wireless Sensor Networks
Maen Takruri
Centre for Real-Time Information Networks (CRIN)
University of Technology, Sydney
Australia
Subhash Challa
NICTA Victoria Research Laboratory
The University of Melbourne
Australia
1. Introduction
Wireless Sensor Networks (WSNs) emerged as an important research area (Estrin et al., 2001).
This development was encouraged by the dramatic advances in sensor technology, wireless
communications, digital electronics and computer networks, enabling the development of low
cost, low power, multi-functional sensor nodes that are small in size and can communicate
over short distances (Akyildiz et al., 2002). When they work as a group, these nodes can
accomplish far more complex tasks and inferences than more powerful nodes in isolation.
This led to a wide spectrum of possible military and civilian applications, such as battleeld
surveillance, home automation, smart environments and forest re detection.
On the down side, the wireless sensors are usually left unattended for long periods of time
in the eld, which makes them prone to failures. This is due to either sensors running out
of energy, ageing or harsh environmental conditions surrounding them. Besides the random
noise, these cheap sensors tend to develop drift in their measurements as they age. We dene
the drift as a slow, unidirectional long-term change in the sensor measurement. This poses
a major problem for end applications, as the data from the network becomes progressively
useless. An early detection of such drift is essential for the successful operation of the sensor
network. In this process, the sensors, which otherwise would have been deemed unusable,
can continue to be used, thus prolonging the effective life span of the sensor network and
optimising the cost effectiveness of the solutions.
A common problem faced in large scale sensor networks is that sensors can suffer from bias
in their measurements (Bychkovskiy et al., 2003). The bias and drift errors (systematic errors)
have a direct impact on the effectiveness of the associated decision support systems. Cali-
brating the sensors to account for these errors is a costly and time consuming process. Tra-
ditionally, such errors are corrected by site visits where an accurate, calibrated sensor is used
to calibrate other sensors. This process is manually intensive and is only effective when the
number of sensors deployed is small and the calibration is infrequent. In a large scale sensor
18
network, constituted of cheap sensors, there is a need for frequent recalibration. Due to the
size of such networks, it is impractical and cost prohibitive to manually calibrate them. Hence,
there is a signicant need for auto calibration (Takruri & Challa, 2007) in sensor networks.
The sensor drift problem and its effects on sensor inferences is addressed in this work under
the assumption that neighbouring sensors in a network observe correlated data, i.e., the mea-
surements of one sensor is related to the measurements of its neighbours. Furthermore, the
physical phenomenon that these sensors observe also follows some spatial correlation. More-
over, the faults of the neighbouring nodes are likely to be uncorrelated (Krishnamachari &
Iyengar, 2004). Hence, in principle, it is possible to predict the data of one sensor using the
data from other closely situated sensors (Krishnamachari & Iyengar, 2004; Takruri & Challa,
2007). This predicted data provides a suitable basis to correct anomalies in a sensors reported
measurements. At this point, it is important to differentiate between the measurement of the
sensor or the reported data which may contain bias and/or drift, and the corrected reading
which is evaluated by the error correction algorithms. The early detection of anomalous data
enables us not only to detect drift in sensor readings, but also to correct it.
In this work, we present a general and comprehensive framework for detecting and correcting
both the systematic (drift and bias) and random errors in sensor measurements. The solution
addresses the sparse deployment scenario of WSNs. Statistical modelling rather than physical
modelling is used to model the spatio-temporal cross correlations among sensors measure-
ments. This makes the framework presented here likely to be applicable to most sensing prob-
lems with minor changes. The proposed algorithm is tested on real data obtained from the
Intel Berkeley Research Laboratory sensor deployment. The results show that our algorithm
successfully detects and corrects drifts and noise developed in sensors and thereby prolongs
the effective lifetime of the network.
The rest of the chapter is organised as follows. Section 2 presents the related work on error de-
tection and correction in WSNs literature. We present our network structure and the problem
statement in Section 3. Sections 4 and 5 formulate the Support Vector Regression and Un-
scented Kalman Filter framework for error correction in sensor networks. Section 6 evaluates
the proposed algorithm using real data and section 7 concludes with future work.
2. Related Work
The sensor bias and drift problems and their effects on sensor inferences have rarely been
addressed in the sensor networks literature. In contrast, the bias correction problem has been
well studiedin the context of the multi-radar tracking problem. In the target tracking literature
the problem is usually referred to as the registration problem (Okello & Challa, 2003; Okello &
Pulford, 1996). When the same target is observed by two sensors (radars) from two different
angles, the data fromthose two sensors can be fused to estimate the bias in both sensors. In the
context of image processing of moving objects, the problem is referred to as image registration,
which is the process of overlaying two or more images of the same scene taken at different
times, from different viewpoints, and/or by different cameras. It geometrically aligns two
images: the reference and sensed images (Brown, 1992). Image registration is a crucial step
in all image analysis tasks in which the nal information is gained from the combination of
various data sources like in image fusion (Zitova & Flusser, 2003). That is, in order to fuse
two sensor readings, in this case two images, the readings must rst be put into a common
coordinates systems before being fused. The essential idea brought forth by the solution to the
registration problemis the augmentation of the state vector with the bias components. In other
words, the problem is enlarged to estimate not only the states of the targets, using the radar
measurements for example, but also the biases of the radars. This is the approach we consider
in the case of sensor networks. Target tracking lters, in conjunction with sensor drift models
are used to estimate the sensor drift in real time. The estimate is used for correction and as a
feedback to the next estimation step. The presented methodology is a robust framework for
auto calibration of sensors in a WSN.
A straightforward approach to bias calibration is to apply a known stimulus to the sensor
network and measure the response. Then comparing the ground truth input to the response
will result in nding the gain and offset for the linear drifts case (Hoadley, 1970). This method
is referred to by (Balzano & Nowak, 2007) as non-blind calibration since the ground truth is
used to calibrate the sensors. Another form of non-blind calibration is manually calibrating
a subset of sensors in the sensor network and then allowing the non-calibrated sensors to
adjust their readings based on the calibrated subset. The calibrated subset in this context
form a reference point to the ground truth (Bychkovskiy, 2003; Bychkovskiy et al., 2003). The
above mentioned methods are impractical and cost prohibitive in the case of large scale sensor
networks.
The calibration problem of the sensor network was also tackled by (Balzano & Nowak, 2007;
2008) in a different fashion. They stated that after sensors were calibrated to the factory set-
tings, when deployed, their measurements would differ linearly from the ground truth by
certain gains and offsets for each sensor. They presented a method for estimating these gains
and offsets using subspace matching. The method only required routine measurements to be
collected by the sensors and did not need ground truth measurements for comparison. They
referred to this problem as blind calibration of sensor networks. The method did not require
dense deployment of the sensors or a controlled stimulus. However, It required that the sen-
sor measurements are at least slightly correlated over space i.e. the network over sampled the
underlying signals of interest. The theoretical analysis of their work did not take noise into
consideration and assumed linear calibration functions. Therefore, the solution might not be
robust in noisy conditions and will probably result in wrong estimates if applied in a scenario
where the relationship between the measurement and the ground truth is nonlinear. The eval-
uations they presented showed that the method worked better in a controlled environment.
An earlier work on blind calibration of sensor nodes in a sensor network was presented in
(Bychkovskiy, 2003; Bychkovskiy et al., 2003). They assumed that the sensors of the network
under consideration were sufciently densely deployed that they observed the same phe-
nomenon. They used the temporal correlation of signals received by neighbouring sensors
when the signals were highly correlated to derive a function relating the bias in their am-
plitudes. Another method for calibration was considered by (Feng et al., 2003). They used
geometrical and physical constraints on the behaviour of a point light source to calibrate light
sensors without the need for comparing the measurement with an accurate sensor (ground
truth). They assumed that the light sensors under consideration suffered form a constant bias
with time.
The authors in (Whitehouse & Culler, 2002; 2003) argued that calibrating the sensors in sensor
networks is a problematic task since it comprises large number of sensor that are deployed
in partially unobservable and dynamic environments and may themselves be unobservable.
They suggested that the calibration problemin sensor/actuator networks should be expressed
as a parameter estimation problemon the network scale. Therefore, instead of calibrating each
sensor individually to optimise its measurement, the sensors of the network are calibrated to
optimise the overall response of the network. The joint calibration method they presented cal-
ibrated sensors in a controlled environment. The method was tested on an ad-hoc localisation
network, constituted of cheap sensors, there is a need for frequent recalibration. Due to the
size of such networks, it is impractical and cost prohibitive to manually calibrate them. Hence,
there is a signicant need for auto calibration (Takruri & Challa, 2007) in sensor networks.
The sensor drift problem and its effects on sensor inferences is addressed in this work under
the assumption that neighbouring sensors in a network observe correlated data, i.e., the mea-
surements of one sensor is related to the measurements of its neighbours. Furthermore, the
physical phenomenon that these sensors observe also follows some spatial correlation. More-
over, the faults of the neighbouring nodes are likely to be uncorrelated (Krishnamachari &
Iyengar, 2004). Hence, in principle, it is possible to predict the data of one sensor using the
data from other closely situated sensors (Krishnamachari & Iyengar, 2004; Takruri & Challa,
2007). This predicted data provides a suitable basis to correct anomalies in a sensors reported
measurements. At this point, it is important to differentiate between the measurement of the
sensor or the reported data which may contain bias and/or drift, and the corrected reading
which is evaluated by the error correction algorithms. The early detection of anomalous data
enables us not only to detect drift in sensor readings, but also to correct it.
In this work, we present a general and comprehensive framework for detecting and correcting
both the systematic (drift and bias) and random errors in sensor measurements. The solution
addresses the sparse deployment scenario of WSNs. Statistical modelling rather than physical
modelling is used to model the spatio-temporal cross correlations among sensors measure-
ments. This makes the framework presented here likely to be applicable to most sensing prob-
lems with minor changes. The proposed algorithm is tested on real data obtained from the
Intel Berkeley Research Laboratory sensor deployment. The results show that our algorithm
successfully detects and corrects drifts and noise developed in sensors and thereby prolongs
the effective lifetime of the network.
The rest of the chapter is organised as follows. Section 2 presents the related work on error de-
tection and correction in WSNs literature. We present our network structure and the problem
statement in Section 3. Sections 4 and 5 formulate the Support Vector Regression and Un-
scented Kalman Filter framework for error correction in sensor networks. Section 6 evaluates
the proposed algorithm using real data and section 7 concludes with future work.
2. Related Work
The sensor bias and drift problems and their effects on sensor inferences have rarely been
addressed in the sensor networks literature. In contrast, the bias correction problem has been
well studiedin the context of the multi-radar tracking problem. In the target tracking literature
the problem is usually referred to as the registration problem (Okello & Challa, 2003; Okello &
Pulford, 1996). When the same target is observed by two sensors (radars) from two different
angles, the data fromthose two sensors can be fused to estimate the bias in both sensors. In the
context of image processing of moving objects, the problem is referred to as image registration,
which is the process of overlaying two or more images of the same scene taken at different
times, from different viewpoints, and/or by different cameras. It geometrically aligns two
images: the reference and sensed images (Brown, 1992). Image registration is a crucial step
in all image analysis tasks in which the nal information is gained from the combination of
various data sources like in image fusion (Zitova & Flusser, 2003). That is, in order to fuse
two sensor readings, in this case two images, the readings must rst be put into a common
coordinates systems before being fused. The essential idea brought forth by the solution to the
registration problemis the augmentation of the state vector with the bias components. In other
words, the problem is enlarged to estimate not only the states of the targets, using the radar
measurements for example, but also the biases of the radars. This is the approach we consider
in the case of sensor networks. Target tracking lters, in conjunction with sensor drift models
are used to estimate the sensor drift in real time. The estimate is used for correction and as a
feedback to the next estimation step. The presented methodology is a robust framework for
auto calibration of sensors in a WSN.
A straightforward approach to bias calibration is to apply a known stimulus to the sensor
network and measure the response. Then comparing the ground truth input to the response
will result in nding the gain and offset for the linear drifts case (Hoadley, 1970). This method
is referred to by (Balzano & Nowak, 2007) as non-blind calibration since the ground truth is
used to calibrate the sensors. Another form of non-blind calibration is manually calibrating
a subset of sensors in the sensor network and then allowing the non-calibrated sensors to
adjust their readings based on the calibrated subset. The calibrated subset in this context
form a reference point to the ground truth (Bychkovskiy, 2003; Bychkovskiy et al., 2003). The
above mentioned methods are impractical and cost prohibitive in the case of large scale sensor
networks.
The calibration problem of the sensor network was also tackled by (Balzano & Nowak, 2007;
2008) in a different fashion. They stated that after sensors were calibrated to the factory set-
tings, when deployed, their measurements would differ linearly from the ground truth by
certain gains and offsets for each sensor. They presented a method for estimating these gains
and offsets using subspace matching. The method only required routine measurements to be
collected by the sensors and did not need ground truth measurements for comparison. They
referred to this problem as blind calibration of sensor networks. The method did not require
dense deployment of the sensors or a controlled stimulus. However, It required that the sen-
sor measurements are at least slightly correlated over space i.e. the network over sampled the
underlying signals of interest. The theoretical analysis of their work did not take noise into
consideration and assumed linear calibration functions. Therefore, the solution might not be
robust in noisy conditions and will probably result in wrong estimates if applied in a scenario
where the relationship between the measurement and the ground truth is nonlinear. The eval-
uations they presented showed that the method worked better in a controlled environment.
An earlier work on blind calibration of sensor nodes in a sensor network was presented in
(Bychkovskiy, 2003; Bychkovskiy et al., 2003). They assumed that the sensors of the network
under consideration were sufciently densely deployed that they observed the same phe-
nomenon. They used the temporal correlation of signals received by neighbouring sensors
when the signals were highly correlated to derive a function relating the bias in their am-
plitudes. Another method for calibration was considered by (Feng et al., 2003). They used
geometrical and physical constraints on the behaviour of a point light source to calibrate light
sensors without the need for comparing the measurement with an accurate sensor (ground
truth). They assumed that the light sensors under consideration suffered form a constant bias
with time.
The authors in (Whitehouse & Culler, 2002; 2003) argued that calibrating the sensors in sensor
networks is a problematic task since it comprises large number of sensor that are deployed
in partially unobservable and dynamic environments and may themselves be unobservable.
They suggested that the calibration problemin sensor/actuator networks should be expressed
as a parameter estimation problemon the network scale. Therefore, instead of calibrating each
sensor individually to optimise its measurement, the sensors of the network are calibrated to
optimise the overall response of the network. The joint calibration method they presented cal-
ibrated sensors in a controlled environment. The method was tested on an ad-hoc localisation
system and resulted in reducing the error in the measured distance from 74.6% to 10.1%. The
authors claimed that the joint calibration method could be transformed into an auto calibra-
tion technique for WSNs in an uncontrolled environment i.e. some form of blind calibration
where the value of the ground truth measurement (here the distance) is unknown. They for-
mulated the problemas a quadratic programming problem. Similar to (Whitehouse & Culler,
2002; 2003), blindly calibrating range measurements for localisation purposes between sensors
using received signal strength and/or time delay were considered in (Ihler et al., 2004; Taylor
et al., 2006).
The work of (Elnahrawy & Nath, 2003) aimed to reduce the uncertainties in the sensors read-
ings. It introduced a Bayesian framework for online cleaning of noisy sensor data in WSNs.
The solution was designed to reduce the inuence of random errors in sensors measurements
on the inferences of the sensor network but did not address systematic errors. The framework
was applied in a centralised fashion and on synthetic data set and showed promising results.
The author of (Balzano, 2007) described a method for in-situ blind calibration of moisture
sensors in a sensor network. She used the Ensemble Kalman Filter (EnKF) to correct the values
measured by the sensors, or in other words, to estimate the true moisture at each sensor. The
state equation was governed by a physical model of moisture used in environmental and civil
engineering and the measurements were assumed to be related to the real state by a certain
offset and gain. The state (moisture) vector was augmented with the calibration parameters
(gain and offset) and then the gains and offsets were estimated to recover the correct state
from the measurements.
Another method for detecting a single sensor failure that is a part of an automation system (a
sort of wired sensor network) was proposed by (Sallans et al., 2005). Using the incoming sen-
sor measurement, a model for the sensor behaviour was constructed and then optimised using
an online maximum likelihood algorithm. Sensor readings were compared with the model.
In event that the sensor reading deviated from the modelled value by a certain threshold, the
system labelled this sensor as faulty. On the other hand, when the difference was small, the
system automatically adapted to it. This made the system capable of adapting to slow drifts.
A neural network-based instrument surveillance, calibration and verication system for a
chemical processing system (a sort of wired sensor network) was introduced in (Xu et al.,
1998). The neural network used the correlation in the measurements of the interconnected
sensors to correct the drifting sensors readings. The sensors that were discovered to be faulty
were replaced automatically with the best neural network estimate thus restoring the correct
signal. The performance of the system depended on the degree of correlation of the sensors
readings. It was also found that the robustness of the monitoring network was related to the
amount of signal redundancies and the degree of signal correlations. The authors concluded
that their system could be used to continuously monitor sensors for faults in a plant. How-
ever, they noted that retraining the entire network may be necessary for major changes in
plant operating conditions
Support Vector Machines (SVM) were used in (Rajasegarar et al., 2007) to detect anomalies
and faulty sensors of a sensor network. The data reported by the sensors were mapped from
the input space (the space where the features are observed) to the feature space ( higher di-
mensional space) using kernels. The projected data were then classied into clusters and the
data points that did not lie in a normal data cluster were considered anomalous. The sensor
that always reported anomalous data was considered faulty.
The authors of (Guestrin et al., 2004) presented a method for in-network modelling of sensor
data in a WSN. The method used kernel linear regressionto t functions to the data measured
by the sensors along a time window. The basis functions used were known by the sensors.
Therefore, if a sensor knew the weights of its neighbour, it would be able to answer any query
about the neighbour within the time window. So instead of sending the measured data of the
whole window period from one sensor to another, sending the weights would considerably
reduce the communication overhead. This was one of the aims of the method. The other
aim was to enable any sensor in the network to estimate the measured variable at points
within the network where there were no sensors using the spatial correlation in the network.
An application for the introduced method is computing contour levels of sensor values as in
(Nowak & Mitra, 2003). Even that the work in (Guestrin et al., 2004) considered the unreliable
communication between distant sensors and the noise in sensor readings, it did not address
the systematic errors (drift and bias) which can build up along time and propagate among
sensors causing the continuously modelled functions to produce estimates that deviate from
the ground truth values.
In addition to its superb capabilities in generalisation, function estimation and curve tting,
Support Vector Machines (SVR) is used in other applications such as forecasting and estimat-
ing the physical parameters of a certain phenomenon. In (Wang et al., 2003), SVR was utilised
in medical imaging for nonlinear estimation and modelling of functional magnetic resonance
imaging (fMRI) data to reect their intrinsic spatio-temporal autocorrelations. Moreover, SVR
was used in (Gill et al., 2006) to successfully predict the ground moisture at a site using me-
teorological parameters such as relative humidity, temperature average solar radiation, and
moisture measurements collected from spatially distinct locations. A similar experiment to
predict ground moisture was reported in (Gill et al., 2007). In addition to using the SVR to pre-
dict the moisture measurements ahead in time, they introduced the use of an EnKF to correct
or match the predicted values with the real measurements at certain points of time (whenever
measurements are available) to keep the predicted values close to the measurements taken on
site and eventually reduce the prediction error.
The above survey, has introduced most of the work undertaken in the area of fault detec-
tion and fault detection/correction in wireless sensor networks. This research approaches the
problem in a more comprehensive manner resulting in several novel solutions for detecting
and correcting drift and bias in WSNs. It does not assume linearity of the sensor faults (drift)
with time and addresses smooth drifts and drifts with sudden changes and jumps. It also
considers the cases when the sensors of the network are densely and sparsely (non densely)
deployed. Moreover, it introduces recursive online algorithms for the continuous calibration
of the sensors. In addition to all of that, the solutions presented are decentralised to reduce
the communication overhead. Some of the papers that have arisen from this research are
surveyed below: (Takruri & Challa, 2007) introduced the idea of drift aware wireless sensor
network which detects and corrects sensors drifts and eventually extends the functional life
time of the network. A formal statistical procedure for tracking and detecting smooth sen-
sors drifts using decentralised Kalman Filter (KF) algorithm in a densely deployed network
was introduced in (Takruri, Aboura & Challa, 2008; Takruri, Challa & Chacravorty, 2010). The
sensors of the network were close enough to have similar temperature readings and the av-
erage of their measurements was taken as a sensible estimate to be used by each sensor to
self-assess. As an upgrade for this work, the KFs were replaced in (Takruri, Challa & Chacra-
vorty, 2010; Takruri, Challa & Chakravorty, 2008) by interacting multiple model (IMM) based
lters to deal with unsmooth drifts. A more general solution was considered in (Takruri, Ra-
jasegarar, Challa, Leckie & Palaniswami, 2008). The assumption of dense sensor deployment
was relaxed. Therefore, each sensor in the network ran an SVR algorithm on its neighbours
system and resulted in reducing the error in the measured distance from 74.6% to 10.1%. The
authors claimed that the joint calibration method could be transformed into an auto calibra-
tion technique for WSNs in an uncontrolled environment i.e. some form of blind calibration
where the value of the ground truth measurement (here the distance) is unknown. They for-
mulated the problemas a quadratic programming problem. Similar to (Whitehouse & Culler,
2002; 2003), blindly calibrating range measurements for localisation purposes between sensors
using received signal strength and/or time delay were considered in (Ihler et al., 2004; Taylor
et al., 2006).
The work of (Elnahrawy & Nath, 2003) aimed to reduce the uncertainties in the sensors read-
ings. It introduced a Bayesian framework for online cleaning of noisy sensor data in WSNs.
The solution was designed to reduce the inuence of random errors in sensors measurements
on the inferences of the sensor network but did not address systematic errors. The framework
was applied in a centralised fashion and on synthetic data set and showed promising results.
The author of (Balzano, 2007) described a method for in-situ blind calibration of moisture
sensors in a sensor network. She used the Ensemble Kalman Filter (EnKF) to correct the values
measured by the sensors, or in other words, to estimate the true moisture at each sensor. The
state equation was governed by a physical model of moisture used in environmental and civil
engineering and the measurements were assumed to be related to the real state by a certain
offset and gain. The state (moisture) vector was augmented with the calibration parameters
(gain and offset) and then the gains and offsets were estimated to recover the correct state
from the measurements.
Another method for detecting a single sensor failure that is a part of an automation system (a
sort of wired sensor network) was proposed by (Sallans et al., 2005). Using the incoming sen-
sor measurement, a model for the sensor behaviour was constructed and then optimised using
an online maximum likelihood algorithm. Sensor readings were compared with the model.
In event that the sensor reading deviated from the modelled value by a certain threshold, the
system labelled this sensor as faulty. On the other hand, when the difference was small, the
system automatically adapted to it. This made the system capable of adapting to slow drifts.
A neural network-based instrument surveillance, calibration and verication system for a
chemical processing system (a sort of wired sensor network) was introduced in (Xu et al.,
1998). The neural network used the correlation in the measurements of the interconnected
sensors to correct the drifting sensors readings. The sensors that were discovered to be faulty
were replaced automatically with the best neural network estimate thus restoring the correct
signal. The performance of the system depended on the degree of correlation of the sensors
readings. It was also found that the robustness of the monitoring network was related to the
amount of signal redundancies and the degree of signal correlations. The authors concluded
that their system could be used to continuously monitor sensors for faults in a plant. How-
ever, they noted that retraining the entire network may be necessary for major changes in
plant operating conditions
Support Vector Machines (SVM) were used in (Rajasegarar et al., 2007) to detect anomalies
and faulty sensors of a sensor network. The data reported by the sensors were mapped from
the input space (the space where the features are observed) to the feature space ( higher di-
mensional space) using kernels. The projected data were then classied into clusters and the
data points that did not lie in a normal data cluster were considered anomalous. The sensor
that always reported anomalous data was considered faulty.
The authors of (Guestrin et al., 2004) presented a method for in-network modelling of sensor
data in a WSN. The method used kernel linear regressionto t functions to the data measured
by the sensors along a time window. The basis functions used were known by the sensors.
Therefore, if a sensor knew the weights of its neighbour, it would be able to answer any query
about the neighbour within the time window. So instead of sending the measured data of the
whole window period from one sensor to another, sending the weights would considerably
reduce the communication overhead. This was one of the aims of the method. The other
aim was to enable any sensor in the network to estimate the measured variable at points
within the network where there were no sensors using the spatial correlation in the network.
An application for the introduced method is computing contour levels of sensor values as in
(Nowak & Mitra, 2003). Even that the work in (Guestrin et al., 2004) considered the unreliable
communication between distant sensors and the noise in sensor readings, it did not address
the systematic errors (drift and bias) which can build up along time and propagate among
sensors causing the continuously modelled functions to produce estimates that deviate from
the ground truth values.
In addition to its superb capabilities in generalisation, function estimation and curve tting,
Support Vector Machines (SVR) is used in other applications such as forecasting and estimat-
ing the physical parameters of a certain phenomenon. In (Wang et al., 2003), SVR was utilised
in medical imaging for nonlinear estimation and modelling of functional magnetic resonance
imaging (fMRI) data to reect their intrinsic spatio-temporal autocorrelations. Moreover, SVR
was used in (Gill et al., 2006) to successfully predict the ground moisture at a site using me-
teorological parameters such as relative humidity, temperature average solar radiation, and
moisture measurements collected from spatially distinct locations. A similar experiment to
predict ground moisture was reported in (Gill et al., 2007). In addition to using the SVR to pre-
dict the moisture measurements ahead in time, they introduced the use of an EnKF to correct
or match the predicted values with the real measurements at certain points of time (whenever
measurements are available) to keep the predicted values close to the measurements taken on
site and eventually reduce the prediction error.
The above survey, has introduced most of the work undertaken in the area of fault detec-
tion and fault detection/correction in wireless sensor networks. This research approaches the
problem in a more comprehensive manner resulting in several novel solutions for detecting
and correcting drift and bias in WSNs. It does not assume linearity of the sensor faults (drift)
with time and addresses smooth drifts and drifts with sudden changes and jumps. It also
considers the cases when the sensors of the network are densely and sparsely (non densely)
deployed. Moreover, it introduces recursive online algorithms for the continuous calibration
of the sensors. In addition to all of that, the solutions presented are decentralised to reduce
the communication overhead. Some of the papers that have arisen from this research are
surveyed below: (Takruri & Challa, 2007) introduced the idea of drift aware wireless sensor
network which detects and corrects sensors drifts and eventually extends the functional life
time of the network. A formal statistical procedure for tracking and detecting smooth sen-
sors drifts using decentralised Kalman Filter (KF) algorithm in a densely deployed network
was introduced in (Takruri, Aboura & Challa, 2008; Takruri, Challa & Chacravorty, 2010). The
sensors of the network were close enough to have similar temperature readings and the av-
erage of their measurements was taken as a sensible estimate to be used by each sensor to
self-assess. As an upgrade for this work, the KFs were replaced in (Takruri, Challa & Chacra-
vorty, 2010; Takruri, Challa & Chakravorty, 2008) by interacting multiple model (IMM) based
lters to deal with unsmooth drifts. A more general solution was considered in (Takruri, Ra-
jasegarar, Challa, Leckie & Palaniswami, 2008). The assumption of dense sensor deployment
was relaxed. Therefore, each sensor in the network ran an SVR algorithm on its neighbours
corrected readings to obtain a predicted value for its measurements. It then used this pre-
dicted data to self-assess its measurement, detect (track) its drift using a KF and then correct
the measurement.
A more robust and reliable decentralised algorithm for online sensor calibration in sparsely
deployed wireless sensor networks was presented in (Takruri, Rajasegarar, Challa, Leckie
& Palaniswami, 2010). The algorithm represents a substantial improvement of method in
(Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2008). By using an Unscented Kalman
Filter (UKF) instead of the KF, the bias in the estimated temperature (system error) was
dramatically reduced compared to that reported in (Takruri, Rajasegarar, Challa, Leckie &
Palaniswami, 2008). This is justied by the fact that UKF is a better approximation method
for propagating the mean and covariance of a random variable through a nonlinear trans-
formation than the KF is. The algorithm was then upgraded in (Takruri et al., 2009) to be-
come more adaptable for under sampled sensor measurements and consequently, allowing
for reducing the communication between sensors and maintain the calibration. This led to
reducing the energy consumed from the batteries. Unlike the work in (Balzano, 2007), sta-
tistical modelling rather than physical relations was used to model the spatio-temporal cross
correlations among the sensors measurements. Similar to (Takruri, Rajasegarar, Challa, Leckie
& Palaniswami, 2008), statistical modelling was achieved by applying SVR. This in principal
made the framework applicable to most sensing problems without needing to nd the phys-
ical model that describes the phenomenon under observation, and without the need to abide
by the constraints of that physical formulation. The algorithm runs recursively and is fully
decentralised. It does not make assumptions regarding the linearity of the drifts as opposed
the work in (Balzano & Nowak, 2007). The implementation of the algorithm on real data ob-
tained from the Intel Berkeley research laboratory (IBRL) showed a great success in detecting
and correcting sensors drifts and extending the functional lifetime of the network.
In this chapter, we present another model for error detection and correction in sparsely de-
ployed WSNs. Similar to (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010), SVR is
used to model the spatio-temporal cross correlations among the sensors measurements to ob-
tain a predicted value for the actual ground truth measurements and Unscented Kalman Filter
is used to estimate the corrected sensors readings. However, both algorithms are substantially
different in terms of the training data set used for training the SVR framework, the dynamic
equations that govern the models and the estimated variables. The state transition function in
the new model is taken to be linear resulting in much lower computational complexity than
(Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010) and comparable results.
3. Network Structure and Problem Statement
Consider a wireless sensor network with a large number of sensors distributed randomly in
a certain area of deployment such as the one shown in Figure 1. The sensors are grouped
in clusters (sub-networks) according to their spatial proximity. Each sensor measures a phe-
nomenon such as ambient temperature, chemical concentration, noise or atmospheric pres-
sure. The measurement, say temperature, is considered to be a function of time and space.
As a result, the measurements of sensors that lie within the same cluster can be different from
each other. For example, a sensor closer to a heat source or near direct sunlight will have
readings higher than those in a shaded region or away from the heat source. An example of a
cluster is shown using a circle in Figure 1. The sensors within the cluster are considered to be
capable of communicating their readings among each other.
0 10 20 30 40 50 60 70 80 90 100 110
0
10
20
30
40
50
60
70
80
90
100
110
Length(m)
W
i
d
t
h

(
m
)
Fig. 1. Wireless sensor area with encircled sub-network
As time progresses, some nodes may start experiencing drift in their readings. If these read-
ings are collected and used fromthese nodes, they will cause the users of the network to draw
erroneous conclusions. After some level of unreliability is reached, the network inferences
become untrustworthy. Consequently, the sensor network becomes useless. In order to miti-
gate this problem of drift, each sensor node in the network has to detect and correct its own
drift using the feedback obtained from its neighbouring nodes. This is based on the principle
that the data from nodes that lie within a cluster are correlated, while their faults or drifts
instantiations are likely to be uncorrelated. The ability of the sensor nodes to auto-detect and
correct their drifts helps to extend the effective (useful) lifetime of the network. In addition to
the drift problem, we also consider the inherent bias that may exist within some sensor nodes.
There is a distinct difference between these two types of errors. The former changes with time
and often becomes accentuated, while the latter, is considered to be a constant error from the
beginning of the operation. This error is usually caused by a possible manufacturing defect or
a faulty calibration.
The sensor drift that we consider in this work is slow smooth drift that we model as linear
and/or exponential function of time. It is dependent on the environmental conditions, and
strongly relate to the manufacturing process of the sensor. It is highly unlikely that two elec-
tronic components fail in a correlated manner unless they are fromthe same integrated circuit.
Therefore, we assume that the instantiations of drifts are different from one sensor to another
in a sensor neighbourhood or a cluster. Figure 2 shows examples of the theoretical models for
smooth drift.
Consider a sensor sub-network that consists of n sensors deployed randomly in a certain area
of interest. Without loss of generality, we choose a sensor network measuring temperature,
even though this is generally applicable to all other types of sensors that suffer from drift
and bias problems. Let T be the ground truth temperature. T varies with time and space.
Therefore, we denote the temperature at a certain time instance and sensor location as T
i,k
where i is the sensor number and k is the time index. At each time instant k, node i in the sub-
network measures a reading r
i,k
of T
i,k
. It then estimates and reports a drift corrected value
x
i,k
to its neighbours. The corrected value x
i,k
should ideally be equal to the ground truth
temperature T
i,k
. If all nodes are perfect, r
i,k
will be equal to the T
i,k
, and the reported values
will ideally be equal to the readings, i.e., x
i,k
= r
i,k
.
corrected readings to obtain a predicted value for its measurements. It then used this pre-
dicted data to self-assess its measurement, detect (track) its drift using a KF and then correct
the measurement.
A more robust and reliable decentralised algorithm for online sensor calibration in sparsely
deployed wireless sensor networks was presented in (Takruri, Rajasegarar, Challa, Leckie
& Palaniswami, 2010). The algorithm represents a substantial improvement of method in
(Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2008). By using an Unscented Kalman
Filter (UKF) instead of the KF, the bias in the estimated temperature (system error) was
dramatically reduced compared to that reported in (Takruri, Rajasegarar, Challa, Leckie &
Palaniswami, 2008). This is justied by the fact that UKF is a better approximation method
for propagating the mean and covariance of a random variable through a nonlinear trans-
formation than the KF is. The algorithm was then upgraded in (Takruri et al., 2009) to be-
come more adaptable for under sampled sensor measurements and consequently, allowing
for reducing the communication between sensors and maintain the calibration. This led to
reducing the energy consumed from the batteries. Unlike the work in (Balzano, 2007), sta-
tistical modelling rather than physical relations was used to model the spatio-temporal cross
correlations among the sensors measurements. Similar to (Takruri, Rajasegarar, Challa, Leckie
& Palaniswami, 2008), statistical modelling was achieved by applying SVR. This in principal
made the framework applicable to most sensing problems without needing to nd the phys-
ical model that describes the phenomenon under observation, and without the need to abide
by the constraints of that physical formulation. The algorithm runs recursively and is fully
decentralised. It does not make assumptions regarding the linearity of the drifts as opposed
the work in (Balzano & Nowak, 2007). The implementation of the algorithm on real data ob-
tained from the Intel Berkeley research laboratory (IBRL) showed a great success in detecting
and correcting sensors drifts and extending the functional lifetime of the network.
In this chapter, we present another model for error detection and correction in sparsely de-
ployed WSNs. Similar to (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010), SVR is
used to model the spatio-temporal cross correlations among the sensors measurements to ob-
tain a predicted value for the actual ground truth measurements and Unscented Kalman Filter
is used to estimate the corrected sensors readings. However, both algorithms are substantially
different in terms of the training data set used for training the SVR framework, the dynamic
equations that govern the models and the estimated variables. The state transition function in
the new model is taken to be linear resulting in much lower computational complexity than
(Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010) and comparable results.
3. Network Structure and Problem Statement
Consider a wireless sensor network with a large number of sensors distributed randomly in
a certain area of deployment such as the one shown in Figure 1. The sensors are grouped
in clusters (sub-networks) according to their spatial proximity. Each sensor measures a phe-
nomenon such as ambient temperature, chemical concentration, noise or atmospheric pres-
sure. The measurement, say temperature, is considered to be a function of time and space.
As a result, the measurements of sensors that lie within the same cluster can be different from
each other. For example, a sensor closer to a heat source or near direct sunlight will have
readings higher than those in a shaded region or away from the heat source. An example of a
cluster is shown using a circle in Figure 1. The sensors within the cluster are considered to be
capable of communicating their readings among each other.
0 10 20 30 40 50 60 70 80 90 100 110
0
10
20
30
40
50
60
70
80
90
100
110
Length(m)
W
i
d
t
h

(
m
)
Fig. 1. Wireless sensor area with encircled sub-network
As time progresses, some nodes may start experiencing drift in their readings. If these read-
ings are collected and used fromthese nodes, they will cause the users of the network to draw
erroneous conclusions. After some level of unreliability is reached, the network inferences
become untrustworthy. Consequently, the sensor network becomes useless. In order to miti-
gate this problem of drift, each sensor node in the network has to detect and correct its own
drift using the feedback obtained from its neighbouring nodes. This is based on the principle
that the data from nodes that lie within a cluster are correlated, while their faults or drifts
instantiations are likely to be uncorrelated. The ability of the sensor nodes to auto-detect and
correct their drifts helps to extend the effective (useful) lifetime of the network. In addition to
the drift problem, we also consider the inherent bias that may exist within some sensor nodes.
There is a distinct difference between these two types of errors. The former changes with time
and often becomes accentuated, while the latter, is considered to be a constant error from the
beginning of the operation. This error is usually caused by a possible manufacturing defect or
a faulty calibration.
The sensor drift that we consider in this work is slow smooth drift that we model as linear
and/or exponential function of time. It is dependent on the environmental conditions, and
strongly relate to the manufacturing process of the sensor. It is highly unlikely that two elec-
tronic components fail in a correlated manner unless they are fromthe same integrated circuit.
Therefore, we assume that the instantiations of drifts are different from one sensor to another
in a sensor neighbourhood or a cluster. Figure 2 shows examples of the theoretical models for
smooth drift.
Consider a sensor sub-network that consists of n sensors deployed randomly in a certain area
of interest. Without loss of generality, we choose a sensor network measuring temperature,
even though this is generally applicable to all other types of sensors that suffer from drift
and bias problems. Let T be the ground truth temperature. T varies with time and space.
Therefore, we denote the temperature at a certain time instance and sensor location as T
i,k
where i is the sensor number and k is the time index. At each time instant k, node i in the sub-
network measures a reading r
i,k
of T
i,k
. It then estimates and reports a drift corrected value
x
i,k
to its neighbours. The corrected value x
i,k
should ideally be equal to the ground truth
temperature T
i,k
. If all nodes are perfect, r
i,k
will be equal to the T
i,k
, and the reported values
will ideally be equal to the readings, i.e., x
i,k
= r
i,k
.
0 10 20 30 40 50 60 70 80 90 100
3
2
1
0
1
2
3
4
Time steps
D
r
i
f
t
Fig. 2. Examples of smooth drifts
To estimate the corrected value x
i,k
, each node i rst nds a predictedvalue x
i,k
for its tempera-
ture as a function of the corrected measurements collected fromits neighbours in the previous
time step using x
i,k
= f ({x
j,k1
}
n
j=1,j=i
). Then it fuses this predicted value together with its
measurement r
i,k
and the projected drift d
i,k
to result in an error corrected sensor measurement
x
i,k
. In practice, each sensor reading comes with an associated random reading error (noise),
and a drift d
i,k
. This drift may be null or insignicant during the initial period of deployment,
depending on the nature of the sensor and the deployment environment. The problem we
address here is how to account for the drift in each sensor node i, using the predicted value
x
i,k
, so that the reading r
i,k
is corrected and reported as x
i,k
.
In the following sections, x
i,k
is computed using a support vector regression (SVR) modelled
function that takes into account the temporal and spatial correlations of the sensor measure-
ments. In this work, SVR approximates x
i,k
using the previous corrected readings of all the
sensors in the neighbourhood (cluster) excluding the sensor itself x
i,k
= f ({x
j,k1
}
n
j=1,j=i
).
4. Modelling and predicting measurements using Support Vector Regression
The purpose of using Support Vector Regression (SVR) is to predict the actual sensor mea-
surements x
i,k
of a sensor node i at time instant k using the corrected measurements from
neighbouring sensors. The intention is that each sensor learns a model function f (.) that can
be used for predicting its subsequent actual (error free) measurements through out the whole
period of the experiment. SVR implements this in two phases, namely the training phase and
the running phase. During the training phase, sensor measurements collected during the initial
deployment period (training data set) are used to model the function f (.). During the running
phase, the trained model f (.) is used to predict the subsequent actual sensor measurements
x
i,k
.
We assume that the training data (collected during the initial periods of deployment) is void of
any drift and can be used for training the SVR at each node. This is a reasonable assumption
in practice, as the sensors are usually calibrated before deployment to ensure that they are
working in order. Similar to our work in (Takruri, Rajasegarar, Challa, Leckie & Palaniswami,
2010), we use the widely used Gaussian kernel SVR for our evaluations (Scholkopf & Smola,
2002). However, the training data set used here is slightly different in that it comprises the
corrected readings of the neigbours and does not take into consideration the corrected reading
of node i itself. The training data set at each node i is given by X
s
= (TrX, TrZ), where
TrX = {x
j,k1
: j = 1...n, k = 1...m, j = i}, TrZ = {x
i,k
: k = 1...m} and m is number of
training data vectors. A detailed explanation of our implementation of the SVR can be found
in (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010).
The model obtained via SVR training is then used during the running phase for predicting
subsequent actual measurements x
i,k
. The difference between the sensor reading r
i,k
and the
SVR modelled value x
i,k
, y
(2)
i,k
, which we refer to as the drift measurement of node i at time
instant k, is used by an Unscented Kalman Filter together with r
i,k
to estimate the corrected
reading x
i,k
and the drift d
i,k
as will be shown in the following section.
5. Iterative measurement estimation and correction using an SVR-UKF framework
The solution to the smooth drift problem consists of the following iterative steps. At stage k,
a reading r
i,k
is made by node i. The node also has a prediction for its corrected measurement
(actual temperature at this sensor), x
i,k
= f ({x
j,k1
}
n
j=1,j=i
), as a function of the corrected
measurements of all neighbouring sensors in the cluster from the previous time step. Using
this predicted value ( x
i,k
) together with r
i,k
, the corrected reading x
i,k
and the drift value d
i,k
are estimated. The node then sends the corrected sensor value x
i,k
to its neighbours. After
that, each node collects the neighbourhood corrected measurements and computes x
i,k
and so
on. It is important here to emphasise that our main objective is to estimate x
i,k
the corrected
reading which represents our estimate for the ground truth value T
i,k
at node i. Assuming
that x
i,k
and d
i,k
change slowly with time the dynamics of x
i,k
and d
i,k
are mathematically
described by:
x
i,k
= x
i,k1
+
(1)
i,k

(1)
i,k
N(0, Q
(1)
i,k
) (1)
d
i,k
= d
i,k1
+
(2)
i,k

(2)
i,k
N(0, Q
(2)
i,k
) (2)
where
(1)
i,k
and
(2)
i,k
are the process noises. They are taken to be uncorrelated Gaussian noises
with zero means and variances Q
(1)
i,k
and Q
(2)
i,k
, respectively.
The value x
i,k
is never sensed or measured. What is really measured is r
i,k
, the reading of the
sensor. As we argued earlier, r
i,k
deviates from x
i,k
by both systematic and randomerrors. The
random error is taken to be a Gaussian noise w
i,k
N(0, R
i,k
) with zero mean and variance
R
i,k
(measurement noise variance). The systematic error is referred to as the drift d
i,k
. This
leads to (3).
y
(1)
i,k
= r
i,k
= x
i,k
+ d
i,k
+ w
i,k
w
i,k
N(0, R
i,k
) (3)
We also dene y
(2)
i,k
as the difference between the measurement r
i,k
and the SVR modelled
value x
i,k
and refer to y
(2)
i,k
as the drift measurement of node i at time instant k.
y
(2)
i,k
= y
(1)
i,k
f ({x
j,k1
}
n
j=1,j=i
)
= x
i,k
+ d
i,k
+ w
i,k
f ({x
j,k1
}
n
j=1,j=i
)
= x
i,k
+ d
i,k
+ w
i,k
x
i,k
w
i,k
N(0, R
i,k
) (4)
0 10 20 30 40 50 60 70 80 90 100
3
2
1
0
1
2
3
4
Time steps
D
r
i
f
t
Fig. 2. Examples of smooth drifts
To estimate the corrected value x
i,k
, each node i rst nds a predictedvalue x
i,k
for its tempera-
ture as a function of the corrected measurements collected fromits neighbours in the previous
time step using x
i,k
= f ({x
j,k1
}
n
j=1,j=i
). Then it fuses this predicted value together with its
measurement r
i,k
and the projected drift d
i,k
to result in an error corrected sensor measurement
x
i,k
. In practice, each sensor reading comes with an associated random reading error (noise),
and a drift d
i,k
. This drift may be null or insignicant during the initial period of deployment,
depending on the nature of the sensor and the deployment environment. The problem we
address here is how to account for the drift in each sensor node i, using the predicted value
x
i,k
, so that the reading r
i,k
is corrected and reported as x
i,k
.
In the following sections, x
i,k
is computed using a support vector regression (SVR) modelled
function that takes into account the temporal and spatial correlations of the sensor measure-
ments. In this work, SVR approximates x
i,k
using the previous corrected readings of all the
sensors in the neighbourhood (cluster) excluding the sensor itself x
i,k
= f ({x
j,k1
}
n
j=1,j=i
).
4. Modelling and predicting measurements using Support Vector Regression
The purpose of using Support Vector Regression (SVR) is to predict the actual sensor mea-
surements x
i,k
of a sensor node i at time instant k using the corrected measurements from
neighbouring sensors. The intention is that each sensor learns a model function f (.) that can
be used for predicting its subsequent actual (error free) measurements through out the whole
period of the experiment. SVR implements this in two phases, namely the training phase and
the running phase. During the training phase, sensor measurements collected during the initial
deployment period (training data set) are used to model the function f (.). During the running
phase, the trained model f (.) is used to predict the subsequent actual sensor measurements
x
i,k
.
We assume that the training data (collected during the initial periods of deployment) is void of
any drift and can be used for training the SVR at each node. This is a reasonable assumption
in practice, as the sensors are usually calibrated before deployment to ensure that they are
working in order. Similar to our work in (Takruri, Rajasegarar, Challa, Leckie & Palaniswami,
2010), we use the widely used Gaussian kernel SVR for our evaluations (Scholkopf & Smola,
2002). However, the training data set used here is slightly different in that it comprises the
corrected readings of the neigbours and does not take into consideration the corrected reading
of node i itself. The training data set at each node i is given by X
s
= (TrX, TrZ), where
TrX = {x
j,k1
: j = 1...n, k = 1...m, j = i}, TrZ = {x
i,k
: k = 1...m} and m is number of
training data vectors. A detailed explanation of our implementation of the SVR can be found
in (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010).
The model obtained via SVR training is then used during the running phase for predicting
subsequent actual measurements x
i,k
. The difference between the sensor reading r
i,k
and the
SVR modelled value x
i,k
, y
(2)
i,k
, which we refer to as the drift measurement of node i at time
instant k, is used by an Unscented Kalman Filter together with r
i,k
to estimate the corrected
reading x
i,k
and the drift d
i,k
as will be shown in the following section.
5. Iterative measurement estimation and correction using an SVR-UKF framework
The solution to the smooth drift problem consists of the following iterative steps. At stage k,
a reading r
i,k
is made by node i. The node also has a prediction for its corrected measurement
(actual temperature at this sensor), x
i,k
= f ({x
j,k1
}
n
j=1,j=i
), as a function of the corrected
measurements of all neighbouring sensors in the cluster from the previous time step. Using
this predicted value ( x
i,k
) together with r
i,k
, the corrected reading x
i,k
and the drift value d
i,k
are estimated. The node then sends the corrected sensor value x
i,k
to its neighbours. After
that, each node collects the neighbourhood corrected measurements and computes x
i,k
and so
on. It is important here to emphasise that our main objective is to estimate x
i,k
the corrected
reading which represents our estimate for the ground truth value T
i,k
at node i. Assuming
that x
i,k
and d
i,k
change slowly with time the dynamics of x
i,k
and d
i,k
are mathematically
described by:
x
i,k
= x
i,k1
+
(1)
i,k

(1)
i,k
N(0, Q
(1)
i,k
) (1)
d
i,k
= d
i,k1
+
(2)
i,k

(2)
i,k
N(0, Q
(2)
i,k
) (2)
where
(1)
i,k
and
(2)
i,k
are the process noises. They are taken to be uncorrelated Gaussian noises
with zero means and variances Q
(1)
i,k
and Q
(2)
i,k
, respectively.
The value x
i,k
is never sensed or measured. What is really measured is r
i,k
, the reading of the
sensor. As we argued earlier, r
i,k
deviates from x
i,k
by both systematic and randomerrors. The
random error is taken to be a Gaussian noise w
i,k
N(0, R
i,k
) with zero mean and variance
R
i,k
(measurement noise variance). The systematic error is referred to as the drift d
i,k
. This
leads to (3).
y
(1)
i,k
= r
i,k
= x
i,k
+ d
i,k
+ w
i,k
w
i,k
N(0, R
i,k
) (3)
We also dene y
(2)
i,k
as the difference between the measurement r
i,k
and the SVR modelled
value x
i,k
and refer to y
(2)
i,k
as the drift measurement of node i at time instant k.
y
(2)
i,k
= y
(1)
i,k
f ({x
j,k1
}
n
j=1,j=i
)
= x
i,k
+ d
i,k
+ w
i,k
f ({x
j,k1
}
n
j=1,j=i
)
= x
i,k
+ d
i,k
+ w
i,k
x
i,k
w
i,k
N(0, R
i,k
) (4)
The model is expressed in vector notation as follows:
X
i,k
=
x
i,k
d
i,k
1 0
0 1

x
i,k1
d
i,k1

(1)
i,k
(2)
i,k
(5)
Y
i,k
=
y
(1)
i,k
y
(2)
i,k
1 1
1 1

x
i,k
d
i,k
w
i,k
w
i,k
0
x
i,k
(6)
The noise component associated with X
i,k
is Gaussian with mean vector
X
i,k
= [0 0]
T
and
covariance matrix Qx
i,k
=
Q
(1)
i,k
0
0 Q
(2)
i,k
. The noise component associated with Y

i,k
has a
mean vector
Y
i,k
= [0 0]
T
and covariance matrix Ry
i,k
=
R
i,k
R
i,k
R
i,k
R
i,k
which indicates that

it is not White Gaussian. The system is clearly observable when x
i,k
= x
i,k
, i.e. when x
i,k
is a
true, bias free, representation of x
i,k
and the difference between x
i,k
and x
i,k
is zero.
Since the noise component associated with Y
i,k
is not White Gaussian, the KF cannot be used
(Lu et al., 2007) to estimate x
i,k
and d
i,k
. Another lter that can be used for solving such a
problem is the Particle Filter. Unfortunately, the high computational complexity of the Par-
ticle Filter makes it unsuitable for the use in WSNs, where the sensors are limited in their
energy and computational capabilities. A better alternative is to use the UKF. The Unscented
Transformation (UT) was introduced by Julier et al. in (Julier et al., 1995) as an approximation
method for propagating the mean and covariance of a random variable through a nonlinear
transformation. This method was used to derive UKF in (Julier & Uhlmann, 1997). UKF can
deal with versatile and complicated nonlinear sensor models and non-Gaussian noise that
are not necessarily additive (Challa et al., 2008) with a comparable computational complexity
to the Extended Kalman Filter (EKF) (Wan & van der Merwe, 2000). It also outperforms the
EKF since it provides better estimation for the posterior mean and covariance to the third or-
der Taylor series expansion when the input is Gaussian, whereas, the EKF, only achieves the
rst order Taylor series expansion (Wan & van der Merwe, 2000). Below, we explain the UKF
algorithm in detail.
The UT as mentioned before is a method for nding the statistics of a random variable
Z = g(X) which undergoes nonlinear transformation. Let X of dimension L be the ran-
dom variable that is propagated through the nonlinear function Z = g(X). Assume that X
has a mean

X and a covariance P. According to (Challa et al., 2008), to nd the statistics of
Z using the scaled unscented transformation, which was introduced in (Julier, 2002), the fol-
lowing steps must be followed: First, 2L +1 (where L is the dimension of vector X) weighted
samples or sigma points
i
= {W
i
, X
i
} are deterministically chosen to completely capture the
true mean and covariance of the random variable X. Then, the sigma points are propagated
through the function g(X) to capture the statistics (mean and covariance) of Z. A selection
scheme that satises the requirement is given below:
X
0
=

X, W
m
0
=

+ L
W
c
0
=

+ L
+ (1
2
+ )
X
i
=

X + (
(L + )P)
i
, W
i
=
1
2( + L)
X
L+i
=

X (
(L + )P)
i
, W
L+i
=
1
2( + L)
(7)
where i = 1, ..., L and =
2
(L + ) L is a scaling parameter. determines the spread of
the sigma points around the mean

X and is usually set to a small positive value (e.g., 0.001).
is a secondary scaling parameter which is usually set to 0, and is used to incorporate prior
knowledge of the distribution of X. The optimal value of for a Gaussian distribution is = 2
as stated in (Wan & van der Merwe, 2000). The term (
(L + )P)
i
is the ith row of the matrix
square root of matrix (L +)P. In our work here , and are taken to be equal to 0.001, 0, 2,
respectively. The UKF is used to estimate X
i,k
for sensor i at time step k. The dimension L of
X
i,k
is equal to 2. This means that we only have ve sigma points for each node i. The steps of
the UKF algorithm are given below as in (Challa et al., 2008):
Let

X
i,k1|k1
be the prior mean of the state variable and P
i,k1|k1
be the associated covari-
ance for node i. To simplify the notation we write the prior mean of the state variable and
the associated covariance as

X
k1|k1
and P
k1|k1
(without showing the sensor number i)
keeping in mind that they refer to a certain sensor node i. This also applies for all the other
parameters we use in describing the UKF algorithm.
The sigma points are calculated from (7) and then propagated through the state equation
function g(.). This results in X
0,k|k1
, X
1,k|k1
, X
2,k|k1
, X
3,k|k1
and X
4,k|k1
as shown in (8).
X
k|k1
= g(X
k1
) = X
k1
(8)
The predicted mean and covariance of the state variable are given by (9) and (10), respectively.
X
k|k1
= W
m
0
X
0,k|k1
+
2L
i=1
W
i
X
i,k|k1
(9)
P
k|k1
= W
c
0
(X
0,k|k1

X
k|k1
)(X
0,k|k1

X
k|k1
)
T
+
2L
i=1
W
i
(X
i,k|k1

X
k|k1
)(X
i,k|k1

X
k|k1
)
T
+ Qx
k
(10)
The propagated sigma points are then passed through the measurement function h(.)
as shown in (11).
Y
k|k1
= h(X
k|k1
) (11)
Then the predicted mean and covariance of each sensor measurement are given by (12) and
(13), respectively.
Y
k|k1
= W
m
0
Y
0,k|k1
+
2L
i=1
W
i
Y
i,k|k1
(12)
The model is expressed in vector notation as follows:
X
i,k
=
x
i,k
d
i,k
1 0
0 1

x
i,k1
d
i,k1

(1)
i,k
(2)
i,k
(5)
Y
i,k
=
y
(1)
i,k
y
(2)
i,k
1 1
1 1

x
i,k
d
i,k
w
i,k
w
i,k
0
x
i,k
(6)
The noise component associated with X
i,k
is Gaussian with mean vector
X
i,k
= [0 0]
T
and
covariance matrix Qx
i,k
=
Q
(1)
i,k
0
0 Q
(2)
i,k
. The noise component associated with Y

i,k
has a
mean vector
Y
i,k
= [0 0]
T
and covariance matrix Ry
i,k
=
R
i,k
R
i,k
R
i,k
R
i,k
which indicates that

it is not White Gaussian. The system is clearly observable when x
i,k
= x
i,k
, i.e. when x
i,k
is a
true, bias free, representation of x
i,k
and the difference between x
i,k
and x
i,k
is zero.
Since the noise component associated with Y
i,k
is not White Gaussian, the KF cannot be used
(Lu et al., 2007) to estimate x
i,k
and d
i,k
. Another lter that can be used for solving such a
problem is the Particle Filter. Unfortunately, the high computational complexity of the Par-
ticle Filter makes it unsuitable for the use in WSNs, where the sensors are limited in their
energy and computational capabilities. A better alternative is to use the UKF. The Unscented
Transformation (UT) was introduced by Julier et al. in (Julier et al., 1995) as an approximation
method for propagating the mean and covariance of a random variable through a nonlinear
transformation. This method was used to derive UKF in (Julier & Uhlmann, 1997). UKF can
deal with versatile and complicated nonlinear sensor models and non-Gaussian noise that
are not necessarily additive (Challa et al., 2008) with a comparable computational complexity
to the Extended Kalman Filter (EKF) (Wan & van der Merwe, 2000). It also outperforms the
EKF since it provides better estimation for the posterior mean and covariance to the third or-
der Taylor series expansion when the input is Gaussian, whereas, the EKF, only achieves the
rst order Taylor series expansion (Wan & van der Merwe, 2000). Below, we explain the UKF
algorithm in detail.
The UT as mentioned before is a method for nding the statistics of a random variable
Z = g(X) which undergoes nonlinear transformation. Let X of dimension L be the ran-
dom variable that is propagated through the nonlinear function Z = g(X). Assume that X
has a mean

X and a covariance P. According to (Challa et al., 2008), to nd the statistics of
Z using the scaled unscented transformation, which was introduced in (Julier, 2002), the fol-
lowing steps must be followed: First, 2L +1 (where L is the dimension of vector X) weighted
samples or sigma points
i
= {W
i
, X
i
} are deterministically chosen to completely capture the
true mean and covariance of the random variable X. Then, the sigma points are propagated
through the function g(X) to capture the statistics (mean and covariance) of Z. A selection
scheme that satises the requirement is given below:
X
0
=

X, W
m
0
=

+ L
W
c
0
=

+ L
+ (1
2
+ )
X
i
=

X + (
(L + )P)
i
, W
i
=
1
2( + L)
X
L+i
=

X (
(L + )P)
i
, W
L+i
=
1
2( + L)
(7)
where i = 1, ..., L and =
2
(L + ) L is a scaling parameter. determines the spread of
the sigma points around the mean

X and is usually set to a small positive value (e.g., 0.001).
is a secondary scaling parameter which is usually set to 0, and is used to incorporate prior
knowledge of the distribution of X. The optimal value of for a Gaussian distribution is = 2
as stated in (Wan & van der Merwe, 2000). The term (
(L + )P)
i
is the ith row of the matrix
square root of matrix (L +)P. In our work here , and are taken to be equal to 0.001, 0, 2,
respectively. The UKF is used to estimate X
i,k
for sensor i at time step k. The dimension L of
X
i,k
is equal to 2. This means that we only have ve sigma points for each node i. The steps of
the UKF algorithm are given below as in (Challa et al., 2008):
Let

X
i,k1|k1
be the prior mean of the state variable and P
i,k1|k1
be the associated covari-
ance for node i. To simplify the notation we write the prior mean of the state variable and
the associated covariance as

X
k1|k1
and P
k1|k1
(without showing the sensor number i)
keeping in mind that they refer to a certain sensor node i. This also applies for all the other
parameters we use in describing the UKF algorithm.
The sigma points are calculated from (7) and then propagated through the state equation
function g(.). This results in X
0,k|k1
, X
1,k|k1
, X
2,k|k1
, X
3,k|k1
and X
4,k|k1
as shown in (8).
X
k|k1
= g(X
k1
) = X
k1
(8)
The predicted mean and covariance of the state variable are given by (9) and (10), respectively.
X
k|k1
= W
m
0
X
0,k|k1
+
2L
i=1
W
i
X
i,k|k1
(9)
P
k|k1
= W
c
0
(X
0,k|k1

X
k|k1
)(X
0,k|k1

X
k|k1
)
T
+
2L
i=1
W
i
(X
i,k|k1

X
k|k1
)(X
i,k|k1

X
k|k1
)
T
+ Qx
k
(10)
The propagated sigma points are then passed through the measurement function h(.)
as shown in (11).
Y
k|k1
= h(X
k|k1
) (11)
Then the predicted mean and covariance of each sensor measurement are given by (12) and
(13), respectively.
Y
k|k1
= W
m
0
Y
0,k|k1
+
2L
i=1
W
i
Y
i,k|k1
(12)
P
Y
k
Y
k
= W
c
0
(Y
0,k|k1

Y
k|k1
)(Y
0,k|k1

Y
k|k1
)
T
+
2L
i=1
W
i
(Y
i,k|k1

Y
k|k1
)(Y
i,k|k1

Y
k|k1
)
T
+ Ry
k
(13)
The cross covariance of the predicted state and sensor measurement is found by (14).
P
X
k
Y
k
= W
c
0
(X
0,k|k1

X
k|k1
)(Y
0,k|k1

Y
k|k1
)
T
+
2L
i=1
[W
i
(X
i,k|k1

X
k|k1
)(Y
i,k|k1

Y
k|k1
)
T
] (14)
where
K
k
= P
X
k
Y
k
P
1
Y
k
Y
k
(15)
The updated posterior mean and covariance of the state are then estimated by (16) and (17),
respectively.
X
k|k
=

X
k|k1
+ K
k
(Y
k

Y
k|k1
) (16)
P
k|k
= P
k|k1
+ K
k
P
Y
k
Y
k
K
T
k
(17)
where

X
k|k
and P
k|k
are the mean and covariance of the state of node i at time step k.
Figure 3 shows a block diagram of our drift correction algorithm. It clearly summarises the
stages of the error detection and correction framework in one of the nodes in the cluster. The
steps of the algorithm are stated below:
Decentralised error correction algorithm using the SVR-UKF framework
At time step k
Each node i nds its predicted corrected measurement x
i,k
=f ({x
j,k1
}
n
j=1,j=i
).
Each node i obtains its reading y
(1)
i,k
= r
i,k
.
Each node i calculates the drift measurement y
(2)
i,k
.
Each node i nds the sigma points
i
= {W
i
, X
i
} from

X
i,k1|k1
= [ x
i,k1|k1
d
i,k1|k1
]
T
.
For each node i, the sigma points are propagated through the state equation function
g(.).
The UKF estimates the corrected measurement and the drift using (9)-(17) and then
sends the result to the neighbouring nodes.
The algorithm reiterates.
6. Evaluation
Our aim is to evaluate the ability of our proposed framework to correct the drift experienced
in sensor nodes and to extend the functional life of the sensor network. The data in our eval-
uation are a set of real sensor measurements gathered from a deployment of wireless sensors
in the IBRL (2006, Accessed on 07/09/2006).
In 2004, a set of wireless sensors with 55 sensor nodes (including a gateway node) were de-
ployed in the IBRL lab for monitoring the lab environment (refer to Figure 4). They recorded
Fig. 3. The SVR-UKF Measurement correction framework at node i.
temperature, humidity, light and voltage measurements at 30 seconds intervals during the
period starting from 28
th
February 2004 to 5
th
April 2004.
The data from the sensor nodes are re-sampled at seven minute intervals and the rst 2000
samples are used for our evaluation purposes. This corresponds to the data collected during
a ten day period from 28
th
February 2004 to 9
th
March 2004. We use the rst 1000 samples
(this corresponds to the rst ve days data) as the training set for use in the training phase. An
exponential drift is introduced to the real data in each node, starting randomly after the rst
1000 samples. The data after 1000 samples and up to 2000 samples are used in the running
phase for testing our algorithm for drift correction. These samples correspond to the next ve
days of the IBRL data. Temperature measurements are used in all our evaluations.
We formed a network of sensors using nodes selected from the IBRL deployment using six-
teen sensor nodes. The node IDs used are {1,2,3,4,6,7,8,9,10,31,31,33,34,35,36,37}. Each sensor
communicates only with its closest 8 neighbours.
Fig. 4. Sensor nodes in the IBRL deployment. Nodes are shown in black with their corre-
sponding node-IDs. Node 0 is the gateway node (2006, Accessed on 07/09/2006).
P
Y
k
Y
k
= W
c
0
(Y
0,k|k1

Y
k|k1
)(Y
0,k|k1

Y
k|k1
)
T
+
2L
i=1
W
i
(Y
i,k|k1

Y
k|k1
)(Y
i,k|k1

Y
k|k1
)
T
+ Ry
k
(13)
The cross covariance of the predicted state and sensor measurement is found by (14).
P
X
k
Y
k
= W
c
0
(X
0,k|k1

X
k|k1
)(Y
0,k|k1

Y
k|k1
)
T
+
2L
i=1
[W
i
(X
i,k|k1

X
k|k1
)(Y
i,k|k1

Y
k|k1
)
T
] (14)
where
K
k
= P
X
k
Y
k
P
1
Y
k
Y
k
(15)
The updated posterior mean and covariance of the state are then estimated by (16) and (17),
respectively.
X
k|k
=

X
k|k1
+ K
k
(Y
k

Y
k|k1
) (16)
P
k|k
= P
k|k1
+ K
k
P
Y
k
Y
k
K
T
k
(17)
where

X
k|k
and P
k|k
are the mean and covariance of the state of node i at time step k.
Figure 3 shows a block diagram of our drift correction algorithm. It clearly summarises the
stages of the error detection and correction framework in one of the nodes in the cluster. The
steps of the algorithm are stated below:
Decentralised error correction algorithm using the SVR-UKF framework
At time step k
Each node i nds its predicted corrected measurement x
i,k
=f ({x
j,k1
}
n
j=1,j=i
).
Each node i obtains its reading y
(1)
i,k
= r
i,k
.
Each node i calculates the drift measurement y
(2)
i,k
.
Each node i nds the sigma points
i
= {W
i
, X
i
} from

X
i,k1|k1
= [ x
i,k1|k1
d
i,k1|k1
]
T
.
For each node i, the sigma points are propagated through the state equation function
g(.).
The UKF estimates the corrected measurement and the drift using (9)-(17) and then
sends the result to the neighbouring nodes.
The algorithm reiterates.
6. Evaluation
Our aim is to evaluate the ability of our proposed framework to correct the drift experienced
in sensor nodes and to extend the functional life of the sensor network. The data in our eval-
uation are a set of real sensor measurements gathered from a deployment of wireless sensors
in the IBRL (2006, Accessed on 07/09/2006).
In 2004, a set of wireless sensors with 55 sensor nodes (including a gateway node) were de-
ployed in the IBRL lab for monitoring the lab environment (refer to Figure 4). They recorded
Fig. 3. The SVR-UKF Measurement correction framework at node i.
temperature, humidity, light and voltage measurements at 30 seconds intervals during the
period starting from 28
th
February 2004 to 5
th
April 2004.
The data from the sensor nodes are re-sampled at seven minute intervals and the rst 2000
samples are used for our evaluation purposes. This corresponds to the data collected during
a ten day period from 28
th
February 2004 to 9
th
March 2004. We use the rst 1000 samples
(this corresponds to the rst ve days data) as the training set for use in the training phase. An
exponential drift is introduced to the real data in each node, starting randomly after the rst
1000 samples. The data after 1000 samples and up to 2000 samples are used in the running
phase for testing our algorithm for drift correction. These samples correspond to the next ve
days of the IBRL data. Temperature measurements are used in all our evaluations.
We formed a network of sensors using nodes selected from the IBRL deployment using six-
teen sensor nodes. The node IDs used are {1,2,3,4,6,7,8,9,10,31,31,33,34,35,36,37}. Each sensor
communicates only with its closest 8 neighbours.
Fig. 4. Sensor nodes in the IBRL deployment. Nodes are shown in black with their corre-
sponding node-IDs. Node 0 is the gateway node (2006, Accessed on 07/09/2006).
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
3 sensors drifting
6 sensors drifting
9 sensors drifting
12 sensors
drifting
15 sensors
drifting
Threshold line
Fig. 5. Mean Absolute Error for the network without correction.
Our algorithm is implemented in MatLab, utilising the SVR toolbox from (Canu et al., 2005)
and the UKF toolbox from (Srkk & Hartikainen, 2007). For comparison purposes, we run
the algorithm on two data sets. One, the data without the introduced drift (WOD), and the
other, the data with drift introduced (WD). Initially, the SVR of each node is trained on the
rst 1000 samples of its neighbours readings.
The UKF parameters , and are set to the default values as explained in Section 5. Through
out our evaluations, we take Q
(1)
i,k
= Q
(2)
i,k
= Q
i,k
. Q
i,k
and R
i,k
are tuned using trial and error
for both cases. The values used in our evaluation are Q
i,k
= 0.001 and R
i,k
= 0.02. If R
i,k
is
set to a high value, the estimated temperature will follow the reading (which may have drift)
whereas if R
i,k
is set to a small value, the estimated temperature will not be able to follow the
real temperature. Thus, it will not totally correct the error. On the other hand, a high value for
Q
i,k
will result in oscillatory estimates and lead to an unstable state. Hence, a trade off has to
be considered in selecting the values for Q
i,k
and R
i,k
to obtain the best results.
We have conducted two simulations using two data sets. One data set has no drifts intro-
duced. We denote this data set by R-WOD, which stands for Readings Without Drift and
represents the sensor measurements that only suffer fromnoise. The other is the same data set
with drifts introduced in several scenarios. We denote the readings of this data set by R-WD,
which stands for Readings With Drift and represents the sensor measurements that suffer
from both drift and noise. The drift scenarios considered in R-WD are as follows: scenario 1
(SCN 1) being one node drifting, scenario 2 (SCN 2) being two nodes drifting and so on until
the last scenario (SCN 16) having all nodes drifting. The resulting corrected measurements
obtained when the algorithm is run on the R-WD data sets are denoted by DCM-WD, which
stands for Drift Corrected Measurement for readings With Drift. Similarly, the corrected
measurements obtained using data set R-WOD are denoted by DCM-WOD, which stands for
Drift Corrected Measurement for readings WithOut Drift.
To evaluate the performance of our algorithm from the networks point of view, we compare
the average absolute error of all the sensors of the network with and without implementing
our drift correction algorithm.
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
Threshold line 6 Sensors
drifting
9 Sensors
drifting
12 Sensors
drifting
15 Sensors
drifting
3 Sensors
drifting
Fig. 6. Mean Absolute Error for the network with correction for 2001 samples in 10 days.
Figure 5 shows the mean absolute error between the true temperatures (R-WOD) and the
values reported by the sensors (R-WD) for the whole network, for ve different scenarios. The
mean absolute error of the network is computed for each scenario as follows: for each node, at
each instant of time, the absolute error between the true temperature (R-WOD) and the value
reported by the sensors (R-WD) is computed. The average for all these nodes absolute errors
is then found. This gives the mean absolute error of the network. Similarly, the mean absolute
error between the true temperatures (R-WOD) and the drift corrected measurements (DCM-
WD) is calculated at each instant of time and plotted in Figure 6. By comparing Figures 5 and
6 it is evident that applying the drift correction algorithm results in less measurements error
for all of the scenarios. For our evaluation purposes we assumed that the maximum mean
absolute error that can be tolerated in the network is 1
o
C. If the mean absolute error of the
network exceeds that limit, the network is deemed to be useless or has broken down. This
maximum limit is shown by a horizontal threshold line in Figures 5 and 6. The choice of the
threshold is dependent on the error tolerance allowed by the application.
In Figure 5, it is evident that the curves for scenarios 6, 9, 12 and 15 cross the threshold line
after the 6
th
day of the experiment. In contrast, in Figure 6, the curves for scenarios 6 and 9
do not cross the threshold line at all for the whole period of the experiment, while the curves
of scenarios 12 and 15 cross the threshold line on the 8
th
day and the 7
th
day, respectively.
This demonstrates that our algorithm extends the operational life of the network for all of the
scenarios.
In another simulation we repeated the experiment after doubling the sampling rate. This
resulted in 4001 samples for the 10 days experiment. Figure 7 shows the mean absolute error
between the true temperatures (R-WOD) and the drift corrected measurements (DCM-WD)
for the whole network for ve different scenarios. The error is computed in a similar way to
the method used for the Figures 5 and 6. By comparing Figures 5 and 7, it is evident that the
application of the error correction algorithm results in less measurements errors for all of the
scenarios.
Looking at Figures 6 and 7, we can notice that the performance when using 4001 samples is
better for scenario 9 since the absolute error curve does not cross the 1
o
C threshold line as it
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
3 sensors drifting
6 sensors drifting
9 sensors drifting
12 sensors
drifting
15 sensors
drifting
Threshold line
Fig. 5. Mean Absolute Error for the network without correction.
Our algorithm is implemented in MatLab, utilising the SVR toolbox from (Canu et al., 2005)
and the UKF toolbox from (Srkk & Hartikainen, 2007). For comparison purposes, we run
the algorithm on two data sets. One, the data without the introduced drift (WOD), and the
other, the data with drift introduced (WD). Initially, the SVR of each node is trained on the
rst 1000 samples of its neighbours readings.
The UKF parameters , and are set to the default values as explained in Section 5. Through
out our evaluations, we take Q
(1)
i,k
= Q
(2)
i,k
= Q
i,k
. Q
i,k
and R
i,k
are tuned using trial and error
for both cases. The values used in our evaluation are Q
i,k
= 0.001 and R
i,k
= 0.02. If R
i,k
is
set to a high value, the estimated temperature will follow the reading (which may have drift)
whereas if R
i,k
is set to a small value, the estimated temperature will not be able to follow the
real temperature. Thus, it will not totally correct the error. On the other hand, a high value for
Q
i,k
will result in oscillatory estimates and lead to an unstable state. Hence, a trade off has to
be considered in selecting the values for Q
i,k
and R
i,k
to obtain the best results.
We have conducted two simulations using two data sets. One data set has no drifts intro-
duced. We denote this data set by R-WOD, which stands for Readings Without Drift and
represents the sensor measurements that only suffer fromnoise. The other is the same data set
with drifts introduced in several scenarios. We denote the readings of this data set by R-WD,
which stands for Readings With Drift and represents the sensor measurements that suffer
from both drift and noise. The drift scenarios considered in R-WD are as follows: scenario 1
(SCN 1) being one node drifting, scenario 2 (SCN 2) being two nodes drifting and so on until
the last scenario (SCN 16) having all nodes drifting. The resulting corrected measurements
obtained when the algorithm is run on the R-WD data sets are denoted by DCM-WD, which
stands for Drift Corrected Measurement for readings With Drift. Similarly, the corrected
measurements obtained using data set R-WOD are denoted by DCM-WOD, which stands for
Drift Corrected Measurement for readings WithOut Drift.
To evaluate the performance of our algorithm from the networks point of view, we compare
the average absolute error of all the sensors of the network with and without implementing
our drift correction algorithm.
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
Threshold line 6 Sensors
drifting
9 Sensors
drifting
12 Sensors
drifting
15 Sensors
drifting
3 Sensors
drifting
Figure 5 shows the mean absolute error between the true temperatures (R-WOD) and the
values reported by the sensors (R-WD) for the whole network, for ve different scenarios. The
mean absolute error of the network is computed for each scenario as follows: for each node, at
each instant of time, the absolute error between the true temperature (R-WOD) and the value
reported by the sensors (R-WD) is computed. The average for all these nodes absolute errors
is then found. This gives the mean absolute error of the network. Similarly, the mean absolute
error between the true temperatures (R-WOD) and the drift corrected measurements (DCM-
WD) is calculated at each instant of time and plotted in Figure 6. By comparing Figures 5 and
6 it is evident that applying the drift correction algorithm results in less measurements error
for all of the scenarios. For our evaluation purposes we assumed that the maximum mean
absolute error that can be tolerated in the network is 1
o
C. If the mean absolute error of the
network exceeds that limit, the network is deemed to be useless or has broken down. This
maximum limit is shown by a horizontal threshold line in Figures 5 and 6. The choice of the
threshold is dependent on the error tolerance allowed by the application.
In Figure 5, it is evident that the curves for scenarios 6, 9, 12 and 15 cross the threshold line
after the 6
th
day of the experiment. In contrast, in Figure 6, the curves for scenarios 6 and 9
do not cross the threshold line at all for the whole period of the experiment, while the curves
of scenarios 12 and 15 cross the threshold line on the 8
th
day and the 7
th
day, respectively.
This demonstrates that our algorithm extends the operational life of the network for all of the
scenarios.
In another simulation we repeated the experiment after doubling the sampling rate. This
resulted in 4001 samples for the 10 days experiment. Figure 7 shows the mean absolute error
between the true temperatures (R-WOD) and the drift corrected measurements (DCM-WD)
for the whole network for ve different scenarios. The error is computed in a similar way to
the method used for the Figures 5 and 6. By comparing Figures 5 and 7, it is evident that the
application of the error correction algorithm results in less measurements errors for all of the
scenarios.
Looking at Figures 6 and 7, we can notice that the performance when using 4001 samples is
better for scenario 9 since the absolute error curve does not cross the 1
o
C threshold line as it
does in the 2001 samples case. This means that the operational lifetime has been extended
from around 8 days in the case of 2001 samples, to more than 9 days for the case of 4001
samples. Moreover, we notice in Figure 7 that the curves for each scenario are smoother than
the corresponding curves in Figure 6 and that the observed occasional jumps and peaks are
smaller. The jumps in the curves are caused by the fast changes in the readings or the ambient
temperature at some instants of time. An effective way of reducing the size of the jumps is
to increase the sampling rate as we noticed in Figure 7. However, that would be at a cost of
the increased communication overhead due to the increased data transmissions among the
sensors. This means that a trade off between the smoothness of the curves and the commu-
nication overhead has to be made. Another important thing to note in both Figures 6 and 7
is that the mean absolute error of the networks estimated temperatures is proportional to the
number of sensors developing drift.
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
Threshold line
15 Sensors
drifting
12 Sensors
drifting
9 Sensors
drifting
6 Sensors
drifting
3 Sensors
drifting
The choice of R
i,k
and Q
i,k
is crucial. It affects the accuracy of estimating the temperature and
the induced error. In general, we can say that increasing R
i,k
improves the tracking of drift in
the drifting sensors. However, it also increases both the induced error in drift estimation in
the non drifting sensors and the uctuations in the drifting sensors. Since, the error caused
by the fast changes in temperature in the case of 4001 samples is less than that for the case of
2001 samples (as explained previously), R
i,k
is taken to be 0.05 for the case of 4001 and 0.02
for the 2001 samples case. This way, the drift tracking is improved for the 4001 samples case
keeping error levels comparable to the case of 2001 samples. On the other hand, increasing
Q
i,k
increases the uctuations in the estimated drift in both drifting and non-drifting sensors
and causes the response to become less stable. The Q
i,k
used in all our simulations is equal to
0.001.
It is important here to note that for comparison purposes, the data set used in the evalua-
tions of this chapter is the same as the data set used in (Takruri, Rajasegarar, Challa, Leckie
& Palaniswami, 2010) evaluations. Comparing gure 8, which is quoted from (Takruri, Ra-
jasegarar, Challa, Leckie & Palaniswami, 2010), with gure 6, where both of them are for 2001
samples, it can be clearly noticed that SVR-UKF framework presented in (Takruri, Rajasegarar,
Challa, Leckie & Palaniswami, 2010) outperforms the algorithm presented in this chapter in
correcting sensor readings errors as it manages to keep the absolute error for the case of 9 sen-
sors drifting below the threshold line. However, this is at the cost of substantially increased
computational complexity. The state transition function used in this chapter is linear and
the number of sigma points is 5 whereas the state transition function used in the algorithm
in (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010) is the SVR modelled function
which is highly nonliner and the number of sigma points is 19. This results in a computa-
tional complexity of several orders of magnitude.
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
12 sensors
drifting
9 sensors
drifting
threshold line
6 sensors
drifting
3 sensors
drifting
15 sensors
drifting
Quoted from (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010).
The error correction performance for a sensor node in a cluster is dependent on the correlation
of the actual temperature at the sensor under consideration with the actual temperatures at
the neighbours. The SVR at a sensor predicts the actual temperature at the sensor x
i,k
using
previous estimates of the neighbourhood { x
j,k1|k1
}
n
j=1,j=i
. Therefore, low correlation will
lead to a poor prediction, and result in poor estimate of the actual temperature at the sensor
under consideration. In practice, the correlation among the nodes may change depending on
their spatial proximity within the cluster and with the change in the observed phenomenon
along time.
It can be observed in the IBRL sensor deployment that not all the sensors were subject to
the same conditions. This is because of their physical locations. Some of the nodes were
closer to air conditioning. Some were closer to windows and hence were affected by the sun.
Some were closer to the kitchen and thus affected by the heat and humidity coming from
there. Furthermore, the patterns followed by sensor measurements changed seasonally. As
an example, during a week period, the pattern followed in week days was different than that
followed in weekends. That was because the air conditioning was reduced or turned off in
the laboratory on the weekends. This caused the interrelationship among the sensors to vary
not only with their spatial locations, but also with time.
A solution to overcome such a problem is to choose the neighbour sensors of each node so
that they are physically close and subject to similar conditions. An alternative solution is to
does in the 2001 samples case. This means that the operational lifetime has been extended
from around 8 days in the case of 2001 samples, to more than 9 days for the case of 4001
samples. Moreover, we notice in Figure 7 that the curves for each scenario are smoother than
the corresponding curves in Figure 6 and that the observed occasional jumps and peaks are
smaller. The jumps in the curves are caused by the fast changes in the readings or the ambient
temperature at some instants of time. An effective way of reducing the size of the jumps is
to increase the sampling rate as we noticed in Figure 7. However, that would be at a cost of
the increased communication overhead due to the increased data transmissions among the
sensors. This means that a trade off between the smoothness of the curves and the commu-
nication overhead has to be made. Another important thing to note in both Figures 6 and 7
is that the mean absolute error of the networks estimated temperatures is proportional to the
number of sensors developing drift.
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
Threshold line
15 Sensors
drifting
12 Sensors
drifting
9 Sensors
drifting
6 Sensors
drifting
3 Sensors
drifting
The choice of R
i,k
and Q
i,k
is crucial. It affects the accuracy of estimating the temperature and
the induced error. In general, we can say that increasing R
i,k
improves the tracking of drift in
the drifting sensors. However, it also increases both the induced error in drift estimation in
the non drifting sensors and the uctuations in the drifting sensors. Since, the error caused
by the fast changes in temperature in the case of 4001 samples is less than that for the case of
2001 samples (as explained previously), R
i,k
is taken to be 0.05 for the case of 4001 and 0.02
for the 2001 samples case. This way, the drift tracking is improved for the 4001 samples case
keeping error levels comparable to the case of 2001 samples. On the other hand, increasing
Q
i,k
increases the uctuations in the estimated drift in both drifting and non-drifting sensors
and causes the response to become less stable. The Q
i,k
used in all our simulations is equal to
0.001.
It is important here to note that for comparison purposes, the data set used in the evalua-
tions of this chapter is the same as the data set used in (Takruri, Rajasegarar, Challa, Leckie
& Palaniswami, 2010) evaluations. Comparing gure 8, which is quoted from (Takruri, Ra-
jasegarar, Challa, Leckie & Palaniswami, 2010), with gure 6, where both of them are for 2001
samples, it can be clearly noticed that SVR-UKF framework presented in (Takruri, Rajasegarar,
Challa, Leckie & Palaniswami, 2010) outperforms the algorithm presented in this chapter in
correcting sensor readings errors as it manages to keep the absolute error for the case of 9 sen-
sors drifting below the threshold line. However, this is at the cost of substantially increased
computational complexity. The state transition function used in this chapter is linear and
the number of sigma points is 5 whereas the state transition function used in the algorithm
in (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010) is the SVR modelled function
which is highly nonliner and the number of sigma points is 19. This results in a computa-
tional complexity of several orders of magnitude.
4 5 6 7 8 9
0
0.5
1
1.5
2
2.5
3
3.5
Time (Days)
M
e
a
n

A
b
s
o
l
u
t
e

E
r
r
o
r
12 sensors
drifting
9 sensors
drifting
threshold line
6 sensors
drifting
3 sensors
drifting
15 sensors
drifting
Quoted from (Takruri, Rajasegarar, Challa, Leckie & Palaniswami, 2010).
The error correction performance for a sensor node in a cluster is dependent on the correlation
of the actual temperature at the sensor under consideration with the actual temperatures at
the neighbours. The SVR at a sensor predicts the actual temperature at the sensor x
i,k
using
previous estimates of the neighbourhood { x
j,k1|k1
}
n
j=1,j=i
. Therefore, low correlation will
lead to a poor prediction, and result in poor estimate of the actual temperature at the sensor
under consideration. In practice, the correlation among the nodes may change depending on
their spatial proximity within the cluster and with the change in the observed phenomenon
along time.
It can be observed in the IBRL sensor deployment that not all the sensors were subject to
the same conditions. This is because of their physical locations. Some of the nodes were
closer to air conditioning. Some were closer to windows and hence were affected by the sun.
Some were closer to the kitchen and thus affected by the heat and humidity coming from
there. Furthermore, the patterns followed by sensor measurements changed seasonally. As
an example, during a week period, the pattern followed in week days was different than that
followed in weekends. That was because the air conditioning was reduced or turned off in
the laboratory on the weekends. This caused the interrelationship among the sensors to vary
not only with their spatial locations, but also with time.
A solution to overcome such a problem is to choose the neighbour sensors of each node so
that they are physically close and subject to similar conditions. An alternative solution is to
upgrade the model to become incremental with time to account for phenomenal changes. This
can be achieved using incremental learning of the SVR. The learning process can then be per-
formed at each time step (incrementally) or at predened short intervals, depending on how
severe the change is. Incremental SVR learning algorithms (Platt, April 1998; Shilton et al.,
Jan. 2005) can be utilised with the UKF to perform adaptive drift correction in the network.
Devising an adaptive drift correction framework by incorporating incremental learning is a
direction for future work.
7. Conclusion
In this chapter we have proposed a formal statistical algorithm for detecting and correcting
sensor measurement errors in sparsely deployed WSN based on the assumption that sensor
nodes in a neighbourhoodobserving a certain physical phenomenon have correlatedmeasure-
ments and uncorrelated drifts and biases. We have used SVR to model the spatio-temporal
correlations in the neighbouring sensor measurements to obtain predictions of the future sen-
sor measurements. The predicted data have then been used by a UKF to estimate the actual
value of the measured variable at the sensor under consideration. The algorithm runs recur-
sively and is fully decentralised. Extensive evaluations of the presented algorithm on real
data obtained from the IBRL have proved that it is effective in detecting and correcting sensor
errors and extending the effective life of the network.
In future, we plan to upgrade our algorithm to become more adaptive to any phenomenal
changes that may occur in the network deployment area by implementing an incremental
SVR to periodically re-train the SVR. This will also be tested on sensor networks deployed
both in a controlled Lab environment and uncontrolled outdoor environments.
8. References
2006 (Accessed on 07/09/2006). http://db.lcs.mit.edu/labdata/labdata.html, [online].
URL: http://db.lcs.mit.edu/labdata/labdata.html
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y. & Cayirci, E. (2002). Wireless sensor networks:
a survey, Comp. Networks 38: 393422.
Balzano, L. (2007). Addressing fault and calibration in wireless sensor networks, Masters thesis,
University of California, Los Angeles, California.
Balzano, L. & Nowak, R. (2007). Blind calibration of sensor networks, Information Processing in
Sensor Networks .
Balzano, L. & Nowak, R. (2008). Blind calibration of networks of sensors: Theory and algo-
rithms, Networked Sensing Information and Control, Springer US, pp. 937.
Brown, L. (1992). A survey of image registration techniques, ACM Computing Surveys
24(4): 326376.
Bychkovskiy, V. (2003). Distributed in-place calibration in sensor networks, Masters thesis, Uni-
versity of California, Los Angeles, California.
Bychkovskiy, V., Megerian, S., Estrin, D. & Potkonjak, M. (2003). A collaborative approach to
in-place sensor calibration, Int. Workshop on Information Processing in Sensor Networks
pp. 301316.
Canu, S., Grandvalet, Y., Guigue, V. & Rakotomamonjy, A. (2005). Svm and kernel meth-
ods matlab toolbox, Perception Systmes et Information, INSA de Rouen, Rouen,
France.
URL: http://asi.insa-rouen.fr/enseignants/ arakotom/toolbox/index.html
Challa, S., Evans, R., Morelande, M. & Musicki, D. (2008). Fundamentals of Object Tracking,
Cambridge University Press.
Elnahrawy, E. & Nath, B. (2003). Cleaning and querying noisy sensors, in Proceedings of ACM
WSNA03.
Estrin, D., Girod, L., Pottie, G. & Srivastava, M. (2001). Instrumenting the world with wireless
sensor networks, Int. Conference on Acoustics, Speech, and Signal Processing) .
Feng, J., Megerian, S. & Potkonjak, M. (2003). Model-based calibration for sensor networks,
Sensors pp. 737 742.
Gill, M. K., Asefa, T., Kemblowski, M. W. & McKee, M. (2006). Soil moisture prediction using
support vector machines1, J. of the American Water Resources Association 42(4): 1033
1046.
URL: http://www.blackwell-synergy.com/doi/abs/10.1111/j.1752-1688.2006.tb04512.x
Gill, M. K., Kemblowski, M. W. & McKee, M. (2007). Soil moisture data assimilation using
support vector machines and ensemble kalman lter, J. of the American Water Resources
Association 43(4): 10041015.
Guestrin, C., Bodik, P., Thibaux, R., Paskin, M. & Madden, S. (2004). Distributed regression:
an efcient framework for modeling sensor network data, In IPSN04, pp. 110.
Hoadley, B. (1970). A baysian look at inverse linear regression, J. of the American Stats. Associ-
ation 65(329): 356369.
Ihler, A., Fisher, J., Moses, R. & Willsky, A. (2004). Nonparametric belief propagation for self-
calibration in sensor networks, In Proceedings of the Third international Symposium on
Information Processing in Sensor Networks.
Julier, S. (2002). The scaled unscented transformation, American Control Conference 6: 4555
4559.
Julier, S. J., Uhlmann, J. K. & Durrant-Whyte, H. F. (1995). A new approach for ltering non-
linear systems, American control Conference pp. 1628 1632.
Julier, S. & Uhlmann, J. (1997). A new extension of the Kalman lter to nonlinear systems, Int.
Symp. Aerospace/Defense Sensing, Simul. and Controls.
Krishnamachari, B. & Iyengar, S. (2004). Distributed bayesian algorithms for fault-tolerant
event region detection in wireless sensor networks, IEEE Tran. Computers 53(3): 241
250.
Lu, S., Cai, L., Lu, D. & Chen, J. (2007). Two efcient implementation forms of unscented
kalman lter, IEEE Int. Conference on Control and Automation pp. 761 764.
Nowak, R. & Mitra, U. (2003). Boundary estimation in sensor networks: Theory and methods,
In IPSN, pp. 8095.
Okello, N. & Challa, S. (2003). Simultaneous registration and track fusion for networked
trackers, conference on information fusion, Montral, Canada.
Okello, N. & Pulford, G. (1996). Simultaneous registration and tracking for multiple radars
with cluttered measurements, IEEE Signal Processing Workshop on Statistical Signal and
Array Processing pp. 6063.
Platt, J. (April 1998). Sequential minimal optimization: A fast algorithm for training support
vector machines, Technical Report 98-14, Microsoft Research, Redmond, Washington. .
URL: citeseer.ist.psu.edu/platt98sequential.html
Rajasegarar, S., Leckie, C., Palaniswami, M. & Bezdek, J. (2007). Quarter sphere based dis-
tributed anomaly detection in wireless sensor networks, Proceedings of the IEEE Inter-
national Conference Communications (IEEE ICC 07), UK.
upgrade the model to become incremental with time to account for phenomenal changes. This
can be achieved using incremental learning of the SVR. The learning process can then be per-
formed at each time step (incrementally) or at predened short intervals, depending on how
severe the change is. Incremental SVR learning algorithms (Platt, April 1998; Shilton et al.,
Jan. 2005) can be utilised with the UKF to perform adaptive drift correction in the network.
Devising an adaptive drift correction framework by incorporating incremental learning is a
direction for future work.
7. Conclusion
In this chapter we have proposed a formal statistical algorithm for detecting and correcting
sensor measurement errors in sparsely deployed WSN based on the assumption that sensor
nodes in a neighbourhoodobserving a certain physical phenomenon have correlatedmeasure-
ments and uncorrelated drifts and biases. We have used SVR to model the spatio-temporal
correlations in the neighbouring sensor measurements to obtain predictions of the future sen-
sor measurements. The predicted data have then been used by a UKF to estimate the actual
value of the measured variable at the sensor under consideration. The algorithm runs recur-
sively and is fully decentralised. Extensive evaluations of the presented algorithm on real
data obtained from the IBRL have proved that it is effective in detecting and correcting sensor
errors and extending the effective life of the network.
In future, we plan to upgrade our algorithm to become more adaptive to any phenomenal
changes that may occur in the network deployment area by implementing an incremental
SVR to periodically re-train the SVR. This will also be tested on sensor networks deployed
both in a controlled Lab environment and uncontrolled outdoor environments.
8. References
2006 (Accessed on 07/09/2006). http://db.lcs.mit.edu/labdata/labdata.html, [online].
URL: http://db.lcs.mit.edu/labdata/labdata.html
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y. & Cayirci, E. (2002). Wireless sensor networks:
a survey, Comp. Networks 38: 393422.
Balzano, L. (2007). Addressing fault and calibration in wireless sensor networks, Masters thesis,
University of California, Los Angeles, California.
Balzano, L. & Nowak, R. (2007). Blind calibration of sensor networks, Information Processing in
Sensor Networks .
Balzano, L. & Nowak, R. (2008). Blind calibration of networks of sensors: Theory and algo-
rithms, Networked Sensing Information and Control, Springer US, pp. 937.
Brown, L. (1992). A survey of image registration techniques, ACM Computing Surveys
24(4): 326376.
Bychkovskiy, V. (2003). Distributed in-place calibration in sensor networks, Masters thesis, Uni-
versity of California, Los Angeles, California.
Bychkovskiy, V., Megerian, S., Estrin, D. & Potkonjak, M. (2003). A collaborative approach to
in-place sensor calibration, Int. Workshop on Information Processing in Sensor Networks
pp. 301316.
Canu, S., Grandvalet, Y., Guigue, V. & Rakotomamonjy, A. (2005). Svm and kernel meth-
ods matlab toolbox, Perception Systmes et Information, INSA de Rouen, Rouen,
France.
URL: http://asi.insa-rouen.fr/enseignants/ arakotom/toolbox/index.html
Challa, S., Evans, R., Morelande, M. & Musicki, D. (2008). Fundamentals of Object Tracking,
Cambridge University Press.
Elnahrawy, E. & Nath, B. (2003). Cleaning and querying noisy sensors, in Proceedings of ACM
WSNA03.
Estrin, D., Girod, L., Pottie, G. & Srivastava, M. (2001). Instrumenting the world with wireless
sensor networks, Int. Conference on Acoustics, Speech, and Signal Processing) .
Sensors pp. 737 742.
Gill, M. K., Asefa, T., Kemblowski, M. W. & McKee, M. (2006). Soil moisture prediction using
support vector machines1, J. of the American Water Resources Association 42(4): 1033
1046.
URL: http://www.blackwell-synergy.com/doi/abs/10.1111/j.1752-1688.2006.tb04512.x
Gill, M. K., Kemblowski, M. W. & McKee, M. (2007). Soil moisture data assimilation using
support vector machines and ensemble kalman lter, J. of the American Water Resources
Association 43(4): 10041015.
Guestrin, C., Bodik, P., Thibaux, R., Paskin, M. & Madden, S. (2004). Distributed regression:
an efcient framework for modeling sensor network data, In IPSN04, pp. 110.
Hoadley, B. (1970). A baysian look at inverse linear regression, J. of the American Stats. Associ-
ation 65(329): 356369.
Ihler, A., Fisher, J., Moses, R. & Willsky, A. (2004). Nonparametric belief propagation for self-
calibration in sensor networks, In Proceedings of the Third international Symposium on
Information Processing in Sensor Networks.
Julier, S. (2002). The scaled unscented transformation, American Control Conference 6: 4555
4559.
Julier, S. J., Uhlmann, J. K. & Durrant-Whyte, H. F. (1995). A new approach for ltering non-
linear systems, American control Conference pp. 1628 1632.
Julier, S. & Uhlmann, J. (1997). A new extension of the Kalman lter to nonlinear systems, Int.
Symp. Aerospace/Defense Sensing, Simul. and Controls.
Krishnamachari, B. & Iyengar, S. (2004). Distributed bayesian algorithms for fault-tolerant
event region detection in wireless sensor networks, IEEE Tran. Computers 53(3): 241
250.
Lu, S., Cai, L., Lu, D. & Chen, J. (2007). Two efcient implementation forms of unscented
kalman lter, IEEE Int. Conference on Control and Automation pp. 761 764.
Nowak, R. & Mitra, U. (2003). Boundary estimation in sensor networks: Theory and methods,
In IPSN, pp. 8095.
Okello, N. & Challa, S. (2003). Simultaneous registration and track fusion for networked
trackers, conference on information fusion, Montral, Canada.
Okello, N. & Pulford, G. (1996). Simultaneous registration and tracking for multiple radars
with cluttered measurements, IEEE Signal Processing Workshop on Statistical Signal and
Array Processing pp. 6063.
Platt, J. (April 1998). Sequential minimal optimization: A fast algorithm for training support
vector machines, Technical Report 98-14, Microsoft Research, Redmond, Washington. .
URL: citeseer.ist.psu.edu/platt98sequential.html
Rajasegarar, S., Leckie, C., Palaniswami, M. & Bezdek, J. (2007). Quarter sphere based dis-
tributed anomaly detection in wireless sensor networks, Proceedings of the IEEE Inter-
national Conference Communications (IEEE ICC 07), UK.
Sallans, B., Bruckner, D. & Russ, G. (2005). Statistical model-based sensor diagnostic for au-
tomation systems, in M. L. Chavez (ed.), Fieldbus systems and their applications, Else-
vier, pp. 239246.
Scholkopf, B. & Smola, A. (2002). Learning with Kernels, MIT Press.
Shilton, A., Palaniswami, M., Ralph, D. & Tsoi, A. C. (Jan. 2005). Incremental training of
support vector machines, IEEE Tran. on Neural Networks 16(1): 114131.
Srkk, S. & Hartikainen, J. (2007). Ekf/ukf toolbox for matlab v1.2, Centre of Excellence
in Computational Complex Systems Research, Helsinki University of Technology
(HUT), Finland. http://www.lce.hut./research/mm/ekfukf/.
Takruri, M., Aboura, K. & Challa, S. (2008). Distributed recursive algorithm for auto cali-
bration in drift aware wireless sensor networks, in K. Elleithy (ed.), Innovations and
Advanced Techniques in Systems, Computing Sciences and Software Engineering, Springer,
pp. 2125.
Takruri, M. & Challa, S. (2007). Drift aware wireless sensor networks, Proceedings of the 10th
international conference on information fusion, Quebec City, Canada.
Takruri, M., Challa, S. & Chacravorty, R. (2010). Recursive bayesian approaches for auto cali-
bration in drift aware wireless sensor networks, Journal of Networks 5(7): 823832.
Takruri, M., Challa, S. & Chakravorty, R. (2008). Auto calibration in drift aware wireless sen-
sor networks using the interacting multiple model algorithm, Mosharaka International
Conference on Communications, Computers and Applications (MIC-CCA 2008), Amman,
Jordan.
Takruri, M., Challa, S. & Yunis, R. (2009). Data fusion techniques for auto calibration in wire-
less sensor networks, Proceedings of the 12th International conference on information fu-
sion, Seattle, USA.
Takruri, M., Rajasegarar, S., Challa, S., Leckie, C. & Palaniswami, M. (2008). Online drift
correction in wireless sensor networks using spatio-temporal modeling, Proceedings
of the 11th international conference on information fusion, Cologne, Germany.
Takruri, M., Rajasegarar, S., Challa, S., Leckie, C. & Palaniswami, M. (2010). Spatio-temporal
modelling based drift aware wireless sensor networks, IET Wirel. Sens. Syst. (under
review) .
Taylor, C., Rahimi, A., Bachrach, J., Shrobe, H. & Grue, A. (2006). Simultaneous localization,
calibration, and tracking in an ad hoc sensor network, In Proceedings of 5th Interna-
tional Conference on Information Processing in Sensor Networks (IPSN06), pp. 2733.
Wan, E. & van der Merwe, R. (2000). The unscented kalman lter for nonlinear estimation,
IEEE Symposium 2000 (AS-SPCC) .
Wang, Y. M., Schultz, R. T., Constable, R. T. & Staib1, L. H. (2003). Nonlinear estimation and
modeling of fmri data using spatio-temporal support vector regression, Information
Processing in Medical Imaging 2732: 647659.
Whitehouse, K. & Culler, D. (2002). Calibration as parameter estimation in sensor networks,
ACM International Workshop on Wireless Sensor Networks and Applications (WSNA02).
Whitehouse, K. & Culler, D. (2003). Macro-calibration in sensor/actuator networks, Mobile
Networks and Applications Journal (MONET), Special Issue on Wireless Sensor Networks .
Xu, X., Hines, J. W. & Uhrig, R. E. (1998). On-line sensor calibration monitoring and fault
detection for chemical processes, Maintenance and Reliability Conference (MARCON
98), pp. 1214.
Zitova, B. & Flusser, J. (2003). Image registration methods: a survey, Image and Vision Comput-
ing 21(1): 9771000.
Target Tracking in Wireless Sensor Networks
Jianxun LI and Yan ZHOU
X

Target Tracking in Wireless Sensor Networks

Jianxun LI* and Yan ZHOU**,*
* Department of Automation, Shanghai Jiao Tong University, Shanghai 200240, China
** College of Information Engineering, Xiangtan University, Xiangtan 411105, China

1. Introduction
Wireless sensor networks (WSNs) have gained worldwide attention in recent years,
particularly with the proliferation in Micro-Electro-Mechanical Systems (MEMS) technology
which has facilitated the development of smart sensors (Akyildiz et al., 2006; Akyildiz et al.,
2007; Yick et al., 2008). Target tracking in WSNs is an important problem with a large
spectrum of applications (Akyildiz et al., 2006; Zhao et al., 2002), such as surveillance
(Valera & Velastin, 2005), natural disaster relief (Wang et al., 2003), traffic monitoring (Li et
al., 2009), pursuit evasion games, etc.

1.1 Opportunities and challenges
A target tracking system through WSNs can have several advantages (Veeravalli &
Chamberland, 2007): (i) qualitative and fidelity observations; (ii) signal processing
accurately and timely; and (iii) increased system robustness and tracking accuracy.
However, the use of sensor networks for target tracking presents a number of new
challenges. These challenges include limited energy supply and communication bandwidth,
distributed algorithms and control, and handling the fundamental performance limits of
sensor nodes, especially as the size of the network becomes large. Unlike traditional
networks, a WSN has its own design and resource constraints. Resource constraints include
a limited amount of energy, short communication range, low bandwidth, and limited
processing and storage in each node. Design constraints are application dependent and are
based on the monitored environment. The environment plays a key role in determining the
size of the network, the deployment scheme, and the network topology.
Power consumption is the most important design factor for WSNs (Shorey et al., 2006).
Commonly, saving power during the operation of the electronic device could be achieved
on more than one protocol level. Plenty of research work is dedicated to the design of power
efficient schemes for target tracking which try to explore good trade-off between power
consumption and tracking accuracy (see e.g. Lee et al. 2007; Xu & Lee, 2003; Walchli et al.,
2007; Tsai et al., 2007, and the references therein).
Besides, the traditional target tracking methodologies make use of a centralized approach.
As the number of sensors rise in the network, more messages are passed on towards the sink
and will consume additional bandwidth. Thus traditional approaches are not fault tolerant
as there is single point of failure and does not scale well. However, in sensor networks,
19

hundreds, and in the extreme, hundreds of thousands of sensors are deployed in a large
geographical area. In some cases dropped from airplanes, or deployed using artillery shells.
Requiring that every node must work in order for the network to operate is difficult to
achieve. The network must have a high level of fault tolerance in order to be of any practical
value (Hoblos et al., 2000).
In a word, target tracking algorithm considering the tradeoff between the tracking accuracy
and network resources such as energy, bandwidth, and communication/computation
burden is challenging.

1.2 Contributions and chapter organization
In this chapter, a self-contained overview of tracking approaches through a WSN is given
according the architecture of the networks. As can be seen soon, WSNs are typically
classified into two main categories (Sohraby et al., 2006): hierarchical network and peer-to-
peer network. For the former, nave, tree, cluster, and hybrid network based methods are
reviewed in detail. While for the latter, average consensus is usually adopted to achieve
network-wide agreement on target estimate and two approaches commonly used including
the dynamic consensus filter and alternating direction method. Then advantage and
limitations of these approaches are compared.
Considering the stringent energy and bandwidth limitation, quantized messages based
tracking method is discussed separately. Local data quantization/compression is usually
adopted so as to reduce the required expenditure of resources. Then the signal processing
unit, i.e. the fusion center (FC), or cluster head (CH), or any node in the network, combines
the quantized messages from local sensors to produce a final estimation of the target state.
In the quantized scenario, Shannons rate-distortion bound that notionally is a curve of
possible and impossible points on distortion (e.g. MSE versus bit-rate axes) is important and
practical. In the literature, target tracking algorithms using quantized information can be
categorized to mainly quantized measurement based and quantized innovation based
tracking. Both categories will be overview in tail with the characteristics discussed.
The remainder of this chapter is organized as follows. Section 2 gives a taxonomy on target
tracking approaches through a WSN. In Section 3, details of target tracking approaches in
hierarchical network are highlighted focusing on: nave activation based tracking, tree-based
tracking methods, cluster-based tracking approaches, and hybrid tracking methods. In
Section 4, target tracking in peer-to-peer sensor networks are reviewed with both embedded
filter based consensus and alternating direction based consensus methods discussed.
Comparison criterions are given in Section 5 and the methods aforementioned are
compared. Section 6 discusses the methods based on quantized information in detail, which
followed by Section 7 that gives the future research directions and concluding remarks.

2. A taxonomy on tracking methods
In the literature, various algorithms and approaches for target tracking are presented but
there is a deficiency of well defined classification of solutions of this well known application
of WSN. This chapter presents a taxonomy of target tracking approaches as well as
discussing each method under the appropriate category.
The taxonomy on the tracking techniques presented is based on the network architecture. In
this chapter, we classify the WSNs into two categories (Veeravalli & Chamberland, 2007):

hierarchical (category 1 WSNs) and peer-to-peer (category 2 WSNs). For the category 1
WSNs (see Fig. 1 (a)), almost invariably mesh-based systems with multihop radio
connectivity among or between wireless nodes are employed. The sensors in the vicinity of
an event must be able to monitor the event of interest and report back to the sink. A sink
sensor node has capability to communicate with outside world such as laptop, base station.
The important characterizations of the category 1 WSNs are that (i) sensor nods can support
communications on behalf of other sensor nodes by acting as repeaters; (ii) the forwarding
node can support data processing or information fusion on behalf of the senor nodes. The
category 2 WSNs (see Fig. 1 (b)), which are also referred as flat, or point-to-point network,
generally are with single-hop radio connectivity to wireless nodes utilizing static routing
over the wireless network. The main features of the category 2 WSNs are that (i) the
forwarding node only supports static routing; (ii) each node only communicates with its
neighbouring node(s) and network-wide consensus can be achieved through information
exchange between neighbours.
As illustrated in Fig. 2, we taxonomize the tracking algorithms into two aspects according to
the aforementioned two categories of network architecture. One is hierarchical network
based tracking, the other is peer-to-peer network based tracking. The former can be further
classified into four schemes, which are: Nave activation based tracking, tree-based tracking,
cluster-based tracking, and hybrid methods. In tree-based target tracking, nodes in a
network may be organized in a hierarchical tree or represented as a graph in which vertices
represent sensor nodes and edges are links between nodes that can directly communicate
with each other.
The cluster-based methods provide scalability and better usage of bandwidth than other
types of methods. If CH is formed via local network processing, extra messages are reduced
and fewer messages are transmitted towards base station thus providing security as well as
less usage of bandwidth (Rapaka & Madria, 2007). In the conventional cluster architecture,
clusters are formed statistically at the time of network deployment and the properties of
each cluster are fixed such as number of members, area covered, etc. Static clustering has
several drawbacks regardless of its simplicity, for example, static membership is not robust

Fig. 1. Architecture of wireless sensor networks
Higher
Lever
Lower
Lever
Backbone Sensor node
Sensor node
(a) Hierarchical network
(b) Peer-to-peer network

hundreds, and in the extreme, hundreds of thousands of sensors are deployed in a large
geographical area. In some cases dropped from airplanes, or deployed using artillery shells.
Requiring that every node must work in order for the network to operate is difficult to
achieve. The network must have a high level of fault tolerance in order to be of any practical
value (Hoblos et al., 2000).
In a word, target tracking algorithm considering the tradeoff between the tracking accuracy
and network resources such as energy, bandwidth, and communication/computation
burden is challenging.

1.2 Contributions and chapter organization
In this chapter, a self-contained overview of tracking approaches through a WSN is given
according the architecture of the networks. As can be seen soon, WSNs are typically
classified into two main categories (Sohraby et al., 2006): hierarchical network and peer-to-
peer network. For the former, nave, tree, cluster, and hybrid network based methods are
reviewed in detail. While for the latter, average consensus is usually adopted to achieve
network-wide agreement on target estimate and two approaches commonly used including
the dynamic consensus filter and alternating direction method. Then advantage and
limitations of these approaches are compared.
Considering the stringent energy and bandwidth limitation, quantized messages based
tracking method is discussed separately. Local data quantization/compression is usually
adopted so as to reduce the required expenditure of resources. Then the signal processing
unit, i.e. the fusion center (FC), or cluster head (CH), or any node in the network, combines
the quantized messages from local sensors to produce a final estimation of the target state.
In the quantized scenario, Shannons rate-distortion bound that notionally is a curve of
possible and impossible points on distortion (e.g. MSE versus bit-rate axes) is important and
practical. In the literature, target tracking algorithms using quantized information can be
categorized to mainly quantized measurement based and quantized innovation based
tracking. Both categories will be overview in tail with the characteristics discussed.
The remainder of this chapter is organized as follows. Section 2 gives a taxonomy on target
tracking approaches through a WSN. In Section 3, details of target tracking approaches in
hierarchical network are highlighted focusing on: nave activation based tracking, tree-based
tracking methods, cluster-based tracking approaches, and hybrid tracking methods. In
Section 4, target tracking in peer-to-peer sensor networks are reviewed with both embedded
filter based consensus and alternating direction based consensus methods discussed.
Comparison criterions are given in Section 5 and the methods aforementioned are
compared. Section 6 discusses the methods based on quantized information in detail, which
followed by Section 7 that gives the future research directions and concluding remarks.

2. A taxonomy on tracking methods
In the literature, various algorithms and approaches for target tracking are presented but
there is a deficiency of well defined classification of solutions of this well known application
of WSN. This chapter presents a taxonomy of target tracking approaches as well as
discussing each method under the appropriate category.
The taxonomy on the tracking techniques presented is based on the network architecture. In
this chapter, we classify the WSNs into two categories (Veeravalli & Chamberland, 2007):

hierarchical (category 1 WSNs) and peer-to-peer (category 2 WSNs). For the category 1
WSNs (see Fig. 1 (a)), almost invariably mesh-based systems with multihop radio
connectivity among or between wireless nodes are employed. The sensors in the vicinity of
an event must be able to monitor the event of interest and report back to the sink. A sink
sensor node has capability to communicate with outside world such as laptop, base station.
The important characterizations of the category 1 WSNs are that (i) sensor nods can support
communications on behalf of other sensor nodes by acting as repeaters; (ii) the forwarding
node can support data processing or information fusion on behalf of the senor nodes. The
category 2 WSNs (see Fig. 1 (b)), which are also referred as flat, or point-to-point network,
generally are with single-hop radio connectivity to wireless nodes utilizing static routing
over the wireless network. The main features of the category 2 WSNs are that (i) the
forwarding node only supports static routing; (ii) each node only communicates with its
neighbouring node(s) and network-wide consensus can be achieved through information
exchange between neighbours.
As illustrated in Fig. 2, we taxonomize the tracking algorithms into two aspects according to
the aforementioned two categories of network architecture. One is hierarchical network
based tracking, the other is peer-to-peer network based tracking. The former can be further
classified into four schemes, which are: Nave activation based tracking, tree-based tracking,
cluster-based tracking, and hybrid methods. In tree-based target tracking, nodes in a
network may be organized in a hierarchical tree or represented as a graph in which vertices
represent sensor nodes and edges are links between nodes that can directly communicate
with each other.
The cluster-based methods provide scalability and better usage of bandwidth than other
types of methods. If CH is formed via local network processing, extra messages are reduced
and fewer messages are transmitted towards base station thus providing security as well as
less usage of bandwidth (Rapaka & Madria, 2007). In the conventional cluster architecture,
clusters are formed statistically at the time of network deployment and the properties of
each cluster are fixed such as number of members, area covered, etc. Static clustering has
several drawbacks regardless of its simplicity, for example, static membership is not robust

Fig. 1. Architecture of wireless sensor networks
Higher
Lever
Lower
Lever
Backbone Sensor node
Sensor node
(a) Hierarchical network
(b) Peer-to-peer network

from fault-tolerance point of view and it prevents sensor in different clusters from sharing
information. In contrast, dynamic clustering offers several advantages where clusters are
formed dynamically depending on occurrence of certain events, for instance, when a node
with sufficient battery and computational power detects an event, it comes forward to act as
a CH. To make sure only one CH remains active for target tracking, some decentralized
mechanism is adapted. The CH invites nearby sensor nodes and makes them members of
that cluster. Since sensors dont statistically form a cluster, they may belong to altered
cluster at different timings. As only one cluster is active at a time, redundant data and
interference is reduced.
According to the methods approaching consensus, the peer-to-peer networks based target
tracking systems can be further classified into embedded filter based tracking and
alternating-direction based tracking. Since network-wide consensus can be achieved
through only information exchange between neighbours, considerable communication
energy is reduced and this makes the peer-to-peer networks based tracking scalable for
large-scale sensor network.

Fig. 2. Taxonomy of target tracking algorithms in WSNs.

3. Tracking methods for hierarchical networks
3.1 Nave activation based tracking
Nave activation (or direct communication) based tracking scheme (Guo et al., 2003) is the
simplest approach, for which all nodes are in tracking mode all the time. Each node sends
the local measurement to the sink node or base station. Then the base station estimates and
predicts the target state according to the received local measurements. Since it offers the best
tracking results, it is a useful baseline for comparison. However, this strategy offers the
worst energy efficiency and it inflects heavy communication and computation burden on
the base station of sink node. This makes the nave approach not robust against base station
failure especially for the case of link failure and channel congestion.

Peer-to-peer
Network
Target
Tracking
In WSNs
Static clustering
Dynamic clustering

Space-time clustering
Hierarchical
Network
Nave activation based
Tree-based tracking
Hybrid method
Cluster-based tracking
Alternating-direction
based consensus
Embedded filter
based consensus

3.2 Tree-based tracking
Centralized target tracking approaches are both time and energy consuming, to avoid this
limitation tree-based tracking methods are proposed, for example, Scalable Tracking Using
Networked sensors (STUN) (Kung & Vlah, 2003), Dynamic Convoy Tree-Based
Collaboration (DCTC) (Zhang & Cao, 2004a; Zhang & Cao, 2004b), Deviation Avoidance
Tree (DAT) (Lin et al., 2004), and Dynamic Object Tracking (DOT) approach (Tsai et al.,
2007).
Specifically, STUN (Kung & Vlah, 2003) is a tree-based approach in which a cost is assigned
to each link calculated by Euclidean distance between the two nodes. The leaf nodes are
used for tracking the moving object and then sending collected data to the sink through
intermediate nodes. The intermediate nodes keep a record of detected object and whenever
there is a change in that record, they send updated information to the sink. However, STUN
has two drawbacks. First, drain and balance tree does not replicate physical sensor network
as it is a logical tree, hence an edge may consist of multiple communication hops and may
raise communication cost. Second, the construction of DAB tree does not consider query
cost.
In (Zhang & Cao, 2004a; Zhang & Cao, 2004b), a spanning tree rooted at the sensor node
close to a target is used for target tracking, with the target position estimated by the location
of the root sensor. More statistically-oriented algorithms for mobile target identification and
localization are proposed therein, which allows the designer to directly model the
distributional properties of sensor signals.
A network aggregation model by organizing sensor nodes in logical tree is intended in (Lin
et al., 2004). As physical topology of the network is considered, thus reducing the total
communication cost. The object tracking involves two steps: update and query. In first step;
location update cost is reduced by Deviation Avoidance Tree (DAT) algorithm and in
second step; query cost is reduced by query cost reduction algorithm.
DOT (Tsai et al., 2007), a unique protocol reports the tracking information of moving object
to moving source. First of all, the face neighbours are identified by Gabriel graph. In target
discovery step, source sends request to sensor nodes and the node close to the target replies
back. To detect moving target continuously, the spatial neighbours of near sensor node are
waken up. In target tracking step, source send query to beacon node (node keeping track
information), which reply back targets next location and the source moves towards next
beacon node. The process is repeated until the source catches the target.
Lin et al. propose a CLOUD framework to track the region-based event (Lin et al., 2005). The
basic idea is to dynamically form a tree-based collective structure for each event region in
each time slot. However, both of their approaches in (Lin et al., 2005; Jiang et al., 2004) are
limited by the dependence on the tree structure for the network topology.
In (Jin & Nittel, 2006), an R-tree sensor network topology is adopted for the detection and
tracking of region-based targets. Two approaches: forward-all and forward-description
methods are proposed for the detection of event regions. Furthermore, the authors describe
three detailed algorithms: boundary detection algorithm, merging algorithm and
description improvement algorithm to deal with the problems of how to detect an event
boundary, how to merge the event region obtained from the child nodes in the R-tree, and
how to simplify and smooth the event boundary.


from fault-tolerance point of view and it prevents sensor in different clusters from sharing
information. In contrast, dynamic clustering offers several advantages where clusters are
formed dynamically depending on occurrence of certain events, for instance, when a node
with sufficient battery and computational power detects an event, it comes forward to act as
a CH. To make sure only one CH remains active for target tracking, some decentralized
mechanism is adapted. The CH invites nearby sensor nodes and makes them members of
that cluster. Since sensors dont statistically form a cluster, they may belong to altered
cluster at different timings. As only one cluster is active at a time, redundant data and
interference is reduced.
According to the methods approaching consensus, the peer-to-peer networks based target
tracking systems can be further classified into embedded filter based tracking and
alternating-direction based tracking. Since network-wide consensus can be achieved
through only information exchange between neighbours, considerable communication
energy is reduced and this makes the peer-to-peer networks based tracking scalable for
large-scale sensor network.

Fig. 2. Taxonomy of target tracking algorithms in WSNs.

3. Tracking methods for hierarchical networks
3.1 Nave activation based tracking
Nave activation (or direct communication) based tracking scheme (Guo et al., 2003) is the
simplest approach, for which all nodes are in tracking mode all the time. Each node sends
the local measurement to the sink node or base station. Then the base station estimates and
predicts the target state according to the received local measurements. Since it offers the best
tracking results, it is a useful baseline for comparison. However, this strategy offers the
worst energy efficiency and it inflects heavy communication and computation burden on
the base station of sink node. This makes the nave approach not robust against base station
failure especially for the case of link failure and channel congestion.

Peer-to-peer
Network
Target
Tracking
In WSNs
Static clustering
Dynamic clustering

Space-time clustering
Hierarchical
Network
Nave activation based
Tree-based tracking
Hybrid method
Cluster-based tracking
Alternating-direction
based consensus
Embedded filter
based consensus

3.2 Tree-based tracking
Centralized target tracking approaches are both time and energy consuming, to avoid this
limitation tree-based tracking methods are proposed, for example, Scalable Tracking Using
Networked sensors (STUN) (Kung & Vlah, 2003), Dynamic Convoy Tree-Based
Collaboration (DCTC) (Zhang & Cao, 2004a; Zhang & Cao, 2004b), Deviation Avoidance
Tree (DAT) (Lin et al., 2004), and Dynamic Object Tracking (DOT) approach (Tsai et al.,
2007).
Specifically, STUN (Kung & Vlah, 2003) is a tree-based approach in which a cost is assigned
to each link calculated by Euclidean distance between the two nodes. The leaf nodes are
used for tracking the moving object and then sending collected data to the sink through
intermediate nodes. The intermediate nodes keep a record of detected object and whenever
there is a change in that record, they send updated information to the sink. However, STUN
has two drawbacks. First, drain and balance tree does not replicate physical sensor network
as it is a logical tree, hence an edge may consist of multiple communication hops and may
raise communication cost. Second, the construction of DAB tree does not consider query
cost.
In (Zhang & Cao, 2004a; Zhang & Cao, 2004b), a spanning tree rooted at the sensor node
close to a target is used for target tracking, with the target position estimated by the location
of the root sensor. More statistically-oriented algorithms for mobile target identification and
localization are proposed therein, which allows the designer to directly model the
distributional properties of sensor signals.
A network aggregation model by organizing sensor nodes in logical tree is intended in (Lin
et al., 2004). As physical topology of the network is considered, thus reducing the total
communication cost. The object tracking involves two steps: update and query. In first step;
location update cost is reduced by Deviation Avoidance Tree (DAT) algorithm and in
second step; query cost is reduced by query cost reduction algorithm.
DOT (Tsai et al., 2007), a unique protocol reports the tracking information of moving object
to moving source. First of all, the face neighbours are identified by Gabriel graph. In target
discovery step, source sends request to sensor nodes and the node close to the target replies
back. To detect moving target continuously, the spatial neighbours of near sensor node are
waken up. In target tracking step, source send query to beacon node (node keeping track
information), which reply back targets next location and the source moves towards next
beacon node. The process is repeated until the source catches the target.
Lin et al. propose a CLOUD framework to track the region-based event (Lin et al., 2005). The
basic idea is to dynamically form a tree-based collective structure for each event region in
each time slot. However, both of their approaches in (Lin et al., 2005; Jiang et al., 2004) are
limited by the dependence on the tree structure for the network topology.
In (Jin & Nittel, 2006), an R-tree sensor network topology is adopted for the detection and
tracking of region-based targets. Two approaches: forward-all and forward-description
methods are proposed for the detection of event regions. Furthermore, the authors describe
three detailed algorithms: boundary detection algorithm, merging algorithm and
description improvement algorithm to deal with the problems of how to detect an event
boundary, how to merge the event region obtained from the child nodes in the R-tree, and
how to simplify and smooth the event boundary.


3.3 Cluster-based tracking
To facilitate collaborative data processing in target tracking-centric sensor networks, the
cluster architecture is usually used in which sensors are organized into clusters, with each
cluster consisting of a CH and several slave nodes (members). Hierarchical (clustering)
techniques can aid in reducing useful energy consumption (Heinzelman et al., 2002).
Clustering is particularly useful for applications that require scalability to hundreds or
thousands of nodes. Scalability in this context implies the need for load balancing and
efficient resource utilization. Clustering can be extremely effective in one-to-many, many-to-
one, one-to-any, or one-to-all (broadcast) communication. For example, in many-to-one
communication, clustering can support data fusion and reduce communication interference
(Younis & Fahmy, 2004).

3.3.1 Static clustering
Conventionally, clusters are formed statically at the time of network deployment. The
attributes of each cluster, such as the size of a cluster, the area it covers, and the members it
possesses, are static. In spite of its simplicity, the static cluster architecture suffers from
several drawbacks. First, fixed membership is not robust from the perspective of fault
tolerance. If a CH dies of power depletion, all the sensors in the cluster render useless.
Second, fixed membership prevents sensor nodes in different clusters from sharing
information and collaborating on data processing. Finally, fixed membership cannot adapt
to highly dynamic scenarios in which sensors in the region of high (low) event concentration
may be instrumented to stay awake (go to sleep).

3.3.2 Dynamic clustering
Dynamic cluster architectures, on the other hand, offer several desirable features (Chen et
al., 2003). Formation of a cluster is triggered by certain events of interest (e.g., detection of
an approaching target with acoustic sounds). When a sensor with sufficient battery and
computational power detects (with a high signal-to-noise ratio, SNR) signals of interest, it
volunteers to act as a CH. No explicit leader (CH) election is required and, hence, no
excessive message exchanges are incurred. As more than one powerful sensors may detect
the signal, multiple volunteers may exist. A judicious, decentralized approach has to be
applied to ensure that only one CH is active in the vicinity of a target to be tracked with
high probability. Sensors in the vicinity of the active CH are invited to become members
of the cluster and report their measurements to the CH. Compared with the static clustering
approaches, dynamic clustering networked sensors do not statically belong to a cluster and
may support different clusters at different times. Moreover, as only one cluster is active in
the vicinity of a target with high probability, redundant data is suppressed and potential
interference and contention at the MAC level is mitigated.
Examples of dynamic cluster-based tracking are information-driven sensor querying (IDSQ)
(Zhao et al., 2002), DELTA (Walchli et al., 2007), and RARE (Olule et al., 2007).
Zhao et al. addressed the dynamic sensor collaboration problem in distributed tracking to
determine dynamically which sensor is most appropriate to perform the sensing, what
needs to be sensed, and to whom to communicate the information (Zhao et al., 2002). They
developed the IDSQ approach, enabling collaboration based on resource constraints and the
const of transmitting information. Information utility functions employed include entropy,

Mahalanobis distance, and a measure on expected posterior distribution. This approach
assumes that each node in the network can locally estimate the cost of sensing, processing
and communicating data to another node. Although the approach is power efficient (since
only few nodes are active at any given time), it is applied for tracking a single object only.
Walchli et al. present DELTA (Walchli et al., 2007), a distributive algorithm for tracking a
person moving at constant speed by dynamically making a cluster and selecting CH based
on light measurement. The CH is responsible to reliably monitor moving object and
collaborate with sensor nodes. The limitation of DELTA algorithm is that it can only deal
with constant speed, whereas, varying speed is not considered.
Energy aware probabilistic target localization algorithm for a single target using cluster-
based WSN is proposed in (Zou & Chakrabarty, 2003), where a two step protocol for
communication between CH and sensors in the cluster is put forward. In the first step,
sensors detecting the target report to the CH by a short message. Then the CH executes
localization procedure to determine the subset of sensors in the vicinity of target and query
detailed target information from them.
Olule et al. investigate an energy efficient target tracking protocol based on two algorithms,
ARE-Area (Reduced Area Reporting) and RARE-Node (Reduction of Active node
Redundancy) via static clustering (Olule et al., 2007). RARE-Area reduces number of nodes
participating in tracking by inhibiting far away nodes from taking part in tracking. RARE-
node reduces redundant information by identifying overlapping sensors. Cluster is formed
dynamically by prediction during target tracking (Jin et al., 2006), thus reducing number of
nodes involved in tracking. Although the method consumes low energy, the missing target
recovery procedure is not well defined.
Quantized measurements are usually adopted in such a network to attack the problem of
limited power supply and communication bandwidth. Very recently, the problem of target
tracking in a WSN that consists of randomly distributed range-only sensors is considered in
(Zhou et al., 2010)). The posterior Cramr-Rao lower bounds (CRLB) on the mean squared
error (MSE) on target tracking with quantized range-only measurements are derived. Due to
the analytical difficulties, particle filter is applied to approximate the theoretical bounds. In
this paper, recursion of posterior CRLB on tracking based on both constant velocity (CV)
and constant acceleration (CA) model for target dynamics and a general range-only
measuring model for local sensors are obtained. More details on tracking using quantized
messages can be found in Section 6.

3.3.3 Space-time clustering
In order to present the event processing with high accuracy, Phoha et al. propose the
dynamic space-time clustering (DSTC) (Phoha et al., 2003a). In this architecture, clusters of
space-time neighbouring nodes are dynamically organized to present the event around by
combining the local information among nodes in the inner space-time cluster. The type and
track of the target then are estimated by the CH.
Phoha et al. propose two methods by combining the DSTC and beamforming: one is DSTC
beamforming controlled, the other is DSTC logic controlled beamforming (Phoha et al.,
2003b). The former is composed of hundreds of low-cost DSTC nodes and a few
beamforming nodes, which estimate the target position through triangulation. In the case of
failure of beamforming nodes, the DSTC nodes are activated to localize the target. The latter

3.3 Cluster-based tracking
To facilitate collaborative data processing in target tracking-centric sensor networks, the
cluster architecture is usually used in which sensors are organized into clusters, with each
cluster consisting of a CH and several slave nodes (members). Hierarchical (clustering)
techniques can aid in reducing useful energy consumption (Heinzelman et al., 2002).
Clustering is particularly useful for applications that require scalability to hundreds or
thousands of nodes. Scalability in this context implies the need for load balancing and
efficient resource utilization. Clustering can be extremely effective in one-to-many, many-to-
one, one-to-any, or one-to-all (broadcast) communication. For example, in many-to-one
communication, clustering can support data fusion and reduce communication interference
(Younis & Fahmy, 2004).

3.3.1 Static clustering
Conventionally, clusters are formed statically at the time of network deployment. The
attributes of each cluster, such as the size of a cluster, the area it covers, and the members it
possesses, are static. In spite of its simplicity, the static cluster architecture suffers from
several drawbacks. First, fixed membership is not robust from the perspective of fault
tolerance. If a CH dies of power depletion, all the sensors in the cluster render useless.
Second, fixed membership prevents sensor nodes in different clusters from sharing
information and collaborating on data processing. Finally, fixed membership cannot adapt
to highly dynamic scenarios in which sensors in the region of high (low) event concentration
may be instrumented to stay awake (go to sleep).

3.3.2 Dynamic clustering
Dynamic cluster architectures, on the other hand, offer several desirable features (Chen et
al., 2003). Formation of a cluster is triggered by certain events of interest (e.g., detection of
an approaching target with acoustic sounds). When a sensor with sufficient battery and
computational power detects (with a high signal-to-noise ratio, SNR) signals of interest, it
volunteers to act as a CH. No explicit leader (CH) election is required and, hence, no
excessive message exchanges are incurred. As more than one powerful sensors may detect
the signal, multiple volunteers may exist. A judicious, decentralized approach has to be
applied to ensure that only one CH is active in the vicinity of a target to be tracked with
high probability. Sensors in the vicinity of the active CH are invited to become members
of the cluster and report their measurements to the CH. Compared with the static clustering
approaches, dynamic clustering networked sensors do not statically belong to a cluster and
may support different clusters at different times. Moreover, as only one cluster is active in
the vicinity of a target with high probability, redundant data is suppressed and potential
interference and contention at the MAC level is mitigated.
Examples of dynamic cluster-based tracking are information-driven sensor querying (IDSQ)
(Zhao et al., 2002), DELTA (Walchli et al., 2007), and RARE (Olule et al., 2007).
Zhao et al. addressed the dynamic sensor collaboration problem in distributed tracking to
determine dynamically which sensor is most appropriate to perform the sensing, what
needs to be sensed, and to whom to communicate the information (Zhao et al., 2002). They
developed the IDSQ approach, enabling collaboration based on resource constraints and the
const of transmitting information. Information utility functions employed include entropy,

Mahalanobis distance, and a measure on expected posterior distribution. This approach
assumes that each node in the network can locally estimate the cost of sensing, processing
and communicating data to another node. Although the approach is power efficient (since
only few nodes are active at any given time), it is applied for tracking a single object only.
Walchli et al. present DELTA (Walchli et al., 2007), a distributive algorithm for tracking a
person moving at constant speed by dynamically making a cluster and selecting CH based
on light measurement. The CH is responsible to reliably monitor moving object and
collaborate with sensor nodes. The limitation of DELTA algorithm is that it can only deal
with constant speed, whereas, varying speed is not considered.
Energy aware probabilistic target localization algorithm for a single target using cluster-
based WSN is proposed in (Zou & Chakrabarty, 2003), where a two step protocol for
communication between CH and sensors in the cluster is put forward. In the first step,
sensors detecting the target report to the CH by a short message. Then the CH executes
localization procedure to determine the subset of sensors in the vicinity of target and query
detailed target information from them.
Olule et al. investigate an energy efficient target tracking protocol based on two algorithms,
ARE-Area (Reduced Area Reporting) and RARE-Node (Reduction of Active node
Redundancy) via static clustering (Olule et al., 2007). RARE-Area reduces number of nodes
participating in tracking by inhibiting far away nodes from taking part in tracking. RARE-
node reduces redundant information by identifying overlapping sensors. Cluster is formed
dynamically by prediction during target tracking (Jin et al., 2006), thus reducing number of
nodes involved in tracking. Although the method consumes low energy, the missing target
recovery procedure is not well defined.
Quantized measurements are usually adopted in such a network to attack the problem of
limited power supply and communication bandwidth. Very recently, the problem of target
tracking in a WSN that consists of randomly distributed range-only sensors is considered in
(Zhou et al., 2010)). The posterior Cramr-Rao lower bounds (CRLB) on the mean squared
error (MSE) on target tracking with quantized range-only measurements are derived. Due to
the analytical difficulties, particle filter is applied to approximate the theoretical bounds. In
this paper, recursion of posterior CRLB on tracking based on both constant velocity (CV)
and constant acceleration (CA) model for target dynamics and a general range-only
measuring model for local sensors are obtained. More details on tracking using quantized
messages can be found in Section 6.

3.3.3 Space-time clustering
In order to present the event processing with high accuracy, Phoha et al. propose the
dynamic space-time clustering (DSTC) (Phoha et al., 2003a). In this architecture, clusters of
space-time neighbouring nodes are dynamically organized to present the event around by
combining the local information among nodes in the inner space-time cluster. The type and
track of the target then are estimated by the CH.
Phoha et al. propose two methods by combining the DSTC and beamforming: one is DSTC
beamforming controlled, the other is DSTC logic controlled beamforming (Phoha et al.,
2003b). The former is composed of hundreds of low-cost DSTC nodes and a few
beamforming nodes, which estimate the target position through triangulation. In the case of
failure of beamforming nodes, the DSTC nodes are activated to localize the target. The latter

determine a cluster to track the target according to DSTC logic, while the member nodes run
the beamforming algorithm to estimate the target state.

3.4 Hybrid method
Hybrid methods are referred to the tracking algorithms that fulfill the requirements of more
than one types of target tracking. Examples include distributed predictive tracking (DPT)
(Yang & Sikdor, 2003), DCAT (Chen et al., 2003), and Hierarchical prediction strategy (HPS)
(Wang et al., 2008).
The DPT adopts a clustering based approach for scalability and a prediction based tracking
mechanism to provide a distributed and energy efficient solution (Yang & Sikdor, 2003). The
protocol is proven to be robust against node or prediction failures which may result in
temporary loss of the target and recovers from such scenarios quickly and with very little
additional energy use.
A decentralized dynamic clustering algorithm for single target tracking (Here we referred as
dynamic clustering for acoustic tracking, DCAT) is proposed in (Chen et al., 2003). Using
Voronoi Diagrams, clusters are formed and only one CH becomes active when the acoustic
signal strength detected by CH exceeds a pre-determined threshold. The CH then asks the
sensors in its vicinity to join cluster by sending a broadcast packet. The sensor based on the
probabilistic distance estimates between itself and target, decides whether it should reply to
CH. Afterwards, CH executes a localization method to estimate location of target based on
sensor replies and sends result to the sink.
In HPS, cluster is formed using Voronoi division and a target next location is predicted via
Least Square Method but overheads are not well defined. HVE protocol uses cluster
structure and prediction for estimating shape and size of forwarding zone and delivering
mobicast messages.
A location model determines the granularity of location information and the prediction
model processes the historical data to predict next movement of mobile object. An
interesting example of multiple targets tracking using prediction is given in (Chong et al.,
2003).

4. Tracking methods for peer-to-peer networks
For the tree- or cluster-based methods, sensing task is usually performed by several nodes at
a time and inflicts heavy computation burden on the root node or the CH. This makes the
tree- or cluster-based WSN tracking systems lack of robustness in case of root node or the
CH failures. On the contrary, another architecture for target tracking is the peer-to-peer
WSN. As it can guarantee that sensors obtain the desired estimates and rely only on single-
hop communications between neighbouring nodes, the limitations mentioned above are not
encountered in peer-to-peer WSN based target tracking systems.
On the other hand, the well-known strategy concerning estimation and tracking is
decentralized Kalman filtering or nonlinear filtering scheme, e.g. extended Kalman filtering
(EKF), unscented Kalman filtering (UKF), and particle filtering (PF), which involve state
estimation using a set of local filters that communicate with all other nodes (see e.g. Li &
Wang, 2000; Mutambara, 1998; Vercauteren & Wang, 2005, and the references therein). The
information flow in the traditional decentralized Kalman filtering (see e.g. Mutambara,
1998) or unscented Kalman filtering scheme (Vercauteren & Wang, 2005) is all-to-all with

communication complexity of O(N2) (here N is the number of sensors in the network),
which is not scalable for sensor networks (Speyer et al., 2004). On the contrary, the peer-to-
peer network tracking is usually based on average consensus algorithms that have proven to
be effective tools for performing network-wide distributed computation task ranging from
flocking to robot rendezvous as in the papers (Olfati-Saber & Murry, 2004; Tanner et al.,
2007; Kar & Moura, 2009), and the references therein. Hence, we refer this kind of methods
as average consensus based tracking (AC tracking).

4.1 Embedded filter based consensus
Distributed estimation using peer-to-peer WSNs is based on successive refinements of local
estimates maintained at individual sensors. In a nutshell, each iteration of the algorithm
comprises a communication step where the sensors interchange information with their
neighbours, and an update step where each sensor uses this information to refine its local
estimate. In this context, estimation of deterministic parameters in linear data models, via
decentralized computation of the BLUE or the sample average estimator, was considered in
(Olfati-Saber & Murry, 2004; Scherber & Papadopoulos, 2005; Xiao & Boyd, 2004) using the
notion of consensus averaging. Decentralized estimation of Gaussian random parameters
was reported in (Delouille et al., 2004) for stationary environments, while the dynamic case
was considered in (Spanos et al., 2005).
Olfati-Saber introduces a distributed Kalman filtering (DKF) algorithm that uses dynamic
consensus strategy in (Olfati-Saber, 2005; Olfati-Saber, 2007). The DKF algorithm consists of
a network of micro-Kalman filters each embedded with a high-gain high-pass consensus
filter (or consensus protocol). The role of consensus filters is to estimate of global
information contribution using only local and neighbouring information. Recently, the
problem of estimating a simpler scenario with a scalar state of a dynamical system from
distributed noisy measurements based on consensus strategies is considered in (Carli et al.,
2006), the focuses are with the interaction between the consensus matrix, the number of
messages exchanged per sampling time, and the Kalman gain for scalar systems.
Very recently, the distributed and scalable robust filtering problem using average consensus
strategy in a sensor network is investigated in (Zhou & Li, 2009a). Specifically, based on the
information form robust filter, every node estimates the global average information
contribution using local and neighbours information rather than using the information
from whole network. Due to the adoption of iterations of robust filter, the proposed
algorithm relaxes the necessity to have the prior knowledge of the noise statistics. Moreover,
the proposed algorithm is applicable to large-scale sensor network since each node
broadcasts message only to its neighbouring nodes.
The aforementioned embedded filter based consensus for distributed target tracking is
proposed for linear systems with Gaussian or energy bounded noises, there is little result on
tracking algorithm for nonlinear dynamic systems and/or nonlinear observations. In (Zhou
& Li, 2009b), a distributed scalable Sigma-Point Kalman filter (DS2PKF) is proposed for
distributed target tracking in a sensor network based on the dynamic consensus strategy.
The main idea is to use dynamic consensus strategy to the information form sigma-point
Kalman filter (ISPKF) that derived from weighted statistical linearization perspective. Each
node estimates the global average information contribution by using local and neighbours
information rather than by the information from all nodes in the network. Therefore, the
proposed DSPKF algorithm is completely distributed and applicable to large-scale sensor

determine a cluster to track the target according to DSTC logic, while the member nodes run
the beamforming algorithm to estimate the target state.

3.4 Hybrid method
Hybrid methods are referred to the tracking algorithms that fulfill the requirements of more
than one types of target tracking. Examples include distributed predictive tracking (DPT)
(Yang & Sikdor, 2003), DCAT (Chen et al., 2003), and Hierarchical prediction strategy (HPS)
(Wang et al., 2008).
The DPT adopts a clustering based approach for scalability and a prediction based tracking
mechanism to provide a distributed and energy efficient solution (Yang & Sikdor, 2003). The
protocol is proven to be robust against node or prediction failures which may result in
temporary loss of the target and recovers from such scenarios quickly and with very little
additional energy use.
A decentralized dynamic clustering algorithm for single target tracking (Here we referred as
dynamic clustering for acoustic tracking, DCAT) is proposed in (Chen et al., 2003). Using
Voronoi Diagrams, clusters are formed and only one CH becomes active when the acoustic
signal strength detected by CH exceeds a pre-determined threshold. The CH then asks the
sensors in its vicinity to join cluster by sending a broadcast packet. The sensor based on the
probabilistic distance estimates between itself and target, decides whether it should reply to
CH. Afterwards, CH executes a localization method to estimate location of target based on
sensor replies and sends result to the sink.
In HPS, cluster is formed using Voronoi division and a target next location is predicted via
Least Square Method but overheads are not well defined. HVE protocol uses cluster
structure and prediction for estimating shape and size of forwarding zone and delivering
mobicast messages.
A location model determines the granularity of location information and the prediction
model processes the historical data to predict next movement of mobile object. An
interesting example of multiple targets tracking using prediction is given in (Chong et al.,
2003).

4. Tracking methods for peer-to-peer networks
For the tree- or cluster-based methods, sensing task is usually performed by several nodes at
a time and inflicts heavy computation burden on the root node or the CH. This makes the
tree- or cluster-based WSN tracking systems lack of robustness in case of root node or the
CH failures. On the contrary, another architecture for target tracking is the peer-to-peer
WSN. As it can guarantee that sensors obtain the desired estimates and rely only on single-
hop communications between neighbouring nodes, the limitations mentioned above are not
encountered in peer-to-peer WSN based target tracking systems.
On the other hand, the well-known strategy concerning estimation and tracking is
decentralized Kalman filtering or nonlinear filtering scheme, e.g. extended Kalman filtering
(EKF), unscented Kalman filtering (UKF), and particle filtering (PF), which involve state
estimation using a set of local filters that communicate with all other nodes (see e.g. Li &
Wang, 2000; Mutambara, 1998; Vercauteren & Wang, 2005, and the references therein). The
information flow in the traditional decentralized Kalman filtering (see e.g. Mutambara,
1998) or unscented Kalman filtering scheme (Vercauteren & Wang, 2005) is all-to-all with

communication complexity of O(N2) (here N is the number of sensors in the network),
which is not scalable for sensor networks (Speyer et al., 2004). On the contrary, the peer-to-
peer network tracking is usually based on average consensus algorithms that have proven to
be effective tools for performing network-wide distributed computation task ranging from
flocking to robot rendezvous as in the papers (Olfati-Saber & Murry, 2004; Tanner et al.,
2007; Kar & Moura, 2009), and the references therein. Hence, we refer this kind of methods
as average consensus based tracking (AC tracking).

4.1 Embedded filter based consensus
Distributed estimation using peer-to-peer WSNs is based on successive refinements of local
estimates maintained at individual sensors. In a nutshell, each iteration of the algorithm
comprises a communication step where the sensors interchange information with their
neighbours, and an update step where each sensor uses this information to refine its local
estimate. In this context, estimation of deterministic parameters in linear data models, via
decentralized computation of the BLUE or the sample average estimator, was considered in
(Olfati-Saber & Murry, 2004; Scherber & Papadopoulos, 2005; Xiao & Boyd, 2004) using the
notion of consensus averaging. Decentralized estimation of Gaussian random parameters
was reported in (Delouille et al., 2004) for stationary environments, while the dynamic case
was considered in (Spanos et al., 2005).
Olfati-Saber introduces a distributed Kalman filtering (DKF) algorithm that uses dynamic
consensus strategy in (Olfati-Saber, 2005; Olfati-Saber, 2007). The DKF algorithm consists of
a network of micro-Kalman filters each embedded with a high-gain high-pass consensus
filter (or consensus protocol). The role of consensus filters is to estimate of global
information contribution using only local and neighbouring information. Recently, the
problem of estimating a simpler scenario with a scalar state of a dynamical system from
distributed noisy measurements based on consensus strategies is considered in (Carli et al.,
2006), the focuses are with the interaction between the consensus matrix, the number of
messages exchanged per sampling time, and the Kalman gain for scalar systems.
Very recently, the distributed and scalable robust filtering problem using average consensus
strategy in a sensor network is investigated in (Zhou & Li, 2009a). Specifically, based on the
information form robust filter, every node estimates the global average information
contribution using local and neighbours information rather than using the information
from whole network. Due to the adoption of iterations of robust filter, the proposed
algorithm relaxes the necessity to have the prior knowledge of the noise statistics. Moreover,
the proposed algorithm is applicable to large-scale sensor network since each node
broadcasts message only to its neighbouring nodes.
The aforementioned embedded filter based consensus for distributed target tracking is
proposed for linear systems with Gaussian or energy bounded noises, there is little result on
tracking algorithm for nonlinear dynamic systems and/or nonlinear observations. In (Zhou
& Li, 2009b), a distributed scalable Sigma-Point Kalman filter (DS2PKF) is proposed for
distributed target tracking in a sensor network based on the dynamic consensus strategy.
The main idea is to use dynamic consensus strategy to the information form sigma-point
Kalman filter (ISPKF) that derived from weighted statistical linearization perspective. Each
node estimates the global average information contribution by using local and neighbours
information rather than by the information from all nodes in the network. Therefore, the
proposed DSPKF algorithm is completely distributed and applicable to large-scale sensor

network. A novel dynamic consensus filter is proposed, and its asymptotical convergence
performance and stability are discussed.

4.2 Alternating-direction based consensus
Alternating-direction method of multipliers (Bertsekas & Tsitsiklis, 1999) is proven to be
efficient in solving the distributed estimation (Schizas et al., 2008a; Schizas et al., 2008b).
Recently, decentralized estimation of random signals in arbitrary nonlinear and non-
Gaussian setups was considered in (Schizas & Giannakis, 2006), while distributed estimation
of stationary Markov random fields was pursued in (Dogandzic & Zhang, 2006).
Adaptive algorithms based on in-network processing of distributed observations are well-
motivated for online parameter estimation and tracking of (non)stationary signals using
peer-to-peer WSNs. To this end, a fully distributed least mean-square (D-LMS) algorithm is
developed in (Schizas et al., 2009), offering simplicity and flexibility while solely requiring
single-hop communications among sensors. The resultant estimator minimizes a pertinent
squared-error cost by resorting to i) the alternating-direction method of multipliers so as to
gain the desired degree of parallelization and ii) a stochastic approximation iteration to cope
with the time-varying statistics of the process under consideration. Information is efficiently
percolated across the WSN using a subset of bridge sensors, which further tradeoff
communication cost for robustness to sensor failures. For a linear data model and under
mild assumptions aligned with those considered in the centralized LMS, stability of the
novel D-LMS algorithm is established to guarantee that local sensor estimation error norms
remain bounded most of the time.
Forero et al. develop a decentralized expectation-maximization (EM) algorithm to estimate
the parameters of a mixture density model for use in distributed learning tasks performed
with data collected at spatially deployed wireless sensors (Forero et al., 2008). The E-step in
the novel iterative scheme relies on local information available to individual sensors, while
during the M-step sensors exchange information only with their single hop neighbours to
reach consensus and eventually percolate the global information needed to estimate the
wanted parameters across the WSN.

5. Analysis and comparison
All the methods mentioned above are compared in Table 1 in terms of tracking accuracy,
communicational burden, scalability, computational complexity, and fault tolerance, etc. In
Table 1, we rate the method into four levels, i.e. A-D, according to the performance
criterions mention above. We note that criterions, such as communicational burden, tracking
accuracy and fault tolerance, are proportional to energy utilization for target tracking
through WSNs. If communicational burden is high for cluster formation, more energy is
consumed. High tracking accuracy demand will ultimately end with additional energy
usage. Similarly fault tolerance will increase overheads and energy consumption. The total
energy consumption and bandwidth usage during target tracking is the key concern in the
majority of the methods since the network is with strictly limited energy and bandwidth.
The energy consumption of a sensor node can be divided into three main domains, radio
communication, sensing and data processing.
It is also worth pointing out that all the rating levels are relative since different methods are
proposed within different network scenarios. For example, the AC tracking is mainly for the

peer-to-peer network to improve the scalability. However, the cluster-based tracking such as
IDSQ is mainly for the energy consumption and the lifetime.

Method
Tracking
accuracy
Scalability
Computational
complexity
Communicational
burden
Fault
tolerance
DC A D B D A
STUN C C C C B
DCTC B C C B C
DAT D C C C C
DOT D C C C C
R-tree B B D C C
IDSQ C B A A B
DELTA C B B B B
RARE C B C B C
DSTC B B B B B
DPT D C B C C
DCAT A C A C B
HPS C C B C C
AC
tracking
B A B A A
Table 1. Comparison of target tracking methods in WSNs

6. Quantized scenario
In the WSN tracking system, ach sensor node acquires measurements which are noisy linear
or nonlinear transformations of the target state. The sensors then transmit measurements to
the fusion center (for the FC-based WSNs) or the neighbouring nodes (for the distributed
peer-to-peer WSNs) in order to form a state estimate. If measurements were available at a
common location, minimum mean-square error (MMSE) estimates could be obtained using
a Kalman filter, or nonlinear estimation methods, such as UKF and PF. However, since
measurements are distributed in space and there is limited communication bandwidth, the
measurements have to be quantized before transmission. Thus, the original estimation
problem is transformed into decentralized state estimation based on quantized
measurements. The problem is further complicated by the harsh environment typical of
WSNs; see e.g., Chong & Kumar, 2003, and Culler et al., 2004.
The problem of decentralized estimation based on quantized measurements has been
studied in early works such as Gubner, 1993, and Lam & Reibman, 1993. Recently, universal
decentralized estimation taking into account local signal-to-noise ratio (SNR) and the
channel path loss in sensor network is studied (Xiao et al., 2005). When the noise
probabilistic density function (PDF) is unknown, the problem of estimation based on
severely quantized data has been also addressed in (Luo, 2005).
In this section, we category the tracking methods based on quantized information into
quantized measurements and quantized innovations. The latter is usually with higher
accuracy when using the same quantization bit rate. It is because that the range of
innovations is commonly little that causes little quantization noise.

network. A novel dynamic consensus filter is proposed, and its asymptotical convergence
performance and stability are discussed.

4.2 Alternating-direction based consensus
Alternating-direction method of multipliers (Bertsekas & Tsitsiklis, 1999) is proven to be
efficient in solving the distributed estimation (Schizas et al., 2008a; Schizas et al., 2008b).
Recently, decentralized estimation of random signals in arbitrary nonlinear and non-
Gaussian setups was considered in (Schizas & Giannakis, 2006), while distributed estimation
of stationary Markov random fields was pursued in (Dogandzic & Zhang, 2006).
Adaptive algorithms based on in-network processing of distributed observations are well-
motivated for online parameter estimation and tracking of (non)stationary signals using
peer-to-peer WSNs. To this end, a fully distributed least mean-square (D-LMS) algorithm is
developed in (Schizas et al., 2009), offering simplicity and flexibility while solely requiring
single-hop communications among sensors. The resultant estimator minimizes a pertinent
squared-error cost by resorting to i) the alternating-direction method of multipliers so as to
gain the desired degree of parallelization and ii) a stochastic approximation iteration to cope
with the time-varying statistics of the process under consideration. Information is efficiently
percolated across the WSN using a subset of bridge sensors, which further tradeoff
communication cost for robustness to sensor failures. For a linear data model and under
mild assumptions aligned with those considered in the centralized LMS, stability of the
novel D-LMS algorithm is established to guarantee that local sensor estimation error norms
remain bounded most of the time.
Forero et al. develop a decentralized expectation-maximization (EM) algorithm to estimate
the parameters of a mixture density model for use in distributed learning tasks performed
with data collected at spatially deployed wireless sensors (Forero et al., 2008). The E-step in
the novel iterative scheme relies on local information available to individual sensors, while
during the M-step sensors exchange information only with their single hop neighbours to
reach consensus and eventually percolate the global information needed to estimate the
wanted parameters across the WSN.

5. Analysis and comparison
All the methods mentioned above are compared in Table 1 in terms of tracking accuracy,
communicational burden, scalability, computational complexity, and fault tolerance, etc. In
Table 1, we rate the method into four levels, i.e. A-D, according to the performance
criterions mention above. We note that criterions, such as communicational burden, tracking
accuracy and fault tolerance, are proportional to energy utilization for target tracking
through WSNs. If communicational burden is high for cluster formation, more energy is
consumed. High tracking accuracy demand will ultimately end with additional energy
usage. Similarly fault tolerance will increase overheads and energy consumption. The total
energy consumption and bandwidth usage during target tracking is the key concern in the
majority of the methods since the network is with strictly limited energy and bandwidth.
The energy consumption of a sensor node can be divided into three main domains, radio
communication, sensing and data processing.
It is also worth pointing out that all the rating levels are relative since different methods are
proposed within different network scenarios. For example, the AC tracking is mainly for the

peer-to-peer network to improve the scalability. However, the cluster-based tracking such as
IDSQ is mainly for the energy consumption and the lifetime.

Method
Tracking
accuracy
Scalability
Computational
complexity
Communicational
burden
Fault
tolerance
DC A D B D A
STUN C C C C B
DCTC B C C B C
DAT D C C C C
DOT D C C C C
R-tree B B D C C
IDSQ C B A A B
DELTA C B B B B
RARE C B C B C
DSTC B B B B B
DPT D C B C C
DCAT A C A C B
HPS C C B C C
AC
tracking
B A B A A
Table 1. Comparison of target tracking methods in WSNs

6. Quantized scenario
In the WSN tracking system, ach sensor node acquires measurements which are noisy linear
or nonlinear transformations of the target state. The sensors then transmit measurements to
the fusion center (for the FC-based WSNs) or the neighbouring nodes (for the distributed
peer-to-peer WSNs) in order to form a state estimate. If measurements were available at a
common location, minimum mean-square error (MMSE) estimates could be obtained using
a Kalman filter, or nonlinear estimation methods, such as UKF and PF. However, since
measurements are distributed in space and there is limited communication bandwidth, the
measurements have to be quantized before transmission. Thus, the original estimation
problem is transformed into decentralized state estimation based on quantized
measurements. The problem is further complicated by the harsh environment typical of
WSNs; see e.g., Chong & Kumar, 2003, and Culler et al., 2004.
The problem of decentralized estimation based on quantized measurements has been
studied in early works such as Gubner, 1993, and Lam & Reibman, 1993. Recently, universal
decentralized estimation taking into account local signal-to-noise ratio (SNR) and the
channel path loss in sensor network is studied (Xiao et al., 2005). When the noise
probabilistic density function (PDF) is unknown, the problem of estimation based on
severely quantized data has been also addressed in (Luo, 2005).
In this section, we category the tracking methods based on quantized information into
quantized measurements and quantized innovations. The latter is usually with higher
accuracy when using the same quantization bit rate. It is because that the range of
innovations is commonly little that causes little quantization noise.

6.1 Quantized measurement based tracking
Quantizing measurements to estimate a parameter of interest is not the same as quantizing a
signal for later reconstruction (Gray, 2006). Instead of a reconstruction algorithm, the
objective is finding, e.g., MMSE optimal, estimators using quantized observations
(Papadopoulos et al., 2001; Ribeiro & Giannakis, 2006). Furthermore, optimal quantizers for
reconstruction are, generally, different from optimal quantizers for estimation. State
estimation using quantized observations is a nonlinear estimation problem that can be
solved using e.g., EKF, UKF, or PF.
From the measurement fusion perspective, the problem for target tracking using quantized
information in WSNs is investigated in (Zhou & Li, 2009c) and (Zhou et al., 2009a). Due to
the limited energy and bandwidth, each activated node quantizes and then transmits the
local measurements by probabilistic quantization strategy. The FC estimates the target state
in a dimension compression way instead of merging all the quantized messages to a vector
(augmented scheme). A closed-form solution to the optimization problem for bandwidth
scheduling is given, where the total energy consumption measure is minimized subject to a
constraint on the mean square error (MSE) incurred by quasi-best linear unbiased estimation
(Quasi-BLUE) fusion. The results are extended to the case of tracking maneuvering target
and correlation noise in (Zhou & Li, 2009d) and (Zhou et al., 2009b), respectively.
Quantizing measurements is an efficient way that gives tradeoff between the
bandwidth/energy constraints and tacking accuracy. However, if the values of
measurements are large, quantizing measurements will bring large information loss under
the limited bandwidth, which means that the variance of the quantization noise is large. In
this scenario, the quantized measurements based tracking will have a low filtering accuracy.
To reduce the information loss and improve the filtering accuracy, quantized innovations
based tracking has been extensively investigated recently. Since the values of innovation
data are smaller than those of measured data, quantizing innovations will bring smaller
information loss than quantizing measurements under the same bandwidth constraint.

6.2 Quantized innovation based tracking
Surprisingly, for the case where quantized observations are defined as the sign of the
innovation (SOI) sequence, it is possible to derive a filter with complexity and performance
very close to the clairvoyant KF based on the analog-amplitude observations (Ribeiro et al.,
2006). Even though promising, the approach of (Ribeiro et al., 2006) is limited to a particular
1-bit per observation quantizer. Msechu et al. introduce two novel decentralized KF
estimators based on quantized measurement innovations (Msechu et al., 2008). In the first
quantization approach, the region of an observation is partitioned into contiguous, non-
overlapping intervals where each partition is binary encoded using a block of bits. Analysis
and Monte Carlo simulations reveal that with minimal communication overhead, the mean-
square error (MSE) of a novel decentralized KF tracker based on 2-3 bits comes stunningly
close to that of the clairvoyant KF. In the second quantization approach, if intersensor
communications can afford bits at time , then the th bit is iteratively formed using the sign
of the difference between the observation and its estimate based on past observations (up to
time 1) along with previous bits (up to 1) of the current observation.
Recently, by optimizing the filter with respect to the quantization levels, a multiple-level
quantized innovation Kalman filter (MLQ-KF) for estimation of linear dynamic stochastic
systems is proposed in (You et al., 2008). Furthermore, Sukhavasi and Hassibi propose a

particle filter that approximates the optimal nonlinear filer and observe that the error
covariance of the particle filter follows the modified Riccati recursion (Sukhavasi, & Hassibi,
2009).
Very recently, Zhou et al. investigate the decentralized collaborative target tracking problem
in a WSN from the fusion of quantized innovations perspective (Zhou et al., 2009c). A
hierarchical fusion structure with feedback from the FC to each deployed sensor is proposed
for tracking a target with nonlinear Gaussian dynamics. Probabilistic quantization strategy
is employed in the local sensor node to quantize the innovation. After the FC received the
quantized innovations, it estimates the state of the target using the Sigma-Point Kalman
Filtering (SPKF). To attack the energy/power source and communication bandwidth
constraints, the tradeoff between the communication energy and the global tracking
accuracy is considered in (Zhou et al., 2009d). By Lagrange multiplier, a closed-form
solution to the optimization problem for bandwidth scheduling is given, where the total
energy consumption measure is minimized subject to a constraint on the covariance of the
quantization noises. Simulation example is given to illustrate the proposed scheme obtains
average percentage of communication energy saving up to 41.5% compared with the
uniform quantization, while keeping tracking accuracy very closely to the clairvoyant UKF
that relies on analog-amplitude measurements. In (Ozdemir et al., 2009), a new framework
for target tracking in a wireless sensor network using particle filters is proposed. Under this
framework, the imperfect nature of the wireless communication channels between sensors
and the FC along with some physical layer design parameters of the network are
incorporated in the tracking algorithm based on particle filters. It is call channel-aware
particle filtering that derived for different wireless channel models and receiver
architectures. Furthermore, the posterior CRLBs for the proposed channel-aware particle
filters are also given.

7. Concluding remarks and open research directions
The extensively research of target tracking through WSNs inspired us to present a literature
survey. In this chapter, we have explored the categories of target tracking methods,
including tree-based, cluster-based, hybrid, and consensus-based tracking algorithm.
Considering the stringent limitation on energy supply, the quantized messages based
tracking has been discussed separately.
The emergence of WSN in the variety of application areas brought many open issues to
researchers. The open research issues for target tracking in WSNs include, channel-aware
tracking, mobile node aided tracking, multitarget association & tracking, cross-layer design,
and fault tolerant tracking methods, etc.
First, wireless communication channels between sensors and the FC or base station are not
perfect. Incorporating the statistics of the channel imperfection to the tracking algorithm is
expected to improve the tracking accuracy. Second, the scenario becomes complicated in the
presence of multiple targets and their tracking with mobile sensors which leads to intend
more realistic solutions. Message transmission consumes more energy than local processing,
thus, well organized computing and nominal transmission of messages without degradation
of performance must be considered while designing a target tracking method (Rapaka &
Madria, 2007). Data association is an important problem when multiple targets are present
in a small region. Each node must associate its measurements of the environment with

6.1 Quantized measurement based tracking
Quantizing measurements to estimate a parameter of interest is not the same as quantizing a
signal for later reconstruction (Gray, 2006). Instead of a reconstruction algorithm, the
objective is finding, e.g., MMSE optimal, estimators using quantized observations
(Papadopoulos et al., 2001; Ribeiro & Giannakis, 2006). Furthermore, optimal quantizers for
reconstruction are, generally, different from optimal quantizers for estimation. State
estimation using quantized observations is a nonlinear estimation problem that can be
solved using e.g., EKF, UKF, or PF.
From the measurement fusion perspective, the problem for target tracking using quantized
information in WSNs is investigated in (Zhou & Li, 2009c) and (Zhou et al., 2009a). Due to
the limited energy and bandwidth, each activated node quantizes and then transmits the
local measurements by probabilistic quantization strategy. The FC estimates the target state
in a dimension compression way instead of merging all the quantized messages to a vector
(augmented scheme). A closed-form solution to the optimization problem for bandwidth
scheduling is given, where the total energy consumption measure is minimized subject to a
constraint on the mean square error (MSE) incurred by quasi-best linear unbiased estimation
(Quasi-BLUE) fusion. The results are extended to the case of tracking maneuvering target
and correlation noise in (Zhou & Li, 2009d) and (Zhou et al., 2009b), respectively.
Quantizing measurements is an efficient way that gives tradeoff between the
bandwidth/energy constraints and tacking accuracy. However, if the values of
measurements are large, quantizing measurements will bring large information loss under
the limited bandwidth, which means that the variance of the quantization noise is large. In
this scenario, the quantized measurements based tracking will have a low filtering accuracy.
To reduce the information loss and improve the filtering accuracy, quantized innovations
based tracking has been extensively investigated recently. Since the values of innovation
data are smaller than those of measured data, quantizing innovations will bring smaller
information loss than quantizing measurements under the same bandwidth constraint.

6.2 Quantized innovation based tracking
Surprisingly, for the case where quantized observations are defined as the sign of the
innovation (SOI) sequence, it is possible to derive a filter with complexity and performance
very close to the clairvoyant KF based on the analog-amplitude observations (Ribeiro et al.,
2006). Even though promising, the approach of (Ribeiro et al., 2006) is limited to a particular
1-bit per observation quantizer. Msechu et al. introduce two novel decentralized KF
estimators based on quantized measurement innovations (Msechu et al., 2008). In the first
quantization approach, the region of an observation is partitioned into contiguous, non-
overlapping intervals where each partition is binary encoded using a block of bits. Analysis
and Monte Carlo simulations reveal that with minimal communication overhead, the mean-
square error (MSE) of a novel decentralized KF tracker based on 2-3 bits comes stunningly
close to that of the clairvoyant KF. In the second quantization approach, if intersensor
communications can afford bits at time , then the th bit is iteratively formed using the sign
of the difference between the observation and its estimate based on past observations (up to
time 1) along with previous bits (up to 1) of the current observation.
Recently, by optimizing the filter with respect to the quantization levels, a multiple-level
quantized innovation Kalman filter (MLQ-KF) for estimation of linear dynamic stochastic
systems is proposed in (You et al., 2008). Furthermore, Sukhavasi and Hassibi propose a

particle filter that approximates the optimal nonlinear filer and observe that the error
covariance of the particle filter follows the modified Riccati recursion (Sukhavasi, & Hassibi,
2009).
Very recently, Zhou et al. investigate the decentralized collaborative target tracking problem
in a WSN from the fusion of quantized innovations perspective (Zhou et al., 2009c). A
hierarchical fusion structure with feedback from the FC to each deployed sensor is proposed
for tracking a target with nonlinear Gaussian dynamics. Probabilistic quantization strategy
is employed in the local sensor node to quantize the innovation. After the FC received the
quantized innovations, it estimates the state of the target using the Sigma-Point Kalman
Filtering (SPKF). To attack the energy/power source and communication bandwidth
constraints, the tradeoff between the communication energy and the global tracking
accuracy is considered in (Zhou et al., 2009d). By Lagrange multiplier, a closed-form
solution to the optimization problem for bandwidth scheduling is given, where the total
energy consumption measure is minimized subject to a constraint on the covariance of the
quantization noises. Simulation example is given to illustrate the proposed scheme obtains
average percentage of communication energy saving up to 41.5% compared with the
uniform quantization, while keeping tracking accuracy very closely to the clairvoyant UKF
that relies on analog-amplitude measurements. In (Ozdemir et al., 2009), a new framework
for target tracking in a wireless sensor network using particle filters is proposed. Under this
framework, the imperfect nature of the wireless communication channels between sensors
and the FC along with some physical layer design parameters of the network are
incorporated in the tracking algorithm based on particle filters. It is call channel-aware
particle filtering that derived for different wireless channel models and receiver
architectures. Furthermore, the posterior CRLBs for the proposed channel-aware particle
filters are also given.

7. Concluding remarks and open research directions
The extensively research of target tracking through WSNs inspired us to present a literature
survey. In this chapter, we have explored the categories of target tracking methods,
including tree-based, cluster-based, hybrid, and consensus-based tracking algorithm.
Considering the stringent limitation on energy supply, the quantized messages based
tracking has been discussed separately.
The emergence of WSN in the variety of application areas brought many open issues to
researchers. The open research issues for target tracking in WSNs include, channel-aware
tracking, mobile node aided tracking, multitarget association & tracking, cross-layer design,
and fault tolerant tracking methods, etc.
First, wireless communication channels between sensors and the FC or base station are not
perfect. Incorporating the statistics of the channel imperfection to the tracking algorithm is
expected to improve the tracking accuracy. Second, the scenario becomes complicated in the
presence of multiple targets and their tracking with mobile sensors which leads to intend
more realistic solutions. Message transmission consumes more energy than local processing,
thus, well organized computing and nominal transmission of messages without degradation
of performance must be considered while designing a target tracking method (Rapaka &
Madria, 2007). Data association is an important problem when multiple targets are present
in a small region. Each node must associate its measurements of the environment with

individual targets. Combining the track association and tracking becomes more
complicated, especially in circumstance of low cost sensor network with limited
computation capacity and communication bandwidth (Li et al., 2010).
Another interesting issue for target tracking is the consideration of node failure. The sensor
nodes are usually deployed in harsh environments so various nodes may fail, may be
attacked or node energy may be depleted due to obstacles. Therefore, fault tolerant target
tracking algorithms and protocols must be designed for wireless sensor networks as the
fault tolerant approaches developed for traditional wired or wireless networks are not well
suited for WSN because of various differences between these networks (Ding & Cheng,
2009).
The cross-layered approach in WSN is more effective and energy efficient than in traditional
layered approach. While traditional layered approach endures more transfer overhead,
cross-layered approach minimizes these overhead by having data shared among layers
(Melodia et al., 2006; Kwon et al., 2006; Song & Hatzinakos, 2007). In the cross-layered
approach, the protocol stack is treated as a system and not individual layers, independent of
each other. Layers share information from the system. The development of various protocols
and services in a cross-layered approach is optimized and improved as a whole.
In last decades, the problem of decentralized information fusion has been discussed
extensively in the literature. However, the algorithms developed are free of energy and
communication constraints, see e.g. Sun & Deng, 2004; Li & Wang, 2000; Zhou & Li, 2008a;
Zhou & Li, 2008b. Novel fusion approaches include practical constraints in WSNs while
keeping high fusion performance must be investigated (Ruan et al., 2008). Moreover,
tracking with adaptive quantization thresholds and/or allocated bandwidth is another
promising research direction since the communicational condition dependent quantization
will definitely improve the estimation accuracy while using less communicational energy
(Zhou et al., 2011; Xu & Li, 2010).
Finally, WSNs have the potential to enhance and change the way people interact with
technology and the world (Aboelaze & Aloul, 2005). The direction of future WSNs also lies
in identifying real business and industry needs. Interactions between research and
development are necessary to bridge the gap between existing technology and the
development of business solutions. Applying sensor technology to different applications
will improve business processes as well as open up more problems for researchers.

8. Acknowledgements
The work was jointly supported by the National Natural Science Foundation of China
(Under Grant 60874104, 60935001); 973Project (2009CB824900, 2010CB734103); Shanghai Key
Basic Research Foundation (08JC1411800).

9. References
Aboelaze, M. & Aloul, F. (2005). Current and future trends in sensor networks: a survey,
Proceedings of the 14th IEEE Intl. Conf. on Wireless and Optical Communication, New
York, USA, pp. 133-138
Akyildiz, I.F.; Su, W. Sankarasubramaniam, Y. & Cayirci, E. (2002). Wireless Sensor
Network: A Survey. Computer Networks, vol. 38, no. 4, pp. 393422

Akyildiz, I.F.; Melodia, T. & Chowdury, K.R. (2007). Wireless multimedia sensor networks:
A survey. IEEE Wireless Communications, vol. 14, no. 6, pp. 32-39
Bertsekas, D. P. & Tsitsiklis, J. N. (1999). Parallel and Distributed Computation: Numerical
Methods, 2nd ed. Belmont, MA: Athena Scientific
Carli, R. Chiuso, A. Schenato, L. & Zampieri, S. (2006). Distributed Kalman filtering based
on consensus strategies. IEEE J. Selected Areas in Communications, vol. 26, no. 4, pp.
622-632
Chen, W.P.; Hou, J.C. & Sha, L. (2003). Dynamic clustering for acoustic target tracking in
wireless sensor networks, Proceedings of 11th IEEE International Conf. Network
Protocols, Atlanta, Georgia, USA, pp. 284294
wireless sensor networks. IEEE Transactions on Mobile Computing, vol. 3, no. 3, pp.
258-273
Chong, C.Y. & Kumar, S. (2003). Sensor networks: Evolution, opportunities, and challenges.
Proc. IEEE, vol. 91, pp. 2741
Chong, C.Y.; Zhao, F. Mori, S. & Kumar, S. (2003). Distributed tracking in wireless Ad Hoc
sensor networks, Proceedings Sixth Intl. Conf. on Information Fusion, Cairns,
Australia, pp. 431438
Culler, D.; Estrin, D. & Srivastava, M. (2004). Overview of sensor networks. Computer, vol.
37, no. 8, pp. 4149
Delouille, V.; Neelamani, R. & Baraniuk, R. (2004). Robust distributed estimation in sensor
networks using the embedded polygons algorithm, Proceedings of the 3rd Int. Symp.
Info. Processing Sensor Networks, Berkeley, CA, pp. 405413
Ding, M. & Cheng, X. (2009). Fault tolerant target tracking in sensor networks, Proceedings of
the tenth ACM international symposium on Mobile ad hoc networking and computing,
New Orleans, LA, USA
Dogandzic, A. & Zhang, B. (2006). Distributed estimation and detection for sensor networks
using hidden Markov random field models. IEEE Trans. Signal Process., vol. 54, no.
8, pp. 32003215
Forero, P.A.; Cano, A. & Giannakis, G.B. (2008). Consensus-based distributed expectation-
maximization algorithm for density estimation and classification using wireless
sensor networks, Proceedings of the IEEE Intl Conf. Acoustics, Speech and Signal
Processing, pp. 1989-1992
Gray, R. M. (2006). Quantization in task-driven sensing and distributed processing,
Proceedings Int. Conf. Acoustics, Speech, Signal Processing, Toulouse, France, vol. 5,
pp. V-1049V-1052
Gubner, J. (1993). Distributed estimation and quantization. IEEE Trans. Information Theory,
vol. 39, no. 5, pp.1456-1459
Guo, W.H.; Liu, Z.Y. & Wu, G.B. (2003). An energy-balanced transmission scheme for sensor
st
Intl. Conf. Embedded Networked Sensor Systems. Los
Angeles, CA, USA, pp. 300-301
Heinzelman, W. R.; Chandrakasan, A. & Balakrishnan, H. (2002). An application-specific
protocol architecture for wireless microsensor networks. IEEE Transactions on
Wireless Communications, vol. 1, no. 4, pp. 660670
Hoblos, G.; Staroswiecki, M. & Aitouche, A. (2000). Optimal design of fault tolerant sensor
networks, Proceedings of the IEEE Intl Conf. on Control Applications, pp. 467-472

individual targets. Combining the track association and tracking becomes more
complicated, especially in circumstance of low cost sensor network with limited
computation capacity and communication bandwidth (Li et al., 2010).
Another interesting issue for target tracking is the consideration of node failure. The sensor
nodes are usually deployed in harsh environments so various nodes may fail, may be
attacked or node energy may be depleted due to obstacles. Therefore, fault tolerant target
tracking algorithms and protocols must be designed for wireless sensor networks as the
fault tolerant approaches developed for traditional wired or wireless networks are not well
suited for WSN because of various differences between these networks (Ding & Cheng,
2009).
The cross-layered approach in WSN is more effective and energy efficient than in traditional
layered approach. While traditional layered approach endures more transfer overhead,
cross-layered approach minimizes these overhead by having data shared among layers
(Melodia et al., 2006; Kwon et al., 2006; Song & Hatzinakos, 2007). In the cross-layered
approach, the protocol stack is treated as a system and not individual layers, independent of
each other. Layers share information from the system. The development of various protocols
and services in a cross-layered approach is optimized and improved as a whole.
In last decades, the problem of decentralized information fusion has been discussed
extensively in the literature. However, the algorithms developed are free of energy and
communication constraints, see e.g. Sun & Deng, 2004; Li & Wang, 2000; Zhou & Li, 2008a;
Zhou & Li, 2008b. Novel fusion approaches include practical constraints in WSNs while
keeping high fusion performance must be investigated (Ruan et al., 2008). Moreover,
tracking with adaptive quantization thresholds and/or allocated bandwidth is another
promising research direction since the communicational condition dependent quantization
will definitely improve the estimation accuracy while using less communicational energy
(Zhou et al., 2011; Xu & Li, 2010).
Finally, WSNs have the potential to enhance and change the way people interact with
technology and the world (Aboelaze & Aloul, 2005). The direction of future WSNs also lies
in identifying real business and industry needs. Interactions between research and
development are necessary to bridge the gap between existing technology and the
development of business solutions. Applying sensor technology to different applications
will improve business processes as well as open up more problems for researchers.

8. Acknowledgements
The work was jointly supported by the National Natural Science Foundation of China
(Under Grant 60874104, 60935001); 973Project (2009CB824900, 2010CB734103); Shanghai Key
Basic Research Foundation (08JC1411800).

9. References
Aboelaze, M. & Aloul, F. (2005). Current and future trends in sensor networks: a survey,
Proceedings of the 14th IEEE Intl. Conf. on Wireless and Optical Communication, New
York, USA, pp. 133-138
Akyildiz, I.F.; Su, W. Sankarasubramaniam, Y. & Cayirci, E. (2002). Wireless Sensor
Network: A Survey. Computer Networks, vol. 38, no. 4, pp. 393422

Akyildiz, I.F.; Melodia, T. & Chowdury, K.R. (2007). Wireless multimedia sensor networks:
A survey. IEEE Wireless Communications, vol. 14, no. 6, pp. 32-39
Bertsekas, D. P. & Tsitsiklis, J. N. (1999). Parallel and Distributed Computation: Numerical
Methods, 2nd ed. Belmont, MA: Athena Scientific
Carli, R. Chiuso, A. Schenato, L. & Zampieri, S. (2006). Distributed Kalman filtering based
on consensus strategies. IEEE J. Selected Areas in Communications, vol. 26, no. 4, pp.
622-632
wireless sensor networks, Proceedings of 11th IEEE International Conf. Network
Protocols, Atlanta, Georgia, USA, pp. 284294
wireless sensor networks. IEEE Transactions on Mobile Computing, vol. 3, no. 3, pp.
258-273
Chong, C.Y. & Kumar, S. (2003). Sensor networks: Evolution, opportunities, and challenges.
Proc. IEEE, vol. 91, pp. 2741
Chong, C.Y.; Zhao, F. Mori, S. & Kumar, S. (2003). Distributed tracking in wireless Ad Hoc
sensor networks, Proceedings Sixth Intl. Conf. on Information Fusion, Cairns,
Australia, pp. 431438
Culler, D.; Estrin, D. & Srivastava, M. (2004). Overview of sensor networks. Computer, vol.
37, no. 8, pp. 4149
Delouille, V.; Neelamani, R. & Baraniuk, R. (2004). Robust distributed estimation in sensor
networks using the embedded polygons algorithm, Proceedings of the 3rd Int. Symp.
Info. Processing Sensor Networks, Berkeley, CA, pp. 405413
Ding, M. & Cheng, X. (2009). Fault tolerant target tracking in sensor networks, Proceedings of
the tenth ACM international symposium on Mobile ad hoc networking and computing,
New Orleans, LA, USA
Dogandzic, A. & Zhang, B. (2006). Distributed estimation and detection for sensor networks
using hidden Markov random field models. IEEE Trans. Signal Process., vol. 54, no.
8, pp. 32003215
Forero, P.A.; Cano, A. & Giannakis, G.B. (2008). Consensus-based distributed expectation-
maximization algorithm for density estimation and classification using wireless
sensor networks, Proceedings of the IEEE Intl Conf. Acoustics, Speech and Signal
Gray, R. M. (2006). Quantization in task-driven sensing and distributed processing,
Proceedings Int. Conf. Acoustics, Speech, Signal Processing, Toulouse, France, vol. 5,
pp. V-1049V-1052
Gubner, J. (1993). Distributed estimation and quantization. IEEE Trans. Information Theory,
vol. 39, no. 5, pp.1456-1459
Guo, W.H.; Liu, Z.Y. & Wu, G.B. (2003). An energy-balanced transmission scheme for sensor
st
Intl. Conf. Embedded Networked Sensor Systems. Los
Angeles, CA, USA, pp. 300-301
Heinzelman, W. R.; Chandrakasan, A. & Balakrishnan, H. (2002). An application-specific
protocol architecture for wireless microsensor networks. IEEE Transactions on
Wireless Communications, vol. 1, no. 4, pp. 660670
Hoblos, G.; Staroswiecki, M. & Aitouche, A. (2000). Optimal design of fault tolerant sensor
networks, Proceedings of the IEEE Intl Conf. on Control Applications, pp. 467-472

Jiang, C.; Dong, G. & Wang, B. (2005). Detection and tracking of region-based evolving
targets in sensor betworks, Proceedings of 14th Int. Conf. on Computer Communications
and Networks, ICCCN 2005, pp:563 - 568
Jin, G. & Nittel, S. (2006). NED: An efficient noise-tolerant event and event boundary
detection algorithm in wireless sensor networks, Proceedings of the 7th Int. Conf. on
Mobile Data Management
Jin, G.Y.; Lu, X.Y. & Park, M.S. (2006). Dynamic clustering for object tracking in wireless
sensor networks. Ubiquitous Computing Systems, pp.200-209
Kar, S. & Moura, J.M.F. (2009). Distributed consensus algorithms in sensor networks with
imperfect communication: link failures and channel noise. IEEE Trans. Signal
Processing, vol. 57, no. 1, pp. 355-369
Kung, H.T. & Vlah, D. (2003). Efficient location tracking using sensor networks, Proceedings
of the IEEE Wireless Communications and Networking Conference, New Orleans,
Louisiana, USA
Kwon, H.; Kim, T.H. Choi, S. & Lee, B.G. (2006). A cross-layer strategy for energy-efficient
reliable delivery in wireless sensor networks. IEEE Trans. on Wireless
Communications, vol. 5, no. 12, pp. 3689-3699
Lam, W. & Reibman, A. (1993). Quantizer design for decentralized systems with
communication constrains. IEEE Trans. Communications, vol. 41, no. 8, pp.1602-1605
Lee, S.M.; Cha, H. & Ha, R. (2007). Energy-aware location error handling for object tracking
applications. Wireless Sensor Networks, Computer Communication, vol. 30, pp. 1443-
1450
Li, X.; Shu, W. Li, M. Huang, H.Y. Luo P.E. & Wu, M.Y. (2009). Performance evaluation of
vehicle-based mobile sensor networks for traffic monitoring. IEEE Trans. Vehic.
Tech., vol. 58, no. 4, pp. 1647-1653
Li, X. R. & Wang, J. (2000). Unified optimal linear estimation fusion Part II: Discussions
and examples, Proceedings of the 3rd ISIF Conf. On Information Fusion, MoC2/18-
MoC2/25
Li, X.R.; Zhu, Y. & Han, C. (2000). Unified optimal linear estimation fusion, Proceedings of the
39th IEEE Conf. On Decision and Control, Sydney, Australia, 10-25
Li, Z.; Chen, S. Leung, H. & Bosse, E. (2010). Joint data association, registration, and fusion
using EM-KF. IEEE Trans. Aerospace and Electronic Systems, Vol. 46, no. 2, pp. 496-
507
Lin, C.; King, C. & Hsiao, H. (2005). Region abstraction for event tracking in wireless Sensor
networks, Proceedings of the 8th Int. Symposium on Parallel Architectures, Algorithms,
and Networks
Lin, C.Y.; Peng, W.C. & Tseng, Y.C. (2004). Efficient in-network moving object tracking in
wireless sensor networks. Department of Computer Science and Information
Engineering - National Chiao Tung University
Luo, Z. (2005). Universal decentralized estimation in a bandwidth constrained sensor
network. IEEE Trans. Information Theory, vol. 51, pp.2210-2219
Melodia, T.; Vuran, M.C. & Pompili, D. (2006). The state of art in cross-layer design for
wireless sensor networks. in Network Architect. In Next Generation Internet (M.
Cesana, L. Fratta, eds.) Berlin Heidelberg: Springer-Verlag

Msechu, E.J.; Roumeliotis, S.I. Ribeiro, A. & Giannakis, G.B. (2008). Decentralized quantized
Kalman filtering with scalable communication cost, IEEE Trans. Signal Process., vol.
56, no. 8, pp. 37273741
Mutambara, A.G.O. (1998). Decentralized estimation and control for multisensory systems, Boca
Raton, FL: CRC Press
Olfati-Saber, R. (2005). Distributed Kalman filter with embedded consensus filters,
Proceedings of the 44
th
IEEE Conf. on Decision and Control, and European Control Conf.,
pp. 8179-8184
Olfati-Saber, R. (2007). Distributed Kalman filtering for sensor networks. 46
th
IEEE Conf. on
Decision and Control, New Orleans, USA
Olfati-Saber, R. & Murry, R.M. (2004). Consensus problems in network of agents with
switching topology and time-delays. IEEE Trans. Automat. Control, vol. 49, no.9, pp.
101-115
Olule, E.; Wang, G. Guo, M. & Dong, M. (2007). RARE: An energy efficient target tracking
protocol for wireless sensor networks, Proceedings of the Intl. Conf. on Parallel
Ozdemir, O.; Niu, R. & Varshney, P.K. (2009). Tracking in wireless sensor networks using
particle filtering: physical layer considerations. IEEE Trans. Signal Processing, bol.
57, no. 5, pp. 1987-1999
Papadopoulos, H.; Wornell, G. & Oppenheim, A. (2001). Sequential signal encoding from
noisy measurements using quantizers with dynamic bias control. IEEE Trans. Inf.
Theory, vol. 47, no. 3, pp. 9781002
Phoha, S.; Javobson, N. & Friedlander, D. (2003a). Sensor network based localization and
target tracking hybridization in the operational domains of beamforming and
dynamic space-time clustering. Proceedings of the Global Telecommunications
Conference. Michigan, USA, pp. 2952- 2956
Phoha, S.; Jacobson, N. Friedlander, D. & Brooks, R. (2003b). Sensor network based
localization and tracking through hybridization in the operational domains of
beamforming and dynamic space-time clustering. Proceedings of the Conf. Global
Telecommunications. Michigan, USA, pp. 1137- 1140
Rapaka, A. & Madria, S. (2007). Two energy efficient algorithms for tracking objects in a
sensor network. Wireless Communication Mobile Computing, vol. 12, no. 7, pp.809-819
Ribeiro, A. & Giannakis, G. B. (2006). Bandwidth-constrained distributed estimation for
wireless sensor networks, Part II: Unknown PDF. IEEE Trans. Signal Process., vol.
54, no. 3, pp. 11311143
Ribeiro, A.; Giannakis, G. B. & Roumeliotis, S. I. (2006). SOI-KF: Distributed Kalman filtering
with low-cost communications using the sign of innovations. IEEE Trans. Signal
Process., vol. 54, no. 12, pp. 47824795
Ruan, Y.; Willett, P. Marrs, A. Palmieri, F. & Marano, S. (2008). Practical fusion of quantized
measurements via particle filtering. IEEE Trans. Aerosp. Electron. Syst., vol. 44, no. 1,
pp. 15-29
Scherber, D. & Papadopoulos, H. C. (2005). Distributed computation of averages over ad hoc
networks. IEEE J. Sel. Areas Commun., vol. 23, no. 4, pp. 776787
Schizas, I. D. & Giannakis, G. B. (2006). Consensus-based distributed estimation of random
signals with wireless sensor networks, Proceedings of the 40th Asilomar Conf. Signals,
Systems, Computers, Monterey, CA

Jiang, C.; Dong, G. & Wang, B. (2005). Detection and tracking of region-based evolving
targets in sensor betworks, Proceedings of 14th Int. Conf. on Computer Communications
and Networks, ICCCN 2005, pp:563 - 568
Jin, G. & Nittel, S. (2006). NED: An efficient noise-tolerant event and event boundary
detection algorithm in wireless sensor networks, Proceedings of the 7th Int. Conf. on
Mobile Data Management
Jin, G.Y.; Lu, X.Y. & Park, M.S. (2006). Dynamic clustering for object tracking in wireless
sensor networks. Ubiquitous Computing Systems, pp.200-209
Kar, S. & Moura, J.M.F. (2009). Distributed consensus algorithms in sensor networks with
imperfect communication: link failures and channel noise. IEEE Trans. Signal
Kung, H.T. & Vlah, D. (2003). Efficient location tracking using sensor networks, Proceedings
of the IEEE Wireless Communications and Networking Conference, New Orleans,
Louisiana, USA
Kwon, H.; Kim, T.H. Choi, S. & Lee, B.G. (2006). A cross-layer strategy for energy-efficient
reliable delivery in wireless sensor networks. IEEE Trans. on Wireless
Communications, vol. 5, no. 12, pp. 3689-3699
Lam, W. & Reibman, A. (1993). Quantizer design for decentralized systems with
communication constrains. IEEE Trans. Communications, vol. 41, no. 8, pp.1602-1605
Lee, S.M.; Cha, H. & Ha, R. (2007). Energy-aware location error handling for object tracking
applications. Wireless Sensor Networks, Computer Communication, vol. 30, pp. 1443-
1450
Li, X.; Shu, W. Li, M. Huang, H.Y. Luo P.E. & Wu, M.Y. (2009). Performance evaluation of
vehicle-based mobile sensor networks for traffic monitoring. IEEE Trans. Vehic.
Tech., vol. 58, no. 4, pp. 1647-1653
Li, X. R. & Wang, J. (2000). Unified optimal linear estimation fusion Part II: Discussions
and examples, Proceedings of the 3rd ISIF Conf. On Information Fusion, MoC2/18-
MoC2/25
Li, X.R.; Zhu, Y. & Han, C. (2000). Unified optimal linear estimation fusion, Proceedings of the
39th IEEE Conf. On Decision and Control, Sydney, Australia, 10-25
Li, Z.; Chen, S. Leung, H. & Bosse, E. (2010). Joint data association, registration, and fusion
using EM-KF. IEEE Trans. Aerospace and Electronic Systems, Vol. 46, no. 2, pp. 496-
507
Lin, C.; King, C. & Hsiao, H. (2005). Region abstraction for event tracking in wireless Sensor
networks, Proceedings of the 8th Int. Symposium on Parallel Architectures, Algorithms,
and Networks
Lin, C.Y.; Peng, W.C. & Tseng, Y.C. (2004). Efficient in-network moving object tracking in
wireless sensor networks. Department of Computer Science and Information
Engineering - National Chiao Tung University
Luo, Z. (2005). Universal decentralized estimation in a bandwidth constrained sensor
network. IEEE Trans. Information Theory, vol. 51, pp.2210-2219
Melodia, T.; Vuran, M.C. & Pompili, D. (2006). The state of art in cross-layer design for
wireless sensor networks. in Network Architect. In Next Generation Internet (M.
Cesana, L. Fratta, eds.) Berlin Heidelberg: Springer-Verlag

Msechu, E.J.; Roumeliotis, S.I. Ribeiro, A. & Giannakis, G.B. (2008). Decentralized quantized
Kalman filtering with scalable communication cost, IEEE Trans. Signal Process., vol.
56, no. 8, pp. 37273741
Mutambara, A.G.O. (1998). Decentralized estimation and control for multisensory systems, Boca
Raton, FL: CRC Press
Olfati-Saber, R. (2005). Distributed Kalman filter with embedded consensus filters,
th
IEEE Conf. on Decision and Control, and European Control Conf.,
pp. 8179-8184
Olfati-Saber, R. (2007). Distributed Kalman filtering for sensor networks. 46
th
IEEE Conf. on
Decision and Control, New Orleans, USA
Olfati-Saber, R. & Murry, R.M. (2004). Consensus problems in network of agents with
switching topology and time-delays. IEEE Trans. Automat. Control, vol. 49, no.9, pp.
101-115
Olule, E.; Wang, G. Guo, M. & Dong, M. (2007). RARE: An energy efficient target tracking
protocol for wireless sensor networks, Proceedings of the Intl. Conf. on Parallel
Ozdemir, O.; Niu, R. & Varshney, P.K. (2009). Tracking in wireless sensor networks using
particle filtering: physical layer considerations. IEEE Trans. Signal Processing, bol.
57, no. 5, pp. 1987-1999
Papadopoulos, H.; Wornell, G. & Oppenheim, A. (2001). Sequential signal encoding from
noisy measurements using quantizers with dynamic bias control. IEEE Trans. Inf.
Theory, vol. 47, no. 3, pp. 9781002
Phoha, S.; Javobson, N. & Friedlander, D. (2003a). Sensor network based localization and
target tracking hybridization in the operational domains of beamforming and
dynamic space-time clustering. Proceedings of the Global Telecommunications
Conference. Michigan, USA, pp. 2952- 2956
Phoha, S.; Jacobson, N. Friedlander, D. & Brooks, R. (2003b). Sensor network based
localization and tracking through hybridization in the operational domains of
beamforming and dynamic space-time clustering. Proceedings of the Conf. Global
Telecommunications. Michigan, USA, pp. 1137- 1140
Rapaka, A. & Madria, S. (2007). Two energy efficient algorithms for tracking objects in a
sensor network. Wireless Communication Mobile Computing, vol. 12, no. 7, pp.809-819
Ribeiro, A. & Giannakis, G. B. (2006). Bandwidth-constrained distributed estimation for
wireless sensor networks, Part II: Unknown PDF. IEEE Trans. Signal Process., vol.
54, no. 3, pp. 11311143
Ribeiro, A.; Giannakis, G. B. & Roumeliotis, S. I. (2006). SOI-KF: Distributed Kalman filtering
with low-cost communications using the sign of innovations. IEEE Trans. Signal
Process., vol. 54, no. 12, pp. 47824795
Ruan, Y.; Willett, P. Marrs, A. Palmieri, F. & Marano, S. (2008). Practical fusion of quantized
measurements via particle filtering. IEEE Trans. Aerosp. Electron. Syst., vol. 44, no. 1,
pp. 15-29
Scherber, D. & Papadopoulos, H. C. (2005). Distributed computation of averages over ad hoc
networks. IEEE J. Sel. Areas Commun., vol. 23, no. 4, pp. 776787
Schizas, I. D. & Giannakis, G. B. (2006). Consensus-based distributed estimation of random
signals with wireless sensor networks, Proceedings of the 40th Asilomar Conf. Signals,
Systems, Computers, Monterey, CA

Schizas, I.D.; Mateos, G. & Giannakis, G.B. (2009). Distributed LMS for consensus-based in-
network adaptive processing. IEEE Trans. Signal Process., vol. 57, no. 6, pp. 2365
2382
Shorey, R.; Ananda, A. L. Chan, M. C. & Ooi, W. T. (2006). Mobile, Wireless, and Sensor
Networks: Technology, Applications, and Future Directions, IEEE Press, Piscataway, pp.
173-196
Schizas, I.D.; Ribeiro, A. & Giannakis, G.B. (2008a). Consensus in Ad Hoc WSNs with noisy
linkspart I: distributed estimation of deterministic signals. IEEE Trans. Signal
Schizas, I.D.; Ribeiro, A. & Giannakis, G.B. (2008b). Consensus in Ad Hoc WSNs with noisy
linkspart II: distributed estimation and smoothing of random signals. IEEE Trans.
Signal Processing, vol. 56, no. 4, pp. 1650-1666
Sohraby, K.; Minoli, D. Znati, T. (2006). Wireless sensor networks: Technology, protocols, and
applications. Wiley-interscience, New Jersey
Song, L. & Hatzinakos, D. (2007). A cross-layer architecture of wireless sensor networks for
target tracking. IEEE/ACM Transactions on networking, vol. 15, no. 1, pp. 145-158
Spanos, D. P.; Olfati-Saber, R. & Murray, R. J. (2005). Distributed sensor fusion using
dynamic consensus, Proceedings of the 16th IFAC World Congr., Prague, Czech
Republic
Speyer, J.L. (2004). Computation and transmission requirements for a decentralized linear-
quadratic-Gaussian control problem. IEEE Trans. Automatic Control, vol. 49, no. 9,
pp.1453-1464
Sukhavasi, R.T. & Hassibi, B. (2009). Particle filtering for quantized innovations, Proceedings
of the IEEE Int. Conf. Acout., Speech and Signal Process., Taiwan, China
Sun, S.L. & Deng, Z.L. (2004). Multi-sensor optimal information fusion Kalman filter.
Automatica, vol. 40, no. 5, pp. 1017-1023
Tanner, H.G.; Jadbabaie, A. & Pappas, G.J. (2007). Flocking in fixed and switching networks.
IEEE Trans. Automat. Control, vol. 52, no. 5, pp. 863-868
Tsai, H.W.; Chu, C.P. & Chen, T.S. (2007). Mobile object tracking in wireless sensor
networks. Computer Communications, vol. 30, pp. 18111825
Valera M. & Velastin, S.A. (2005). Intelligent distributed surveillance systems: a review. IEE
Proc. Visual Image Signal Process., vol. 152, no. 2, pp. 192-204
Veeravalli V.V. & Chamberland, J.F. (2007). Detection in sensor networks. in Swami, A.
Zhao, Q. Hong, Y.W. & Tong, L. Eds, Wireless Sensor Networks: Signal Processing And
Communications Perspectives, pp. 119-148, John Wiley & Sons
Vercauteren, T. & Wang, X. (2005). Decentralized sigma-point information filters for target
tracking in collaborative sensor networks. IEEE T. Signal Processing, vol. 53, no. 8,
pp. 2997-3009
Walchli, M.; Skoczylas, P. Meer M. & Braun, T. (2007). Distributed event localization and
tracking with wireless sensors, Proceedings of the 5
th
Intl. Conf. on Wired/Wireless
internet Communications
Wang, Q.X.; Chen, W.P. Zheng, R. Lee, K. & Sha, L. (2003). Acoustic target tracking using
tiny wireless sensor devices, Proceedings of the Int. Workshop on Information Processing
in Sensor Networks

Wang, Z.; Li, H. Shen, X. Sun, X. & Wang, Z. (2008). Tracking and predicting moving targets
in hierarchical sensor networks, Proceedings of the IEEE Intl. Conf. on Networking,
Sensing and Control, pp. 1169-1173
Xiao, J.J.; Cui, S. Luo, Z. Q. & Goldsmith, A.J. (2006). Power scheduling of universal
decentralized estimation in sensor networks. IEEE Trans. Signal Process., vol. 54, no.
2 , pp. 413-421
Xiao, L. & Boyd, S. (2004). Fast linear iterations for distributed averaging. Syst. Control Lett.,
vol. 53, pp. 6578
Xu, J. & Li, J. (2010). State estimation with quantized sensor information in wireless sensor
networks. IET Signal Processing, (peer reviewed, to appear).
Xu Y. & Lee, W.C. (2003). On localized prediction for power efficient object tracking in
sensor networks, Proceedings of the 1st Intl. Workshop on Mobile Distributed computing,
Providence RI, pp. 434439
Yang, H. & Sikdor, B. (2003). A protocol for tracking mobile targets using sensor network,
Proceedings of the First IEEE International Workshop on Sensor Network Protocols and
Applications, Anchorage, Alaska, pp. 7181
Yick, J.; Mukherjee, B. & Ghosal, D. (2008). Wireless sensor network survey. Computer
Networks, vol. 52, no. 12, pp. 2292-2330
You, K.; Xie, L. Sun, S. & Xiao, W. (2008). Multiple-level quantized innovation Kalman filter,
th
IFAC World Congress, Seoul, Korea
Younis, O. & Fahmy, S. (2004). Distributed clustering in ad-hoc sensor networks: a hybrid,
energy-efficient approach, Proceedings of the 23
rd
Annual Joint Conf. IEEE Computer &
Communications Societies, INFOCOM
Zhang, W. & Cao, G. (2004a). Optimizing tree reconfiguration for mobile target tracking in
sensor networks. IEEE INFOCOM, HongKong, China
Zhang, W. & Cao, G. (2004b). DCTC: Dynamic convoy tree-based collaboration for target
tracking in sensor networks. IEEE Transactions on Wireless Communications, vol. 3,
no. 5, pp. 1689-1701
Zhao, F.; Shin, J. & Reich, J. (2002). Information-driven dynamic sensor collaboration for
tracking applications. IEEE Signal Proces. Mag., vol.19, no. 2, pp. 61-72
Zhou, Y. & Li, J. (2008a). Data fusion of unknown correlations using internal ellipsoidal
approximation, Proceedings 17
th
International Foundation Automatic Control
World Congress, Seoul, Korea
Zhou, Y. & Li, J. (2008b). Robust decentralized data fusion based on internal ellipsoid
approximation, Proceedings of the 17
th
Zhou, Y. & Li, J. (2009a). Average consensus based scalable robust filtering for sensor
network, Proceedings of the 5th Intl. Conf. Wireless Communications, Networking and
Mobile Computing, Beijing, China
Zhou, Y. & Li, J. (2009b). Scalable distributed sigma-point Kalman filtering for sensor
networks: dynamic consensus approach, Proceedings of the IEEE Intl. Conf. on
Systems, Man, and Cybernetics, San Antonio, Taxas, USA
Zhou, Y. & Li, J. (2009c). Quantized measurement fusion for target tracking in wireless
sensor networks, Proceedings of the 48th IEEE Conf. on Decision and Control and 28th
Chinese Control Conference, pp. 6835-6840, Shanghai, China

Schizas, I.D.; Mateos, G. & Giannakis, G.B. (2009). Distributed LMS for consensus-based in-
network adaptive processing. IEEE Trans. Signal Process., vol. 57, no. 6, pp. 2365
2382
Shorey, R.; Ananda, A. L. Chan, M. C. & Ooi, W. T. (2006). Mobile, Wireless, and Sensor
Networks: Technology, Applications, and Future Directions, IEEE Press, Piscataway, pp.
173-196
Schizas, I.D.; Ribeiro, A. & Giannakis, G.B. (2008a). Consensus in Ad Hoc WSNs with noisy
linkspart I: distributed estimation of deterministic signals. IEEE Trans. Signal
Schizas, I.D.; Ribeiro, A. & Giannakis, G.B. (2008b). Consensus in Ad Hoc WSNs with noisy
linkspart II: distributed estimation and smoothing of random signals. IEEE Trans.
Signal Processing, vol. 56, no. 4, pp. 1650-1666
Sohraby, K.; Minoli, D. Znati, T. (2006). Wireless sensor networks: Technology, protocols, and
applications. Wiley-interscience, New Jersey
Song, L. & Hatzinakos, D. (2007). A cross-layer architecture of wireless sensor networks for
target tracking. IEEE/ACM Transactions on networking, vol. 15, no. 1, pp. 145-158
Spanos, D. P.; Olfati-Saber, R. & Murray, R. J. (2005). Distributed sensor fusion using
dynamic consensus, Proceedings of the 16th IFAC World Congr., Prague, Czech
Republic
Speyer, J.L. (2004). Computation and transmission requirements for a decentralized linear-
quadratic-Gaussian control problem. IEEE Trans. Automatic Control, vol. 49, no. 9,
pp.1453-1464
Sukhavasi, R.T. & Hassibi, B. (2009). Particle filtering for quantized innovations, Proceedings
of the IEEE Int. Conf. Acout., Speech and Signal Process., Taiwan, China
Sun, S.L. & Deng, Z.L. (2004). Multi-sensor optimal information fusion Kalman filter.
Automatica, vol. 40, no. 5, pp. 1017-1023
Tanner, H.G.; Jadbabaie, A. & Pappas, G.J. (2007). Flocking in fixed and switching networks.
IEEE Trans. Automat. Control, vol. 52, no. 5, pp. 863-868
Tsai, H.W.; Chu, C.P. & Chen, T.S. (2007). Mobile object tracking in wireless sensor
networks. Computer Communications, vol. 30, pp. 18111825
Valera M. & Velastin, S.A. (2005). Intelligent distributed surveillance systems: a review. IEE
Proc. Visual Image Signal Process., vol. 152, no. 2, pp. 192-204
Veeravalli V.V. & Chamberland, J.F. (2007). Detection in sensor networks. in Swami, A.
Zhao, Q. Hong, Y.W. & Tong, L. Eds, Wireless Sensor Networks: Signal Processing And
Communications Perspectives, pp. 119-148, John Wiley & Sons
Vercauteren, T. & Wang, X. (2005). Decentralized sigma-point information filters for target
tracking in collaborative sensor networks. IEEE T. Signal Processing, vol. 53, no. 8,
pp. 2997-3009
Walchli, M.; Skoczylas, P. Meer M. & Braun, T. (2007). Distributed event localization and
tracking with wireless sensors, Proceedings of the 5
th
Intl. Conf. on Wired/Wireless
internet Communications
Wang, Q.X.; Chen, W.P. Zheng, R. Lee, K. & Sha, L. (2003). Acoustic target tracking using
tiny wireless sensor devices, Proceedings of the Int. Workshop on Information Processing
in Sensor Networks

Wang, Z.; Li, H. Shen, X. Sun, X. & Wang, Z. (2008). Tracking and predicting moving targets
in hierarchical sensor networks, Proceedings of the IEEE Intl. Conf. on Networking,
Sensing and Control, pp. 1169-1173
Xiao, J.J.; Cui, S. Luo, Z. Q. & Goldsmith, A.J. (2006). Power scheduling of universal
decentralized estimation in sensor networks. IEEE Trans. Signal Process., vol. 54, no.
2 , pp. 413-421
Xiao, L. & Boyd, S. (2004). Fast linear iterations for distributed averaging. Syst. Control Lett.,
vol. 53, pp. 6578
Xu, J. & Li, J. (2010). State estimation with quantized sensor information in wireless sensor
networks. IET Signal Processing, (peer reviewed, to appear).
Xu Y. & Lee, W.C. (2003). On localized prediction for power efficient object tracking in
sensor networks, Proceedings of the 1st Intl. Workshop on Mobile Distributed computing,
Providence RI, pp. 434439
Yang, H. & Sikdor, B. (2003). A protocol for tracking mobile targets using sensor network,
Proceedings of the First IEEE International Workshop on Sensor Network Protocols and
Applications, Anchorage, Alaska, pp. 7181
Yick, J.; Mukherjee, B. & Ghosal, D. (2008). Wireless sensor network survey. Computer
Networks, vol. 52, no. 12, pp. 2292-2330
You, K.; Xie, L. Sun, S. & Xiao, W. (2008). Multiple-level quantized innovation Kalman filter,
th
IFAC World Congress, Seoul, Korea
Younis, O. & Fahmy, S. (2004). Distributed clustering in ad-hoc sensor networks: a hybrid,
energy-efficient approach, Proceedings of the 23
rd
Annual Joint Conf. IEEE Computer &
Communications Societies, INFOCOM
Zhang, W. & Cao, G. (2004a). Optimizing tree reconfiguration for mobile target tracking in
sensor networks. IEEE INFOCOM, HongKong, China
Zhang, W. & Cao, G. (2004b). DCTC: Dynamic convoy tree-based collaboration for target
tracking in sensor networks. IEEE Transactions on Wireless Communications, vol. 3,
no. 5, pp. 1689-1701
Zhao, F.; Shin, J. & Reich, J. (2002). Information-driven dynamic sensor collaboration for
tracking applications. IEEE Signal Proces. Mag., vol.19, no. 2, pp. 61-72
Zhou, Y. & Li, J. (2008a). Data fusion of unknown correlations using internal ellipsoidal
approximation, Proceedings 17
th
Zhou, Y. & Li, J. (2008b). Robust decentralized data fusion based on internal ellipsoid
approximation, Proceedings of the 17
th
Zhou, Y. & Li, J. (2009a). Average consensus based scalable robust filtering for sensor
network, Proceedings of the 5th Intl. Conf. Wireless Communications, Networking and
Mobile Computing, Beijing, China
Zhou, Y. & Li, J. (2009b). Scalable distributed sigma-point Kalman filtering for sensor
networks: dynamic consensus approach, Proceedings of the IEEE Intl. Conf. on
Zhou, Y. & Li, J. (2009c). Quantized measurement fusion for target tracking in wireless
sensor networks, Proceedings of the 48th IEEE Conf. on Decision and Control and 28th
Chinese Control Conference, pp. 6835-6840, Shanghai, China

Zhou, Y. & Li, J. (2009d). Collaborative maneuvering target tracking in wireless sensor
network with quantized measurements, Proceedings of the IEEE Intl. Conf. on
Zhou, Y. Li, J. & Wang, D. (2009a). Weighted average approach to quantized measurement
fusion in wireless sensor network, Proceedings of the 5th Intl. Conf. Wireless
Communications, Networking and Mobile Computing, Beijing, China
Zhou, Y.; Li, J. & Wang, D. (2009b). Quantized measurement fusion in wireless sensor
network with correlated sensor noises, Proceedings of the 7
th
IEEE Intl. Conf. on
Control & Automation, pp. 1868-1873, Christchurch, New Zealand
Zhou, Y.; Li, J. & Wang, D. (2009c). Collaborative Target Tracking in Wireless Sensor
Networks using Quantized Innovations and Sigma-Point Kalman Filtering,
Proceedings of the IEEE Intl. Symp. Industrial Electronics, Seoul, Korea
Zhou, Y.; Li, J. & Wang, D. (2009d). Unscented Kalman filtering based quantized innovation
fusion for target tracking in WSN with feedback, Proceedings of the Intl. Conf.
Machine Learning and Cybernetics, Baoding, China
Zhou, Y.; Li, J. & Wang, D. (2010). Posterior Cramr-Rao lower bounds for target tracking in
sensor networks with quantized range-only measurements. IEEE Signal Processing
Letters , vol. 17, no. 2, pp. 157-160
Zhou, Y.; Li, J. & Wang, D. (2011). Target tracking in wireless sensor networks using
adaptive measurement quantization. Science In China Series F: Information Sciences
(peer reviewed, to appear).
Zou, Y. & Chakrabarty, K. (2003). Target localization based on energy considerations in
distributed sensor networks, Proceedings of the 1st IEEE Intl. Workshop on Sensor
Network Protocols and Applications, Anchorage, Alaska, pp.5158

A Gaussian Mixture Model-based Event-Driven
Continuous Boundary Detection in 3D Wireless Sensor Networks 393
A Gaussian Mixture Model-based Event-Driven Continuous Boundary
Detection in 3D Wireless Sensor Networks
Jiehui Chen, Mariam B.Salim and Mitsuji Matsumoto
X

Continuous Boundary Detection in 3D Wireless
Sensor Networks

Jiehui Chen
1,2
, Mariam B.Salim
2
and Mitsuji Matsumoto
2

1
Global COE Program International Research and Education Center for Ambient SoC
sponsored by MEXT,Japan
2
Graduate School of Global Information and Telecommunication Studies
Waseda University, Tokyo, Japan

1. Introduction
Wireless sensor networks (WSNs) may consist of tiny, energy efficient sensor nodes
communicating via wireless channels, performing distributed sensing and collaborative
tasks for a variety of monitoring applications. One of the critical problems in sensor
applications is detecting boundary sensors in a complex sensor network environment where
sensed data is often required to be associated with spatial coordinates. In (Zhong, et al, 2007)
a COBOM protocol that monitors the boundary of a continuous object was proposed. Sensor
nodes are assigned with a Boundary sensor Node (BN) array to store BN information. The
boundary monitoring is based on the changes to the observations in the BN array. As a
updated version, (Kim,J.H. et al,2008) presented the DEMOCO protocol that enhanced
COBOM by considering sensor nodes on one side of the boundary line called the IN
range, and ignoring those on the other side of the boundary line called the OUT range
which theoretically reduces approximately by half of the number of the selected BNs. Others
like (Basu, et al, 2006; Eren,T. et al, 2004; He,T. et al,2003; Nissanka,B. et al, 2003) also
involve two-dimensional (2D) sensor localizations. To address the issues of adaptive sensor
coverage and tracking for dynamic network topology, the authors of (Guo, et al, 2008)
utilized a Gaussian mixture model to characterize the mixture distribution of object
locations and proposed a novel methodology to adaptively update sensor node placement
according to the ML estimates of mass object locations with a distributed implementation of
an EM algorithm to reduce communication costs. Moreover, (Olfati-Saber, et al, 2007)
discussed a flocking-base mobility model for Distributed Kalman Filtering (DKF) in mobile
sensor networks and (Funke, et al, 2006; Funke, et al, 2007) demonstrated efficient boundary
detection algorithms with only the connectivity information.
In fact, the boundary detection problem has been mostly considered for 2D sensor networks
and the case of 3D sensor networks has gone practically unnoticed. Despite the fact that
difference between the normal 2D and the more realistic 3D scenario is only one extra
dimension, network topology could be much more complex and the location scheme has to
20
be more robust towards network irregularities. Taking a step further to expand from 2D to
3D sensor applications, several neighborhood-measurement (Peng, et al, 2006) based 3D
range-free boundary detection models (Lance, et al, 2001; Hu, et al, 2004; Ji, et al, 2004; Yi, et
al, 2003; Yi, et al, 2004, Andreas, et al, 2002) have been proposed. However, their tight
dependence on sensor node densities and availability of sufficient neighbors are too
optimistic for real 3D sensor applications due to their non-uniform sensor node densities
and topology randomization. On the other hand, a range-based model such as in (Zhang, et
al, 2006) does not make any assumption about sensor node densities and network topology.
Instead it introduced a strong entity called mobile location assistants (LAs) that enables each
location-unaware sensor node to easily estimate its own position using the measurable
AOAs (Peng, et al, 2006) and RSS (Samitha, et al, 2010). Similar approaches like (He, et al,
2003; Nissanka, et al, 2003; Bulusu, et al, 2001) assume that a small fraction of sensor nodes
called anchors or beacons have a priori knowledge of their location and (Liu&Wang, et al,
2009) proposed a range-based positioning method using beacon signals, that doesnt require
time synchronization since the beacon sensor nodes estimate the range based on frequency
differences instead of time differences. To conclude, all the aforementioned approaches
either introduced strong entities or made irrational assumptions. Furthermore, (Liu&Manli,
et al, 2009) presented a new high precision WSN positioning method with reasonable
implementation cost for a 3D case. Reference sensor nodes with known locations transmit
linear frequency modulation continuous waves (FMCWs), while other sensor nodes
estimate the range difference to them based on the received signals' frequency difference,
called time frequency difference arrival (TFDA).
Motivated by all above observations, instead of introducing miraculous assisting entities,
our range-free Gaussian Mixture Model (GMM)-based approach performs a connectivity
information-based segmentation algorithm (Zhu, et al, 2009) that partitions an irregular
sensor field into nicely shaped pieces, associated with an enhanced BN Array and efficient
distributed in-network information extraction virtual Thick Section Model (TSM); to the best
of our knowledge, this is the first work that presents a principled algorithmic approach
integrating computational geometry constructs adopted simultaneously for boundary
detection in both 2D and 3D network areas. It is promising that our new statistical Gaussian
mixture model (Mclachlan, et al, 2000)-based method in this paper is capable of fusing
multivariate real-valued sensor inputs to detect boundaries of events in a mathematically
principled manner. More precisely, the distribution of sensor readings within each sensor
nodes spatial neighborhood is mathematically formulated using most popular finite
GMMs. The model selection techniques (Figueiredo, et al, 2002;Akaike, et al, 1973; Schwarz,
et al, 1978; Solla, et al, 2000) can then effectively identify the correct number of modes for
finite mixture models. Therefore, Boundary and Non-Boundary sensor nodes can be
consequently distinguished from their neighboring sensor node data distributions.
The remainder of this paper is organized as follows: the next section details enhancement to
the BN Array concept; Section 3 simply describes general problems in boundary detection;
Section 4 presents the proposed robust Boundary Detection scheme for 3D (BD3D) sensor
networks in detail; Section 5 proves BD3D by simulation results; Finally, Section 6 concludes
the paper with future work.

2. Enhancement to BN Array
In (Chintalapudi, et al, 2003) three different schemes which can only take inputs of the 0/1
decision predicates from neighboring sensor nodes are proposed. (Jin, et al,2006) presents a
noise-tolerant algorithm named NED for event and event boundary detection. In NED, the
moving mean of the readings of the neighboring sensor node set is used as the estimate for a
certain sensor node. The authors of (Min, et al, 2005) propose Median-based approaches for
outlying classification and event frontline detection, where the median is a useful and
robust estimator which works directly with continuous numbers, rather than binary 0/1
readings. An extra description of the BN-Array of COBOM (Zhong, et al, 2007) and
DECOMO (Kim, et al, 2008) is given in this section. Suppose we have a sensor node v (N
)
and its neighbors (N
) = N
(k is the potential number of neighbors) (k = 6 in Figure 1).

Let us consider the BN array in (Kim, et al, 2008; Zhong, et al, 2007):

Table 1. BN Array of N
[Note: 0 and 1 are sensor readings (sample)].

Fig. 1. Readings of neighbors in BN Array of N
.

In Figure 1, the sensor readings of (N
) only indicate the relative locations of its neighbors

only. Correspondingly, there is no own sensor reading, as a result, N
judges itself by
inquiring (N
) in a time and energy consuming way. In our model, we applied a head with
1 byte more space for the BN Array to store its own sensor reading as well (see Table 2) for
self-judgment as a EBN or non-EBN. Here, we denote a BN inside object as Event BN (EBN),
and a BN outside object as non-EBN. That is very important for monitoring applications in
the sensor network because an Event sensor Node (EN) is usually highly responsible for
sending and receiving the aggregated data should be constantly aware of own status.

Table 2. BD3D BN Array of N
.

Figure 2(a,b) show the expected boundary lines in COBOM and DEMOCO, respectively.
Despite the fact that the shape of the expected boundary line in the 2D model of BD3D (see
be more robust towards network irregularities. Taking a step further to expand from 2D to
3D sensor applications, several neighborhood-measurement (Peng, et al, 2006) based 3D
range-free boundary detection models (Lance, et al, 2001; Hu, et al, 2004; Ji, et al, 2004; Yi, et
al, 2003; Yi, et al, 2004, Andreas, et al, 2002) have been proposed. However, their tight
dependence on sensor node densities and availability of sufficient neighbors are too
optimistic for real 3D sensor applications due to their non-uniform sensor node densities
and topology randomization. On the other hand, a range-based model such as in (Zhang, et
al, 2006) does not make any assumption about sensor node densities and network topology.
Instead it introduced a strong entity called mobile location assistants (LAs) that enables each
location-unaware sensor node to easily estimate its own position using the measurable
AOAs (Peng, et al, 2006) and RSS (Samitha, et al, 2010). Similar approaches like (He, et al,
2003; Nissanka, et al, 2003; Bulusu, et al, 2001) assume that a small fraction of sensor nodes
called anchors or beacons have a priori knowledge of their location and (Liu&Wang, et al,
2009) proposed a range-based positioning method using beacon signals, that doesnt require
time synchronization since the beacon sensor nodes estimate the range based on frequency
differences instead of time differences. To conclude, all the aforementioned approaches
either introduced strong entities or made irrational assumptions. Furthermore, (Liu&Manli,
et al, 2009) presented a new high precision WSN positioning method with reasonable
implementation cost for a 3D case. Reference sensor nodes with known locations transmit
linear frequency modulation continuous waves (FMCWs), while other sensor nodes
estimate the range difference to them based on the received signals' frequency difference,
called time frequency difference arrival (TFDA).
Motivated by all above observations, instead of introducing miraculous assisting entities,
our range-free Gaussian Mixture Model (GMM)-based approach performs a connectivity
information-based segmentation algorithm (Zhu, et al, 2009) that partitions an irregular
sensor field into nicely shaped pieces, associated with an enhanced BN Array and efficient
distributed in-network information extraction virtual Thick Section Model (TSM); to the best
of our knowledge, this is the first work that presents a principled algorithmic approach
integrating computational geometry constructs adopted simultaneously for boundary
detection in both 2D and 3D network areas. It is promising that our new statistical Gaussian
mixture model (Mclachlan, et al, 2000)-based method in this paper is capable of fusing
multivariate real-valued sensor inputs to detect boundaries of events in a mathematically
principled manner. More precisely, the distribution of sensor readings within each sensor
nodes spatial neighborhood is mathematically formulated using most popular finite
GMMs. The model selection techniques (Figueiredo, et al, 2002;Akaike, et al, 1973; Schwarz,
et al, 1978; Solla, et al, 2000) can then effectively identify the correct number of modes for
finite mixture models. Therefore, Boundary and Non-Boundary sensor nodes can be
consequently distinguished from their neighboring sensor node data distributions.
The remainder of this paper is organized as follows: the next section details enhancement to
the BN Array concept; Section 3 simply describes general problems in boundary detection;
Section 4 presents the proposed robust Boundary Detection scheme for 3D (BD3D) sensor
networks in detail; Section 5 proves BD3D by simulation results; Finally, Section 6 concludes
the paper with future work.

2. Enhancement to BN Array
In (Chintalapudi, et al, 2003) three different schemes which can only take inputs of the 0/1
decision predicates from neighboring sensor nodes are proposed. (Jin, et al,2006) presents a
noise-tolerant algorithm named NED for event and event boundary detection. In NED, the
moving mean of the readings of the neighboring sensor node set is used as the estimate for a
certain sensor node. The authors of (Min, et al, 2005) propose Median-based approaches for
outlying classification and event frontline detection, where the median is a useful and
robust estimator which works directly with continuous numbers, rather than binary 0/1
readings. An extra description of the BN-Array of COBOM (Zhong, et al, 2007) and
DECOMO (Kim, et al, 2008) is given in this section. Suppose we have a sensor node v (N
)
and its neighbors (N
) = N
(k is the potential number of neighbors) (k = 6 in Figure 1).

Let us consider the BN array in (Kim, et al, 2008; Zhong, et al, 2007):

Table 1. BN Array of N
[Note: 0 and 1 are sensor readings (sample)].

Fig. 1. Readings of neighbors in BN Array of N
.

In Figure 1, the sensor readings of (N
) only indicate the relative locations of its neighbors

only. Correspondingly, there is no own sensor reading, as a result, N
judges itself by
inquiring (N
) in a time and energy consuming way. In our model, we applied a head with
1 byte more space for the BN Array to store its own sensor reading as well (see Table 2) for
self-judgment as a EBN or non-EBN. Here, we denote a BN inside object as Event BN (EBN),
and a BN outside object as non-EBN. That is very important for monitoring applications in
the sensor network because an Event sensor Node (EN) is usually highly responsible for
sending and receiving the aggregated data should be constantly aware of own status.

.

Figure 2(a,b) show the expected boundary lines in COBOM and DEMOCO, respectively.
Despite the fact that the shape of the expected boundary line in the 2D model of BD3D (see
Figure 3) is similar to that of DEMOCO, the knowledge about Boundary sensor Nodes (BNs)
promises to be different because we can clearly distinguish EBN and non-EBN as well.

Fig. 2. Expected boundary lines (Kim, et al, 2008).

3. Problem
We first present the problems before outlining how our proposal can benefit dynamic
boundary detection for 2D and 3D sensor networks in the coming sections. To generally
analyze the existing problems for superior boundary detection in a 3D impediment scenario,
sensor nodes in the network usually have slight mobility which makes it difficult to
establish their locations. Figure 3 illustrates the possible boundary line changes in a 2D
scenario when the object is shrunk or expanded which becomes a big problem that involves
frequent inquiries among BNs and massive modifications to BN arrays.

Fig. 3. Possible boundary line changes when the object shrunk or expanded.
4. Boundary Detection for a 3D Sensor Network (BD3D)
This section involves the main objective of achieving a flexible and energy-efficient 3D
continuous boundary detection with a clear knowledge of EBN and non-EBN. Assume that
sensor nodes are randomly deployed over 3-dimensional terrain. Each sensor node has
limited resources (CPU, battery, etc), and is equipped with an omni-directional antenna. For
the radio model, E
is for running the transmitter or receiver circuitry and
is for the
transmit amplifier. To transmit a -bit message a distance l using this radio model, the radio
expends (E
), to receive the message, the radio expands (E
)
(Heinzelman, et al, 2000). This energy model assumes a continuous energy consumption
function. Moreover, we currently assume that sensor node failures are primarily caused by
energy depletion. Note that in our model, no assumptions are made about (1) homogeneity
of sensor node distribution; (2) network and BN density; (3) proximity of querying
observers and sensor node synchronization.
Our major contribution could be creating a statistical property of the finite mixture model,
especially the Gaussian mixture model (GMM) and adopting it to distributed sensing
scenarios. Suppose that we have a set of data observations
= {
,,
}, n N (N is the
total number of sensor nodes in the network) with each
representing a D-dimensional
random vector. Assume that
follows a k-component finite mixture distribution

(Mclachlan, et al,2000) as follows:

(
|) =
), j = 1,2,,k; I = 1,2,,n. (1)

subject to
=1
where
is the mixing weight or sometimes called the prior weight and
is the set of
parameters of the jth mixture component (
|). Denote = {
,,
}. The
objective function of estimating from
is to maximize the log-likelihood criterion as

follows:
Log
|
N
= log
(2)

Therefore, the maximum likelihood estimator of is:

ML
=arg max
{log
} (3)

Obviously,
ML
cannot be computed analytically from the above equation. Instead, GMM is
applied as its general solver to iteratively find the maximum likelihood solution of
ML
.
GMM is the most important class of finite mixture densities. GMM is formulated by using a
Gaussian density (
) with its mean vector
and covariance matrix
to replace the
general probability density function (
) in the finite mixture model:

(
|) =
) (4)

where a multi-dimensional multivariate Gaussian distribution is defined as:

|, =
||
(5)
Figure 3) is similar to that of DEMOCO, the knowledge about Boundary sensor Nodes (BNs)
promises to be different because we can clearly distinguish EBN and non-EBN as well.

Fig. 2. Expected boundary lines (Kim, et al, 2008).

3. Problem
We first present the problems before outlining how our proposal can benefit dynamic
boundary detection for 2D and 3D sensor networks in the coming sections. To generally
analyze the existing problems for superior boundary detection in a 3D impediment scenario,
sensor nodes in the network usually have slight mobility which makes it difficult to
establish their locations. Figure 3 illustrates the possible boundary line changes in a 2D
scenario when the object is shrunk or expanded which becomes a big problem that involves
frequent inquiries among BNs and massive modifications to BN arrays.

Fig. 3. Possible boundary line changes when the object shrunk or expanded.
4. Boundary Detection for a 3D Sensor Network (BD3D)
This section involves the main objective of achieving a flexible and energy-efficient 3D
continuous boundary detection with a clear knowledge of EBN and non-EBN. Assume that
sensor nodes are randomly deployed over 3-dimensional terrain. Each sensor node has
limited resources (CPU, battery, etc), and is equipped with an omni-directional antenna. For
the radio model, E
is for running the transmitter or receiver circuitry and
is for the
transmit amplifier. To transmit a -bit message a distance l using this radio model, the radio
expends (E
), to receive the message, the radio expands (E
)
(Heinzelman, et al, 2000). This energy model assumes a continuous energy consumption
function. Moreover, we currently assume that sensor node failures are primarily caused by
energy depletion. Note that in our model, no assumptions are made about (1) homogeneity
of sensor node distribution; (2) network and BN density; (3) proximity of querying
observers and sensor node synchronization.
Our major contribution could be creating a statistical property of the finite mixture model,
especially the Gaussian mixture model (GMM) and adopting it to distributed sensing
scenarios. Suppose that we have a set of data observations
= {
,,
}, n N (N is the
total number of sensor nodes in the network) with each
representing a D-dimensional
random vector. Assume that
follows a k-component finite mixture distribution

(Mclachlan, et al,2000) as follows:

(
|) =
), j = 1,2,,k; I = 1,2,,n. (1)

subject to
=1
where
is the mixing weight or sometimes called the prior weight and
is the set of
parameters of the jth mixture component (
|). Denote = {
,,
}. The
objective function of estimating from
is to maximize the log-likelihood criterion as

follows:
Log
|
N
= log
(2)

Therefore, the maximum likelihood estimator of is:

ML
=arg max
{log
} (3)

Obviously,
ML
cannot be computed analytically from the above equation. Instead, GMM is
applied as its general solver to iteratively find the maximum likelihood solution of
ML
.
GMM is the most important class of finite mixture densities. GMM is formulated by using a
Gaussian density (
) with its mean vector
and covariance matrix
to replace the
general probability density function (
) in the finite mixture model:

(
|) =
) (4)

where a multi-dimensional multivariate Gaussian distribution is defined as:

|, =
||
(5)
The Bayesian Information Criterion (BIC) (Schwarz, et al,1978) is one of the most popular
model selection criteria based on penalty terms of model complexity. In this paper, we use
BIC for GMM model selection:

BIC () = 2log ((
| )) + Klog (m) (6)

where, m is the data sample number, and K is the total number of parameters to be
estimated in GMM.
In this paper, we provide an algorithm for classifying EBNs. Given a sensor network {S
}, we
assume that sensor nodes are deployed with moderate density in the spatial terrain. From a
mathematical perspective, sensor readings provide a dense, but discrete sampling of the
underlying continuous distribution. To check whether or not N
is a sensor node lying on

the boundary of an event, we put the data {
} from readings of the sensor nodes in (N
)
and then build our best GMM based on {
}.
In more detail, we first set the upper bound of the mixture component number to be K. Then
for each j = 1,2,,K, the data set {
}is fed into (5) (6) for estimation of (J). Let BM denote
the number of mixture components of the best model. We select BM where
BIC ((BM)) = min
K
BIC ((J)). Therefore our final is (BM) or
,,,BM
.
To classify if N
is a EBN, the conditional probability for
given model (BM) is computed

by
(
| (BM) ) =
BM
) (7)

then (
| (BM) )< , N
is classified as a EBN. Where is used as a threshold to measure

EBN which has significantly low probability density values given the final model (BM).
The threshold is set as = 0.25, the upper bound of the component number is set as K = 5.
These parameters are used as the default in Sections 4 and 5, unless otherwise stated.
To dynamically update the estimates of observations by conducting (Zivkovic, et al,2004),
we have the following dynamic evolvement and observation equations:

= f (
) + w
(8)

= g(
) + v
(9)

where f(.) is the linear or nonlinear state evolvement function and g(.) is highly nonlinear
observation function. w
and v
are the standard deviation (noise sequences). For example

in static sensor node location, where
remain the same after deployment because of the

governance of Equation (8). Therefore, we get the expression:

=f (
) + w
(10)

where, w
model the small position perturbation or other effects.

4.1. BD3D scheme in 2D model [D = 2 in (5)]
Suppose n total sensor nodes are randomly distributed in a 2D terrain, with network density
enough to perform a boundary detection application. BD3D provides simultaneous
selection of EBN and non-EBN during BN selection process by tactfully using the proposed
BD3D BN Array (see Table 3) and GMM-based mathematic model.

Sensor reading of N
v
(head) Sensor readings of (N
v
)(rear)
.

Note: (N
) = N
(see Section 2) and both head and rear are initialized with 0; Rule: a
sensor node is EN if its own reading equals to 1 and vice versa.
Although the determination of sensor node status e.g., EBN or non-EBN etc. is practically
meaningful, in literature fewer works are focusing on this issue. In BD3D, we tactfully
utilize a BD3D BN array to adequately energy-friendly determine sensor node status (see
Table 4).

BD3D
BN
Array

EN

Non-EN
BN
Non-BN EBN Non-EBN
Head 1 0 1 1 0 1
HR random random All 0 &
random
All 1 All
0
All 1
Table 4. Head &HR based sensor node status determination.

*EBN EN, EBN non-EBN = BN (see Figure 4) and random means it is either all 1 or all 0.

Fig. 4. EBN and non-EBN on BL in BD3D 2D model when object expanded or shrunk.

head=
1 EN, EBNBN, non BN
0 non EN, non BN

HR=
all 1 non EBNBN, non BN
all 0 EBNBN, non BN
random EN, non EN or EBNBN

The Bayesian Information Criterion (BIC) (Schwarz, et al,1978) is one of the most popular
model selection criteria based on penalty terms of model complexity. In this paper, we use
BIC for GMM model selection:

BIC () = 2log ((
| )) + Klog (m) (6)

where, m is the data sample number, and K is the total number of parameters to be
estimated in GMM.
In this paper, we provide an algorithm for classifying EBNs. Given a sensor network {S
}, we
assume that sensor nodes are deployed with moderate density in the spatial terrain. From a
mathematical perspective, sensor readings provide a dense, but discrete sampling of the
underlying continuous distribution. To check whether or not N
is a sensor node lying on

the boundary of an event, we put the data {
} from readings of the sensor nodes in (N
)
and then build our best GMM based on {
}.
In more detail, we first set the upper bound of the mixture component number to be K. Then
for each j = 1,2,,K, the data set {
}is fed into (5) (6) for estimation of (J). Let BM denote
the number of mixture components of the best model. We select BM where
BIC ((BM)) = min
K
BIC ((J)). Therefore our final is (BM) or
,,,BM
.
To classify if N
is a EBN, the conditional probability for
given model (BM) is computed

by
(
| (BM) ) =
BM
) (7)

then (
| (BM) )< , N
is classified as a EBN. Where is used as a threshold to measure

EBN which has significantly low probability density values given the final model (BM).
The threshold is set as = 0.25, the upper bound of the component number is set as K = 5.
These parameters are used as the default in Sections 4 and 5, unless otherwise stated.
To dynamically update the estimates of observations by conducting (Zivkovic, et al,2004),
we have the following dynamic evolvement and observation equations:

= f (
) + w
(8)

= g(
) + v
(9)

where f(.) is the linear or nonlinear state evolvement function and g(.) is highly nonlinear
observation function. w
and v
are the standard deviation (noise sequences). For example

in static sensor node location, where
remain the same after deployment because of the

governance of Equation (8). Therefore, we get the expression:

=f (
) + w
(10)

where, w
model the small position perturbation or other effects.

4.1. BD3D scheme in 2D model [D = 2 in (5)]
Suppose n total sensor nodes are randomly distributed in a 2D terrain, with network density
enough to perform a boundary detection application. BD3D provides simultaneous
selection of EBN and non-EBN during BN selection process by tactfully using the proposed
BD3D BN Array (see Table 3) and GMM-based mathematic model.

Sensor reading of N
v
(head) Sensor readings of (N
v
)(rear)
.

Note: (N
) = N
(see Section 2) and both head and rear are initialized with 0; Rule: a
sensor node is EN if its own reading equals to 1 and vice versa.
Although the determination of sensor node status e.g., EBN or non-EBN etc. is practically
meaningful, in literature fewer works are focusing on this issue. In BD3D, we tactfully
utilize a BD3D BN array to adequately energy-friendly determine sensor node status (see
Table 4).

BD3D
BN
Array

EN

Non-EN
BN
Non-BN EBN Non-EBN
Head 1 0 1 1 0 1
HR random random All 0 &
random
All 1 All
0
All 1
Table 4. Head &HR based sensor node status determination.

*EBN EN, EBN non-EBN = BN (see Figure 4) and random means it is either all 1 or all 0.

Fig. 4. EBN and non-EBN on BL in BD3D 2D model when object expanded or shrunk.

head=
1 EN, EBNBN, non BN
0 non EN, non BN

HR=
all 1 non EBNBN, non BN
all 0 EBNBN, non BN
random EN, non EN or EBNBN

1 0 1 1 1 1 1 0 1
Table 5. Example of BD3D BN Array of N
.

N
is determined to be EN or EBN (BN) if head is 1 and HR is random based on the values in

Table 5.
Expanding from 2D to 3D, we find that not only the sensing area of sensor node but the
network topology is getting more complex, therefore when talking about the relative
position of sensor nodes, we need 3D sense of space to construct the model.

4.2. BD3D 3D model [D = 3 in (5)]

Fig. 5. Position of N
in 3D co-ordinate.

Define the state variable as 3D position for a specific sensor node modeled in Figure 5.

Possible data transmission in the 2D plane
Fig. 6. Concept of 2D plane for 3D sensing space.

Fig. 7. TSM concept with d = r for explanation simplicity.

In a 3D sensing space, sensor nodes are randomly distributed to form a network. To simplify
the complicated operations in dealing with sensor node localization in 3D model, we apply
a new concept of 2D plane that each 3D space can be divided into n 2D planes, where
(see Figure 6). The methodology of selection and representation of the 2D plane is
described as:

Randomly pick up to three sensor nodes { N
( x
, y
, z
), N
( x
, y
, z
), N

(x
, y
,z
)}(see Figure 7) from the 3D sensing space to form a 2D plane (either the 2D
plane in blue or green in Figure 6).
Suppose N
and N
are arbitrary points (sensor nodes) on the formed 2D plane

called , the plane representation is (see Figure 7).

=
x
Z
x
Z
x
Z
(11)

Strictly speaking, a 2D plane is definitely as a 2D section (see Figure 7). Therefore, one 3D
sensor network needs n 2D sections ( ) to reconstruct. Due to the impossibility of
computing n in programming, we introduce a virtual Thick Section Model (TSM).
1 0 1 1 1 1 1 0 1
Table 5. Example of BD3D BN Array of N
.

N
is determined to be EN or EBN (BN) if head is 1 and HR is random based on the values in

Table 5.
Expanding from 2D to 3D, we find that not only the sensing area of sensor node but the
network topology is getting more complex, therefore when talking about the relative
position of sensor nodes, we need 3D sense of space to construct the model.

4.2. BD3D 3D model [D = 3 in (5)]

Fig. 5. Position of N
in 3D co-ordinate.

Define the state variable as 3D position for a specific sensor node modeled in Figure 5.

Possible data transmission in the 2D plane
Fig. 6. Concept of 2D plane for 3D sensing space.

Fig. 7. TSM concept with d = r for explanation simplicity.

In a 3D sensing space, sensor nodes are randomly distributed to form a network. To simplify
the complicated operations in dealing with sensor node localization in 3D model, we apply
a new concept of 2D plane that each 3D space can be divided into n 2D planes, where
(see Figure 6). The methodology of selection and representation of the 2D plane is
described as:

Randomly pick up to three sensor nodes { N
( x
, y
, z
), N
( x
, y
, z
), N

(x
, y
,z
)}(see Figure 7) from the 3D sensing space to form a 2D plane (either the 2D
plane in blue or green in Figure 6).
Suppose N
and N
are arbitrary points (sensor nodes) on the formed 2D plane

called , the plane representation is (see Figure 7).

=
x
Z
x
Z
x
Z
(11)

Strictly speaking, a 2D plane is definitely as a 2D section (see Figure 7). Therefore, one 3D
sensor network needs n 2D sections ( ) to reconstruct. Due to the impossibility of
computing n in programming, we introduce a virtual Thick Section Model (TSM).
Figure 7 may help understand the concept of TSM. A 2D section is modeled as a thick plane
() with section thickness (d) with a set of representative points { N
(x
, y
, z
), N
(x
,
z
),N
(x
, z
) }describing the elements of the section. In our model the boundaries are
actually modeled as parametric line segments and points taking into account not only the
position of the plane but also the uncertainties of the plane contour.
Suppose the 3D network area is
cube ( is pre-determined in programming) and d r.

Therefore the selected 2D section could be those in Figure 8:

Fig. 8. Possible 2D sections in 3D network area and (c) is the model used in simulations.

The 2D section in our simulation model actually is a (
*d) area [see Figure 8 (c)]. (the

section thickness (d) is determined a priori to programming).Thus, the simulator only need
to perform TSM (/d times that significantly improve the maneuverability.

5. Simulation
In this section, we evaluated the performance of BD3D 2D and 3D model implemented in
Matlab respectively. The simulation parameters are given in the following table Sensor
nodes make local observations every 2 time slots:

Parameter Value
Network Area (2D,3D)
Number of sensor nodes(2D,3D)
The sink (2D,3D)
Transmission range(2D)
Time slots
Initial Energy
Message size
E

E
DA

(100 m)
2
(100 m)
3

2,500,10,000
(50,175), (50,175,50)
10 m
100 seconds
2J/battery
100 Bytes
50 nJ/bit
10 pJ/bit/m2
0.0013 pJ/bit/m4
5 nJ/bit/signal

5.1. Simulation model
(2D model)
Design a regular variation object: a circle initially centered at (50, 50) and continually
expand it by increasing its radius by 10 meters every 10 time slots. (see Figure 9.)
2D section
Design an irregular variation object: the initial ENs that adequately covers a area
{
=R
} to initiate the event. At every time slot, EN propagates

by picking up a random number of neighbors to join the event. In this way, the
network is guaranteed to be fully connected. (see Figure 9.)
(3D model)
Design a regular variation 3D object: the object center is (50, 50, 50) and continually
expand its radius by 10 meters every 10 time slots.
Design an irregular variation 3D object: the initial ENs are within a spherical area
{
+

= R
. EN propagates in a similar way as that

used for the irregular variation object in 2D model.

(a) Regular variation object (b) Irregular variation object
Fig. 9. Sample of BD3D 2D model with regular variation and irregular variation object.

Fig. 10. Rebound and boundary distances for BD3D.

The BD3D is flexible enough to be used in a clustered network or a non-clustered network
since it does not put any constraints on cluster architecture. However, BNs are usually
heavily utilized to send aggregated data associated with the object/network boundary
information to cluster head (in clustered networks) or the sink (non-clustered networks),
they would run out of energy more quickly. Therefore, achieving a reasonable amount of
BNs (the less the better) benefits energy saving.
Figure 7 may help understand the concept of TSM. A 2D section is modeled as a thick plane
() with section thickness (d) with a set of representative points { N
(x
, y
, z
), N
(x
,
z
),N
(x
, z
) }describing the elements of the section. In our model the boundaries are
actually modeled as parametric line segments and points taking into account not only the
position of the plane but also the uncertainties of the plane contour.
Suppose the 3D network area is
cube ( is pre-determined in programming) and d r.

Therefore the selected 2D section could be those in Figure 8:

Fig. 8. Possible 2D sections in 3D network area and (c) is the model used in simulations.

The 2D section in our simulation model actually is a (
*d) area [see Figure 8 (c)]. (the

section thickness (d) is determined a priori to programming).Thus, the simulator only need
to perform TSM (/d times that significantly improve the maneuverability.

5. Simulation
In this section, we evaluated the performance of BD3D 2D and 3D model implemented in
Matlab respectively. The simulation parameters are given in the following table Sensor
nodes make local observations every 2 time slots:

Parameter Value
Network Area (2D,3D)
Number of sensor nodes(2D,3D)
The sink (2D,3D)
Transmission range(2D)
Time slots
Initial Energy
Message size
E

E
DA

(100 m)
2
(100 m)
3

2,500,10,000
(50,175), (50,175,50)
10 m
100 seconds
2J/battery
100 Bytes
50 nJ/bit
10 pJ/bit/m2
0.0013 pJ/bit/m4
5 nJ/bit/signal

5.1. Simulation model
(2D model)
Design a regular variation object: a circle initially centered at (50, 50) and continually
expand it by increasing its radius by 10 meters every 10 time slots. (see Figure 9.)
2D section
Design an irregular variation object: the initial ENs that adequately covers a area
{
=R
} to initiate the event. At every time slot, EN propagates

by picking up a random number of neighbors to join the event. In this way, the
network is guaranteed to be fully connected. (see Figure 9.)
(3D model)
Design a regular variation 3D object: the object center is (50, 50, 50) and continually
expand its radius by 10 meters every 10 time slots.
Design an irregular variation 3D object: the initial ENs are within a spherical area
{
+

= R
. EN propagates in a similar way as that

used for the irregular variation object in 2D model.

(a) Regular variation object (b) Irregular variation object
Fig. 9. Sample of BD3D 2D model with regular variation and irregular variation object.

Fig. 10. Rebound and boundary distances for BD3D.

The BD3D is flexible enough to be used in a clustered network or a non-clustered network
since it does not put any constraints on cluster architecture. However, BNs are usually
heavily utilized to send aggregated data associated with the object/network boundary
information to cluster head (in clustered networks) or the sink (non-clustered networks),
they would run out of energy more quickly. Therefore, achieving a reasonable amount of
BNs (the less the better) benefits energy saving.
5.2. BD3D 2D model
This section discusses the performance evaluations based on BD3D 2D model. Figure 11
demonstrates the performances of the BD3D 2D model and DEMOCO and COBOM in terms
of the average residual energy per sensor node at 50 and 100 time slots of operation
respectively. Obviously, the performance of BD3D 2D is apparently better than DEMOCO
and COBOM. However at the meanwhile, it shows the good stability of energy load
balancing among the sensor nodes over the individual residual energy differences.

(a)

(b)
Fig. 11. Average energy level status of 2,500 sensor nodes after 50 and 100 time slots
operation.
Comparison of the number of BNs for a regular variation object with COBOM and
DEMOCO is shown in Figure 12 (a). To increase the comparability, the network is only
operated during 50 time slots. Figure 12 (a) shows that the BD3D 2D model consistently
provides less than half of the BNs selected by COBOM and reduces approximately by 1/3
those achieved by DEMOCO in the same environment. This could be due to the use of the
BD3D BN array (see Section 2) and GMM that helps selecting potential BNs easier than the
aforementioned COBOM and DEMOCO. Consequently, this avoids low data delivery rates and
excessive energy consumption by frequent flooding of inquiring packets

(a)

(b)
Fig. 12. Performance evaluation by using BD3D 2D when r = 10 m. (a) Comparison with
DEMOCO and COBOM on the number of BNs based on the number of sensor nodes
(regular variation object case for 50 time slots) (b) Number of BNs (EBNs and non-EBNs)
based on time slots (both regular variation and irregular variation object case).

However, due to the elusive ways proposed to expand the irregular variation object, we can
hardly do comparison with COBOM and DEMOCO anymore. Figure 12 (b) shows only the
0 500 1000 1500 2000 2500
0
50
100
150
200
250
Number of sensors
N
u
m
b
e
r

o
f

B
N
s

COBOM
DEMOCO
BD3D 2D model
0 10 20 30 40 50 60 70 80 90 100
0
5
10
15
20
25
30
35
40
45
50
Time slots
N
u
m
b
e
r

o
f

B
N
s

BN (smoothed)
EBN (smoothed)
non- EBN (smoothed)
BN (unsmoothed)
EBN (unsmoothed)
non- EBN (unsmoothed)
EBN
5.2. BD3D 2D model
This section discusses the performance evaluations based on BD3D 2D model. Figure 11
demonstrates the performances of the BD3D 2D model and DEMOCO and COBOM in terms
of the average residual energy per sensor node at 50 and 100 time slots of operation
respectively. Obviously, the performance of BD3D 2D is apparently better than DEMOCO
and COBOM. However at the meanwhile, it shows the good stability of energy load
balancing among the sensor nodes over the individual residual energy differences.

(a)

(b)
Fig. 11. Average energy level status of 2,500 sensor nodes after 50 and 100 time slots
operation.
Comparison of the number of BNs for a regular variation object with COBOM and
DEMOCO is shown in Figure 12 (a). To increase the comparability, the network is only
operated during 50 time slots. Figure 12 (a) shows that the BD3D 2D model consistently
provides less than half of the BNs selected by COBOM and reduces approximately by 1/3
those achieved by DEMOCO in the same environment. This could be due to the use of the
BD3D BN array (see Section 2) and GMM that helps selecting potential BNs easier than the
aforementioned COBOM and DEMOCO. Consequently, this avoids low data delivery rates and
excessive energy consumption by frequent flooding of inquiring packets

(a)

(b)
Fig. 12. Performance evaluation by using BD3D 2D when r = 10 m. (a) Comparison with
DEMOCO and COBOM on the number of BNs based on the number of sensor nodes
(regular variation object case for 50 time slots) (b) Number of BNs (EBNs and non-EBNs)
based on time slots (both regular variation and irregular variation object case).

However, due to the elusive ways proposed to expand the irregular variation object, we can
hardly do comparison with COBOM and DEMOCO anymore. Figure 12 (b) shows only the
0 500 1000 1500 2000 2500
0
50
100
150
200
250
Number of sensors
N
u
m
b
e
r

o
f

B
N
s

COBOM
DEMOCO
BD3D 2D model
0 10 20 30 40 50 60 70 80 90 100
0
5
10
15
20
25
30
35
40
45
50
Time slots
N
u
m
b
e
r

o
f

B
N
s

BN (smoothed)
EBN (smoothed)
non- EBN (smoothed)
BN (unsmoothed)
EBN (unsmoothed)
non- EBN (unsmoothed)
EBN
performance evaluation of the BD3D 2D model. As promised, EBNs and non-EBNs for
regular variation and irregular variation object cases are clearly found. From our analysis,
the value of BN (irregular variation) tends to be affected by irregular BL movements due to
the elusive change of object compared to that of BN (regular variation) that looks more
euphemistic. When the regular variation object expands over the rebound distance which
indicates the end of saturated distribution around the BL, Figure 12 (b) shows a rapid
increase of EBN (regular variation) and decrease of non-EBN (regular variation) until the
object covers the whole network. On the other hand, due to the non-determinacy of
irregular variation object shape changes, trajectories of both EBN (irregular variation) and
non-EBN (irregular variation) are always difficult to size up. However, we find similarity or
resemblance as to be essentially interchangeable for the first 50 time slots of operation
between regular variation and irregular variation object cases. For the second 50 time slots,
the performance of EBN (irregular variation) and non-EBN (irregular variation) are going to
split up, but show no direct relationship with rebound distance and boundary distance.

5.3. BD3D 3D model
Figure 13 shows a vertical section view of 3D sensor network area using TSMa
combinational view of three conditions { d < r, d > r, d >> r }

Fig. 13. A combinational vertical section view of 3D sensor network with {d < r, d > r, d >> r}

In this section, we modeled the BD3D 3D with different values of r and d by using TSM for
regular variation and irregular variation objects, respectively. Figure 14(a) compares the
number of BNs based on the value of d with r = 10m. As d r, the values we got are
approximately the same. Moreover, we varied the value of d (d r) for simulating the cases
with the significant existence of BA (see Figure 13), many BNs of highly possible BA got lost,
resulting in decrease of the number of BNs. Figure 14(b) compares the number of BNs based
on the value of r with d fixed at 8 m. As d r, the performance shows the comparatively
worst. By increasing r to meet d r, it shows the significant improvement on the
performances with very imperceptible distinctions. It is easy to guess that if communication
range of a sensor node is large, there will be many neighbors that it can communicate with,
which will result in more BNs.

(a)
(b)
Fig. 14. Comparison for regular variation object case using BD3D 3D model. (a) Number of
BNs based on time slots via varying d (r = 10 m), (b) Number of BNs based on time slots via
varying r (d = 8 m).

Meanwhile, we set the same parameter environment in the BD3D 3D model for evaluating
the number of BNs in the network in Figure 15. The most interesting feature is that the
network based on d r apparently performs better than that with d r. As a result, it can be
alleged that there is no strong relationship between the number of BNs and the
communication range (r) using TSM once d r. This occurrence can be clarified by the
analysis illustrated in Figure 13. Another interesting feature we can observe from Figure
15(b) is that when r = 5 and d = 8, it undergoes a very slow increase first and then
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
500
Time slots
N
u
m
b
e
r

o
f

B
N
s

d=4, r=10
d=8, r=10
d=12, r=10
d=16, r=10
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
500
Time slots
N
u
m
b
e
r

o
f

B
N
s

r=20,d=8
r=15,d=8
r=10,d=8
r=5,d=8
performance evaluation of the BD3D 2D model. As promised, EBNs and non-EBNs for
regular variation and irregular variation object cases are clearly found. From our analysis,
the value of BN (irregular variation) tends to be affected by irregular BL movements due to
the elusive change of object compared to that of BN (regular variation) that looks more
euphemistic. When the regular variation object expands over the rebound distance which
indicates the end of saturated distribution around the BL, Figure 12 (b) shows a rapid
increase of EBN (regular variation) and decrease of non-EBN (regular variation) until the
object covers the whole network. On the other hand, due to the non-determinacy of
irregular variation object shape changes, trajectories of both EBN (irregular variation) and
non-EBN (irregular variation) are always difficult to size up. However, we find similarity or
resemblance as to be essentially interchangeable for the first 50 time slots of operation
between regular variation and irregular variation object cases. For the second 50 time slots,
the performance of EBN (irregular variation) and non-EBN (irregular variation) are going to
split up, but show no direct relationship with rebound distance and boundary distance.

5.3. BD3D 3D model
Figure 13 shows a vertical section view of 3D sensor network area using TSMa
combinational view of three conditions { d < r, d > r, d >> r }

Fig. 13. A combinational vertical section view of 3D sensor network with {d < r, d > r, d >> r}

In this section, we modeled the BD3D 3D with different values of r and d by using TSM for
regular variation and irregular variation objects, respectively. Figure 14(a) compares the
number of BNs based on the value of d with r = 10m. As d r, the values we got are
approximately the same. Moreover, we varied the value of d (d r) for simulating the cases
with the significant existence of BA (see Figure 13), many BNs of highly possible BA got lost,
resulting in decrease of the number of BNs. Figure 14(b) compares the number of BNs based
on the value of r with d fixed at 8 m. As d r, the performance shows the comparatively
worst. By increasing r to meet d r, it shows the significant improvement on the
performances with very imperceptible distinctions. It is easy to guess that if communication
range of a sensor node is large, there will be many neighbors that it can communicate with,
which will result in more BNs.

(a)
(b)
Fig. 14. Comparison for regular variation object case using BD3D 3D model. (a) Number of
BNs based on time slots via varying d (r = 10 m), (b) Number of BNs based on time slots via
varying r (d = 8 m).

Meanwhile, we set the same parameter environment in the BD3D 3D model for evaluating
the number of BNs in the network in Figure 15. The most interesting feature is that the
network based on d r apparently performs better than that with d r. As a result, it can be
alleged that there is no strong relationship between the number of BNs and the
communication range (r) using TSM once d r. This occurrence can be clarified by the
analysis illustrated in Figure 13. Another interesting feature we can observe from Figure
15(b) is that when r = 5 and d = 8, it undergoes a very slow increase first and then
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
500
Time slots
N
u
m
b
e
r

o
f

B
N
s

d=4, r=10
d=8, r=10
d=12, r=10
d=16, r=10
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
500
Time slots
N
u
m
b
e
r

o
f

B
N
s

r=20,d=8
r=15,d=8
r=10,d=8
r=5,d=8
experiences a sharp increase in the number of BNs after 40 time slots. This was caused by a
phenomenon that the objects expand highly depending on the number of existing BNs.
However, at network initialization, we have relatively fewer existing BNs. As the cardinal
number designating the existence of BNs is over a special value (available at around 40 time
slots), the performance miraculously achieves a sudden improvement.

(a)
(b)
Fig. 15. Performance comparison for irregular variation object case using BD3D 3D model.
(a)Number of BNs based on time slots via varying d (r = 10m); (b)Number of BNs based on
time slots via varying r (d = 8m).

We hereby conclude that our BD3D for continuous boundary detection in 3D case works
well especially when d r using TSM. An in depth study about the impact of localization
impact on various routing protocols and its implications on design of location-dependent
system are left as future work.
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
500
Time slots
N
u
m
b
e
r

o
f

B
N
s

d=4, r=10
d=8, r=10
d=12, r=10
d=16, r=10
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
Time slots
N
u
m
b
e
r

o
f

B
N
s

r=20, d=8
r=15, d=8
r=10, d=8
r=5, d=8
6. Conclusions
This paper has proposed a novel Gaussian Mixture Model-based BD3D scheme for
boundary detection of continuously moving object in a 3D sensor network. We adequately
presented the proposed protocol, and the simulation results shown support our allegation
that the BD3D 2D model surely outperforms COBOM and DEMOCO in terms of average
residual energy per sensor node and the number of selected BNs, and the BD3D 3D model
achieves accurate boundary detections by soundly selecting EBN and non-EBN for both
regular variation and irregular variation object cases. Our future work will include
additional optimization desired to improve the performance of our algorithm and
verification of the precision of the expected boundaries and invention of a new protocol that
considers data losses and route failures due to unpredictable errors such as sensor node
failures, contention, interference and fading (Woo, et al, 2003; Seada, et al, 2004). Moreover,
the more accurate energy and mobility model will be addressed in future work.

Acknowledgements
This research was supported by Waseda University Global COE Program International
Research and Education Center for Ambient SoC sponsored by MEXT, Japan.

7. References
Kim, J.H.; Kim, K.B.; Sajjad, H.C.; Yang, W.C.;&Park, M.S.(2008). DEMOCO: Energy-
Efficient Detection and Monitoring for Continuous Objects in Wireless Sensor
Networks. IEICE Trans. Com. 2008, E91B, pp.3648-3656.
Zhong, C.& Worboys, M.(2007) Energy-efficient continuous boundary monitoring in sensor
networks. Technical Report, 2007. Available online:
http://ilab1.korea.ac.kr/papers/ref2.pdf/ (accessed on 31 July 2010).
Basu, A.; Jie, G.; Joseph, S.B.M.& Girishkumar, S.(2006) Distributed Localization by Noisy
Distance and Angle Information. In Proceedings of ACM MOBIHOC06, Los
Angeles, CA, USA, 2006;pp. 262-273
Eren, T.; Goldenberg, D.K.; Whiteley, W.& Yang, Y.R.(2004). Rigidity, Computation, and
Randomization in Network Localization. In Proceedings of IEEE
INFOCOM04,March 2004, Hongkong, China.
He, T.; Huang C.D.; Blum, B.M.; John A.S.& Tarek, A.(2003) Range-Free Localization
Schemes for Large Scale Sensor Networks. In Proceedings of ACM MOBICOM03,
Annapolis, MD, USA, June 2003; pp. 81-95
Nissanka, B.; Priyantha Hari, B.; Erik, D.& Seth, T.(2003) Anchor-Free Distributed
Localization in Sensor Networks. LCS Technical Report #892; MIT: Cambridge,
MA, USA, April 2003.
Guo, Z.; Zhou, M.& Jiang, G.(2008) Adaptive optimal sensor placement and boundary
estimation for dynamic mass objects. IEEE Trans. Syst. Man Cybern B. Cybern.
2008, 38, 222-32.
Olfati-Saber, R.(2007). Distributed tracking for mobile sensor networks with information
driven mobility. In Proceedings of Amer. Control Conference, New York, NY, USA,
July, 2007; pp. 4606-4612.
experiences a sharp increase in the number of BNs after 40 time slots. This was caused by a
phenomenon that the objects expand highly depending on the number of existing BNs.
However, at network initialization, we have relatively fewer existing BNs. As the cardinal
number designating the existence of BNs is over a special value (available at around 40 time
slots), the performance miraculously achieves a sudden improvement.

(a)
(b)
Fig. 15. Performance comparison for irregular variation object case using BD3D 3D model.
(a)Number of BNs based on time slots via varying d (r = 10m); (b)Number of BNs based on
time slots via varying r (d = 8m).

We hereby conclude that our BD3D for continuous boundary detection in 3D case works
well especially when d r using TSM. An in depth study about the impact of localization
impact on various routing protocols and its implications on design of location-dependent
system are left as future work.
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
500
Time slots
N
u
m
b
e
r

o
f

B
N
s

d=4, r=10
d=8, r=10
d=12, r=10
d=16, r=10
0 10 20 30 40 50 60 70 80
0
50
100
150
200
250
300
350
400
450
Time slots
N
u
m
b
e
r

o
f

B
N
s

r=20, d=8
r=15, d=8
r=10, d=8
r=5, d=8
6. Conclusions
This paper has proposed a novel Gaussian Mixture Model-based BD3D scheme for
boundary detection of continuously moving object in a 3D sensor network. We adequately
presented the proposed protocol, and the simulation results shown support our allegation
that the BD3D 2D model surely outperforms COBOM and DEMOCO in terms of average
residual energy per sensor node and the number of selected BNs, and the BD3D 3D model
achieves accurate boundary detections by soundly selecting EBN and non-EBN for both
regular variation and irregular variation object cases. Our future work will include
additional optimization desired to improve the performance of our algorithm and
verification of the precision of the expected boundaries and invention of a new protocol that
considers data losses and route failures due to unpredictable errors such as sensor node
failures, contention, interference and fading (Woo, et al, 2003; Seada, et al, 2004). Moreover,
the more accurate energy and mobility model will be addressed in future work.

Acknowledgements
This research was supported by Waseda University Global COE Program International
Research and Education Center for Ambient SoC sponsored by MEXT, Japan.

7. References
Kim, J.H.; Kim, K.B.; Sajjad, H.C.; Yang, W.C.;&Park, M.S.(2008). DEMOCO: Energy-
Efficient Detection and Monitoring for Continuous Objects in Wireless Sensor
Networks. IEICE Trans. Com. 2008, E91B, pp.3648-3656.
Zhong, C.& Worboys, M.(2007) Energy-efficient continuous boundary monitoring in sensor
networks. Technical Report, 2007. Available online:
http://ilab1.korea.ac.kr/papers/ref2.pdf/ (accessed on 31 July 2010).
Basu, A.; Jie, G.; Joseph, S.B.M.& Girishkumar, S.(2006) Distributed Localization by Noisy
Distance and Angle Information. In Proceedings of ACM MOBIHOC06, Los
Angeles, CA, USA, 2006;pp. 262-273
Eren, T.; Goldenberg, D.K.; Whiteley, W.& Yang, Y.R.(2004). Rigidity, Computation, and
Randomization in Network Localization. In Proceedings of IEEE
INFOCOM04,March 2004, Hongkong, China.
He, T.; Huang C.D.; Blum, B.M.; John A.S.& Tarek, A.(2003) Range-Free Localization
Schemes for Large Scale Sensor Networks. In Proceedings of ACM MOBICOM03,
Annapolis, MD, USA, June 2003; pp. 81-95
Nissanka, B.; Priyantha Hari, B.; Erik, D.& Seth, T.(2003) Anchor-Free Distributed
Localization in Sensor Networks. LCS Technical Report #892; MIT: Cambridge,
MA, USA, April 2003.
Guo, Z.; Zhou, M.& Jiang, G.(2008) Adaptive optimal sensor placement and boundary
estimation for dynamic mass objects. IEEE Trans. Syst. Man Cybern B. Cybern.
2008, 38, 222-32.
Olfati-Saber, R.(2007). Distributed tracking for mobile sensor networks with information
driven mobility. In Proceedings of Amer. Control Conference, New York, NY, USA,
July, 2007; pp. 4606-4612.
Funke, S. & Klein, C(2006). Hole Detection or: How Much Geometry Hides in Connectivity?
In Proceedings of the Twenty-Second Annual Symposium on Computational
Geometry, SCG 06, ACM Press: New York, NY, USA, 2006; pp. 377-385.
Funke, S.& Milosavljevic, N.(2007). Network sketching or: how much geometry hides in
connectivity?part ii. In Proceedings of the Eighteenth Annual ACM-SIAM
Symposium on Discrete Algorithms (SODA2007), New Orleans, LA, USA, 2007; pp.
958-967.
Peng, R.& Sichitiu, M.L.(2006) Angle of Arrival Localization for Wireless Sensor Networks.
In Proceedings of Third Annual IEEE Communications Society Conference on
Sensor, Mesh and Ad Hoc Communications and Networks (Secon06), Reston, VA,
USA, September 2006; pp. 25-28.
Lance, D.; Kristofer S.J.P.& Laurent EL G.(2001) Convex Position Estimation in Wireless
Sensor Networks. In Proceedings of IEEE INFOCOM01, Anchorage, April 2001,
AK, USA.
Hu, L.X & David, E.(2004) Localization for Mobile Sensor Networks. In Proceedings of ACM
MOBICOM04, Philadelphia, PA, USA, September 2004; pp. 45-57.
Ji, X. & Zha, H.(2004) Sensor Positioning in Wireless Ad-hoc Sensor Networks Using
Multidimensional Scaling. In Proceedings of INFOCOM04, March 2004,
Hongkong, China.
Yi, S.; Wheeler, R.; Zhang, Y.& Markus, P.J.F.(2003) Localization From Mere Connectivity, In
Proceedings of ACM MOBIHOC03, Annapolis, MD, USA, June 2003; pp. 201-212.
Yi, S. & Wheeler, R.(2004) Improved MDS-Based Localization. In Proceedings of IEEE
INFOCOM04, Hongkong, China, March 2004; pp. 2640-2651.
Andreas, S.; Park, H. & Mani, B.S.(2002) The Bits and Flops of the N-hop Multilateration
Primitive for Node Localization Problems. In Proceedings of ACM WSNA02,
Atlanta, GA, USA, September 28, 2002; pp. 112-121.
Zhang, L.Q.; Zhou, X.B. & Cheng, Q.(2006) Landscape-3D: A Robust Localization Scheme for
Sensor Networks over Complex 3D Terrains. In Proceedings of 31st Annual IEEE
Conference on Local Computer Networks (LCN), IEEE Computer Society Press:
Tampa, FL, USA, November 2006;pp. 239-246.
Samitha, E. & Pubudu, P.(2010) RSS Based Technologies in Wireless Sensor Networks,
Mobile and Wireless Communications Network Layer and Circuit Level Design,
Fares, S.A., Fumiyuki Adachi, F., Eds.; INTECH Book: Vienna, Austria, 2010.
Bulusu, N.; Hohn, H. & Deborah, E.(2001) Density Adaptive Algorithms for Beacon
Placement in Wireless Sensor Networks. In Proceedings of IEEE ICDCS01;
Phoenix, April 2001,AZ, USA.
Liu, L.; Wang, Z. & Zhou, M.(2009). An Innovative Beacon-Assisted Bi-Mode Positioning
Method in Wireless Sensor Networks. In Proceedings of IEEE International
Conference on Networking Sensing and Control (ICNSC09), Okayama, Japan,
March 2009, pp. 570-575.
Liu, L.; Manli, E.; Wang, Z.G. & Zhou, M.C.(2009). A 3D Self-positioning Method for
Wireless Sensor Nodes Based on Linear FMCW and TFDA. In Proceedings of IEEE
International Conference on Systems, Man, and Cybernetics, San Antonio, TX,
USA, October 2009; pp. 3069-3074.
Zhu, X.J.; Rik, S. & Gao, J.(2009). Segmenting a Sensor Field: Algorithm and Applications in
Network Design. ACM Trans. Sensor Netw. (TOSN) 2009, 5, 1-31.
McLachlan, G. & Peel, D.(2000). Finite Mixture Models; John Wiley & Sons: New York: NY,
USA, 2000.
Figueiredo, M. & Jain, A.K.(2002). Unsupervised learning of finite mixture models. IEEE
Trans. Patt. Anal. Mach. Int. 2002, 24, 381-396.
Akaike, H.(1973). Information Theory and an Extension of the Maximum Likelihood
Principle. In Proceedings of the Second International Symposium on Information
Theory, Akadmiai Kiad: Budapest, Hungary, 1973; pp. 267-281
Schwarz, G.(1978). Estimating the dimension of a model. Ann. Statist. 1978, 6, 461-464.
Solla, S.A.; Leen, T.K. & Muller, K.R.(2000). The Infinite Gaussian Mixture Model. In
Advances in Neural Information Processing Systems; MIT Press: Cambridge, MA,
USA, 2000; pp. 554-560.
Chintalapudi, K. & Govindan, R.(2003) Localized edge detection in sensor fields. IEEE Ad
Hoc Netw. J. 2003, pp.59-70
Jin, G. & Nittel, S.(2006) NED: An Efficient Noise-Tolerant Event and Event Boundary
Detection Algorithm in Wireless Sensor Networks. In Proceedings of the 7th
International Conferences on Mobile Data Management, Nara, Japan, May, 2006;
pp. 1551-6245.
Min, D.; Chen, D.; Kai, X. & Cheng, X.(2005). Localized Fault-Tolerant Event Boundary
Detection in Sensor Networks. IEEE Infocom. 2005; Miami, FL, USA, March, 2005;
pp. 902-913.
Heinzelman, W.R.; Chandrakasan, A. & Balakrishnan. H.(2000). Energy-Efficient
Communication Protocol for Wireless Microsensor Networks. In the Proceedings of
the Hawaii International Conference on System Sciences, Maui, Hawaii, USA,
January 4-7, 2000; pp.3005-3014.
Schwarz, G.(1978). Estimating the dimension of a model. Ann. Stat. 1978, 6, pp.461-464.
Zivkovic, Z. & van der Heijden, F.(2004). Recursive Unsupervised Learning of Finite
Mixture Models. In Proceedings of IEEE Transactions on Pattern Analysis and
Machine Intelligence, Washington, DC, USA, May 2004; pp. 651-656.
Woo, A.; Tong, T. & Culler, D.(2003). Taming the underlying challenges of reliable multihop
routing in sensor networks. In Proceedings of the 1st International Conference on
Embedded Networked Sensor Systems, Los Angeles, CA, USA, 2003; pp. 14-27.
Seada, A.K.; Zuniga, M.; Helmy, A. & Bhaskar, K.(2004). Energy-Efficient Forwarding
Strategies for Geographic Routing in Lossy Wireless Sensor Networks. In
Proceedings of the 2nd International Conference on Embedded Networked Sensor
Systems, Baltimore, MD, USA, 2004; pp. 108-121.

Funke, S. & Klein, C(2006). Hole Detection or: How Much Geometry Hides in Connectivity?
In Proceedings of the Twenty-Second Annual Symposium on Computational
Geometry, SCG 06, ACM Press: New York, NY, USA, 2006; pp. 377-385.
Funke, S.& Milosavljevic, N.(2007). Network sketching or: how much geometry hides in
connectivity?part ii. In Proceedings of the Eighteenth Annual ACM-SIAM
Symposium on Discrete Algorithms (SODA2007), New Orleans, LA, USA, 2007; pp.
958-967.
Peng, R.& Sichitiu, M.L.(2006) Angle of Arrival Localization for Wireless Sensor Networks.
In Proceedings of Third Annual IEEE Communications Society Conference on
Sensor, Mesh and Ad Hoc Communications and Networks (Secon06), Reston, VA,
USA, September 2006; pp. 25-28.
Lance, D.; Kristofer S.J.P.& Laurent EL G.(2001) Convex Position Estimation in Wireless
Sensor Networks. In Proceedings of IEEE INFOCOM01, Anchorage, April 2001,
AK, USA.
Hu, L.X & David, E.(2004) Localization for Mobile Sensor Networks. In Proceedings of ACM
MOBICOM04, Philadelphia, PA, USA, September 2004; pp. 45-57.
Ji, X. & Zha, H.(2004) Sensor Positioning in Wireless Ad-hoc Sensor Networks Using
Multidimensional Scaling. In Proceedings of INFOCOM04, March 2004,
Hongkong, China.
Yi, S.; Wheeler, R.; Zhang, Y.& Markus, P.J.F.(2003) Localization From Mere Connectivity, In
Proceedings of ACM MOBIHOC03, Annapolis, MD, USA, June 2003; pp. 201-212.
Yi, S. & Wheeler, R.(2004) Improved MDS-Based Localization. In Proceedings of IEEE
INFOCOM04, Hongkong, China, March 2004; pp. 2640-2651.
Andreas, S.; Park, H. & Mani, B.S.(2002) The Bits and Flops of the N-hop Multilateration
Primitive for Node Localization Problems. In Proceedings of ACM WSNA02,
Atlanta, GA, USA, September 28, 2002; pp. 112-121.
Zhang, L.Q.; Zhou, X.B. & Cheng, Q.(2006) Landscape-3D: A Robust Localization Scheme for
Sensor Networks over Complex 3D Terrains. In Proceedings of 31st Annual IEEE
Conference on Local Computer Networks (LCN), IEEE Computer Society Press:
Tampa, FL, USA, November 2006;pp. 239-246.
Samitha, E. & Pubudu, P.(2010) RSS Based Technologies in Wireless Sensor Networks,
Mobile and Wireless Communications Network Layer and Circuit Level Design,
Fares, S.A., Fumiyuki Adachi, F., Eds.; INTECH Book: Vienna, Austria, 2010.
Bulusu, N.; Hohn, H. & Deborah, E.(2001) Density Adaptive Algorithms for Beacon
Placement in Wireless Sensor Networks. In Proceedings of IEEE ICDCS01;
Phoenix, April 2001,AZ, USA.
Liu, L.; Wang, Z. & Zhou, M.(2009). An Innovative Beacon-Assisted Bi-Mode Positioning
Method in Wireless Sensor Networks. In Proceedings of IEEE International
Conference on Networking Sensing and Control (ICNSC09), Okayama, Japan,
March 2009, pp. 570-575.
Liu, L.; Manli, E.; Wang, Z.G. & Zhou, M.C.(2009). A 3D Self-positioning Method for
Wireless Sensor Nodes Based on Linear FMCW and TFDA. In Proceedings of IEEE
International Conference on Systems, Man, and Cybernetics, San Antonio, TX,
USA, October 2009; pp. 3069-3074.
Zhu, X.J.; Rik, S. & Gao, J.(2009). Segmenting a Sensor Field: Algorithm and Applications in
Network Design. ACM Trans. Sensor Netw. (TOSN) 2009, 5, 1-31.
McLachlan, G. & Peel, D.(2000). Finite Mixture Models; John Wiley & Sons: New York: NY,
USA, 2000.
Figueiredo, M. & Jain, A.K.(2002). Unsupervised learning of finite mixture models. IEEE
Trans. Patt. Anal. Mach. Int. 2002, 24, 381-396.
Akaike, H.(1973). Information Theory and an Extension of the Maximum Likelihood
Principle. In Proceedings of the Second International Symposium on Information
Theory, Akadmiai Kiad: Budapest, Hungary, 1973; pp. 267-281
Schwarz, G.(1978). Estimating the dimension of a model. Ann. Statist. 1978, 6, 461-464.
Solla, S.A.; Leen, T.K. & Muller, K.R.(2000). The Infinite Gaussian Mixture Model. In
Advances in Neural Information Processing Systems; MIT Press: Cambridge, MA,
USA, 2000; pp. 554-560.
Chintalapudi, K. & Govindan, R.(2003) Localized edge detection in sensor fields. IEEE Ad
Hoc Netw. J. 2003, pp.59-70
Jin, G. & Nittel, S.(2006) NED: An Efficient Noise-Tolerant Event and Event Boundary
Detection Algorithm in Wireless Sensor Networks. In Proceedings of the 7th
International Conferences on Mobile Data Management, Nara, Japan, May, 2006;
pp. 1551-6245.
Min, D.; Chen, D.; Kai, X. & Cheng, X.(2005). Localized Fault-Tolerant Event Boundary
Detection in Sensor Networks. IEEE Infocom. 2005; Miami, FL, USA, March, 2005;
pp. 902-913.
Heinzelman, W.R.; Chandrakasan, A. & Balakrishnan. H.(2000). Energy-Efficient
Communication Protocol for Wireless Microsensor Networks. In the Proceedings of
the Hawaii International Conference on System Sciences, Maui, Hawaii, USA,
January 4-7, 2000; pp.3005-3014.
Schwarz, G.(1978). Estimating the dimension of a model. Ann. Stat. 1978, 6, pp.461-464.
Zivkovic, Z. & van der Heijden, F.(2004). Recursive Unsupervised Learning of Finite
Mixture Models. In Proceedings of IEEE Transactions on Pattern Analysis and
Machine Intelligence, Washington, DC, USA, May 2004; pp. 651-656.
Woo, A.; Tong, T. & Culler, D.(2003). Taming the underlying challenges of reliable multihop
routing in sensor networks. In Proceedings of the 1st International Conference on
Embedded Networked Sensor Systems, Los Angeles, CA, USA, 2003; pp. 14-27.
Seada, A.K.; Zuniga, M.; Helmy, A. & Bhaskar, K.(2004). Energy-Efficient Forwarding
Strategies for Geographic Routing in Lossy Wireless Sensor Networks. In
Proceedings of the 2nd International Conference on Embedded Networked Sensor
Systems, Baltimore, MD, USA, 2004; pp. 108-121.

Monitoring Wireless Sensor Network Performance by Tracking Node operational Deviation 413
Monitoring Wireless Sensor Network Performance by Tracking Node
operational Deviation
Yaqoob J. Y. Al-raisi and Nazar E. M. Adam
X

Monitoring Wireless Sensor
Network Performance by Tracking
Node operational Deviation

Yaqoob J. Y. Al-raisi
1
and Nazar E. M. Adam
2

1
HIS Department, Sultan Qaboos University Hospital,
Oman
2
Computer Engineering Department, Fahad Bin Sultan University
Saudi Arabia

1. Introduction
Wireless Sensor Network (WSN) is a very powerful tool that enables its users to closely
monitor, understand and control application processes. It is different from traditional wired
sensor networks in that its characteristics make it cheap to manufacture, implement and
deploy. However, this tool is still at an early stage and many aspects need to be addressed in
order to increase its reliability. One of these aspects is the degradation of network
performance as a result of network nodes deviation. This may directly reduces the quality
and the quantity of data collected by the network and may cause, in turn, the monitoring
application to fail or the network lifetime to be reduced.
Deviations in sensor node operations arise as a result of systematic or/and transient errors
(Elnahrawy, 2004). Systematic error is mainly caused by hardware faults, such as calibration
error after prolonged use, a reduction in operating power levels, or a change in operating
conditions; this type of error affects node operations continuously until the problem is
rectified. Transient errors, on the other hand, occur as a result of temporary external or
internal circumstances, such as various random environmental effects, unstable hardware,
software bugs, channel interface, and multi-path effects. This type of error deviates node
operations until the effect disappears.
These two types of error may directly and indirectly affect the quality and the quantity of
data collected by the WSN. They directly affect sensor measurements and cause drift by a
constant value (i.e. bias); they change the difference between a sensor measurement and the
actual value, (i.e. drift); and can cause sensor measurements to remain constant, regardless
of changes in the actual value, (i.e. complete failure). In addition, they affect the
communication and exchange of packets by dropping them. On the other hand, the above-
mentioned errors can have an indirect effect on the networks collaboration function, the
construction of routing tables, the selection of the node reporting rate, and the selection of
data gathering points. Analysis of the data collected by the network (in some practical
deployments, such as (Ramanathan, 2004), (Tolle, 2005)), shows that these error reduces the
21

quality of network collected data by 49%; and in some cases, the network had to be
redeployed in order to collect the data because of the failure of the monitored application.
Analysis also indicate that a 51% overall improvement of WSN functionality can be
expected, as well as an improvement in the quality of the collected data, if real-time
monitoring tools are used.

2. Motivations
To detect and isolate operational deviations in WSNs researchers proposed several data
clearance, fault-tolerance, diagnosis, and performance measurement techniques.
Data cleaning techniques work at a high network level and consider reading impacts from a
deviated sensor on multi-sensor aggregation/fusion such as in (Yao-jung, 2004). Such
research proposes several methods that isolate deviated readings by tracking or predicting
correlation between neighbour node measurements. Most of this research uses complex
methods or models that need a high resource usage to detect and predict sensor
measurements. Moreover, these techniques rectify deviated data after detecting them
without checking their cause and their impact on network functionality.
Fault-tolerance techniques are important in embedded networks which are difficult to access
physically. The advantage of these techniques is their ability to address all network levels;
such as circuit level, logical level, memory level, program level and system level; but due to
WSNs scare recourses these techniques have a limited usage. In general WSNs fault-tolerant
techniques detect faults in fusion and aggregation operation, network deployment and
collaboration, coverage and connectivity, energy consumption, energy event fault tolerance,
reporting rate, network detection, and many others (Song, 2004, Linnyer, 2004, Bhaskar,
2004, Koushanfar, 2003, Luo, 2006). Faults are detected using logical decision predicates
computed in individual sensors (Bhaskar, 2004), faulty node detection (Koushanfar, 2003),
or event region and event boundary detection (Luo, 2006). These methods detect metrics
either at high or low network level without relating them to each other and without
checking their impact on network functionality. The main problem with these techniques is
the impact of deviation on network functionality and collected data accuracy before it is
detected.
Diagnosis techniques use passive or active monitoring to trace, visualize, simulate and
debug historical network log files in real and non real time as discussed in (Jaikaeo, 2001).
These techniques are used to detect faults at high or low network levels after testing their
cause. For example, Nithya at (Ramanathan, 2005) proposed a debugging system that
debugs low network level statistical changes by drawing correlations between seemingly
unrelated, distributed events and producing graphs that highlight those correlations. Most
of these diagnosis techniques are complex and use iteration tests for their detection. These
techniques assume a minimal cost associated with continuously transmitting of debug
information to centralized or distributed monitor nodes and send/receive test packets to
conform the detection of a faultier.
Finally, performance techniques are similar to diagnosis techniques but without iteration
tests and screw pack techniques. Unfortunately there is little literature and research on
systematic measurement and monitoring in wireless sensor networks. Yonggang in
(Yonggang, 2004) studied the effect of packet loss and their impact on network stability and
network processing. He studied the effect of the environmental conditions, traffic load,

network dynamics, collaboration behavior, and constraint recourse on packet delivery
performance using empirical experiments and simulations. Although packet delivery is
important in wireless communication and can predict network performance, it can give
wrong indications of network performance level due to collaboration behavior, and
measurement redundancy which makes a network able to tolerate a certain degree of
changes. Also, Yonggang proposed an energy map aggregation based approach that sends
messages recording significant energy level drops to the sink.
The work in this paper has been motivated by the need to find a tool that uses a very low
level of network resources and detects deviations in the networks operations that affect the
quality and quantity of the data that are collected before they seriously degrade the
networks overall functionality and reduce its lifetime.

3. Project Methodology
3.1 Layout of manuscript
The layout of this paper is organised as follows: Section 2 includes a discussion of related
work on functionality degradation detection in WSNs, followed by an explanation of the
algorithms approach. The fourth section explains the practical implementation of the
algorithm in a TinyOS Surge multi-hop application; results of experiments at the network
level are then discussed. Finally, the paper ends with a conclusion and suggestions for
future work.

3.2 Algorithm Approach
In order to overcome the above-mentioned drawbacks, the Voting Median Base Algorithm
for Approximate Performance Measurements of Wireless Sensor Networks (VMBA)
algorithm is proposed. This algorithm is a passive voting algorithm that collects its metrics
directly from the application by utilizing the overhearing which exists in the
neighbourhood. The algorithm requires only readings of neighbours measurements and
does not rely on any information regarding global topology. This makes it scalable to any
network deployment size. The proposed algorithm uses parameters found in nodes for
other networking and application protocols which makes it much cheaper in terms of
resource usage. It uses only the transceiver to send warning messages if there is a network
performance degradation or when the node disagrees with the warning messages of
neighbours.
The algorithm is divided into four different modules; i.e. listening and filtering, data
analysis and threshold test, decision and confidence control and warning packet exchange.
In this section we give some definitions and then the VMBA functional algorithm is
presented.

A. Listening and Filtering Module
The listening and filtering module is responsible for examining the validity of the received
neighbour nodes measurements by filtering those readings beyond the range of the sensors
physical characteristics; as shown in the pseudo-code in Fig.1 . The module then constructs
neighbour readings tables and builds statistics in the loss table for neighbour readings.


quality of network collected data by 49%; and in some cases, the network had to be
redeployed in order to collect the data because of the failure of the monitored application.
Analysis also indicate that a 51% overall improvement of WSN functionality can be
expected, as well as an improvement in the quality of the collected data, if real-time
monitoring tools are used.

2. Motivations
To detect and isolate operational deviations in WSNs researchers proposed several data
clearance, fault-tolerance, diagnosis, and performance measurement techniques.
Data cleaning techniques work at a high network level and consider reading impacts from a
deviated sensor on multi-sensor aggregation/fusion such as in (Yao-jung, 2004). Such
research proposes several methods that isolate deviated readings by tracking or predicting
correlation between neighbour node measurements. Most of this research uses complex
methods or models that need a high resource usage to detect and predict sensor
measurements. Moreover, these techniques rectify deviated data after detecting them
without checking their cause and their impact on network functionality.
Fault-tolerance techniques are important in embedded networks which are difficult to access
physically. The advantage of these techniques is their ability to address all network levels;
such as circuit level, logical level, memory level, program level and system level; but due to
WSNs scare recourses these techniques have a limited usage. In general WSNs fault-tolerant
techniques detect faults in fusion and aggregation operation, network deployment and
collaboration, coverage and connectivity, energy consumption, energy event fault tolerance,
reporting rate, network detection, and many others (Song, 2004, Linnyer, 2004, Bhaskar,
2004, Koushanfar, 2003, Luo, 2006). Faults are detected using logical decision predicates
computed in individual sensors (Bhaskar, 2004), faulty node detection (Koushanfar, 2003),
or event region and event boundary detection (Luo, 2006). These methods detect metrics
either at high or low network level without relating them to each other and without
checking their impact on network functionality. The main problem with these techniques is
the impact of deviation on network functionality and collected data accuracy before it is
detected.
Diagnosis techniques use passive or active monitoring to trace, visualize, simulate and
debug historical network log files in real and non real time as discussed in (Jaikaeo, 2001).
These techniques are used to detect faults at high or low network levels after testing their
cause. For example, Nithya at (Ramanathan, 2005) proposed a debugging system that
debugs low network level statistical changes by drawing correlations between seemingly
unrelated, distributed events and producing graphs that highlight those correlations. Most
of these diagnosis techniques are complex and use iteration tests for their detection. These
techniques assume a minimal cost associated with continuously transmitting of debug
information to centralized or distributed monitor nodes and send/receive test packets to
conform the detection of a faultier.
Finally, performance techniques are similar to diagnosis techniques but without iteration
tests and screw pack techniques. Unfortunately there is little literature and research on
systematic measurement and monitoring in wireless sensor networks. Yonggang in
(Yonggang, 2004) studied the effect of packet loss and their impact on network stability and
network processing. He studied the effect of the environmental conditions, traffic load,

network dynamics, collaboration behavior, and constraint recourse on packet delivery
performance using empirical experiments and simulations. Although packet delivery is
important in wireless communication and can predict network performance, it can give
wrong indications of network performance level due to collaboration behavior, and
measurement redundancy which makes a network able to tolerate a certain degree of
changes. Also, Yonggang proposed an energy map aggregation based approach that sends
messages recording significant energy level drops to the sink.
The work in this paper has been motivated by the need to find a tool that uses a very low
level of network resources and detects deviations in the networks operations that affect the
quality and quantity of the data that are collected before they seriously degrade the
networks overall functionality and reduce its lifetime.

3. Project Methodology
3.1 Layout of manuscript
The layout of this paper is organised as follows: Section 2 includes a discussion of related
work on functionality degradation detection in WSNs, followed by an explanation of the
algorithms approach. The fourth section explains the practical implementation of the
algorithm in a TinyOS Surge multi-hop application; results of experiments at the network
level are then discussed. Finally, the paper ends with a conclusion and suggestions for
future work.

3.2 Algorithm Approach
In order to overcome the above-mentioned drawbacks, the Voting Median Base Algorithm
for Approximate Performance Measurements of Wireless Sensor Networks (VMBA)
algorithm is proposed. This algorithm is a passive voting algorithm that collects its metrics
directly from the application by utilizing the overhearing which exists in the
neighbourhood. The algorithm requires only readings of neighbours measurements and
does not rely on any information regarding global topology. This makes it scalable to any
network deployment size. The proposed algorithm uses parameters found in nodes for
other networking and application protocols which makes it much cheaper in terms of
resource usage. It uses only the transceiver to send warning messages if there is a network
performance degradation or when the node disagrees with the warning messages of
neighbours.
The algorithm is divided into four different modules; i.e. listening and filtering, data
analysis and threshold test, decision and confidence control and warning packet exchange.
In this section we give some definitions and then the VMBA functional algorithm is
presented.

A. Listening and Filtering Module
The listening and filtering module is responsible for examining the validity of the received
neighbour nodes measurements by filtering those readings beyond the range of the sensors
physical characteristics; as shown in the pseudo-code in Fig.1 . The module then constructs
neighbour readings tables and builds statistics in the loss table for neighbour readings.


1: Each
i
S
senses the phenomenon and wait for
time T to receive N(
i
S
) readings
2: IF t > T THEN
3: For each unreceived
i
j
x
increment
i
j
L
;
4: IF
L
C
>
i
j
x
>
M
C

5: Remove
i
j
x
from data set and increment
i
j
D

6: Calculate
i
med
of the available
i
S
data set
Fig. 1. VMBA Algorithm Module 1

B. Data analysis and Threshold Test Module
The second module; i.e. data analysis and threshold test module; tests the content of these
tables. This is done by evaluating the data with regard to assigned dynamic or static limits
calculated from a reference value or median.
The proposed algorithm has followed a straightforward approach in calculating faulty
deviations in sensor functionality. Its analysis assumes that true measurements of a
phenomenons characteristics, following a Gaussian pdf, centred on the calculated median
of neighbourhood readings. Any deviation is controlled by the correlation expected at the
end of the sensing range of a node, and the sensor nodes measuring accuracy (where most
of the physical processes monitored by WSNs are typically modeled as diffusion models
with varying dispersion functions). This assumption is based on the fact that random errors
are normally distributed with a zero mean and standard deviation is equal to the
specification of the goals designed for the nodes and the network. Any sensor measurement
that is not in this region is considered deviated to a degree equal to the ratio of the distance
from the neighbourhood median value to the median value.

1: IF |
i
med
-
1 i
med

| > med
Increment
i
M
and let
i
med
=
1 i
med

2:
j
d = |
i
med
-
i
j
x |
3: IF
j
d
>
1
and |
i
i
x
-
i
j
x
| <
1

4: Increment
i
j
COV

5: ELSE increment
i
R

6: IF i
R
k
> 40%
7: Increment
i
N

8: IF i
R
k
*
j
d
>
1

9: Increment
i
j
D


In addition, the second module tests the effect of losses on the reliability of the collected data
by calculating the degree of distortion in the neighbourhood data that has occurred because
of its affect on the collected data accuracy and network functionality. This is done by
calculating the ratio of the number of healthy readings to the total number readings as
shown in Fig 2 step 8.

C. Decision Confidence Control Module
The third module; i.e. Decision confidence control module; is concerned with tracking
changes in the health of neighbour nodes in an assigned time window. This is set depending
on the characteristics of the network application and the required response detection time. If
exceeded, a request is sent to module four in order to send a detection message to the sink
identifying suspected node number, the type of fault, the number of times it has been
detected and the effect of the detection on the neighbourhood data and communication. The
function of this module is shown in Fig 3.

1: Calculate
i
ML

2: IF
i
ML
> 60%
3: Send to module 4 a request to send an inefficient
power consumption warning message
4: IF
i
M
>
M

5: Send to module 4 a request to send a
neighbourhood malfunction due to losses
warning message
6: IF
i
j
COV
>
C

7: Send to module 4 a request to send to
detecting node j a coverage
problem message
8: IF distortion >
d
& median of
i
j
L
> 60%
degrade detection in network
functionality message
10: IF
i
j
D
>
w

11: Send to module 4 a request
to send a detection of node
j malfunction message

D. Warning Packet Exchange Module
When module four receives a send request, it checks its neighbours warning exchange
memory to ensure that none of the neighbour nodes have reported the same fault in that
monitoring window period. If none of the neighbours have so reported, it sends a message
or it cancels the request. In addition, this module tests warning messages received from its
neighbours with statistics from module three. If the suspected node flags up a counter
indication smaller than a threshold, a message will be released indicating

1: Each
i
S
senses the phenomenon and wait for
time T to receive N(
i
S
) readings
2: IF t > T THEN
3: For each unreceived
i
j
x
increment
i
j
L
;
4: IF
L
C
>
i
j
x
>
M
C

5: Remove
i
j
x
from data set and increment
i
j
D

6: Calculate
i
med
of the available
i
S
data set

B. Data analysis and Threshold Test Module
The second module; i.e. data analysis and threshold test module; tests the content of these
tables. This is done by evaluating the data with regard to assigned dynamic or static limits
calculated from a reference value or median.
The proposed algorithm has followed a straightforward approach in calculating faulty
deviations in sensor functionality. Its analysis assumes that true measurements of a
phenomenons characteristics, following a Gaussian pdf, centred on the calculated median
of neighbourhood readings. Any deviation is controlled by the correlation expected at the
end of the sensing range of a node, and the sensor nodes measuring accuracy (where most
of the physical processes monitored by WSNs are typically modeled as diffusion models
with varying dispersion functions). This assumption is based on the fact that random errors
are normally distributed with a zero mean and standard deviation is equal to the
specification of the goals designed for the nodes and the network. Any sensor measurement
that is not in this region is considered deviated to a degree equal to the ratio of the distance
from the neighbourhood median value to the median value.

1: IF |
i
med
-
1 i
med

| > med
Increment
i
M
and let
i
med
=
1 i
med

2:
j
d = |
i
med
-
i
j
x |
3: IF
j
d
>
1
and |
i
i
x
-
i
j
x
| <
1

4: Increment
i
j
COV

5: ELSE increment
i
R

6: IF i
R
k
> 40%
7: Increment
i
N

8: IF i
R
k
*
j
d
>
1

9: Increment
i
j
D


In addition, the second module tests the effect of losses on the reliability of the collected data
by calculating the degree of distortion in the neighbourhood data that has occurred because
of its affect on the collected data accuracy and network functionality. This is done by
calculating the ratio of the number of healthy readings to the total number readings as
shown in Fig 2 step 8.

C. Decision Confidence Control Module
The third module; i.e. Decision confidence control module; is concerned with tracking
changes in the health of neighbour nodes in an assigned time window. This is set depending
on the characteristics of the network application and the required response detection time. If
exceeded, a request is sent to module four in order to send a detection message to the sink
identifying suspected node number, the type of fault, the number of times it has been
detected and the effect of the detection on the neighbourhood data and communication. The
function of this module is shown in Fig 3.

1: Calculate
i
ML

2: IF
i
ML
> 60%
3: Send to module 4 a request to send an inefficient
power consumption warning message
4: IF
i
M
>
M

neighbourhood malfunction due to losses
warning message
6: IF
i
j
COV
>
C

7: Send to module 4 a request to send to
detecting node j a coverage
problem message
8: IF distortion >
d
& median of
i
j
L
> 60%
degrade detection in network
functionality message
10: IF
i
j
D
>
w

11: Send to module 4 a request
to send a detection of node
j malfunction message

D. Warning Packet Exchange Module
When module four receives a send request, it checks its neighbours warning exchange
memory to ensure that none of the neighbour nodes have reported the same fault in that
monitoring window period. If none of the neighbours have so reported, it sends a message
or it cancels the request. In addition, this module tests warning messages received from its
neighbours with statistics from module three. If the suspected node flags up a counter
indication smaller than a threshold, a message will be released indicating

NO_FAULT_EVIDENCE regarding the received warning message. On the other hand, if
the threshold is higher or equal to the threshold, then the node cancels any similar warning
message request from module three during that monitoring period. This is to ensure the
reliability of the warning message detection and to correct any incorrect detection that may
occur because of losses or other network circumstances. Moreover, module four reduces the
algorithm warning packets released by checking if any of its neighbours sent the same
message at that time interval. If it been sent the algorithm is going to discard module three
requests as shown in Fig. 4 part 3.

1: Receiving neighbour warning
a) Check received warning with the same module 3 counter of reported node.
b) IF module 3 counter < 30%
c) Release NO-EVIDENCE-OF-FAULT message
d) ELSE flag the stop sending of the same message from the node at this monitoring
time.
2: Receiving module 3 request
a) Test stop flag of received request warning
b) IF flag = 1 discard message
c) IF send message repeated 3 times send stop reporting the fault message and flag stop
fault counter.
d) ELSE send the requested message by module 3.
3: Testing warning packet release
a) IF detected fault returns to normal reset the same fault counters, send
FAULT_CLEAR message and recalculate protocol tables.
b) IF step 2 and 3-a alternate for the same fault three times in a predefined monitoring
window, the module send s an UNSTABLE_DETECTION warning message to
report the detection and flags a permanent fault counter to stop reporting the
same fault.
c) By the end of the predefined period reset all counters.

4. Performance Evaluation
VMBA algorithm performance can be evaluate on eight different aspects: deviation
detection in single and multi-hop levels, algorithm detection threshold, algorithm detection
confidence, algorithm spatial and temporary change tracking for sensor nodes, the impact of
packet losses on algorithm analysis, resource usage at node and network levels, the impact
of algorithm programming location in the protocol stack, and algorithm released warning
messages. In this paper, we considered the empirical performance evaluation of the
algorithm at the network level.

4.1 Algorithm Programming in Protocol Stacks
The algorithm was implemented on a Berkeley (Crossbow) Mica2 sensor motes testbed that
was programmed in nesC on TinyOS operation system. This is done by building the
proposed algorithm on the TinyOS multi-hop routing protocol.

The TinyOS multi-hop protocol consists of MultiHopEngineM; which provides the over all
packet movement logic for multi-hop functionality; and MultiHopLEPSM; which is used to
provide the link estimation and parent selection mechanisms. These two TinyOS
components were modified by added different functions from the proposed algorithm
modules as shown at Figure 5.

Fig. 5. Functions added to multi-hop components and links between the components

In order to send detected warning packets, a new packet type was constructed. This new
packet carries the algorithm detection parameters; as shown at Figure 6. It has a total length
of 20 bytes, the last 8 are used for algorithm detection, while the first 12 follow the multi-hop
protocol configuration. This is to route the released warning packet in the network.

Fig. 6. Algorithm warning message packet


NO_FAULT_EVIDENCE regarding the received warning message. On the other hand, if
the threshold is higher or equal to the threshold, then the node cancels any similar warning
message request from module three during that monitoring period. This is to ensure the
reliability of the warning message detection and to correct any incorrect detection that may
occur because of losses or other network circumstances. Moreover, module four reduces the
algorithm warning packets released by checking if any of its neighbours sent the same
message at that time interval. If it been sent the algorithm is going to discard module three
requests as shown in Fig. 4 part 3.

1: Receiving neighbour warning
a) Check received warning with the same module 3 counter of reported node.
b) IF module 3 counter < 30%
c) Release NO-EVIDENCE-OF-FAULT message
d) ELSE flag the stop sending of the same message from the node at this monitoring
time.
2: Receiving module 3 request
a) Test stop flag of received request warning
b) IF flag = 1 discard message
c) IF send message repeated 3 times send stop reporting the fault message and flag stop
fault counter.
d) ELSE send the requested message by module 3.
3: Testing warning packet release
a) IF detected fault returns to normal reset the same fault counters, send
FAULT_CLEAR message and recalculate protocol tables.
b) IF step 2 and 3-a alternate for the same fault three times in a predefined monitoring
window, the module send s an UNSTABLE_DETECTION warning message to
report the detection and flags a permanent fault counter to stop reporting the
same fault.
c) By the end of the predefined period reset all counters.

4. Performance Evaluation
VMBA algorithm performance can be evaluate on eight different aspects: deviation
detection in single and multi-hop levels, algorithm detection threshold, algorithm detection
confidence, algorithm spatial and temporary change tracking for sensor nodes, the impact of
packet losses on algorithm analysis, resource usage at node and network levels, the impact
of algorithm programming location in the protocol stack, and algorithm released warning
messages. In this paper, we considered the empirical performance evaluation of the
algorithm at the network level.

4.1 Algorithm Programming in Protocol Stacks
The algorithm was implemented on a Berkeley (Crossbow) Mica2 sensor motes testbed that
was programmed in nesC on TinyOS operation system. This is done by building the
proposed algorithm on the TinyOS multi-hop routing protocol.

The TinyOS multi-hop protocol consists of MultiHopEngineM; which provides the over all
packet movement logic for multi-hop functionality; and MultiHopLEPSM; which is used to
provide the link estimation and parent selection mechanisms. These two TinyOS
components were modified by added different functions from the proposed algorithm
modules as shown at Figure 5.

Fig. 5. Functions added to multi-hop components and links between the components

In order to send detected warning packets, a new packet type was constructed. This new
packet carries the algorithm detection parameters; as shown at Figure 6. It has a total length
of 20 bytes, the last 8 are used for algorithm detection, while the first 12 follow the multi-hop
protocol configuration. This is to route the released warning packet in the network.

Fig. 6. Algorithm warning message packet


At the algorithm detection part, the first byte carries the total number of readings, that is the
number of neighbour nodes in addition to the monitoring node. The next two bytes carry
the number of neighbors detected by the node as dead and deviated respectively. This is
followed by a byte that carries the identification number of the detected faulty neighbour
node. The byte after this carries the type of fault codes; as shown in Table 1; and the final
two bytes carry the number of times that the monitoring node detect the reported fault.

5. Experimental Setting and Evaluation Metrics
Several experiments were conducted indoors at the High Speed Network Research Group
Lab in Loughborough University to test the proposed algorithms functionality in real
sensor network scenarios. These experiments were conducted in the presence of other
devices that are able to interfere with the sensor transmission and reduce the antennae
performance; these offer experiments in a dynamic topology and in circumstances of high
packet losses. Some of these experiments were conducted to test the algorithms
functionality under multi-hop and highly dynamic topology configurations. These
experiments used 13 Mica2 sensors, measuring temperature, distributed in an area of about
4mX5m. The nodes were programmed with the output power of -20 dBm and had top bent
antennae to limit their communication range. In this configuration, the nodes were divided
into two groups which overlapped in an area between them; thus, some of the nodes around
the edge could not hear or communicate with each other (as shown in Figure 7). Moreover,
this configuration forced the topology to be highly dynamic. This leads nodes to miss
hearing each other and frequently change their multi-hop routing parents in the sink. These
experiments used Mica2 nodes attached to a MIB510 programming board as a base station
connected to a computer serial port. A snooping node was also added to the network setting
with its power programmed to the maximum (i.e. 5dBm) in order to listen to
communications among all the nodes within the network and to track packet exchanges in
the multi-hop without increasing the usage of resources of the networks sensor nodes.

Fault Type Code
TOPOLOGY_UNSTABLE 0
FAULT_TYPE_DEVIATION 1
FAULT_TYPE_COMMUNICATION 2
FAULT_TYPE_COVERAGE 3
FAULT_TYPE_ENERGY_CONSUMPTION 4
NO_EVEDENCE_OF_FAULT 5
FAULT_MESSAGE_STOP 6
FAULT_TYPE_DEID 7
FAULT_CLEAR 8
NEIGHBORHOOD_MULFUNCTION 9
PROTOCOL_EFFECT 10
Table 1. Codes of detected faults in algorithm warning messages

The metrics used to evaluate the results were, firstly, the percentage of incorrectly released
dead node warnings. This is the ratio of the number of false dead node detections released
by the algorithm as opposed to the total number of packets released by the application. This
indicates the impact of high network dynamics on the algorithms incorrect detection. The
second metric was the percentage of NO-FAULT-EVIDENCE messages released by the
algorithm, which is the ratio of the number of NO-FAULT-EVIDENCE messages to the
total number of packets released by the application. This also indicates the impact of high
network dynamics but on neighbours passive tests of incorrect detections.

Fig. 7. Logical topology of the experiment at a time interval

These experiments tested the impact of the dead node window threshold, and monitoring
window size on the algorithms detection of dead nodes and the number of warning messages
released by it in a highly dynamic network. The algorithm parameters that were tested, as
shown in Table 2,and 3 were changed in different experiments to check their impact on the
deductibility performance of the network and the exchange of warning packets.

Window
Type
Small Monitoring window Big Monitoring
Window
Stop Reporting
Window
Diversion 120 seconds
(70% threshold)
480 seconds(8
minutes)
1920 seconds
(32 minutes)
Distortion 60 seconds (84% loss threshold and larger
than 25% accuracy of the two nodes)
240 seconds(4
minutes)
960 seconds
(16 minutes)
Dead 60 seconds 240 seconds(4
minutes)
960 seconds
(16 minutes)
Table 2. Sizes of monitoring windows in the experiments


At the algorithm detection part, the first byte carries the total number of readings, that is the
number of neighbour nodes in addition to the monitoring node. The next two bytes carry
the number of neighbors detected by the node as dead and deviated respectively. This is
followed by a byte that carries the identification number of the detected faulty neighbour
node. The byte after this carries the type of fault codes; as shown in Table 1; and the final
two bytes carry the number of times that the monitoring node detect the reported fault.

5. Experimental Setting and Evaluation Metrics
Several experiments were conducted indoors at the High Speed Network Research Group
Lab in Loughborough University to test the proposed algorithms functionality in real
sensor network scenarios. These experiments were conducted in the presence of other
devices that are able to interfere with the sensor transmission and reduce the antennae
performance; these offer experiments in a dynamic topology and in circumstances of high
packet losses. Some of these experiments were conducted to test the algorithms
functionality under multi-hop and highly dynamic topology configurations. These
experiments used 13 Mica2 sensors, measuring temperature, distributed in an area of about
4mX5m. The nodes were programmed with the output power of -20 dBm and had top bent
antennae to limit their communication range. In this configuration, the nodes were divided
into two groups which overlapped in an area between them; thus, some of the nodes around
the edge could not hear or communicate with each other (as shown in Figure 7). Moreover,
this configuration forced the topology to be highly dynamic. This leads nodes to miss
hearing each other and frequently change their multi-hop routing parents in the sink. These
experiments used Mica2 nodes attached to a MIB510 programming board as a base station
connected to a computer serial port. A snooping node was also added to the network setting
with its power programmed to the maximum (i.e. 5dBm) in order to listen to
communications among all the nodes within the network and to track packet exchanges in
the multi-hop without increasing the usage of resources of the networks sensor nodes.

Fault Type Code
TOPOLOGY_UNSTABLE 0
FAULT_TYPE_DEVIATION 1
FAULT_TYPE_COMMUNICATION 2
FAULT_TYPE_COVERAGE 3
FAULT_TYPE_ENERGY_CONSUMPTION 4
NO_EVEDENCE_OF_FAULT 5
FAULT_MESSAGE_STOP 6
FAULT_TYPE_DEID 7
FAULT_CLEAR 8
NEIGHBORHOOD_MULFUNCTION 9
PROTOCOL_EFFECT 10
Table 1. Codes of detected faults in algorithm warning messages

The metrics used to evaluate the results were, firstly, the percentage of incorrectly released
dead node warnings. This is the ratio of the number of false dead node detections released
by the algorithm as opposed to the total number of packets released by the application. This
indicates the impact of high network dynamics on the algorithms incorrect detection. The
second metric was the percentage of NO-FAULT-EVIDENCE messages released by the
algorithm, which is the ratio of the number of NO-FAULT-EVIDENCE messages to the
total number of packets released by the application. This also indicates the impact of high
network dynamics but on neighbours passive tests of incorrect detections.

Fig. 7. Logical topology of the experiment at a time interval

These experiments tested the impact of the dead node window threshold, and monitoring
window size on the algorithms detection of dead nodes and the number of warning messages
released by it in a highly dynamic network. The algorithm parameters that were tested, as
shown in Table 2,and 3 were changed in different experiments to check their impact on the
deductibility performance of the network and the exchange of warning packets.

Window
Type
Small Monitoring window Big Monitoring
Window
Stop Reporting
Window
Diversion 120 seconds
(70% threshold)
480 seconds(8
minutes)
1920 seconds
(32 minutes)
Distortion 60 seconds (84% loss threshold and larger
than 25% accuracy of the two nodes)
240 seconds(4
minutes)
960 seconds
(16 minutes)
Dead 60 seconds 240 seconds(4
minutes)
960 seconds
(16 minutes)
Table 2. Sizes of monitoring windows in the experiments


Window Small
windows
Small
window size
Size of Big
window
Number of
small window
at the group
Total
monitoring
window size
1 Linear
increased
240 seconds
(4 minutes)
3 groups 4-8-12 48 minutes
2 Exponential
increased
8-12-16 64 minutes
3 10-14-18 72 minutes
4 14-16-20 80 minutes
Table 3. Size of monitoring windows

5.1 Effect of Network Topology and Packet Losses on the Algorithms Functionality
Figure 8 plots the relationship between the percentage of detected and No_Fault_Evidence
messages released from the algorithm for different application reporting rates. (Please note
that reporting rates logs were used in the figure to plot these). The results of the experiments
showed that at a 1 second reporting rate (a multi-hop protocol leads to congestion and an
overflow of communication), a large amount of wrong suspected dead warnings occurred
(around 3.2% of the total network packet exchange in the application). Furthermore, a large
number of No_Fault_Evidence replies were released from neighbour messages (i.e. around
0.5% of the total packets in the network application). Reducing the applications reporting
rate to 2 seconds reduced the number of suspected dead messages; these decreased sharply
to 0.5% of the total number of packets released by the network application. This happened
alongside a reduction in No_Fault_Evidence messages which reached around 0.01% of the
total number of packets released. Thus, the number of suspected dead messages was
reduced to almost 0% when the applications reporting rate was adjusted to 1 minute, along
with a decrease in No_Fault_Evidence messages released from neighbours. When the
applications reporting rate was increased to 30 minutes, a sharp increase occurred in the
number of suspected dead and No_Fault_Evidence messages, as shown in the figure. Also,
Figure 8 shows that, by increasing the applications reporting rate above 1 minute, the
number of No_Fault_Evidence messages increases so that it becomes higher than the
number of suspected dead messages. This is as a result of the size of the monitoring
windows and the highly dynamic network topology.
From these experiments, it can be concluded that dead node warnings will not disappear
spatially in a monitored network when the network connections are highly dynamic. To
reduce the number of wrong suspected dead messages, different window sizes and
combinations were tested, as shown in Table 3. Figure 9 shows the relation between the
percentage of correct, positive detected (wrong detection) by the algorithm, together with
the negative false dead nodes for different sizes of large monitoring windows. The figure
illustrates that, as the big monitoring window size increased, the confidence of the
algorithms detection of dead neighbour nodes increased, along with a decrease in the
number of packets released by the algorithm. Although increasing window size will reduce
the number of wrong messages, it also increases the response detection time and the
probability of node failure occurring before releasing the warning message.

10
0
10
1
10
2
10
3
0
0.5
1
1.5
2
2.5
3
3.5
Reporting Rate (Seconds)
P
e
r
c
e
n
t
a
g
e

o
f

A
l
g
o
r
i
t
h
m

W
a
r
n
i
n
g

P
a
c
k
e
t
s
Dead Warning
No-Fault-Evedence

Fig. 8. Changing reporting rates with the percentage of warning messages released with the
same window size

0
1
2
1 2 3 4
Big Window
P
e
r
c
e
n
t
a
g
e

o
f

W
a
r
n
i
n
g

M
e
s
s
a
g
e
s
Dead No_Fault_ Evidence
Message Stopped

Fig. 9. Percentage of warning messages released for different window configurations

To solve this problem, the algorithm was programmed such that it would select the
neighbours it would monitor; this selection depends on the amount of received packets. This
configuration reduced the number of wrong packets reported by 80% and reduced
No_Evidence_Fault by 70%, as Figure 10 shows, but it also added additional complexity to
algorithms source code and its functionality. Moreover, there will be uncovered neighbour
nodes in low density networks. In addition, the proposed algorithm was modified to send
warning messages concerning the detection of connectivity problems between neighbour
nodes. This makes the algorithm stop reporting a suspected node if the node is detected as

Window Small
windows
Small
window size
Size of Big
window
Number of
small window
at the group
Total
monitoring
window size
1 Linear
increased
240 seconds
(4 minutes)
3 groups 4-8-12 48 minutes
2 Exponential
increased
8-12-16 64 minutes
3 10-14-18 72 minutes
4 14-16-20 80 minutes
Table 3. Size of monitoring windows

5.1 Effect of Network Topology and Packet Losses on the Algorithms Functionality
Figure 8 plots the relationship between the percentage of detected and No_Fault_Evidence
messages released from the algorithm for different application reporting rates. (Please note
that reporting rates logs were used in the figure to plot these). The results of the experiments
showed that at a 1 second reporting rate (a multi-hop protocol leads to congestion and an
overflow of communication), a large amount of wrong suspected dead warnings occurred
(around 3.2% of the total network packet exchange in the application). Furthermore, a large
number of No_Fault_Evidence replies were released from neighbour messages (i.e. around
0.5% of the total packets in the network application). Reducing the applications reporting
rate to 2 seconds reduced the number of suspected dead messages; these decreased sharply
to 0.5% of the total number of packets released by the network application. This happened
alongside a reduction in No_Fault_Evidence messages which reached around 0.01% of the
total number of packets released. Thus, the number of suspected dead messages was
reduced to almost 0% when the applications reporting rate was adjusted to 1 minute, along
with a decrease in No_Fault_Evidence messages released from neighbours. When the
applications reporting rate was increased to 30 minutes, a sharp increase occurred in the
number of suspected dead and No_Fault_Evidence messages, as shown in the figure. Also,
Figure 8 shows that, by increasing the applications reporting rate above 1 minute, the
number of No_Fault_Evidence messages increases so that it becomes higher than the
number of suspected dead messages. This is as a result of the size of the monitoring
windows and the highly dynamic network topology.
From these experiments, it can be concluded that dead node warnings will not disappear
spatially in a monitored network when the network connections are highly dynamic. To
reduce the number of wrong suspected dead messages, different window sizes and
combinations were tested, as shown in Table 3. Figure 9 shows the relation between the
percentage of correct, positive detected (wrong detection) by the algorithm, together with
the negative false dead nodes for different sizes of large monitoring windows. The figure
illustrates that, as the big monitoring window size increased, the confidence of the
algorithms detection of dead neighbour nodes increased, along with a decrease in the
number of packets released by the algorithm. Although increasing window size will reduce
the number of wrong messages, it also increases the response detection time and the
probability of node failure occurring before releasing the warning message.

10
0
10
1
10
2
10
3
0
0.5
1
1.5
2
2.5
3
3.5
Reporting Rate (Seconds)
P
e
r
c
e
n
t
a
g
e

o
f

A
l
g
o
r
i
t
h
m

W
a
r
n
i
n
g

P
a
c
k
e
t
s
Dead Warning
No-Fault-Evedence

Fig. 8. Changing reporting rates with the percentage of warning messages released with the
same window size

0
1
2
1 2 3 4
Big Window
P
e
r
c
e
n
t
a
g
e

o
f

W
a
r
n
i
n
g

M
e
s
s
a
g
e
s
Dead No_Fault_ Evidence
Message Stopped

Fig. 9. Percentage of warning messages released for different window configurations

To solve this problem, the algorithm was programmed such that it would select the
neighbours it would monitor; this selection depends on the amount of received packets. This
configuration reduced the number of wrong packets reported by 80% and reduced
No_Evidence_Fault by 70%, as Figure 10 shows, but it also added additional complexity to
algorithms source code and its functionality. Moreover, there will be uncovered neighbour
nodes in low density networks. In addition, the proposed algorithm was modified to send
warning messages concerning the detection of connectivity problems between neighbour
nodes. This makes the algorithm stop reporting a suspected node if the node is detected as

dead and if 3 clear dead messages are detected at the stop reporting monitoring window.
Figure 10 plots comparisons between the percentages of the algorithms released dead and
no evidence messages in a neighbourhood with and without the modification covering
connectivity problems. The figure shows that there is a reduction of 20% in the number of
No_Fault_Evidence messages as a result of a 34% reduction in the detection of dead
packets.

0
0.1
0.2
0.3
0.4
0.5
0.6
With Neighbor Selection Without Neighbor Selection
P
a
c
k
e
t

P
e
r
c
e
n
t
a
g
e
Dead No_Fault_Evidence

Fig. 10. Number of exchanged warning packets between selected and not selected neighbour
nodes.

6. Conclusion and Future Work
We proposed a distributed performance algorithm that enables each sensor node at sensor
network to detect the health of nodes at neighbourhood and their collaborative
functionality. This algorithm sends a warning packet to the sink reporting any degradation
detection.
The proposed algorithm tested using TinyOS Surge multi-hop application on Berkely
Mica2 sensor nodes testbed. These empirical experiments showed that the high loss in WSN
causes proposed algorithm wrong detection of neighbour nodes aliveness and released
more NO_EVIDENCE_FAULT messages. This controlled by adjusting the monitoring
window size and reduces the proposed algorithm wrong detection by 80% and the
NO_EVIDENCE_FAULT messages by 70%.
There are numerous aspects that can be considered in the future in order to extend this work
and improve the algorithms functionality, such as checking the impact of the mobility of
sensor nodes on the algorithms functionality. Also, it would be useful to study the impact
of faulty data on individual WSN protocols and compare these results with the proposed
approximate calculation that depends on the number of deviated nodes.

7. References
Elnahrawy Eiman and N. Badri, (2004). Cleaning and Querying Noisy Sensors, The First ACM
Conference on Embedded Networked Sensor Systems (SenSys'03), pp. 78-87.
N. Ramanathan, T. Schoellhammer, D. Estrin, M. Hansen, T. Harmon, E. Kohler, and M.
Srivastava,(2006). The Final Frontier: Embedding Networked Sensors in the Soil, CENS
Technical Report #68, Center for Embedded Networked Sensing, UCLA, USA.

G. Tolle, J. Polastre, R. Szewczyk, D. Culler, N. Turner, K. Tu, S. Burgess, T. Dawson, P.
Buonadonna, D. Gay, and W. Hong, (2005). A Macroscope in the Redwoods, ACM
W. Yao-jung , M. Alice Agogine and G. Kai. (2004). Fuzzy Validation and Fusion for Wireless
Sensor Networks, in ASMEinternational Mechanical Engineering Congress and
RD&D Expo (IMECE2004),Anaheim, California, USA..
H. Song and C. Edward. (2004). Continuous Residual Energy Monitoring in Wireless Sensor
Networks, in International Symposium on Parallel and Distributed Processing and
Applications (ISPA 2004), pp. 169-177.
Linnyer Beatrys Ruiz, Isabela G. Siqueria and Leonardo B. Oliveira. (2004). Fault
Management in Event-driven Wireless Sensor Networks, in MSWiM04, October 4-6,
Venezia, Italy.
K. Bhaskar and S. S. Iyengar. (2004). Distributes Bayesian Algorithms for Fult-tolerant Event
Region Detection in Wireless Sensor Networks,IEEE Transaction on Computers, vol. 53,
pp. 421-250.
F. Koushanfar, M. Potkonjak and A. Sangiovanni-Vincentelli. (2003). On-line Fault Detection
of Sensor Measurements, in Sensors. Proceedings of IEEE, 2003, pp. 974-979.
X. Luo, M. Dong and Y. Huang(2006). "On Distributed Fault-tolerant Detection in Wireless
Sensor Networks," IEEE Transactions on Computers, vol. 55, pp. 58-70.
C. Jaikaeo, C. Srisathapornphat and C. Shen, (2001). Diagnosis of Sensor Networks, in
Communications, 2001. ICC 2001. IEEE International Conference, pp. 1627-1632.
N. Ramanathan, K. Chang, R. Kapur, L. Girod, E. Kohler and D. Estrin, (2005). Sympathy for
the Sensor Network Debugger, in The 3rd ACM Conf. Embedded Networked Sensor
Systems (SenSys 2005), pp. 255-267.
Z. Yonggang, (2004). Measurement and Monitoring in Wireless Sensor Networks, PhD Thesis,
Computer Science Department, University of Southern California, USA, June. 2004.


dead and if 3 clear dead messages are detected at the stop reporting monitoring window.
Figure 10 plots comparisons between the percentages of the algorithms released dead and
no evidence messages in a neighbourhood with and without the modification covering
connectivity problems. The figure shows that there is a reduction of 20% in the number of
No_Fault_Evidence messages as a result of a 34% reduction in the detection of dead
packets.

0
0.1
0.2
0.3
0.4
0.5
0.6
With Neighbor Selection Without Neighbor Selection
P
a
c
k
e
t

P
e
r
c
e
n
t
a
g
e
Dead No_Fault_Evidence

Fig. 10. Number of exchanged warning packets between selected and not selected neighbour
nodes.

6. Conclusion and Future Work
We proposed a distributed performance algorithm that enables each sensor node at sensor
network to detect the health of nodes at neighbourhood and their collaborative
functionality. This algorithm sends a warning packet to the sink reporting any degradation
detection.
The proposed algorithm tested using TinyOS Surge multi-hop application on Berkely
Mica2 sensor nodes testbed. These empirical experiments showed that the high loss in WSN
causes proposed algorithm wrong detection of neighbour nodes aliveness and released
more NO_EVIDENCE_FAULT messages. This controlled by adjusting the monitoring
window size and reduces the proposed algorithm wrong detection by 80% and the
NO_EVIDENCE_FAULT messages by 70%.
There are numerous aspects that can be considered in the future in order to extend this work
and improve the algorithms functionality, such as checking the impact of the mobility of
sensor nodes on the algorithms functionality. Also, it would be useful to study the impact
of faulty data on individual WSN protocols and compare these results with the proposed
approximate calculation that depends on the number of deviated nodes.

7. References
Elnahrawy Eiman and N. Badri, (2004). Cleaning and Querying Noisy Sensors, The First ACM
N. Ramanathan, T. Schoellhammer, D. Estrin, M. Hansen, T. Harmon, E. Kohler, and M.
Srivastava,(2006). The Final Frontier: Embedding Networked Sensors in the Soil, CENS
Technical Report #68, Center for Embedded Networked Sensing, UCLA, USA.

G. Tolle, J. Polastre, R. Szewczyk, D. Culler, N. Turner, K. Tu, S. Burgess, T. Dawson, P.
Buonadonna, D. Gay, and W. Hong, (2005). A Macroscope in the Redwoods, ACM
W. Yao-jung , M. Alice Agogine and G. Kai. (2004). Fuzzy Validation and Fusion for Wireless
Sensor Networks, in ASMEinternational Mechanical Engineering Congress and
RD&D Expo (IMECE2004),Anaheim, California, USA..
H. Song and C. Edward. (2004). Continuous Residual Energy Monitoring in Wireless Sensor
Networks, in International Symposium on Parallel and Distributed Processing and
Applications (ISPA 2004), pp. 169-177.
Linnyer Beatrys Ruiz, Isabela G. Siqueria and Leonardo B. Oliveira. (2004). Fault
Management in Event-driven Wireless Sensor Networks, in MSWiM04, October 4-6,
Venezia, Italy.
K. Bhaskar and S. S. Iyengar. (2004). Distributes Bayesian Algorithms for Fult-tolerant Event
Region Detection in Wireless Sensor Networks,IEEE Transaction on Computers, vol. 53,
pp. 421-250.
F. Koushanfar, M. Potkonjak and A. Sangiovanni-Vincentelli. (2003). On-line Fault Detection
of Sensor Measurements, in Sensors. Proceedings of IEEE, 2003, pp. 974-979.
X. Luo, M. Dong and Y. Huang(2006). "On Distributed Fault-tolerant Detection in Wireless
Sensor Networks," IEEE Transactions on Computers, vol. 55, pp. 58-70.
C. Jaikaeo, C. Srisathapornphat and C. Shen, (2001). Diagnosis of Sensor Networks, in
Communications, 2001. ICC 2001. IEEE International Conference, pp. 1627-1632.
N. Ramanathan, K. Chang, R. Kapur, L. Girod, E. Kohler and D. Estrin, (2005). Sympathy for
the Sensor Network Debugger, in The 3rd ACM Conf. Embedded Networked Sensor
Systems (SenSys 2005), pp. 255-267.
Z. Yonggang, (2004). Measurement and Monitoring in Wireless Sensor Networks, PhD Thesis,
Computer Science Department, University of Southern California, USA, June. 2004.

Building Context Aware Network of Wireless Sensors Using
a Scalable Distributed Estimation Scheme for Real-time Data Manipulation 427
Building Context Aware Network of Wireless Sensors Using a Scalable
Distributed Estimation Scheme for Real-time Data Manipulation
X

Building Context Aware Network
of Wireless Sensors Using a Scalable
Distributed Estimation Scheme for
Real-time Data Manipulation

Monash University
Australia

1. Introduction
Wireless sensor networks are an exciting emerging domain of deeply networked systems of
low-power wireless motes with a tiny amount of CPU and memory and large federated
networks for high-resolution sensing of the environment (Welsh et al., 2004). The capability
to support plethora of new diverse applications has placed Wireless Sensor Network
technology at threshold of an era of significant potential growth. The technology is
advancing rapidly under the push of the new technological developments and the pull of
vast and diverse potential applications. The near ubiquity of the internet coupled with
recent engineering achievements, are opening the door to a new generation of low-cost and
powerful sensor devices which are capable of delivering high-grade spatial and temporal
resolution. In that regard, distributed estimation and tracking is one of the most
fundamental collaborative information processing challenges in wireless sensor networks
(WSNs). Moreover, estimation issues in wireless networks with packet-loss are gaining lion
share of attention over the last few years. However, due to the inherent limitations of WSNs
in terms of power and computational resources, deploying any distributed estimation
technique within WSNs requires modifying the existing methods to address those
limitations effectively. In fact, the problem with current approaches lies in the significant
increase in the computational expenses of the deployed methods as the result of increase in
the size of the network. This increase puts a heavy practical burden on deployment of those
algorithms for resource-constrained wireless sensor networks.
Current decentralized Kalman filtering involves state estimation using a set of local Kalman
filters that communicate with all other nodes. The information flow is all-to-all with
approximate communication complexity of O(n
2
) which is not scalable for WSNs. In this
chapter an attempt is made to explore new ways in provisioning distributed estimation in
WSNs by introducing a light-weight distributed pattern recognition scheme which provides
single-cycle learning and entails a large number of loosely coupled parallel operations. In
fact, the focus of our approach would be on a novel scalable and distributed filtering scheme
in which each node only communicates messages with its neighbours on a network to
22

minimize the communication overhead to a high degree. In order to achieve higher level of
simplicity to increase effectiveness of algorithm deployment in terms of utilizing limited
available resources, a novel approach toward real-time distributed estimation is introduced,
using a simplified graph-based method called Distributed Hierarchical Graph Neuron
(DHGN). The proposed approach not only enjoys from conserving the limited power
resources of resource-constrained sensor nodes, but also can be scaled effectively to address
scalability issues which are of primary concern in wireless sensor networks. The proposed
scheme is not intending to replace existing complex methods which are treated in the
literature, but rather makes an attempt to reduce the computational expenses involved in
the processing of the gathered in-situ information from sensor nodes. Strength of DHGN lies
in the processing of non-uniform data patterns as it implements a finely distributable
framework at the smallest (atomic) logical sub-pattern level. The results are easily obtained
by summation at the overall pattern level. Hence, the algorithm is able to provide some sort
of divide and distribute filtering process throughout the network in a fine-grained manner
for minimizing the energy use. DHGN is also a highly scalable algorithm that allows real-
time in-network data manipulation; essential feature for real-time data processing in WSNs.
The outline for this chapter is as follows. Section 2 provides an overview of WSN technology
and its current research trends. Section 3 provides a discussion on the limitations of existing
approaches for data fusion and distributed estimation within WSNs. We will then introduce
our proposed distributed event detection and pattern classification scheme in Section 4.
Section 5 deals with performance metrics for our novel algorithm. Section 6 entails further
discussion on our proposed scheme and future direction of this research. Finally, section 7
concludes the chapter.

2. WSN Overview
A wireless sensor network (WSN) in its simplest form can be defined as (Chong & Kumar,
2003; Akyildiz, Su, Sankarasubramaniam & Cayirci, 2002; Culler, Estrin & Srivastava, 2004)
a network of (possibly low-size and low-complex) devices denoted as nodes that can sense
the environment and communicate the information gathered from the monitored field (e.g.,
an area or volume) through wireless links; the data is forwarded, possibly via multiple hops
relaying to a sink (sometimes denoted as controller or monitor ) that can use it locally or is
connected to other networks (e.g., the Internet) through a gateway. The nodes can be
stationary or moving. They can be aware of their location or not. They can be homogeneous
or not. A traditional single-sink WSN is illustrated in Figure 1. Almost all scientific papers in
the literature deal with such a definition. This single-sink scenario suffers from the lack of
scalability: by increasing the number of nodes the amount of data gathered by the sink
increases and once its capacity is reached the network size can not be augmented. Moreover,
for reasons related to medium access control (MAC) and routing aspects, network
performance cannot be considered independent from the network size. A more general
scenario includes multiple sinks in the network (see Figure 2). Given a level of node density,
a larger number of sinks will decrease the probability of isolated clusters of nodes that
cannot deliver their data owing to unfortunate signal propagation conditions.

Fig. 1. Traditional single-sink WSN

In principle, a multiple-sink WSN can be scalable (i.e., the same performance can be
achieved even by increasing the number of nodes), while this is clearly not true for a single-
sink network. However, a multi-sink WSN does not represent a trivial extension of a single-
sink case for the network engineer. There might be mainly two different cases: (1) all sinks
are connected through a separate network (either wired or wireless), or (2) the sinks are
disconnected. In the former case, a node needs to forward the data collected to any element
in the set of sinks. From the protocol viewpoint, this means that a selection can be done
based on a suitable criterion (e.g., minimum delay, maximum throughput, minimum
number of hops, etc.). The presence of multiple sinks in this case ensures better network
performance with respect to the single-sink case (assuming the same number of nodes is
deployed over the same area), but the communication protocols must be more complex and
should be designed according to suitable criteria. In the second case, when the sinks are not
connected, the presence of multiple sinks tends to partition the monitored field into smaller
areas; however from the communication protocols viewpoint no significant changes must be
included, apart from simple sink discovery mechanisms. Clearly, the most general and
interesting case (because of the better potential performance) is the first one, with the sinks
connected through any type of mesh network, or via direct links with a common gateway.
Both the single-sink and multiple-sink networks introduced above do not include the
presence of actuators, that is, devices able to manipulate the environment rather than
observe it. WSANs are composed of both sensing nodes and actuators (see Figure 3). Once
more, the inclusion of actuators does not represent a simple extension of a WSN from the
communication protocol viewpoint. In fact the information flow must be reversed in this
case: the protocols should be able to manage many-to-one communications when sensors
provide data, and one-to-many flows when the actuators need to be addressed, or even one-
to-one links if a specific actuator has to be reached.


minimize the communication overhead to a high degree. In order to achieve higher level of
simplicity to increase effectiveness of algorithm deployment in terms of utilizing limited
available resources, a novel approach toward real-time distributed estimation is introduced,
using a simplified graph-based method called Distributed Hierarchical Graph Neuron
(DHGN). The proposed approach not only enjoys from conserving the limited power
resources of resource-constrained sensor nodes, but also can be scaled effectively to address
scalability issues which are of primary concern in wireless sensor networks. The proposed
scheme is not intending to replace existing complex methods which are treated in the
literature, but rather makes an attempt to reduce the computational expenses involved in
the processing of the gathered in-situ information from sensor nodes. Strength of DHGN lies
in the processing of non-uniform data patterns as it implements a finely distributable
framework at the smallest (atomic) logical sub-pattern level. The results are easily obtained
by summation at the overall pattern level. Hence, the algorithm is able to provide some sort
of divide and distribute filtering process throughout the network in a fine-grained manner
for minimizing the energy use. DHGN is also a highly scalable algorithm that allows real-
time in-network data manipulation; essential feature for real-time data processing in WSNs.
The outline for this chapter is as follows. Section 2 provides an overview of WSN technology
and its current research trends. Section 3 provides a discussion on the limitations of existing
approaches for data fusion and distributed estimation within WSNs. We will then introduce
our proposed distributed event detection and pattern classification scheme in Section 4.
Section 5 deals with performance metrics for our novel algorithm. Section 6 entails further
discussion on our proposed scheme and future direction of this research. Finally, section 7
concludes the chapter.

2. WSN Overview
A wireless sensor network (WSN) in its simplest form can be defined as (Chong & Kumar,
2003; Akyildiz, Su, Sankarasubramaniam & Cayirci, 2002; Culler, Estrin & Srivastava, 2004)
a network of (possibly low-size and low-complex) devices denoted as nodes that can sense
the environment and communicate the information gathered from the monitored field (e.g.,
an area or volume) through wireless links; the data is forwarded, possibly via multiple hops
relaying to a sink (sometimes denoted as controller or monitor ) that can use it locally or is
connected to other networks (e.g., the Internet) through a gateway. The nodes can be
stationary or moving. They can be aware of their location or not. They can be homogeneous
or not. A traditional single-sink WSN is illustrated in Figure 1. Almost all scientific papers in
the literature deal with such a definition. This single-sink scenario suffers from the lack of
scalability: by increasing the number of nodes the amount of data gathered by the sink
increases and once its capacity is reached the network size can not be augmented. Moreover,
for reasons related to medium access control (MAC) and routing aspects, network
performance cannot be considered independent from the network size. A more general
scenario includes multiple sinks in the network (see Figure 2). Given a level of node density,
a larger number of sinks will decrease the probability of isolated clusters of nodes that
cannot deliver their data owing to unfortunate signal propagation conditions.

Fig. 1. Traditional single-sink WSN

In principle, a multiple-sink WSN can be scalable (i.e., the same performance can be
achieved even by increasing the number of nodes), while this is clearly not true for a single-
sink network. However, a multi-sink WSN does not represent a trivial extension of a single-
sink case for the network engineer. There might be mainly two different cases: (1) all sinks
are connected through a separate network (either wired or wireless), or (2) the sinks are
disconnected. In the former case, a node needs to forward the data collected to any element
in the set of sinks. From the protocol viewpoint, this means that a selection can be done
based on a suitable criterion (e.g., minimum delay, maximum throughput, minimum
number of hops, etc.). The presence of multiple sinks in this case ensures better network
performance with respect to the single-sink case (assuming the same number of nodes is
deployed over the same area), but the communication protocols must be more complex and
should be designed according to suitable criteria. In the second case, when the sinks are not
connected, the presence of multiple sinks tends to partition the monitored field into smaller
areas; however from the communication protocols viewpoint no significant changes must be
included, apart from simple sink discovery mechanisms. Clearly, the most general and
interesting case (because of the better potential performance) is the first one, with the sinks
connected through any type of mesh network, or via direct links with a common gateway.
Both the single-sink and multiple-sink networks introduced above do not include the
presence of actuators, that is, devices able to manipulate the environment rather than
observe it. WSANs are composed of both sensing nodes and actuators (see Figure 3). Once
more, the inclusion of actuators does not represent a simple extension of a WSN from the
communication protocol viewpoint. In fact the information flow must be reversed in this
case: the protocols should be able to manage many-to-one communications when sensors
provide data, and one-to-many flows when the actuators need to be addressed, or even one-
to-one links if a specific actuator has to be reached.


Fig. 2. Multi-sink WSN

The complexity of the protocols in this case is even larger. Given the very large number of
nodes that can constitute a WSAN (more than hundreds sometimes), it is clear that MAC
and the network layer are very relevant parts of the protocol stack. Tens of proposals
specifically designed for WSANs have been made in the past few years. The communication
protocols of a WSAN should also allow an easy deployment of nodes; the network must be
able to self-organize and self-heal when some local failures are encountered.

Fig. 3. Typical WSAN

2.1 Current and future research on WSANs
Many technical topics of WSANs are still considered by research as the current solutions are
known to be non-optimized, or too much constrained. From the physical layer viewpoint,
standardization is a key issue for success of WSAN markets. Currently the basic options for
building HW/SW platforms for WSANs are Bluetooth, IEEE 802.15.4 and 802.15.4a. At least,
most commercially available platforms use these three standards for the air interface. For
low data rate applications (250 Kbits on the air), IEEE 802.15.4 seems to be the most flexible
technology currently available. Clearly, the need to have low-complexity and low-cost
devices does not push research in the direction of advanced transmission techniques. MAC
and network layer have attracted a lot of attention in the past years and still deserve
investigation. In particular, combined approaches that jointly consider MAC and routing
seem to be very successful.
Topology creation, control and maintenance are very hot topics. Especially with IEEE
802.15.4, which allows creation of several types of topologies (stars, mesh, trees, cluster-
trees), these issues play a very significant role. Transport protocols are needed for WSANs
depending on the specific type of application. However, some of the most relevant issues
investigated by research in WSANs are cross layer, dealing with vertical functionalities:
security, localization, time synchronization. Basically, the research in the field of WSANs
started very recently with respect to other areas of the wireless communication society, as
broadcasting or cellular networks. The first IEEE papers on WSANs were published after the
turn of the Millennium. The first European projects on WSANs were financed after year
2001. In the US the research on WSANs was boosted a few years before. Many theoretical
issues still need a lot of investments.
The e-Mobility technology platform gathers all major players in the area of wireless and
mobile communications. A strategic research agenda was released and updated in 2006.
According to their views, by the year 2020 mobile and wireless communications will play a
central role in all aspects of European citizens lives, not just telephony, and will be a major
influence on Europes economy, wirelessly enabling every conceivable business endeavour
and personal lifestyle. The aim of research in the field can be summarized as follows: The
improvement of the individuals quality of life, achieved through the availability of an
environment for instant provision and access to meaningful, multi-sensory information and
content.
Environment means that the users will strongly interact with the environment that
surrounds them, for example by using devices for personal use, or by having the location as
a basis for many of the services to be used. This implies a totally different structure for the
networks. Also, the context recognized by the system and it acting dynamically on the
information is a major enabler for intelligent applications and services. This also means that
sensor networks and radio frequency identifications (RFIDs) are increasingly important.
Multi-sensory is related to all the users, devices, and also to the fact that the environment
will be capable of sensing the users presence. Also, virtual presence may be considered,
implying more sensory information being communicated, and an ideal of a rich
communication close to the quality achieved in interpersonal communications or direct
communications with another environment; this could also include non-invasive and
context-aware communication characterizing polite human interactions. Therefore, this
stretches mobile and wireless communications beyond radio and computer science into new
areas of science, like biology, medicine, psychology, sociology, and nano-technologies, and

Fig. 2. Multi-sink WSN

The complexity of the protocols in this case is even larger. Given the very large number of
nodes that can constitute a WSAN (more than hundreds sometimes), it is clear that MAC
and the network layer are very relevant parts of the protocol stack. Tens of proposals
specifically designed for WSANs have been made in the past few years. The communication
protocols of a WSAN should also allow an easy deployment of nodes; the network must be
able to self-organize and self-heal when some local failures are encountered.

Fig. 3. Typical WSAN

2.1 Current and future research on WSANs
Many technical topics of WSANs are still considered by research as the current solutions are
known to be non-optimized, or too much constrained. From the physical layer viewpoint,
standardization is a key issue for success of WSAN markets. Currently the basic options for
building HW/SW platforms for WSANs are Bluetooth, IEEE 802.15.4 and 802.15.4a. At least,
most commercially available platforms use these three standards for the air interface. For
low data rate applications (250 Kbits on the air), IEEE 802.15.4 seems to be the most flexible
technology currently available. Clearly, the need to have low-complexity and low-cost
devices does not push research in the direction of advanced transmission techniques. MAC
and network layer have attracted a lot of attention in the past years and still deserve
investigation. In particular, combined approaches that jointly consider MAC and routing
seem to be very successful.
Topology creation, control and maintenance are very hot topics. Especially with IEEE
802.15.4, which allows creation of several types of topologies (stars, mesh, trees, cluster-
trees), these issues play a very significant role. Transport protocols are needed for WSANs
depending on the specific type of application. However, some of the most relevant issues
investigated by research in WSANs are cross layer, dealing with vertical functionalities:
security, localization, time synchronization. Basically, the research in the field of WSANs
started very recently with respect to other areas of the wireless communication society, as
broadcasting or cellular networks. The first IEEE papers on WSANs were published after the
turn of the Millennium. The first European projects on WSANs were financed after year
2001. In the US the research on WSANs was boosted a few years before. Many theoretical
issues still need a lot of investments.
The e-Mobility technology platform gathers all major players in the area of wireless and
mobile communications. A strategic research agenda was released and updated in 2006.
According to their views, by the year 2020 mobile and wireless communications will play a
central role in all aspects of European citizens lives, not just telephony, and will be a major
influence on Europes economy, wirelessly enabling every conceivable business endeavour
and personal lifestyle. The aim of research in the field can be summarized as follows: The
improvement of the individuals quality of life, achieved through the availability of an
environment for instant provision and access to meaningful, multi-sensory information and
content.
Environment means that the users will strongly interact with the environment that
surrounds them, for example by using devices for personal use, or by having the location as
a basis for many of the services to be used. This implies a totally different structure for the
networks. Also, the context recognized by the system and it acting dynamically on the
information is a major enabler for intelligent applications and services. This also means that
sensor networks and radio frequency identifications (RFIDs) are increasingly important.
Multi-sensory is related to all the users, devices, and also to the fact that the environment
will be capable of sensing the users presence. Also, virtual presence may be considered,
implying more sensory information being communicated, and an ideal of a rich
communication close to the quality achieved in interpersonal communications or direct
communications with another environment; this could also include non-invasive and
context-aware communication characterizing polite human interactions. Therefore, this
stretches mobile and wireless communications beyond radio and computer science into new
areas of science, like biology, medicine, psychology, sociology, and nano-technologies, and

also requires full cooperation with other industries not traditionally associated with
communications.
Finally, the information should be multi-sensory and multi-modal, making use of all human
basic senses to properly capture context, mood, state of mind, and, for example, ones state
of health. Clearly, the realization of this vision of mobile and wireless communications
demands multi-disciplinary research and development, crossing the boundaries of the
above sciences and different industries. Also, the number of electronic sensors and RFIDs
surrounding us is quickly increasing. This will increase the amount of data traffic. The
future system will be complex, consisting of a multitude of service and network types
ranging across wireless sensor networks, personal area, local area, home networks, moving
networks to wide area networks. Therefore, the e-Mobility vision emphasizes the key role
played by WSANs as elements of a more complex system linking different types of access
technologies.
ARTEMIS (advanced research & technology for embedded intelligence and systems) is the
technology platform for embedded systems. The term embedded systems describes
electronic products, equipment or more complex systems, where the embedded computing
devices are not visible from the outside and are generally inaccessible by the user. The
sensor and actuator nodes of WSANs are embedded systems. According to the ARTEMIS
strategic research agenda, intelligent functions embedded in components and devices will
be a key factor in revolutionizing industrial production processes, from design to
manufacturing and distribution, particularly in the traditional sectors. These technologies
add intelligence to the control processes in manufacturing shop floors and improve the
logistic and distribution chains, resulting in an increasing productivity in a wide range of
industrial processes. The grand challenge in the area of sensors and actuators relates to the
support of huge amounts of input and output data envisaged in the application contexts
with minimal power requirements and fail-safe operation.

3. Data Fusion Techniques for WSANs
To fully exploit the potential of sensor networks, it is essential to develop energy-efficient
and bandwidth-efficient signal processing algorithms that can also be implemented in a
fully distributed manner. Distributed signal processing in a WSN has a communication
aspect not present in the traditional centralized signal processing framework, thus it differs
in several important aspects.
Sensor measurements are collected in a distributed fashion across the network.
This necessitates data sharing via inter-sensors communication. Given a low
energy budget per sensor, it is unrealistic for sensors to communicate all their full-
precision data samples with one another. Thus, local data compression becomes a
part of the distributed signal processing design. In contrast, in a traditional signal
processing framework where data is centrally collected, there is no need for
distributed data compression.
The design of optimal distributed signal processing algorithms depends on the
models used to describe: the nodes connectivity, the nodes distribution, the
knowledge of sensor noise distributions, the qualities of inter-sensors
communication channels, and the underlying application metrics. Distributed
signal processing over a wireless sensor network requires proper coordination and

planning of sensor computation as well as careful exploitation of the limited
communication capability per sensor. In other words, distributed signal processing
in sensor networks has communication aspects which are not present in the most of
traditional signal processing frameworks.
In a WSN, sensors may enter or leave the network dynamically, resulting in
unpredictable changes in network size and topology. This can be due to failure
between inter-sensors communication (propagation conditions, interference or
non-available communication channels), duty cycling, drained batteries or nodes
damages. This dynamism requires the necessity for distributed signal processing
algorithms to be robust to the changes in network topology or size. These
algorithms and protocols must also be robust to poor time synchronization across
the network and to inaccurate knowledge of sensor locations.
There are many theoretical challenges such as establishing models, metrics, bounds, and
algorithms for distributed multimodal sensor fusion, distributed management of sensor
networks including auto-configuration, energy-efficient application-specific protocol
designs, formal techniques for the study of architectures and protocols, representation of
information requirements, and sensor network capabilities on a common mathematical
framework that would enable efficient information filtering (Luo et al., 2006). From these
aspects, it is clear that the design of sensor networks under energy, bandwidth, and
application-specific constraints spans all layers of the protocol stacks and it is very
important to have a common framework enabling all these points be taken into account
even if with different approximations degrees.
In this view an example of cross-layer methodology of WSN design for environmental
monitoring will be shown in the following with particular emphasis on the impact of
distributed digital signal processing (DDSP) on the spatial process estimation error on one
side and network lifetime on the other side. Depending on the process under monitoring
and the goal of the WSAN, such as detection of distributed binary events and spatial process
estimation, several techniques can be pursued with envisaging of centralized and
distributed processing.

3.1 Distributed Estimation
Distributed estimation and tracking is one of the most fundamental collaborative
information processing problems in wireless sensor networks (WSN). Multi-sensor fusion
and tracking problems have a long history in signal processing, control theory, and robotics.
Moreover, estimation issues in wireless networks with packet-loss have been the center of
much attention lately. In most applications, the intelligent fusion of information from
geographically-dispersed sensor nodes, commonly known as distributed data fusion, is an
important issue. A related problem is the binary decentralized (or distributed) detection
problem, where a large number of identical sensor nodes deployed randomly over a wide
region, together with a global detector or fusion centre (FC), cooperatively undertake the
task of identifying the presence or absence of a phenomenon of interest (PoI) (see Figure 4).
Specifi cally, each node takes a local decision about the presence or absence of the PoI and
sends its decision to the FC which is responsible for the final decision based on the
information gathered from local sensors.


also requires full cooperation with other industries not traditionally associated with
communications.
Finally, the information should be multi-sensory and multi-modal, making use of all human
basic senses to properly capture context, mood, state of mind, and, for example, ones state
of health. Clearly, the realization of this vision of mobile and wireless communications
demands multi-disciplinary research and development, crossing the boundaries of the
above sciences and different industries. Also, the number of electronic sensors and RFIDs
surrounding us is quickly increasing. This will increase the amount of data traffic. The
future system will be complex, consisting of a multitude of service and network types
ranging across wireless sensor networks, personal area, local area, home networks, moving
networks to wide area networks. Therefore, the e-Mobility vision emphasizes the key role
played by WSANs as elements of a more complex system linking different types of access
technologies.
ARTEMIS (advanced research & technology for embedded intelligence and systems) is the
technology platform for embedded systems. The term embedded systems describes
electronic products, equipment or more complex systems, where the embedded computing
devices are not visible from the outside and are generally inaccessible by the user. The
sensor and actuator nodes of WSANs are embedded systems. According to the ARTEMIS
strategic research agenda, intelligent functions embedded in components and devices will
be a key factor in revolutionizing industrial production processes, from design to
manufacturing and distribution, particularly in the traditional sectors. These technologies
add intelligence to the control processes in manufacturing shop floors and improve the
logistic and distribution chains, resulting in an increasing productivity in a wide range of
industrial processes. The grand challenge in the area of sensors and actuators relates to the
support of huge amounts of input and output data envisaged in the application contexts
with minimal power requirements and fail-safe operation.

3. Data Fusion Techniques for WSANs
To fully exploit the potential of sensor networks, it is essential to develop energy-efficient
and bandwidth-efficient signal processing algorithms that can also be implemented in a
fully distributed manner. Distributed signal processing in a WSN has a communication
aspect not present in the traditional centralized signal processing framework, thus it differs
in several important aspects.
Sensor measurements are collected in a distributed fashion across the network.
This necessitates data sharing via inter-sensors communication. Given a low
energy budget per sensor, it is unrealistic for sensors to communicate all their full-
precision data samples with one another. Thus, local data compression becomes a
part of the distributed signal processing design. In contrast, in a traditional signal
processing framework where data is centrally collected, there is no need for
distributed data compression.
The design of optimal distributed signal processing algorithms depends on the
models used to describe: the nodes connectivity, the nodes distribution, the
knowledge of sensor noise distributions, the qualities of inter-sensors
communication channels, and the underlying application metrics. Distributed
signal processing over a wireless sensor network requires proper coordination and

planning of sensor computation as well as careful exploitation of the limited
communication capability per sensor. In other words, distributed signal processing
in sensor networks has communication aspects which are not present in the most of
traditional signal processing frameworks.
In a WSN, sensors may enter or leave the network dynamically, resulting in
unpredictable changes in network size and topology. This can be due to failure
between inter-sensors communication (propagation conditions, interference or
non-available communication channels), duty cycling, drained batteries or nodes
damages. This dynamism requires the necessity for distributed signal processing
algorithms to be robust to the changes in network topology or size. These
algorithms and protocols must also be robust to poor time synchronization across
the network and to inaccurate knowledge of sensor locations.
There are many theoretical challenges such as establishing models, metrics, bounds, and
algorithms for distributed multimodal sensor fusion, distributed management of sensor
networks including auto-configuration, energy-efficient application-specific protocol
designs, formal techniques for the study of architectures and protocols, representation of
information requirements, and sensor network capabilities on a common mathematical
framework that would enable efficient information filtering (Luo et al., 2006). From these
aspects, it is clear that the design of sensor networks under energy, bandwidth, and
application-specific constraints spans all layers of the protocol stacks and it is very
important to have a common framework enabling all these points be taken into account
even if with different approximations degrees.
In this view an example of cross-layer methodology of WSN design for environmental
monitoring will be shown in the following with particular emphasis on the impact of
distributed digital signal processing (DDSP) on the spatial process estimation error on one
side and network lifetime on the other side. Depending on the process under monitoring
and the goal of the WSAN, such as detection of distributed binary events and spatial process
estimation, several techniques can be pursued with envisaging of centralized and
distributed processing.

3.1 Distributed Estimation
Distributed estimation and tracking is one of the most fundamental collaborative
information processing problems in wireless sensor networks (WSN). Multi-sensor fusion
and tracking problems have a long history in signal processing, control theory, and robotics.
Moreover, estimation issues in wireless networks with packet-loss have been the center of
much attention lately. In most applications, the intelligent fusion of information from
geographically-dispersed sensor nodes, commonly known as distributed data fusion, is an
important issue. A related problem is the binary decentralized (or distributed) detection
problem, where a large number of identical sensor nodes deployed randomly over a wide
region, together with a global detector or fusion centre (FC), cooperatively undertake the
task of identifying the presence or absence of a phenomenon of interest (PoI) (see Figure 4).
Specifi cally, each node takes a local decision about the presence or absence of the PoI and
sends its decision to the FC which is responsible for the final decision based on the
information gathered from local sensors.


Fig. 4. Example of distributed detection scenario

Two problems have to be considered: the design of the decision rule at the FC and the
design of the local sensor signal processing strategies. In the case of perfect knowledge of
system parameters the design of the decision rule at the FC is a well-established task. The
design of the local sensor decision rule, that is, the likelihood ratio test (LRT) threshold in
binary detection, is more challenging due to the distributed nature of the system. In fact, the
optimal choice of each sensor LRT threshold is coupled to each other node threshold,
although nodes are not in general fully connected due to propagation effects or energy
constraints. Recently it has been demonstrated that under the asymptotic regime (i.e., large
number of nodes), the identical LRT threshold rule at the sensors provides the optimal error
exponent if local sensor observations are independent and identically distributed (Sung,
Tong & Swami, 2005). For a more complete overview of decentralized detection the reader is
recommend to read Varshney, 1997; Signal Processing Magazine, Special Issue , 2006. Unlike
in classical decentralized detection problems (Blum, Kassam & Poor, 1997; Varshney, 1997),
greater challenges exist in a WSN setting. There are stringent power constraints for each
node, and communication channels from nodes to the FC are severely bandwidth-
constrained. In addition, the communication channels are no longer lossless (e.g, fading,
noise and, possibly, interference are present), and the observation at each sensor node is
spatially varying (Sung et al., 2005; Niu & Varshney, 2005). Recently, there has been great
interest in cooperative communication (Winters, 1987; Sendonaris, Erkip & Aazhang, 2003).
One may also exploit diversity associated with spatially distributed users, or simply
cooperative diversity , in WSNs. In these networks, multiple sensor nodes pool their
resources in a distributed manner to enhance the reliability of the transmission link. Specifi
cally, in the context of decentralized detection, cooperation allows sensor nodes to exchange
information and to continuously update their local decisions until consensus is reached
across the nodes (Quek, Dardari & Win, 2006c; Quek, Dardari & Win, 2006b; Quek, Dardari
& Win, 2006a). For example, cooperation in decentralized detection can be accomplished via
the use of Parley algorithm (Swaszek & Willett, 1995). This algorithm has been shown to
converge to a global decision after sufficient number of iterations when certain conditions
are met. However, without a fully-connected network and given that the sensor
observations are spatially varying, Parley algorithm may result in convergence to a wrong
decision at most of the nodes.

3.2 Kalman Filtering
Although WSNs present attractive features, challenges associated with the scarcity of
bandwidth and power in wireless communications have to be addressed. To perform state
estimation, sensors may share these observations with each other or communicate them to a
fusion center for centralized processing. In either scenario, the communication cost in terms
of bandwidth and power required to convey observations is large enough to merit
attention.. The Kalman filter is a very popular fusion method. Decentralized Kalman
filtering (Speyer, 1979; Rao, Whyte & Sheen, 1993) involves state estimation using a set of
local Kalman filters that communicate with all other nodes. The Kalman filter estimates the
state x of a discrete-time controlled process that is ruled by the state-space model:

x [k]) = A.x [k-1] +w [k-1] (1)

The system is influenced by process noise denoted w. The state dynamics determine the
linear operator A. The state contributes to the observation y, which also includes a
stochastic, additive measurement noise v:

y [k] = C. x[k] +v [k] (2)

The process and measurement noises are assumed to be Normal processes with known
variances W and V. The information flow is all-to-all with approxmiate communication
complexity of O(n
2
) which can cause scalability issues for large-sclae WSNs.

4. Event Classification and Pattern Recognition
The reality in WSNs is that so far tremendous potential of the WSN has only been
demonstrated for humble applications such as meter reading in buildings and basic forms of
ecological monitoring. Reaping the full potential of this technology requires a second level
of algorithms, which at present is missing. Current techniques solve the more immediate
problem of conveying sensory data to a central entity known as the base station for most of
the processing. This approach is not scalable and has potential to cast suspicions of big
brother analysis. The WSN must therefore localize the computations within its monitored
regions and communicate the computed results to small/roaming devices directly. Doing so
will alleviate the scalability issue by localizing the computations and address the privacy
and reliability concern through direct communications occurring within a decentralized
model. Algorithms, such as the ones in this chapter, would help integrate vast networks of
sensors into intelligent macro-scopes for observing our surroundings. These will bring
unprecedented capabilities within our reach that transform the way we deal with
phenomena occurring over large distances and inaccessible regions.
In order to achieve the broad objectives stated above, an initial step will be to develop a
level of computability within the WSN whereby sensor readings can be instantly translated
into event patterns and then rapidly analysed by the network (locally). This approach will
entail two-fold benefit. On one hand it will enhance event detection e.g. surveillance, and
target location capabilities and on the other hand it will aid in development of advanced
threat detection systems for WSN, e.g. ones that can function like our biological immune
system. The challenge is to evolve an approach, which can successfully detect complex real

Fig. 4. Example of distributed detection scenario

Two problems have to be considered: the design of the decision rule at the FC and the
design of the local sensor signal processing strategies. In the case of perfect knowledge of
system parameters the design of the decision rule at the FC is a well-established task. The
design of the local sensor decision rule, that is, the likelihood ratio test (LRT) threshold in
binary detection, is more challenging due to the distributed nature of the system. In fact, the
optimal choice of each sensor LRT threshold is coupled to each other node threshold,
although nodes are not in general fully connected due to propagation effects or energy
constraints. Recently it has been demonstrated that under the asymptotic regime (i.e., large
number of nodes), the identical LRT threshold rule at the sensors provides the optimal error
exponent if local sensor observations are independent and identically distributed (Sung,
Tong & Swami, 2005). For a more complete overview of decentralized detection the reader is
recommend to read Varshney, 1997; Signal Processing Magazine, Special Issue , 2006. Unlike
in classical decentralized detection problems (Blum, Kassam & Poor, 1997; Varshney, 1997),
greater challenges exist in a WSN setting. There are stringent power constraints for each
node, and communication channels from nodes to the FC are severely bandwidth-
constrained. In addition, the communication channels are no longer lossless (e.g, fading,
noise and, possibly, interference are present), and the observation at each sensor node is
spatially varying (Sung et al., 2005; Niu & Varshney, 2005). Recently, there has been great
interest in cooperative communication (Winters, 1987; Sendonaris, Erkip & Aazhang, 2003).
One may also exploit diversity associated with spatially distributed users, or simply
cooperative diversity , in WSNs. In these networks, multiple sensor nodes pool their
resources in a distributed manner to enhance the reliability of the transmission link. Specifi
cally, in the context of decentralized detection, cooperation allows sensor nodes to exchange
information and to continuously update their local decisions until consensus is reached
across the nodes (Quek, Dardari & Win, 2006c; Quek, Dardari & Win, 2006b; Quek, Dardari
& Win, 2006a). For example, cooperation in decentralized detection can be accomplished via
the use of Parley algorithm (Swaszek & Willett, 1995). This algorithm has been shown to
converge to a global decision after sufficient number of iterations when certain conditions
are met. However, without a fully-connected network and given that the sensor
observations are spatially varying, Parley algorithm may result in convergence to a wrong
decision at most of the nodes.

3.2 Kalman Filtering
Although WSNs present attractive features, challenges associated with the scarcity of
bandwidth and power in wireless communications have to be addressed. To perform state
estimation, sensors may share these observations with each other or communicate them to a
fusion center for centralized processing. In either scenario, the communication cost in terms
of bandwidth and power required to convey observations is large enough to merit
attention.. The Kalman filter is a very popular fusion method. Decentralized Kalman
filtering (Speyer, 1979; Rao, Whyte & Sheen, 1993) involves state estimation using a set of
local Kalman filters that communicate with all other nodes. The Kalman filter estimates the
state x of a discrete-time controlled process that is ruled by the state-space model:

x [k]) = A.x [k-1] +w [k-1] (1)

The system is influenced by process noise denoted w. The state dynamics determine the
linear operator A. The state contributes to the observation y, which also includes a
stochastic, additive measurement noise v:

y [k] = C. x[k] +v [k] (2)

The process and measurement noises are assumed to be Normal processes with known
variances W and V. The information flow is all-to-all with approxmiate communication
complexity of O(n
2
) which can cause scalability issues for large-sclae WSNs.

4. Event Classification and Pattern Recognition
The reality in WSNs is that so far tremendous potential of the WSN has only been
demonstrated for humble applications such as meter reading in buildings and basic forms of
ecological monitoring. Reaping the full potential of this technology requires a second level
of algorithms, which at present is missing. Current techniques solve the more immediate
problem of conveying sensory data to a central entity known as the base station for most of
the processing. This approach is not scalable and has potential to cast suspicions of big
brother analysis. The WSN must therefore localize the computations within its monitored
regions and communicate the computed results to small/roaming devices directly. Doing so
will alleviate the scalability issue by localizing the computations and address the privacy
and reliability concern through direct communications occurring within a decentralized
model. Algorithms, such as the ones in this chapter, would help integrate vast networks of
sensors into intelligent macro-scopes for observing our surroundings. These will bring
unprecedented capabilities within our reach that transform the way we deal with
phenomena occurring over large distances and inaccessible regions.
In order to achieve the broad objectives stated above, an initial step will be to develop a
level of computability within the WSN whereby sensor readings can be instantly translated
into event patterns and then rapidly analysed by the network (locally). This approach will
entail two-fold benefit. On one hand it will enhance event detection e.g. surveillance, and
target location capabilities and on the other hand it will aid in development of advanced
threat detection systems for WSN, e.g. ones that can function like our biological immune
system. The challenge is to evolve an approach, which can successfully detect complex real

life patterns arising from heterogeneous data sets (generated by different types of sensors)
in real-time. In this regard, events of interest need to be correlated to specific pattern classes
of our definition. Furthermore the asymptotic limits of the approach need to be carefully
examined. For instance how far will our scheme scale with increases in the complexity of the
sensory patterns and the size of the network? What will be the trade off between the
schemes sensitivity and false alarms? Can the scheme degrade gracefully with disruptions
to network nodes and wireless links?

4.1 Graph Neuron
DHGN extends the functionalities and capabilities of Graph Neuron (GN) (Khan &
Mihailescu, 2004) and Hierarchical Graph Neuron (HGN) (Khan & Nasution, 2008)
algorithms. Figure 5 depicts a GN array, which is capable of converting the spatial/temporal
patterns into a simple graph-based representation, in which input patterns are compared
with the edges in the graph for memorization or recall operations.

Fig. 5. Store and recall within a GN

GN is an associative memory algorithm, which implements a scalable AM device through
its parallel in-network processing framework (Khan & Isreb, 2004). GN has been tested in
pattern recognition applications within different types of distributed environments (Khan,
Muhammad Amin, 2008a). GN makes use of graph-based model for pattern learning and
recognition. One of the peculiarities of this technique is the employment of parallel in-
network processing capabilities to address scalability issues effectively, which are of
primary concern in distributed approaches. Graph neuron may be conceived as a directed
graph in which processing nodes of the GN array are mapped to vertex set V of the graph,
and links between nodes are mapped to the set of edges of the graph, E.
The communication is limited to the neighbouring nodes to decrease the amount of
overhead involved and as a result, any significant increase in the number of nodes will not
increase the communication overhead significantly. The information in the pattern space can
be presented to each of the nodes in the graph in the form of (position, value) pair. Thus, GN
array is capable of converting the spatial/temporal patterns into a simple graph-based
representation in which input patterns are compared with the edges in the graph for
memorization or possible recall operations (See Figure 6). Among various applications of

the GN-based AM, an input pattern in GN pattern recognition may represent bit elements of
an image (Khan & Muhammad Amin, 2007) or a stimulus/signal spike produced within a
network intrusion detection application (Khan, Baig & Baqer, 2006).

Fig. 6. An eight node GN network is in the process of storing patterns: P1 (RED), P2 (BLUE),
P3 (BLACK), and P4 (GREEN)

4.2 Hierarchical Graph Neuron
In order to solve the issue of the crosstalk in GN model due to the limited perspective of
GNs, the capabilities of perceiving GN neighbors in each GN was expanded in a new model
called Hierarchical Graph Neuron (HGN) to prevent pattern interference (Khan & Nasution,
2008). Figure 7 illustrates a HGN scheme for pattern recognition with a pattern of size 7 and
two possible values within the pattern being stored in the network.

life patterns arising from heterogeneous data sets (generated by different types of sensors)
in real-time. In this regard, events of interest need to be correlated to specific pattern classes
of our definition. Furthermore the asymptotic limits of the approach need to be carefully
examined. For instance how far will our scheme scale with increases in the complexity of the
sensory patterns and the size of the network? What will be the trade off between the
schemes sensitivity and false alarms? Can the scheme degrade gracefully with disruptions
to network nodes and wireless links?

4.1 Graph Neuron
DHGN extends the functionalities and capabilities of Graph Neuron (GN) (Khan &
Mihailescu, 2004) and Hierarchical Graph Neuron (HGN) (Khan & Nasution, 2008)
algorithms. Figure 5 depicts a GN array, which is capable of converting the spatial/temporal
patterns into a simple graph-based representation, in which input patterns are compared
with the edges in the graph for memorization or recall operations.

Fig. 5. Store and recall within a GN

GN is an associative memory algorithm, which implements a scalable AM device through
its parallel in-network processing framework (Khan & Isreb, 2004). GN has been tested in
pattern recognition applications within different types of distributed environments (Khan,
Muhammad Amin, 2008a). GN makes use of graph-based model for pattern learning and
recognition. One of the peculiarities of this technique is the employment of parallel in-
network processing capabilities to address scalability issues effectively, which are of
primary concern in distributed approaches. Graph neuron may be conceived as a directed
graph in which processing nodes of the GN array are mapped to vertex set V of the graph,
and links between nodes are mapped to the set of edges of the graph, E.
The communication is limited to the neighbouring nodes to decrease the amount of
overhead involved and as a result, any significant increase in the number of nodes will not
increase the communication overhead significantly. The information in the pattern space can
be presented to each of the nodes in the graph in the form of (position, value) pair. Thus, GN
array is capable of converting the spatial/temporal patterns into a simple graph-based
representation in which input patterns are compared with the edges in the graph for
memorization or possible recall operations (See Figure 6). Among various applications of

the GN-based AM, an input pattern in GN pattern recognition may represent bit elements of
an image (Khan & Muhammad Amin, 2007) or a stimulus/signal spike produced within a
network intrusion detection application (Khan, Baig & Baqer, 2006).

Fig. 6. An eight node GN network is in the process of storing patterns: P1 (RED), P2 (BLUE),
P3 (BLACK), and P4 (GREEN)

4.2 Hierarchical Graph Neuron
In order to solve the issue of the crosstalk in GN model due to the limited perspective of
GNs, the capabilities of perceiving GN neighbors in each GN was expanded in a new model
called Hierarchical Graph Neuron (HGN) to prevent pattern interference (Khan & Nasution,
2008). Figure 7 illustrates a HGN scheme for pattern recognition with a pattern of size 7 and
two possible values within the pattern being stored in the network.

Fig. 7. Hierarchical GNs

The communication between HGN nodes is accomplished through different stages with the
following fashion:
Stage I. Each GN in the base layer will receive input pairs (value, command)
sequentially. In this regard, all GNs are pre-programmed to receive these values. It
should be mentioned that all the GNs in the same column will receive the same
input pair, and all the GNs in the composition will receive the same command,
whether store or recall.
Stage II. Upon receiving the input, all active GNs will send a (column, row) pair to
all the GNs on preceding and succeeding columns. It should be noted that, those
GNs with matched IDs will issue the response; however other GNs enter the
passive state, while still receiving reports from adjacent columns.
Stage III. At this stage, all the active GNs at the edges have received one report
message, while other active GNs would have received two reports from their
adjacent columns. Then, each GN can determine the bias entry using input values,
and report messages. Here, the GN can decide whether to store input in memory or
declare it as a recall based on the command value.
Stage IV. At this stage, after updating the bias array, each active GN will send its
(row, bias index, command) value to all the GNs in the same column in the upper
layer.
Finally, all the active GNs will return the (level, column, bias index) value back as the
response. The process is continued by the next layer above the base layer. After repetitive
execution of stages 2 to 4 by the remaining layers, the process will be stopped at one layer
which is below the top layer. In general, the process consists of (S 3) / 2 number of
iterations for the last three stages, in which S denotes the size of the pattern space.

4.3 Distributed Hierarchical Graph Neuron
We then extended the HGN by dividing and distributing the recognition processes over the
network. This distributed scheme minimizes the number of processing nodes by reducing
the number of levels within the HGN. Figure 8 depicts the divide-and-distribute
transformation from a monolithic HGN composition (top) to a DHGN configuration
(bottom) for processing the same 35-bit patterns.

Fig. 8. Transformation of HGN structure (top) into an equivalent DHGN structure (bottom)

The base of the HGN structure in Figure 8 represents the size of the pattern. Note that the
base of HGN structure is equivalent to the cumulative base of all the DHGN
subnets/clusters. This transformation of HGN into equivalent DHGN composition allows
on the average 80% reduction in the number of processing nodes required for the
recognition process. Therefore, DHGN is able to substantially reduce the computational
resource requirement for pattern recognition process from 648 processing nodes to 126 for
the case shown in Figure 8.
DHGN is in fact a single-cycle learning associative memory (AM) algorithm for pattern
recognition. DHGN employs the collaborative-comparison learning approach (Khan &
Muhammad Amin, 2009) in pattern recognition. It lowers the complexity of recognition
processes by reducing the number of processing nodes (Khan & Muhammad Amin, 2007).
In addition, pattern recognition using DHGN algorithm is improved through a two-level
recognition process, which applies recognition at sub-pattern level and then recognition at
the overall pattern level (See Figure 9). DHGN is already implemented in a grid
environment (Khan & Muhammad Amin, 2008). We plan to adopt DHGN algorithm for our
novel data estimation model in WSNs. DHGN allows the recognition process to be
conducted in a smaller sub-pattern domain, hence minimizing the number of processing
nodes which in turn reduces the complexity of pattern analysis. In addition, the recognition
process performed using DHGN algorithm is unique in a way that each subnet is only
responsible for memorizing a portion of the pattern (rather than the entire pattern). A
collection of these subnets is able to form a distributed memory structure for the entire
pattern. This feature enables recognition to be performed in parallel and independently. The
decoupled nature of the sub-domains (subnets) is the key feature that brings dynamic
scalability to data estimation within WSNs.


Fig. 7. Hierarchical GNs

The communication between HGN nodes is accomplished through different stages with the
following fashion:
Stage I. Each GN in the base layer will receive input pairs (value, command)
sequentially. In this regard, all GNs are pre-programmed to receive these values. It
should be mentioned that all the GNs in the same column will receive the same
input pair, and all the GNs in the composition will receive the same command,
whether store or recall.
Stage II. Upon receiving the input, all active GNs will send a (column, row) pair to
all the GNs on preceding and succeeding columns. It should be noted that, those
GNs with matched IDs will issue the response; however other GNs enter the
passive state, while still receiving reports from adjacent columns.
Stage III. At this stage, all the active GNs at the edges have received one report
message, while other active GNs would have received two reports from their
adjacent columns. Then, each GN can determine the bias entry using input values,
and report messages. Here, the GN can decide whether to store input in memory or
declare it as a recall based on the command value.
Stage IV. At this stage, after updating the bias array, each active GN will send its
(row, bias index, command) value to all the GNs in the same column in the upper
layer.
Finally, all the active GNs will return the (level, column, bias index) value back as the
response. The process is continued by the next layer above the base layer. After repetitive
execution of stages 2 to 4 by the remaining layers, the process will be stopped at one layer
which is below the top layer. In general, the process consists of (S 3) / 2 number of
iterations for the last three stages, in which S denotes the size of the pattern space.

4.3 Distributed Hierarchical Graph Neuron
We then extended the HGN by dividing and distributing the recognition processes over the
network. This distributed scheme minimizes the number of processing nodes by reducing
the number of levels within the HGN. Figure 8 depicts the divide-and-distribute
transformation from a monolithic HGN composition (top) to a DHGN configuration
(bottom) for processing the same 35-bit patterns.

Fig. 8. Transformation of HGN structure (top) into an equivalent DHGN structure (bottom)

The base of the HGN structure in Figure 8 represents the size of the pattern. Note that the
base of HGN structure is equivalent to the cumulative base of all the DHGN
subnets/clusters. This transformation of HGN into equivalent DHGN composition allows
on the average 80% reduction in the number of processing nodes required for the
recognition process. Therefore, DHGN is able to substantially reduce the computational
resource requirement for pattern recognition process from 648 processing nodes to 126 for
the case shown in Figure 8.
DHGN is in fact a single-cycle learning associative memory (AM) algorithm for pattern
recognition. DHGN employs the collaborative-comparison learning approach (Khan &
Muhammad Amin, 2009) in pattern recognition. It lowers the complexity of recognition
processes by reducing the number of processing nodes (Khan & Muhammad Amin, 2007).
In addition, pattern recognition using DHGN algorithm is improved through a two-level
recognition process, which applies recognition at sub-pattern level and then recognition at
the overall pattern level (See Figure 9). DHGN is already implemented in a grid
environment (Khan & Muhammad Amin, 2008). We plan to adopt DHGN algorithm for our
novel data estimation model in WSNs. DHGN allows the recognition process to be
conducted in a smaller sub-pattern domain, hence minimizing the number of processing
nodes which in turn reduces the complexity of pattern analysis. In addition, the recognition
process performed using DHGN algorithm is unique in a way that each subnet is only
responsible for memorizing a portion of the pattern (rather than the entire pattern). A
collection of these subnets is able to form a distributed memory structure for the entire
pattern. This feature enables recognition to be performed in parallel and independently. The
decoupled nature of the sub-domains (subnets) is the key feature that brings dynamic
scalability to data estimation within WSNs.


Fig. 9. DHGN distributed pattern recognition architecture

In fact, we envisage a vast associative memory network of wireless sensors to support user
defined search and classification algorithms. Data in WSN may be viewed as uniform or
non-uniform in nature. A uniform data block will incur the same computational cost as all
the others. However in many instances, at least one data subset might be required to be
known with a better quality of approximation or greater detail than the others, e.g. it may be
invoked in the query process more often than the others. In this case the cost associated with
data subset will be different. Larger variation can lead to inefficient use of the distributed
resources. Strength of DHGN lies in the processing of non-uniform data patterns as it
implements a finely distributable framework at the smallest (atomic) logical sub-pattern
level. The results are easily obtained by summation at the overall pattern level. DHGN is
also a highly scalable algorithm that incorporates content addressable memory within a
clustered framework. Hence algorithmic strengths of the current estimation approaches can
be investigated for the first time in combination with DHGNs single-cycle learning
mechanism. This one shot approach will allow real-time in-network data manipulation;
essential feature for data processing in WSNs.

5. Performance Metrics
By redesigning data processing models, data records are treated as patterns which enable
data storage and retrieval by association over and above the existing simple data referential
mechanisms. Hence, processing the input data and handling the dynamic load is handled by
using a distributed pattern recognition approach that is implemented through the integration
of loosely-coupled computational networks, followed by a divide-and-distribute approach
that allows distribution of these networks within WSNs dynamically.

5.1 Scalable Simulation of the DHGN Application in WSNs:
Several tests are performed to ensure the accuracy of DHGN algorithm while evaluating its
effectiveness in terms of complexity, timing and fault rate. It is essential that these software
programs are fully integrated and tested as proof of concept that our approach will indeed
work when applied to a dynamic WSN. In order to achieve this goal, we will formulate a
wireless sensor network environment for executing our algorithms over very large numbers
of GN nodes. Doing so will allow us to determine the asymptotical limits of our approach,
find any problems that only arise when the algorithms are scaled-up, and test for robustness
and scalability in face of dynamic changes to WSN configuration.
Our scheme relies on communications among adjacent nodes. The decentralized content
location schemes will be implemented for discovering adjacent nodes in minimum number
of hops. A GN based algorithm for optimally distributing DHGN subnets (clusters or sub-
domains) among the WSN nodes will be provided to automate the boot-strapping of the
distributed application over the network. Doing so will also provide means for investigating
dynamic load balancing over the network. For this matter, a database of stored patterns is
constructed. Patterns can be assumed as abstract notations for various parameters in the
network. The user is able to configure the length and number of elements in the pattern, and
based on these parameters, the program will generate a user-specified number of random
patterns. At this stage, the user can manually enter patterns to be checked against the
database. The program is not only capable of determining whether the pattern is matched
with the previous stored ones or not, but is also able to determine the sub-patterns of the
input pattern which have been visited before. This capability provides a tool for
determining the level of distortion in the input pattern, which in turn adds to the robustness
of the proposed algorithm significantly.

5.2 One shot pattern recognition within WSNs:
In contrast with hierarchical models proposed in the literature, DHGNs pattern matching
capability and the small response time, that remains insensitive to the increases in the
number of stored patterns, makes this approach ideal for WSNs. Moreover, the DHGN does
not require definition of rules or manual interventions by the operator for setting of
thresholds to achieve the desired results, nor does it require heuristics entailing iterative
operations for memorization and recall of patterns. In addition, this approach allows
induction of new patterns in a fixed number of steps. Whilst doing so it exhibits a high level
of scalability i.e. the performance and accuracy do not degrade as the number of stored
pattern increases over time. Its pattern recognition capability remains comparable with
contemporary approaches. Furthermore all computations are completed within the pre-
defined number of steps and as such the approach implements one shot, i.e. single-cycle or
single-pass, learning. Tests have shown that the scheme has an acceptable level of pattern
recognition accuracy in this context (Khan & Muhammad Amin, 2008b). As it is clearly
depicted in Figure 10, the systems response (i.e. recognition times) does not trend upwards
with the increase in the number of stored patterns. In other words, performance of the
algorithm was not affected by the increase in the number of memorized patterns or the WSN
network size.


Fig. 9. DHGN distributed pattern recognition architecture

In fact, we envisage a vast associative memory network of wireless sensors to support user
defined search and classification algorithms. Data in WSN may be viewed as uniform or
non-uniform in nature. A uniform data block will incur the same computational cost as all
the others. However in many instances, at least one data subset might be required to be
known with a better quality of approximation or greater detail than the others, e.g. it may be
invoked in the query process more often than the others. In this case the cost associated with
data subset will be different. Larger variation can lead to inefficient use of the distributed
resources. Strength of DHGN lies in the processing of non-uniform data patterns as it
implements a finely distributable framework at the smallest (atomic) logical sub-pattern
level. The results are easily obtained by summation at the overall pattern level. DHGN is
also a highly scalable algorithm that incorporates content addressable memory within a
clustered framework. Hence algorithmic strengths of the current estimation approaches can
be investigated for the first time in combination with DHGNs single-cycle learning
mechanism. This one shot approach will allow real-time in-network data manipulation;
essential feature for data processing in WSNs.

5. Performance Metrics
By redesigning data processing models, data records are treated as patterns which enable
data storage and retrieval by association over and above the existing simple data referential
mechanisms. Hence, processing the input data and handling the dynamic load is handled by
using a distributed pattern recognition approach that is implemented through the integration
of loosely-coupled computational networks, followed by a divide-and-distribute approach
that allows distribution of these networks within WSNs dynamically.

5.1 Scalable Simulation of the DHGN Application in WSNs:
Several tests are performed to ensure the accuracy of DHGN algorithm while evaluating its
effectiveness in terms of complexity, timing and fault rate. It is essential that these software
programs are fully integrated and tested as proof of concept that our approach will indeed
work when applied to a dynamic WSN. In order to achieve this goal, we will formulate a
wireless sensor network environment for executing our algorithms over very large numbers
of GN nodes. Doing so will allow us to determine the asymptotical limits of our approach,
find any problems that only arise when the algorithms are scaled-up, and test for robustness
and scalability in face of dynamic changes to WSN configuration.
Our scheme relies on communications among adjacent nodes. The decentralized content
location schemes will be implemented for discovering adjacent nodes in minimum number
of hops. A GN based algorithm for optimally distributing DHGN subnets (clusters or sub-
domains) among the WSN nodes will be provided to automate the boot-strapping of the
distributed application over the network. Doing so will also provide means for investigating
dynamic load balancing over the network. For this matter, a database of stored patterns is
constructed. Patterns can be assumed as abstract notations for various parameters in the
network. The user is able to configure the length and number of elements in the pattern, and
based on these parameters, the program will generate a user-specified number of random
patterns. At this stage, the user can manually enter patterns to be checked against the
database. The program is not only capable of determining whether the pattern is matched
with the previous stored ones or not, but is also able to determine the sub-patterns of the
input pattern which have been visited before. This capability provides a tool for
determining the level of distortion in the input pattern, which in turn adds to the robustness
of the proposed algorithm significantly.

5.2 One shot pattern recognition within WSNs:
In contrast with hierarchical models proposed in the literature, DHGNs pattern matching
capability and the small response time, that remains insensitive to the increases in the
number of stored patterns, makes this approach ideal for WSNs. Moreover, the DHGN does
not require definition of rules or manual interventions by the operator for setting of
thresholds to achieve the desired results, nor does it require heuristics entailing iterative
operations for memorization and recall of patterns. In addition, this approach allows
induction of new patterns in a fixed number of steps. Whilst doing so it exhibits a high level
of scalability i.e. the performance and accuracy do not degrade as the number of stored
pattern increases over time. Its pattern recognition capability remains comparable with
contemporary approaches. Furthermore all computations are completed within the pre-
defined number of steps and as such the approach implements one shot, i.e. single-cycle or
single-pass, learning. Tests have shown that the scheme has an acceptable level of pattern
recognition accuracy in this context (Khan & Muhammad Amin, 2008b). As it is clearly
depicted in Figure 10, the systems response (i.e. recognition times) does not trend upwards
with the increase in the number of stored patterns. In other words, performance of the
algorithm was not affected by the increase in the number of memorized patterns or the WSN
network size.


Fig. 10. Pattern recognition timings for storing a large number of 16KB images

5.3 One Shot Learning versus Back Propagation Network:
A comparative analysis with a Back-Propagation (BP) network layer and the input layer of
the HGN highlights the ability of the GN layer to assimilate new patterns into the older
patterns constructively (Khan & Nasution, 2008). The GN layer continues to improve its
accuracy, as more and more patterns are stored, and exhibits better accuracy and
performance as compared to the BP algorithm (See Figures 11 and 12). For performing a
comparison approach, a BP perceptron layer was tested for single-cycle learning using a
piecewise-linear activation function with a slope of 2.5 and a threshold of 0 for optimal
results. These results were then compared with a single GN layer as shown in Figures 11
and 12 respectively. The perceptron layer and the GN layer were both trained once-only
using up to 10,000 fully randomized patterns with a uniform distribution. Pattern sizes of 10
and 20 with random distortions (noise) were used in the tests. Each of these pattern sizes
was associated with 16 possible values. Hence these pattern sizes corresponded to 16^10 or
40 bit (2^40) and 16^20 or 80 bit (2^80) patterns respectively. The accuracy was calculated as
the percentage difference between the pattern size and the number of bits with recall errors.

Fig. 11. Recall Accuracy of BP algorithm, after being fed with up to 10,000 fully randomized
patterns

Figure 12. Recall Accuracy of single-cycle GN, after being fed with up to 10,000 fully
randomized patterns (Adapted from [28])


Fig. 10. Pattern recognition timings for storing a large number of 16KB images

5.3 One Shot Learning versus Back Propagation Network:
A comparative analysis with a Back-Propagation (BP) network layer and the input layer of
the HGN highlights the ability of the GN layer to assimilate new patterns into the older
patterns constructively (Khan & Nasution, 2008). The GN layer continues to improve its
accuracy, as more and more patterns are stored, and exhibits better accuracy and
performance as compared to the BP algorithm (See Figures 11 and 12). For performing a
comparison approach, a BP perceptron layer was tested for single-cycle learning using a
piecewise-linear activation function with a slope of 2.5 and a threshold of 0 for optimal
results. These results were then compared with a single GN layer as shown in Figures 11
and 12 respectively. The perceptron layer and the GN layer were both trained once-only
using up to 10,000 fully randomized patterns with a uniform distribution. Pattern sizes of 10
and 20 with random distortions (noise) were used in the tests. Each of these pattern sizes
was associated with 16 possible values. Hence these pattern sizes corresponded to 16^10 or
40 bit (2^40) and 16^20 or 80 bit (2^80) patterns respectively. The accuracy was calculated as
the percentage difference between the pattern size and the number of bits with recall errors.

Fig. 11. Recall Accuracy of BP algorithm, after being fed with up to 10,000 fully randomized
patterns

Figure 12. Recall Accuracy of single-cycle GN, after being fed with up to 10,000 fully
randomized patterns (Adapted from [28])


It may be noted from Figures 11 and 12 that both the BP and the GN networks achieved
comparable accuracy for the first 500 patterns. It can be seen from Figure 12 that the
accuracy of GN in recognizing previously stored patterns increased as more patterns were
stored (greater improvement with more one shot learning experiences). The GN in both the
cases achieved above 80% accuracy after all the 10,000 patterns (with noise) had been
presented. On the other hand, Figure 11 shows that the accuracy of BP network failed to
improve after 2,000 patterns. Also, the recall accuracy of the perceptron layer noticeably
deteriorated as the larger pattern size of 20 (80 bits) was introduced.

5.4 Superior Scalability:
Another important aspect of HGN is that it can remain highly scalable. In fact, its response
time to store or recall operations is not affected by an increase in the size of the stored
pattern database. Figure 13 illustrates the results from a HGN simulation in this regard.
Figure 14 is based on the mathematical model derived in (Khan & Natusion, 2008), which
corroborates the simulation results. The flat slopes in Figures 13 and 14, which show that the
response times remain insensitive to the increase in stored patterns, represent the high
scalability of the scheme. Hence, the issue of computational overhead increase due to the
increase in the size of pattern space or number of stored patterns, as is the case in many
graph-based matching algorithms will be alleviated in HGN, while the solution can be
achieved within fixed number of steps of single cycle learning and recall.

Fig. 13. Actual recall time for an analysed composition

Fig. 14. Estimated recall time for a HGN composition

6. Discussion and Future Research
Our proposed approach is proven to work as a real-time pattern recogniser for the WSN.
This is the only scheme which can operate within resource-constrained nodes of a WSN and
still deliver speed and accuracies comparable to conventional pattern analysers.
Furthermore our approach allows new patterns to be added at anytime without recourse to
retraining. In contrast with contemporary distributed estimation approaches, our approach
allows induction of new input patterns in a fixed number of steps. Whilst doing so it
exhibits a high level of scalability i.e. the performance and accuracy do not degrade as the
number of stored pattern increases over time. Its pattern recognition capability remains
comparable with contemporary approaches. Furthermore all computations are completed
within the pre-defined number of steps and as such the approach implements one shot, i.e.
single-cycle or single-pass, learning. The one shot learning within this method is achieved
by side-stepping the commonly used error/energy minimization and random walk
approaches. The network functions as a matrix that holds all possible solutions for the
problem domain. The network after applying our algorithm can find the solution in a single-
cycle (i.e. fixed number of steps). Our approach finds and refines the initial solution by
passing the results through a pyramidal hierarchy of similar arrays. In doing so it
eliminates/resolves pattern defects, with distortions up to 20% being tolerated. Previously
encountered patterns are revealed whilst new patterns are memorized without loss of stored
information. In fact the pattern recognition accuracy continues to improve as the network
processes more sensory inputs. HGN though highly suited for the WSN is limited by the
well-known problem afflicting the hierarchical schemes i.e. large increases in network nodes
as the pattern size and/or complexity increases. The distributed version of this scheme,
DHGN, retains HGNs one-shot learning characteristic and reduces the computational
complexity of our hierarchal scheme by distributing the recognition process into smaller
clusters i.e. dividing-and-distributing simple pattern recognition processes across the
network. The computational complexity of the scheme is significantly reduced and the
accuracy is generally improved by using this strategy. The distributed scheme also lowers
storage capacity requirements per node and incurs lesser communication cost thus
improving the response-time characteristic. A comparison with the Hopfield network shows
that our approach generally improves speed of recognition and accuracy. The distributed
scheme also allows better control of the network resources varying from coarse grained to
very fine grained networks. This scheme compares well with contemporary approaches
such as SOM and SVM in terms of speed and accuracy.
There are several benefits and advantages in our DHGN implementation for event detection
and distributed estimation within WSN network. Our approach offers low memory
consumption for event data storage using simple bias array representation. Furthermore,
this scheme only stores sub-patterns or patterns that are related to normal event, rather than
keeping the records of all occurring events. We have also shown that our approach is most
effective for small sub-pattern size, since it uses only a small portion of the memory space in
a typical physical sensor node in WSN network. In addition to this efficient memory usage,
DHGN also eliminates the need for complex computations for event classification technique.
With the adoption of single-cycle learning and adjacency comparison approaches, DHGN
implements a non-iterative and lightweight computational mechanism for event recognition
and classification. The results of our performance analysis have also shown that DHGN
recognition time increases linearly with an increase in the number of processing elements

It may be noted from Figures 11 and 12 that both the BP and the GN networks achieved
comparable accuracy for the first 500 patterns. It can be seen from Figure 12 that the
accuracy of GN in recognizing previously stored patterns increased as more patterns were
stored (greater improvement with more one shot learning experiences). The GN in both the
cases achieved above 80% accuracy after all the 10,000 patterns (with noise) had been
presented. On the other hand, Figure 11 shows that the accuracy of BP network failed to
improve after 2,000 patterns. Also, the recall accuracy of the perceptron layer noticeably
deteriorated as the larger pattern size of 20 (80 bits) was introduced.

5.4 Superior Scalability:
Another important aspect of HGN is that it can remain highly scalable. In fact, its response
time to store or recall operations is not affected by an increase in the size of the stored
pattern database. Figure 13 illustrates the results from a HGN simulation in this regard.
Figure 14 is based on the mathematical model derived in (Khan & Natusion, 2008), which
corroborates the simulation results. The flat slopes in Figures 13 and 14, which show that the
response times remain insensitive to the increase in stored patterns, represent the high
scalability of the scheme. Hence, the issue of computational overhead increase due to the
increase in the size of pattern space or number of stored patterns, as is the case in many
graph-based matching algorithms will be alleviated in HGN, while the solution can be
achieved within fixed number of steps of single cycle learning and recall.

Fig. 13. Actual recall time for an analysed composition

Fig. 14. Estimated recall time for a HGN composition

6. Discussion and Future Research
Our proposed approach is proven to work as a real-time pattern recogniser for the WSN.
This is the only scheme which can operate within resource-constrained nodes of a WSN and
still deliver speed and accuracies comparable to conventional pattern analysers.
Furthermore our approach allows new patterns to be added at anytime without recourse to
retraining. In contrast with contemporary distributed estimation approaches, our approach
allows induction of new input patterns in a fixed number of steps. Whilst doing so it
exhibits a high level of scalability i.e. the performance and accuracy do not degrade as the
number of stored pattern increases over time. Its pattern recognition capability remains
comparable with contemporary approaches. Furthermore all computations are completed
within the pre-defined number of steps and as such the approach implements one shot, i.e.
single-cycle or single-pass, learning. The one shot learning within this method is achieved
by side-stepping the commonly used error/energy minimization and random walk
approaches. The network functions as a matrix that holds all possible solutions for the
problem domain. The network after applying our algorithm can find the solution in a single-
cycle (i.e. fixed number of steps). Our approach finds and refines the initial solution by
passing the results through a pyramidal hierarchy of similar arrays. In doing so it
eliminates/resolves pattern defects, with distortions up to 20% being tolerated. Previously
encountered patterns are revealed whilst new patterns are memorized without loss of stored
information. In fact the pattern recognition accuracy continues to improve as the network
processes more sensory inputs. HGN though highly suited for the WSN is limited by the
well-known problem afflicting the hierarchical schemes i.e. large increases in network nodes
as the pattern size and/or complexity increases. The distributed version of this scheme,
DHGN, retains HGNs one-shot learning characteristic and reduces the computational
complexity of our hierarchal scheme by distributing the recognition process into smaller
clusters i.e. dividing-and-distributing simple pattern recognition processes across the
network. The computational complexity of the scheme is significantly reduced and the
accuracy is generally improved by using this strategy. The distributed scheme also lowers
storage capacity requirements per node and incurs lesser communication cost thus
improving the response-time characteristic. A comparison with the Hopfield network shows
that our approach generally improves speed of recognition and accuracy. The distributed
scheme also allows better control of the network resources varying from coarse grained to
very fine grained networks. This scheme compares well with contemporary approaches
such as SOM and SVM in terms of speed and accuracy.
There are several benefits and advantages in our DHGN implementation for event detection
and distributed estimation within WSN network. Our approach offers low memory
consumption for event data storage using simple bias array representation. Furthermore,
this scheme only stores sub-patterns or patterns that are related to normal event, rather than
keeping the records of all occurring events. We have also shown that our approach is most
effective for small sub-pattern size, since it uses only a small portion of the memory space in
a typical physical sensor node in WSN network. In addition to this efficient memory usage,
DHGN also eliminates the need for complex computations for event classification technique.
With the adoption of single-cycle learning and adjacency comparison approaches, DHGN
implements a non-iterative and lightweight computational mechanism for event recognition
and classification. The results of our performance analysis have also shown that DHGN
recognition time increases linearly with an increase in the number of processing elements

(PEs) within the network. This simply Lightweight Event Detection Scheme using
Distributed Hierarchical Graph Neuron in Wireless Sensor Networks reveals that DHGNs
computational complexity is also scalable with an increase in the size of the sub-patterns.
DHGN is a distributed pattern recognition algorithm. By having this distributed
characteristic, DHGN would be readily-deployable over a distributed network. With such
feature, DHGN has the ability to perform as a front-end detection scheme for event
detection within WSN. Through divide-and-distribute approach, complex events could be
perceived as a composition of events occurring at specific time and location. Our approach
eventually would be able to be used in event tracking as our proposed scheme has been
demonstrated to perform efficiently within an event detection scheme such as forest fire
detection using WSN. Despite all its benefits, DHGN has its own limitations. Firstly, DHGN
simple data representation would requires significant advanced pre-processing at the front-
end of the system. This might not be viable for strictly-resource constrained sensor nodes,
where processing capability is very limited. In addition, DHGN single-hop communication
for event detection scheme is not viable for large area monitoring, due to high possibility of
communication error due to data packet loss during transmission. Our existing DHGN
implementation has also been focusing on supervised classification. However, there is a
need for unsupervised classification technique to be deployed for rapid event detection
scheme. Overcoming the DHGN distributed event detection scheme limitations would be
the path of our future research direction. We intend to look into event tracking scheme
using DHGN distributed detection mechanism, as well as providing unsupervised
classification capability for rapid and robust event detection scheme. Furthermore, we are
looking forward into implementation of this scheme in large-area monitoring using multi-
hop communication strategy.

7. Conclusion
The development of distributed estimation scheme within WSN has been made viable with the
advancement in communication, computational, and sensor technologies. However, existing
estimation/recognition algorithms fail to achieve optimum performance in a distributed environment,
due to its tightly-coupled and computationally intensive nature. In this chapter, we have presented our
readily-distributable distributed estimation scheme for WSN network which is known as Distributed
Hierarchical Graph Neuron (DHGN). Throughout our studies, we discover that DHGN is able to
perform recognition and classification processes with limited training data and within a one-shot
learning. These DHGN features have given added-value for implementing this scheme within a
lightweight distributed network such as WSN. Current implementation of DHGN in event detection
using WSN has been focusing on the front-end processing, in which detection could be carried out
earlier using the available wireless sensor nodes. Our approach differs from other existing estimation
schemes in which major processing steps are conducted at the base station. By having a front-end
detection, our proposed scheme is able to alleviate the computational costs experienced by the
centralised-processing undertaken by the base station. In this chapter, we have also discussed the
advantages and limitations of our proposed scheme. The future direction of this research lies in the
development of a complete distributed estimation scheme that incorporates front-end detection and
back-end complex event analysis. We foresee our DHGN distributed estimation scheme as a complete
estimation scheme and recognition tool that is deployable over different types of event detection
schemes on WSN networks.

8. References
Akyildiz , I. F. , Su , W. , Sankarasubramaniam & Cayirci , E. ( 2002 ). A survey on sensor
networks. IEEE Communications Magazin , pp. 102 114 .
Blum , R. S. , Kassam , S. A. & Poor , H. V. ( 1997 ). Distributed detection with multiple
sensors Part II: Advanced topics. Proceedings of IEEE, pp. 64 79.
Chong , C.-Y. & Kumar, S. P. ( 2003 ). Sensor networks: evolution, opportunities, and
challenges . Proceedings of IEEE, pp. 1247 1256 .
Culler , D. , Estrin , D. & Srivastava , M. ( 2004 ). Overview of sensor networks. IEEE
Computer, pp. 41 49 .
Khan, A.I., Mihailescu, P. (2004), Parallel Pattern Recognition Computations within a
Wireless Sensor Network, In Proceedings of the 17th Intl. Conference on Pattern
Recognition (ICPR'04), Cambridge, United Kingdom: IEEE Computer Society.
Khan A.I., Isreb M., Spindler R.S. (2004). A parallel distributed application of the wireless
sensor network, In Proceedings of the 7th Intl Conf. on High Performance Computing
and Grid.
Khan, A.I., Baig Z.A., Baqer M., (2006). A pattern recognition scheme for distributed denial
of service (DDOS) attacks in wireless sensor networks, In Proceedings of the 18th
International Conference on Pattern Recognition.
Khan, A.I., Muhamad Amin, A.H. (2007). One Shot Associative Memory Method for
Distorted Pattern Recognition, In 20th Australian Joint Conference on Artificial
Intelligence, Gold Coast, Australia, pp. 705-709.
Khan, A.I., Muhamad Amin A.H., (2008a). Commodity-Grid Based Distributed Pattern
Recognition Framework, Sixth Australasian Symposium on Grid Computing and e-
Research (AUSGRID 2008), Wollongong, NSW, Australia.
Khan, A.I., Muhamad Amin, A.H., (2008b). Single-cycle image recognition using an adaptive
granularity associative memory network, Lecture Notes in Artificial Intelligence, vol
5360, Springer-Verlag, Berlin Germany, pp. 386-392.
Khan, A.I., Nasution, B. B. (2008) ,IEEE Transactions on Neural Networks, 212-229.
Khan, A.I., Muhamad Amin, A.H., (2009). Collaborative-Comparison Learning for Complex
Event Detection using Distributed Hierarchical Graph Neuron (DHGN) Approach
in Wireless Sensor Network, In 22nd Australasian Joint Conference on Artificial
Intelligence (AI'09), Melbourne, Australia.
Luo , Z-Q. , Gastpar , M. , Liu , J. & Swami , A. ( 2006 ). Distributed signal processing in
sensor networks [from the guest editors] . I EEE Signal Processing Magazine, pp.
14 15.
Niu , R. , Chen , B. & Varshney , P. ( 2006 ). Fusion of decisions transmitted over Rayleigh
fading channels in wireless sensor networks. IEEE Transactions on Signal Processing,
pp. 1018 1026 .
Quek, T. Q.-S., Dardari, D. & Win, M. Z. (2006a ). Cooperation in bandwidth-constrained
wireless sensor networks. In Proceedings of the IEEE International Workshop Wireless
Ad Hoc Sensor Networks, New York, NY, USA .
Quek, T. Q.-S., Dardari, D. & Win, M. Z. (2006 b). Energy efficiency of cooperative dense
wireless sensor networks. In Proceedings of the International Wireless Communication
sand Mobile Computing Conference, IWCMC 2006, Vancouver, Canada , pp. 13231330.

(PEs) within the network. This simply Lightweight Event Detection Scheme using
Distributed Hierarchical Graph Neuron in Wireless Sensor Networks reveals that DHGNs
computational complexity is also scalable with an increase in the size of the sub-patterns.
DHGN is a distributed pattern recognition algorithm. By having this distributed
characteristic, DHGN would be readily-deployable over a distributed network. With such
feature, DHGN has the ability to perform as a front-end detection scheme for event
detection within WSN. Through divide-and-distribute approach, complex events could be
perceived as a composition of events occurring at specific time and location. Our approach
eventually would be able to be used in event tracking as our proposed scheme has been
demonstrated to perform efficiently within an event detection scheme such as forest fire
detection using WSN. Despite all its benefits, DHGN has its own limitations. Firstly, DHGN
simple data representation would requires significant advanced pre-processing at the front-
end of the system. This might not be viable for strictly-resource constrained sensor nodes,
where processing capability is very limited. In addition, DHGN single-hop communication
for event detection scheme is not viable for large area monitoring, due to high possibility of
communication error due to data packet loss during transmission. Our existing DHGN
implementation has also been focusing on supervised classification. However, there is a
need for unsupervised classification technique to be deployed for rapid event detection
scheme. Overcoming the DHGN distributed event detection scheme limitations would be
the path of our future research direction. We intend to look into event tracking scheme
using DHGN distributed detection mechanism, as well as providing unsupervised
classification capability for rapid and robust event detection scheme. Furthermore, we are
looking forward into implementation of this scheme in large-area monitoring using multi-
hop communication strategy.

7. Conclusion
The development of distributed estimation scheme within WSN has been made viable with the
advancement in communication, computational, and sensor technologies. However, existing
estimation/recognition algorithms fail to achieve optimum performance in a distributed environment,
due to its tightly-coupled and computationally intensive nature. In this chapter, we have presented our
readily-distributable distributed estimation scheme for WSN network which is known as Distributed
Hierarchical Graph Neuron (DHGN). Throughout our studies, we discover that DHGN is able to
perform recognition and classification processes with limited training data and within a one-shot
learning. These DHGN features have given added-value for implementing this scheme within a
lightweight distributed network such as WSN. Current implementation of DHGN in event detection
using WSN has been focusing on the front-end processing, in which detection could be carried out
earlier using the available wireless sensor nodes. Our approach differs from other existing estimation
schemes in which major processing steps are conducted at the base station. By having a front-end
detection, our proposed scheme is able to alleviate the computational costs experienced by the
centralised-processing undertaken by the base station. In this chapter, we have also discussed the
advantages and limitations of our proposed scheme. The future direction of this research lies in the
development of a complete distributed estimation scheme that incorporates front-end detection and
back-end complex event analysis. We foresee our DHGN distributed estimation scheme as a complete
estimation scheme and recognition tool that is deployable over different types of event detection
schemes on WSN networks.

8. References
Akyildiz , I. F. , Su , W. , Sankarasubramaniam & Cayirci , E. ( 2002 ). A survey on sensor
networks. IEEE Communications Magazin , pp. 102 114 .
Blum , R. S. , Kassam , S. A. & Poor , H. V. ( 1997 ). Distributed detection with multiple
sensors Part II: Advanced topics. Proceedings of IEEE, pp. 64 79.
Chong , C.-Y. & Kumar, S. P. ( 2003 ). Sensor networks: evolution, opportunities, and
challenges . Proceedings of IEEE, pp. 1247 1256 .
Culler , D. , Estrin , D. & Srivastava , M. ( 2004 ). Overview of sensor networks. IEEE
Computer, pp. 41 49 .
Khan, A.I., Mihailescu, P. (2004), Parallel Pattern Recognition Computations within a
Wireless Sensor Network, In Proceedings of the 17th Intl. Conference on Pattern
Recognition (ICPR'04), Cambridge, United Kingdom: IEEE Computer Society.
Khan A.I., Isreb M., Spindler R.S. (2004). A parallel distributed application of the wireless
sensor network, In Proceedings of the 7th Intl Conf. on High Performance Computing
and Grid.
Khan, A.I., Baig Z.A., Baqer M., (2006). A pattern recognition scheme for distributed denial
of service (DDOS) attacks in wireless sensor networks, In Proceedings of the 18th
International Conference on Pattern Recognition.
Khan, A.I., Muhamad Amin, A.H. (2007). One Shot Associative Memory Method for
Distorted Pattern Recognition, In 20th Australian Joint Conference on Artificial
Intelligence, Gold Coast, Australia, pp. 705-709.
Khan, A.I., Muhamad Amin A.H., (2008a). Commodity-Grid Based Distributed Pattern
Recognition Framework, Sixth Australasian Symposium on Grid Computing and e-
Research (AUSGRID 2008), Wollongong, NSW, Australia.
Khan, A.I., Muhamad Amin, A.H., (2008b). Single-cycle image recognition using an adaptive
granularity associative memory network, Lecture Notes in Artificial Intelligence, vol
5360, Springer-Verlag, Berlin Germany, pp. 386-392.
Khan, A.I., Nasution, B. B. (2008) ,IEEE Transactions on Neural Networks, 212-229.
Khan, A.I., Muhamad Amin, A.H., (2009). Collaborative-Comparison Learning for Complex
Event Detection using Distributed Hierarchical Graph Neuron (DHGN) Approach
in Wireless Sensor Network, In 22nd Australasian Joint Conference on Artificial
Intelligence (AI'09), Melbourne, Australia.
Luo , Z-Q. , Gastpar , M. , Liu , J. & Swami , A. ( 2006 ). Distributed signal processing in
sensor networks [from the guest editors] . I EEE Signal Processing Magazine, pp.
14 15.
Niu , R. , Chen , B. & Varshney , P. ( 2006 ). Fusion of decisions transmitted over Rayleigh
fading channels in wireless sensor networks. IEEE Transactions on Signal Processing,
pp. 1018 1026 .
Quek, T. Q.-S., Dardari, D. & Win, M. Z. (2006a ). Cooperation in bandwidth-constrained
wireless sensor networks. In Proceedings of the IEEE International Workshop Wireless
Ad Hoc Sensor Networks, New York, NY, USA .
Quek, T. Q.-S., Dardari, D. & Win, M. Z. (2006 b). Energy efficiency of cooperative dense
wireless sensor networks. In Proceedings of the International Wireless Communication
sand Mobile Computing Conference, IWCMC 2006, Vancouver, Canada , pp. 13231330.

Quek, T. Q.-S., Dardari, D. & Win, M. Z. (2006c), Energy efficiency of dense wireless sensor
networks: To cooperate or not to cooperate. In Proceedings of the IEEE International
Conference on Communications, ICC 2006, Istanbul, Turkey, Vol. 10, pp. 44794484.
Rao, B. S. Y., Durrant-Whyte, H.F., and Sheen, J. A. (1993). A fully decentralized multi-
sensor system for tracking and surveillance. International Journal of Robotics Research,
2044.
Sendonaris , A. , Erkip , E. & Aazhang , B. ( 2003). User cooperation diversity Part II:
Implementation aspects and performance analysis. IEEE Transactions on
communications, pp. 1939 1948.
Speyer. L. (1979). Computation and transmission requirements for a decentralized linear-
quadratic-gaussian control problem. IEEE Transactions on Automatic Control, 266
269.
Sung , Y. , Tong , L. & Swami , A. ( 2005 ). Asymptotic locally optimal detector for large-scale
sensor networks under the Poisson regime. IEEE Transactions on Signal Processing,
pp. 2005 2017.
Swaszek, P. & Willett , P. ( 1995 ). Parley as an approach to distributed detection. IEEE
Transactions on Aerospace and Electronic Systems, pp. 447 457.
Welsh, M. , Malan, D. , Duncan, B., Fulford-Jones, T., and Moulton, S. (2004). Wireless
Sensor Networks for Emergency Medical Care, GE Global Research Conference,
Harvard University and Boston University School of Medicine, Boston, MA.
Winters, J. H. ( 1987 ). On the capacity of radio communication systems with diversity in
Rayleigh fading environment. IEEE Journal Selected Areas in Communication, pp.
871 878 . SAC

Multimedia Data Processing and Delivery in Wireless Sensor Networks 449
Multimedia Data Processing and Delivery in Wireless Sensor Networks
Javier Molina, Javier M. Mora-Merchan, Julio Barbancho and Carlos Leon
0
Multimedia Data Processing and Delivery
in Wireless Sensor Networks
Javier Molina, Javier M. Mora-Merchan, Julio Barbancho and Carlos Leon
University of Seville (Department of Electronic Technology)
Spain
1. Introduction
In the last fewyears, multimedia processing in Wireless Sensor Networks (WSNs) has become
a promising technology. It has the potential to enable a large class of applications, most of
themrelated to surveillance and locating (i.e. target detection and tracking, border protection,
patient and elderly assistance, people and object identication, environment monitoring, re
detection, industrial control, ...). To achieve an effective Quality of Service (QoS) in multime-
dia applications, special node and network capabilities are required. For example, compared
to normal WSN nodes, multimedia nodes need additional hardware resources for memory,
processing capability, transmission rate and energy. A Wireless Multimedia Sensor Network
(WMSN) is a special WSN made up of several multimedia sensor nodes, specially designed to
retrieve multimedia content such as video and audio streams, still images, and scalar sensor
data from the environment.
In this paper, we focus on a different, but also very practical and common, sort of wireless
network: the Heterogeneous Networks, where multimedia and non-multimedia nodes deliver
data. In this scenario, the non-multimedia node constraints have to be taken into account to
deliver multimedia data through multi-hop paths.
Generally, collaborative processing makes no sense in this case. Although in-sensor multime-
dia processing is a fundamental topic in order to obtain an effective data reduction.
Currently, research challenges in designing multimedia applications on WSNs include, but
are not limited to the following:
QoS requirements. Streaming media, system snapshots, audio/video store and play-
back applications have different requirements with respect to delay, jitter, and loss tol-
erance.
Bandwidth. WMSNs require a bandwidth that is orders of magnitude higher than that
supported by currently available sensors.
Power. Compared to traditional WSNs, power consumption is greater in multimedia
applications because of high volumes of data, high transmission rates, and extensive
processing.
In-network processing support, to efciently extract relevant information from multi-
media data (e.g. panoramic image fusion, target identication and location).
23
In-node processing support, to compress data, signal analysis and features extraction
(singular points, region segmentation, object detection, ...)
Cross-layer design. An effective optimization of all the above parameters involve cross-
layer protocol design ranging from Application to Physical Layer.
New hardware design to better manage the energy with high QoS (i.e. power supply,
microcontroller (MCUs) architectures and energy harvesting).
In this paper, we propose a newWMSNmodel that includes the Heterogeneous Network. For
this reason, main topics of Wireless Sensor Networks are described in section 2, and the most
relevant issues in multimedia networks are summed up in section 3. These issues are: stan-
dard network architectures, node constraints, compression techniques, memory, bandwidth,
and energy consumption.
Different hardware structure alternatives for heterogeneous nodes are introduced in section 4.
Moreover, a simple application algorithm is proposed to enable the low-end MCUs to capture
multimedia data, or collaborate in delivering.
2. General vision on the design of Wireless Sensor Network Applications
Because of cost and size, the nodes of a WSNexhibit resource constraints in terms of CPUpro-
cessing capacity, memory, bandwidth and energy. Figure 1 shows a node hardware structure,
and table 1 sums up a comparative study of popular MCUs, RF chips, and platforms.
Fig. 1. Node hardware architecture (Soro & Heinzelman, 2009)
Many WSNs are designed for environmental sensing applications, acquiring data from scalar
sensors (temperature, humidity, light, etc). In those cases where vector data sensing are re-
quired, the vector dimension is often low, for example movement sensing (accelerometers),
wind data (anemometers), velocity and positioning (GPS sensors). In contrast, multimedia
sensor nodes are characterized by audio and video streaming, and still image data.
Software development for WSNs nodes is a complex issue. Many researchers program the
modes from scratch, using operating system components, specic middleware, or by higher
programming abstractions (Mottola & Picco, 2010). Figure 2 summarizes a general software
model on a WSNs node platform.
Sensor Node Name Micro controller Transceiver Progra+DataMemory
BTnode ATmega128L CC1000 64k+180K
EPIC mote MSP430 CC2420 10k
EyesIFX v2 MSP430F1611 TDA5250 10k+48k
FireFly ATmega128L CC2420 8k
GWnode PIC18LF8722 BiM 64k
Imote 2.0 PXA271ARM CC2420 256k
Mica2 ATmega128L CC1000 64k+180k
TelosB MSP430F1611 CC2420 10k+48k
TinyNode MSP430 XE1205 8k
XYZ ML67 CC2420 32k
Table 1. WSN Platforms
MAC- Medium Access Control is placed jointly to the operating systems because, in contrast
to other wireless technologies where MAC protocol is done in hardware, in WSNs it is typi-
cally implemented mostly in software, using the low-level language associated to the operat-
ing system. MAC protocols must guarantee an efcient access to the communication media
while carefully managing the energy budget available in the node. The latter goal is typi-
cally achieved by switching the radio to a low-power mode based on the current transmission
schedule.
APPLICATION
System Services
Programming Abstractions
MAC OPERATING SYSTEM
HARDWARE
C
o
m
m
u
n
i
c
a
t
i
o
n
S
t
o
r
a
g
e
D
a
t
a

a
d
q
u
i
s
i
t
i
o
n
a
n
d

p
r
r
o
c
e
s
s
i
n
g
T
i
m
e

s
y
n
c
h
r
o
n
i
z
a
t
i
o
n
R
e
p
r
o
g
r
a
m
m
i
n
g
L
o
c
a
l
i
z
a
t
i
o
n
.

.

.
Fig. 2. Node software model
The operating system is essentially a library linked to the application code to generate an
executable binary code. This way, the program memory resource is reduced to a minimum,
and only necessary hardware specic routines are included. So far, many operating systems
In-node processing support, to compress data, signal analysis and features extraction
(singular points, region segmentation, object detection, ...)
Cross-layer design. An effective optimization of all the above parameters involve cross-
layer protocol design ranging from Application to Physical Layer.
New hardware design to better manage the energy with high QoS (i.e. power supply,
microcontroller (MCUs) architectures and energy harvesting).
In this paper, we propose a newWMSNmodel that includes the Heterogeneous Network. For
this reason, main topics of Wireless Sensor Networks are described in section 2, and the most
relevant issues in multimedia networks are summed up in section 3. These issues are: stan-
dard network architectures, node constraints, compression techniques, memory, bandwidth,
and energy consumption.
Different hardware structure alternatives for heterogeneous nodes are introduced in section 4.
Moreover, a simple application algorithm is proposed to enable the low-end MCUs to capture
multimedia data, or collaborate in delivering.
2. General vision on the design of Wireless Sensor Network Applications
Because of cost and size, the nodes of a WSNexhibit resource constraints in terms of CPUpro-
cessing capacity, memory, bandwidth and energy. Figure 1 shows a node hardware structure,
and table 1 sums up a comparative study of popular MCUs, RF chips, and platforms.
Fig. 1. Node hardware architecture (Soro & Heinzelman, 2009)
Many WSNs are designed for environmental sensing applications, acquiring data from scalar
sensors (temperature, humidity, light, etc). In those cases where vector data sensing are re-
quired, the vector dimension is often low, for example movement sensing (accelerometers),
wind data (anemometers), velocity and positioning (GPS sensors). In contrast, multimedia
sensor nodes are characterized by audio and video streaming, and still image data.
Software development for WSNs nodes is a complex issue. Many researchers program the
modes from scratch, using operating system components, specic middleware, or by higher
programming abstractions (Mottola & Picco, 2010). Figure 2 summarizes a general software
model on a WSNs node platform.
Sensor Node Name Micro controller Transceiver Progra+DataMemory
BTnode ATmega128L CC1000 64k+180K
EPIC mote MSP430 CC2420 10k
EyesIFX v2 MSP430F1611 TDA5250 10k+48k
FireFly ATmega128L CC2420 8k
GWnode PIC18LF8722 BiM 64k
Imote 2.0 PXA271ARM CC2420 256k
Mica2 ATmega128L CC1000 64k+180k
TelosB MSP430F1611 CC2420 10k+48k
TinyNode MSP430 XE1205 8k
XYZ ML67 CC2420 32k
Table 1. WSN Platforms
MAC- Medium Access Control is placed jointly to the operating systems because, in contrast
to other wireless technologies where MAC protocol is done in hardware, in WSNs it is typi-
cally implemented mostly in software, using the low-level language associated to the operat-
ing system. MAC protocols must guarantee an efcient access to the communication media
while carefully managing the energy budget available in the node. The latter goal is typi-
cally achieved by switching the radio to a low-power mode based on the current transmission
schedule.
APPLICATION
System Services
Programming Abstractions
MAC OPERATING SYSTEM
HARDWARE
C
o
m
m
u
n
i
c
a
t
i
o
n
S
t
o
r
a
g
e
D
a
t
a

a
d
q
u
i
s
i
t
i
o
n
a
n
d

p
r
r
o
c
e
s
s
i
n
g
T
i
m
e

s
y
n
c
h
r
o
n
i
z
a
t
i
o
n
R
e
p
r
o
g
r
a
m
m
i
n
g
L
o
c
a
l
i
z
a
t
i
o
n
.

.

.
Fig. 2. Node software model
The operating system is essentially a library linked to the application code to generate an
executable binary code. This way, the program memory resource is reduced to a minimum,
and only necessary hardware specic routines are included. So far, many operating systems
SAMPLING PRE-PROCESSING
FILTERING TYPE-CASTING
CONVERTION
SAMPLING RATE
ADC
DATA PROCESSING
FEATURES
EXTRACTION
DATA
FUSION
COMPRESSION
APPICATION-
SPECIFIC
PROCESSING
DATA ADQUISITION AND PROCESSING SERVICES
S
E
N
S
O
R
S
Fig. 3. Data acquisition and processing service
and specic program languages have been proposed. NesC language (Gay et al., 2003) for
TinyOS (Hill et al., 2000) are the most popular. Alternatives include Contiki (Dunkels et al.,
2004), SOS (Han et al., n.d.), Mantis (Abrach et al., 2003), RETOS (Cha et al., 2007), LiteOS
(Cao & Abdelzaher, n.d.), t-Kernel (Gu, 2006), or NANOrk (Eswaran et al., 2005). Some of
the above operating systems (e.g., SOS, LiteOS, and Contiki) also provide dynamic linking
capabilities, i.e., new code modules can be added at run-time to the application running on
a node. Dynamic linking is particularly important in supporting wireless reprogramming of
the WSN.
System Services are programs designed to support applications. Typically, an applica-
tion requires Location, Time Synchronization, Storage, Communication and Data acquisi-
tion/processing services. Location services estimate the own node position or the location of
a target (He et al., 2003; Mao et al., 2007; Patwari et al., 2005). Time synchronization is usually
required for data time-stamping, to measure the TOA (Time of Arrival) of a signal, to man-
age hibernate-wake up cycles, etc. Many algorithms have been proposed for this (Dai & Han,
2004; Elson & Rmer, 2003; Marti et al., 2004). Communication services deliver data reliably,
manage the network and optimize QoS and energy savings. The researchers community has
developed many different algorithms for routing, cluster management, etc. Although, the
industry has developed the Zigbee standard, it is probably the most widely accepted one so
far.
In most applications, sensor networks are deployed once and intended to operate unattended
for a long period of time. Reprogramming capabilities facilitate the management of applica-
tion changes or updates (Rubio et al., 2007). Because of the data memory constraints in the
nodes, it is not possible to save a monolithic program with many applications before being
burned. Instead of this, small packets of code can be reprogrammed and can also be reliably
delivered through the network. There are two basic schemes to reprogram sensor networks:
code dissemination and code acquisition. The rst one is initiated by the administrator to re-
program all the devices in the network. In the second one, the nodes request and reprogram
a packet dynamically. Data acquisition is not always considered and programmed as a ser-
vice. However, it is especially relevant in multimedia applications. Figure 3 shows a software
Phenomena Sample rate (/Hz)
Very low frequency
Atmospheric temperature 0.0171
Barometric pressure 0.0171
Low frequency
Heart rate 0.83.2
Volcanic infra-sound 2080
Natural seismic vibration 0.2100
Mid frequency (100Hz 1000 Hz)
Earthquake vibrations 100-160
ECG (heart electrical activity) 100250
High frequency (>1 kHz)
Breathing sounds 1005k
Industrial vibrations 40k
Audio (human hearing range) 1544k
Audio (muzzle shock-wave) 1M
Video (digital television) 10M
Table 2. Sensor sampling rates of different phenomena.
model for data acquisition and processing service with three stages: Sampling, Pre-processing
and Data processing.
In the sampling stage, a calibrated set of sensors are sampled, and the row data are time-
stamped, if necessary. Sensor calibration is an important issue. Traditional methodologies can
not be applied because of the cost of a manual and node-by-node adjusting. The sampling
process is red by either an event or periodically. Following the Nyquist information theory,
sampling rate depends on phenomena dynamics. In Table 2 are shown data rates for different
applications.
More modern theories, like compressive sensing, allow the reduction of the sampling rate for
specic signals (e,g images).
Sensor calibration is an important issue. Traditional methodologies can not be applied because
of the cost of a manual and node-by-node adjusting. Several collaborative techniques have
been proposed to allow self-calibration (Feng et al., 2003; Ihler et al., 2004; Miluzzo et al.,
2008).
Accommodation stage involves data transformation into engineering units, and data-type
casting. To pick the right data type, value ranges and quantization errors should be taken
into account. Although, using high precision arithmetic, like oating-point, makes the pro-
cessing performance drastically fall in small CPUs (Zoumboulakis & Roussos, n.d.).
Filtering, if present, reduces or removes undesirable time tendencies, spurious interferences
or frequency bands to enhance later processing algorithms. This stage is especially relevant in
audio processing applications.
The application-specic processing stage often includes algorithms for data compression, data
fusion and feature extraction. In-node data fusion allows the qualication and validation of
sensor data, making it easier to associate different physical measures to complex phenomena.
Feature extraction is highly application dependant. Its main goal is to obtain a pattern to
better represent, detect, identify or predict events and complex phenomena.
SAMPLING PRE-PROCESSING
FILTERING TYPE-CASTING
CONVERTION
SAMPLING RATE
ADC
DATA PROCESSING
FEATURES
EXTRACTION
DATA
FUSION
COMPRESSION
APPICATION-
SPECIFIC
PROCESSING
DATA ADQUISITION AND PROCESSING SERVICES
S
E
N
S
O
R
S
Fig. 3. Data acquisition and processing service
and specic program languages have been proposed. NesC language (Gay et al., 2003) for
TinyOS (Hill et al., 2000) are the most popular. Alternatives include Contiki (Dunkels et al.,
2004), SOS (Han et al., n.d.), Mantis (Abrach et al., 2003), RETOS (Cha et al., 2007), LiteOS
(Cao & Abdelzaher, n.d.), t-Kernel (Gu, 2006), or NANOrk (Eswaran et al., 2005). Some of
the above operating systems (e.g., SOS, LiteOS, and Contiki) also provide dynamic linking
capabilities, i.e., new code modules can be added at run-time to the application running on
a node. Dynamic linking is particularly important in supporting wireless reprogramming of
the WSN.
System Services are programs designed to support applications. Typically, an applica-
tion requires Location, Time Synchronization, Storage, Communication and Data acquisi-
tion/processing services. Location services estimate the own node position or the location of
a target (He et al., 2003; Mao et al., 2007; Patwari et al., 2005). Time synchronization is usually
required for data time-stamping, to measure the TOA (Time of Arrival) of a signal, to man-
age hibernate-wake up cycles, etc. Many algorithms have been proposed for this (Dai & Han,
2004; Elson & Rmer, 2003; Marti et al., 2004). Communication services deliver data reliably,
manage the network and optimize QoS and energy savings. The researchers community has
developed many different algorithms for routing, cluster management, etc. Although, the
industry has developed the Zigbee standard, it is probably the most widely accepted one so
far.
In most applications, sensor networks are deployed once and intended to operate unattended
for a long period of time. Reprogramming capabilities facilitate the management of applica-
tion changes or updates (Rubio et al., 2007). Because of the data memory constraints in the
nodes, it is not possible to save a monolithic program with many applications before being
burned. Instead of this, small packets of code can be reprogrammed and can also be reliably
delivered through the network. There are two basic schemes to reprogram sensor networks:
code dissemination and code acquisition. The rst one is initiated by the administrator to re-
program all the devices in the network. In the second one, the nodes request and reprogram
a packet dynamically. Data acquisition is not always considered and programmed as a ser-
vice. However, it is especially relevant in multimedia applications. Figure 3 shows a software
Phenomena Sample rate (/Hz)
Very low frequency
Atmospheric temperature 0.0171
Barometric pressure 0.0171
Low frequency
Heart rate 0.83.2
Volcanic infra-sound 2080
Natural seismic vibration 0.2100
Mid frequency (100Hz 1000 Hz)
Earthquake vibrations 100-160
ECG (heart electrical activity) 100250
High frequency (>1 kHz)
Breathing sounds 1005k
Industrial vibrations 40k
Audio (human hearing range) 1544k
Audio (muzzle shock-wave) 1M
Video (digital television) 10M
Table 2. Sensor sampling rates of different phenomena.
model for data acquisition and processing service with three stages: Sampling, Pre-processing
and Data processing.
In the sampling stage, a calibrated set of sensors are sampled, and the row data are time-
stamped, if necessary. Sensor calibration is an important issue. Traditional methodologies can
not be applied because of the cost of a manual and node-by-node adjusting. The sampling
process is red by either an event or periodically. Following the Nyquist information theory,
sampling rate depends on phenomena dynamics. In Table 2 are shown data rates for different
applications.
More modern theories, like compressive sensing, allow the reduction of the sampling rate for
specic signals (e,g images).
Sensor calibration is an important issue. Traditional methodologies can not be applied because
of the cost of a manual and node-by-node adjusting. Several collaborative techniques have
been proposed to allow self-calibration (Feng et al., 2003; Ihler et al., 2004; Miluzzo et al.,
2008).
Accommodation stage involves data transformation into engineering units, and data-type
casting. To pick the right data type, value ranges and quantization errors should be taken
into account. Although, using high precision arithmetic, like oating-point, makes the pro-
cessing performance drastically fall in small CPUs (Zoumboulakis & Roussos, n.d.).
Filtering, if present, reduces or removes undesirable time tendencies, spurious interferences
or frequency bands to enhance later processing algorithms. This stage is especially relevant in
audio processing applications.
The application-specic processing stage often includes algorithms for data compression, data
fusion and feature extraction. In-node data fusion allows the qualication and validation of
sensor data, making it easier to associate different physical measures to complex phenomena.
Feature extraction is highly application dependant. Its main goal is to obtain a pattern to
better represent, detect, identify or predict events and complex phenomena.
Data storage strategies usually distinguish between short-term and long-term storage. Un-
compressed, lossless compressed and labeled fused data are often necessary in short-term
storage . Otherwise, lossy compression and ancient data loss (amnesic algorithms) are rele-
vant topics in long-term storage (Girao et al., 2007; Sheng et al., 2006).
All the stages can interchange information for feedback and adaptive processing. In-network
processing involves data exchange among several sensor nodes. At the sampling stage, for
example, several collaborative algorithms have been proposed to calibrate the sensors (Feng
et al., 2003; Ihler et al., 2004; Miluzzo et al., 2008). A smart analysis of fused data or features
from different sensor nodes is necessary to process localized phenomena (i.e. target detection,
tracking and positioning (Arora et al., 2004; Cao et al., 2005)). Collaborative in-network pro-
cessing can also increase application reliability and performance (i.e. using voting schemes
and/or analyzing different target features in different nodes (Bnzit, 2009)).
3. Wireless Multimedia Sensor Networks: specic characteristics
Classic Wireless Sensor Networks have been initially thought up to manage ambient data at
low data rates. In general, for multimedia signals, high computing capabilities, high trans-
mission rates, large memory and an unlimited power source supply are necessary. Wireless
Multimedia Sensor Networks do not meet all of these conditions, though they do support
ambient sensing and multimedia applications, but with a limited quality of service.
3.1 Architectures
The node model shown in gures 1 and 2 is also suitable to describe a node of a Wireless
Multimedia Sensor Network (WMSN). However, multimedia signals require higher sampling
rate and more complex data processing. These topics make multimedia applications high
consumers of resources, in terms of memory, bandwidth, processing capabilities and energy.
Limited resources in most hardware platforms are some of the more relevant constraints for
multimedia applications. Other restrictions depend on the application. They can be described
in terms of Quality of Service. For example, in image delivery, topics like transmission reli-
ability or image quality can be relevant, and video streaming can require security, and low
delay. (Faheem et al., 2010)
Overall network architecture is also another relevant topic in order to understand practical
constraints. The research community has proposed three architectures, as described in the
survey (Akyildiz et al., 2008). From this paper, gure 4 depicts these.
Elements of these architectures are described below:
Single-tier at architecture is made up of homogeneous multimedia sensors connected to a
sink that bridges data to a storage hub or a gateway. The function of these elements is de-
scribed below:
Standard Video and Audio Sensors capture sound, still or moving images of the sensed
event.
Scalar Sensors acquire data from physical variables such as temperature, light, humid-
ity, etc. They are very resource-constraint devices.
Multimedia Processing Hubs. These devices have larger resources compared to sensor
nodes. They add data streams and reduce the volume of data delivered to the sink.
Storage Hubs save still-images or multimedia streams for data-mining and feature ex-
traction, even before the data is delivered to the end user.
Fig. 4. Standard WMNs architectures (Akyildiz et al., 2008)
The Sink is responsible for translating high level user queries to network-specic di-
rectives and returning the right chunk of the multimedia stream. Several sinks may be
needed for large and heterogeneous networks.
the Gateway bridges the sink to the Internet. IP is only assigned to the gateway, even
with multiple sinks. It manages geographical information to allocate the right nodes
and sinks for the selected area.
These architectures depict multimedia sensor networks where data ows between devices
with the same hardware capabilities, or from lower-end to higher-end devices. Complex in-
node and in-network processing is also performed in the multimedia processing hubs or high-
end multimedia nodes. Scalar data acquisition and processing are managed in scalar nodes
which are not involved in multimedia data processing or delivery. But, most environmental
sensor networks manage scalar data and, currently it is becoming more frequent to include
audio, sill-image and video delivery when an event has been taken place. This also means
that image and audio processing capabilities are needed. This situation can be described as
the fourth architecture: single-tier at heterogeneous sensors (STFH), where low-end or high-
end nodes must process and deliver data (streams are not often) through the deployed scalar
sensor nodes (low-end devices). This contrasts with standard architectures, where each tier is
in fact a sub-network with different performances.
Figure 5 describes the new architecture.
In the single-tier at heterogeneous architecture, the constraints of the lowest-end nodes limit
the performance of the overall network, no matter if it contains high-end video nodes. For
this reason, only a limited set of services of image and audio processing make sense. This
architecture will be the reference for this study, because it represents both: a very frequent sit-
uation in environmental applications, and the worst and most restrictive case for multimedia
Data storage strategies usually distinguish between short-term and long-term storage. Un-
compressed, lossless compressed and labeled fused data are often necessary in short-term
storage . Otherwise, lossy compression and ancient data loss (amnesic algorithms) are rele-
vant topics in long-term storage (Girao et al., 2007; Sheng et al., 2006).
All the stages can interchange information for feedback and adaptive processing. In-network
processing involves data exchange among several sensor nodes. At the sampling stage, for
example, several collaborative algorithms have been proposed to calibrate the sensors (Feng
et al., 2003; Ihler et al., 2004; Miluzzo et al., 2008). A smart analysis of fused data or features
from different sensor nodes is necessary to process localized phenomena (i.e. target detection,
tracking and positioning (Arora et al., 2004; Cao et al., 2005)). Collaborative in-network pro-
cessing can also increase application reliability and performance (i.e. using voting schemes
and/or analyzing different target features in different nodes (Bnzit, 2009)).
3. Wireless Multimedia Sensor Networks: specic characteristics
Classic Wireless Sensor Networks have been initially thought up to manage ambient data at
low data rates. In general, for multimedia signals, high computing capabilities, high trans-
mission rates, large memory and an unlimited power source supply are necessary. Wireless
Multimedia Sensor Networks do not meet all of these conditions, though they do support
ambient sensing and multimedia applications, but with a limited quality of service.
3.1 Architectures
The node model shown in gures 1 and 2 is also suitable to describe a node of a Wireless
Multimedia Sensor Network (WMSN). However, multimedia signals require higher sampling
rate and more complex data processing. These topics make multimedia applications high
consumers of resources, in terms of memory, bandwidth, processing capabilities and energy.
Limited resources in most hardware platforms are some of the more relevant constraints for
multimedia applications. Other restrictions depend on the application. They can be described
in terms of Quality of Service. For example, in image delivery, topics like transmission reli-
ability or image quality can be relevant, and video streaming can require security, and low
delay. (Faheem et al., 2010)
Overall network architecture is also another relevant topic in order to understand practical
constraints. The research community has proposed three architectures, as described in the
survey (Akyildiz et al., 2008). From this paper, gure 4 depicts these.
Elements of these architectures are described below:
Single-tier at architecture is made up of homogeneous multimedia sensors connected to a
sink that bridges data to a storage hub or a gateway. The function of these elements is de-
scribed below:
Standard Video and Audio Sensors capture sound, still or moving images of the sensed
event.
Scalar Sensors acquire data from physical variables such as temperature, light, humid-
ity, etc. They are very resource-constraint devices.
Multimedia Processing Hubs. These devices have larger resources compared to sensor
nodes. They add data streams and reduce the volume of data delivered to the sink.
Storage Hubs save still-images or multimedia streams for data-mining and feature ex-
traction, even before the data is delivered to the end user.
Fig. 4. Standard WMNs architectures (Akyildiz et al., 2008)
The Sink is responsible for translating high level user queries to network-specic di-
rectives and returning the right chunk of the multimedia stream. Several sinks may be
needed for large and heterogeneous networks.
the Gateway bridges the sink to the Internet. IP is only assigned to the gateway, even
with multiple sinks. It manages geographical information to allocate the right nodes
and sinks for the selected area.
These architectures depict multimedia sensor networks where data ows between devices
with the same hardware capabilities, or from lower-end to higher-end devices. Complex in-
node and in-network processing is also performed in the multimedia processing hubs or high-
end multimedia nodes. Scalar data acquisition and processing are managed in scalar nodes
which are not involved in multimedia data processing or delivery. But, most environmental
sensor networks manage scalar data and, currently it is becoming more frequent to include
audio, sill-image and video delivery when an event has been taken place. This also means
that image and audio processing capabilities are needed. This situation can be described as
the fourth architecture: single-tier at heterogeneous sensors (STFH), where low-end or high-
end nodes must process and deliver data (streams are not often) through the deployed scalar
sensor nodes (low-end devices). This contrasts with standard architectures, where each tier is
in fact a sub-network with different performances.
Figure 5 describes the new architecture.
In the single-tier at heterogeneous architecture, the constraints of the lowest-end nodes limit
the performance of the overall network, no matter if it contains high-end video nodes. For
this reason, only a limited set of services of image and audio processing make sense. This
architecture will be the reference for this study, because it represents both: a very frequent sit-
uation in environmental applications, and the worst and most restrictive case for multimedia
Fig. 5. Proposed WMNs architecture
capturing, processing and delivery. Any algorithm, specically developed for it, will be easily
and successfully applied to other architecture.
In the STFH architecture, any node can be involved in capturing, processing, delivering and
the storage of multimedia data, the exact role depends on the application, but mainly on
the node capabilities. In contrast to the multi-tier architectures, where nodes in the same
tier have similar resources and functions, in STHF these functions can differ among equal
nodes. For this reason, node constraints and applications requirements should be analyzed
and compared.
3.2 Node constraints
Many commercial and research node platforms have been developed and proposed in scien-
tic literature (Akyildiz et al., 2008). For Wireless Multimedia Sensor Network, these plat-
forms can be classied into three categories that match node devices proposed in the general
multimedia architectures (Akyildiz et al., 2008). Table 3 summarizes resources supported for
these three node platform categories:
3.3 Compression
Multimedia data compression, which is one of the most relevant topics, is often present in any
application with multimedia image storage and delivery. There are many image compression
algorithms, but JPEG is in fact the most extensive. It is also included in most CMOS camera
sensor chips. JPEG standard can easily reach from 10:1 up to 20:1 compression ratio without
visible losses in color images. 30:1 to 50:1 compression is possible with small to moderate
defects, while for very-low-quality purposes like indexing or previews, 100:1 compression
L
O
W
-
E
N
D
C
A
T
E
G
O
R
Y
:
A
U
D
I
O
A
N
D
V
I
D
E
O
S
E
N
S
O
R
P
l
a
t
f
o
r
m
C
a
m
e
r
a
r
e
s
o
l
u
t
i
o
n
M
e
m
o
r
y
B
W
/
C
o
l
o
r
M
C
U
M
I
P
S
M
I
C
A
2
+
C
y
c
l
o
p
s
P
o
-
l
a
s
t
r
e
e
t
a
l
.
(
2
0
0
4
)
R
a
h
i
m
i
e
t
a
l
.
(
2
0
0
5
)
3
2
0
x
2
4
0
(
C
I
F
C
M
O
S
)
6
4
K
b
+
5
1
2
K
b
F
l
a
s
h
C
o
l
o
r
(
8
b
i
t
s
)
A
T
m
e
g
a
1
2
8
L
1
6
M
I
D
D
L
E
-
E
N
D
C
A
T
E
G
O
R
Y
:
M
U
L
T
I
M
E
D
I
A
P
R
O
C
E
S
S
I
N
G
H
U
B
P
l
a
t
f
o
r
m
C
a
m
e
r
a
r
e
s
o
l
u
t
i
o
n
M
e
m
o
r
y
B
W
/
C
o
l
o
r
M
C
U
M
I
P
S
I
M
o
t
e
2
+
C
M
U
C
a
m
2
I
M
O
T
E
2
(
n
.
d
.
)
C
a
r
n
e
g
i
e
M
e
l
l
o
n
U
n
i
v
e
r
s
i
t
y
(
2
0
0
7
)
C
I
F
(
3
5
2
x
2
8
8
)
R
G
B
6
4
K
b
R
a
m
+
1
2
8
K
b
F
l
a
s
h
C
o
l
o
r
(
8
b
i
t
s
)
A
R
M
7
T
D
M
I
3
6
M
e
s
h
E
y
e
H
e
n
g
s
t
l
e
r
e
t
a
l
.
(
2
0
0
7
)
V
G
A
C
M
O
S
(
6
4
0
x
4
8
0
)
6
4
K
b
+
2
5
6
K
b
F
l
a
s
h
G
r
a
y
o
r
C
o
l
o
r
(
2
4
b
i
t
s
)
A
t
m
e
l
A
T
9
1
S
A
M
7
S
2
0
0
3
0
x
3
0
G
r
a
y
P
a
n
o
p
t
e
s
F
e
n
g
e
t
a
l
.
(
2
0
0
5
)
6
4
0
x
4
8
0
6
4
M
b
C
o
l
o
r
S
t
r
o
n
g
A
R
M
2
0
6
M
H
z
2
3
5
H
I
G
H
-
E
N
D
C
A
T
E
G
O
R
Y
:
V
I
D
E
O
P
R
O
C
E
S
S
O
R
P
l
a
t
f
o
r
m
C
a
m
e
r
a
r
e
s
o
l
u
t
i
o
n
M
e
m
o
r
y
B
W
/
C
o
l
o
r
M
C
U
M
I
P
S
S
t
a
r
g
a
t
e
X
-
s
c
a
l
e
+
w
e
b
-
c
a
m
S
t
a
r
g
a
t
e
X
S
c
a
l
e
P
l
a
t
-
f
o
r
m
(
n
.
d
.
)
6
4
0
x
4
8
0
6
4
M
b
R
a
m
+
3
2
M
b
F
l
a
s
h
C
o
l
o
r
X
S
c
a
l
e
P
X
A
2
5
5
4
0
0
M
H
z
3
7
0
S
e
n
s
E
y
e
K
u
l
k
a
r
n
i
e
t
a
l
.
(
2
0
0
5
)
(
C
y
c
l
o
p
s
)
1
2
8
x
1
2
8
6
4
K
b
+
5
1
2
K
b
C
o
l
o
r
(
8
b
i
t
s
)
A
T
M
E
G
A
1
2
8
1
6
(
C
M
U
C
A
M
)
3
5
2
x
2
8
8
C
o
l
o
r
(
8
b
i
t
s
)
S
X
5
2
8
0
(
W
e
b
c
a
m
)
6
4
0
x
4
8
0
C
o
l
o
r
S
t
a
r
g
a
t
e
3
7
0
(
S
N
C
-
R
Z
3
0
N
P
T
Z
)
1
0
2
4
x
7
6
8
C
o
l
o
r
E
m
b
e
d
d
e
d
P
C
(
P
4
1
.
6
G
h
z
)
1
2
3
7
T
a
b
l
e
3
.
C
o
n
s
t
r
a
i
n
t
s
o
f
s
e
v
e
r
a
l
p
l
a
t
f
o
r
m
s
Fig. 5. Proposed WMNs architecture
capturing, processing and delivery. Any algorithm, specically developed for it, will be easily
and successfully applied to other architecture.
In the STFH architecture, any node can be involved in capturing, processing, delivering and
the storage of multimedia data, the exact role depends on the application, but mainly on
the node capabilities. In contrast to the multi-tier architectures, where nodes in the same
tier have similar resources and functions, in STHF these functions can differ among equal
nodes. For this reason, node constraints and applications requirements should be analyzed
and compared.
3.2 Node constraints
Many commercial and research node platforms have been developed and proposed in scien-
tic literature (Akyildiz et al., 2008). For Wireless Multimedia Sensor Network, these plat-
forms can be classied into three categories that match node devices proposed in the general
multimedia architectures (Akyildiz et al., 2008). Table 3 summarizes resources supported for
these three node platform categories:
3.3 Compression
Multimedia data compression, which is one of the most relevant topics, is often present in any
application with multimedia image storage and delivery. There are many image compression
algorithms, but JPEG is in fact the most extensive. It is also included in most CMOS camera
sensor chips. JPEG standard can easily reach from 10:1 up to 20:1 compression ratio without
visible losses in color images. 30:1 to 50:1 compression is possible with small to moderate
defects, while for very-low-quality purposes like indexing or previews, 100:1 compression
L
O
W
-
E
N
D
C
A
T
E
G
O
R
Y
:
A
U
D
I
O
A
N
D
V
I
D
E
O
S
E
N
S
O
R
P
l
a
t
f
o
r
m
C
a
m
e
r
a
r
e
s
o
l
u
t
i
o
n
M
e
m
o
r
y
B
W
/
C
o
l
o
r
M
C
U
M
I
P
S
M
I
C
A
2
+
C
y
c
l
o
p
s
P
o
-
l
a
s
t
r
e
e
t
a
l
.
(
2
0
0
4
)
R
a
h
i
m
i
e
t
a
l
.
(
2
0
0
5
)
3
2
0
x
2
4
0
(
C
I
F
C
M
O
S
)
6
4
K
b
+
5
1
2
K
b
F
l
a
s
h
C
o
l
o
r
(
8
b
i
t
s
)
A
T
m
e
g
a
1
2
8
L
1
6
M
I
D
D
L
E
-
E
N
D
C
A
T
E
G
O
R
Y
:
M
U
L
T
I
M
E
D
I
A
P
R
O
C
E
S
S
I
N
G
H
U
B
P
l
a
t
f
o
r
m
C
a
m
e
r
a
r
e
s
o
l
u
t
i
o
n
M
e
m
o
r
y
B
W
/
C
o
l
o
r
M
C
U
M
I
P
S
I
M
o
t
e
2
+
C
M
U
C
a
m
2
I
M
O
T
E
2
(
n
.
d
.
)
C
a
r
n
e
g
i
e
M
e
l
l
o
n
U
n
i
v
e
r
s
i
t
y
(
2
0
0
7
)
C
I
F
(
3
5
2
x
2
8
8
)
R
G
B
6
4
K
b
R
a
m
+
1
2
8
K
b
F
l
a
s
h
C
o
l
o
r
(
8
b
i
t
s
)
A
R
M
7
T
D
M
I
3
6
M
e
s
h
E
y
e
H
e
n
g
s
t
l
e
r
e
t
a
l
.
(
2
0
0
7
)
V
G
A
C
M
O
S
(
6
4
0
x
4
8
0
)
6
4
K
b
+
2
5
6
K
b
F
l
a
s
h
G
r
a
y
o
r
C
o
l
o
r
(
2
4
b
i
t
s
)
A
t
m
e
l
A
T
9
1
S
A
M
7
S
2
0
0
3
0
x
3
0
G
r
a
y
P
a
n
o
p
t
e
s
F
e
n
g
e
t
a
l
.
(
2
0
0
5
)
6
4
0
x
4
8
0
6
4
M
b
C
o
l
o
r
S
t
r
o
n
g
A
R
M
2
0
6
M
H
z
2
3
5
H
I
G
H
-
E
N
D
C
A
T
E
G
O
R
Y
:
V
I
D
E
O
P
R
O
C
E
S
S
O
R
P
l
a
t
f
o
r
m
C
a
m
e
r
a
r
e
s
o
l
u
t
i
o
n
M
e
m
o
r
y
B
W
/
C
o
l
o
r
M
C
U
M
I
P
S
S
t
a
r
g
a
t
e
X
-
s
c
a
l
e
+
w
e
b
-
c
a
m
S
t
a
r
g
a
t
e
X
S
c
a
l
e
P
l
a
t
-
f
o
r
m
(
n
.
d
.
)
6
4
0
x
4
8
0
6
4
M
b
R
a
m
+
3
2
M
b
F
l
a
s
h
C
o
l
o
r
X
S
c
a
l
e
P
X
A
2
5
5
4
0
0
M
H
z
3
7
0
S
e
n
s
E
y
e
K
u
l
k
a
r
n
i
e
t
a
l
.
(
2
0
0
5
)
(
C
y
c
l
o
p
s
)
1
2
8
x
1
2
8
6
4
K
b
+
5
1
2
K
b
C
o
l
o
r
(
8
b
i
t
s
)
A
T
M
E
G
A
1
2
8
1
6
(
C
M
U
C
A
M
)
3
5
2
x
2
8
8
C
o
l
o
r
(
8
b
i
t
s
)
S
X
5
2
8
0
(
W
e
b
c
a
m
)
6
4
0
x
4
8
0
C
o
l
o
r
S
t
a
r
g
a
t
e
3
7
0
(
S
N
C
-
R
Z
3
0
N
P
T
Z
)
1
0
2
4
x
7
6
8
C
o
l
o
r
E
m
b
e
d
d
e
d
P
C
(
P
4
1
.
6
G
h
z
)
1
2
3
7
T
a
b
l
e
3
.
C
o
n
s
t
r
a
i
n
t
s
o
f
s
e
v
e
r
a
l
p
l
a
t
f
o
r
m
s
is quite feasible. Although, gray-scale images can not be compressed by such large factors
because the human eye is much more sensitive to brightness variations than to hue variations.
The threshold of visible loss using JPEGis often around 5:1 compression for gray-scale images.
Table 4 shows comparative results for JPEG compression.
Image resolution Format/Size Size in memory
Raw Low comp. High comp.
Low 128 X 96 (SQCIF) 36K 3K 369
176 x 144 (QCIF) 74K 7K 760
Med 320 x 240 (QVGA) 225K 22K 2K
352 x 288 (CIF-PAL) 297K 29K 2K
High 640 x 480 (VGA) 900K 90K 9K
704 x 576 (4CIF) 1M 118K 11K
Table 4. Size of images with several levels of compression
In many scientic papers, other compression schemes have been proposed, a great deal of
them based on wavelets transform (Lewis & Knowles, 1992; Villasenor et al., 1995). Haar
compression algorithms are very suitable for resource-limited devices like WSN nodes. More-
over, in image delivery, Haar compression allows variable data rates for adaptive transmission
schemes, by assigning different priorities to the packets. In this way, most relevant coefcients
will be reliably delivered and less relevant coefcients can be discarded if necessary.
Recently, compressions based on Wyner-Ziv encoder theory (Ziv & Lempel, 1978) are being
studied (Aaron et al., 2003), especially for collaborative compressions (Chia et al., 2009). The
main advantage is the simplicity and the light-weight computation of the encoder. For the
same reasons, but with a fairly different perspective, Compressive Sensing (CS) is another
promising technology. It exploits the sparse nature of an image to achieve very high compres-
sion rates.
A similar description can be done for audio data. Lossless compressions are more suitable
for generic applications. In this sense, lossy encoders should be correctly picked to optimize
audio data storage. Despite the fact that a vast amount of literature exists on lossless compres-
sion (Sayood, 2005), only fewalgorithms are shown to be feasible in WSNs due the strong HW
constraints. For example: Delta Coding, Human Coding, Lempel-Ziv SS, and their variants:
Delta Coding + Plain Huffman, Adaptive Huffman, Delta Coding+Adaptive Huffman, etc.
The best compression rate results are about 65% (Alippi et al., 2007).
The application of lossy techniques requires specic signal degradation studies, and they
should be picked and applied carefully. For example, many encoders take advantage of
the human psychoacoustic hearing or human speech properties like the popular (mp3), and
clearly, they are not directly applicable to compress vibrations or seismic data. Here again,
Compressive Sensing is a promising technology because no previous signal characteristic is
supposed, but sparsity is a very common property in nature.
3.4 Memory and bandwidth requirements
Memory supports compressed data storage and temporal raw data saving for multimedia
processing and data transmission/relay. In this context, raw data means images or audio data
without compression, with the same size or dimensions in which they have been captured, but
in a suitable format. For example, 8-bit unsigned integer RGB or YCbCr matrices for images,
and normalized oating-point (or 16-bit integer) arrays for audio. Memory occupation of
multimedia data is depicted in 6. In an image or video application, compressed images or
streams share memory with temporal images, packets for image delivery or relay, and local
and global variables.
IMAGE PROC. APPLICATION
COMPRESSED
IMAGES
RAW IMAGES
PACKETS
STORAGE
CAPTURED &
TEMP. IMAGES
DELIVERY/RELAY
FEATURES PROC. RESULTS
VIDEO STREAMING
RAW FRAMES
&
COMPRESSED
CHUNKS
PACKETS
CAPTURED &
TEMP. IMAGES
DELIVERY/RELAY
Fig. 6. Memory occupation of multimedia data
Compressed images memory occupation depends on the application. In general, considering
the size of compressed images (table 4), it is not possible to save them into the microcontroller
internal memory. If no processing or different compression format was necessary, the CMOS
camera sensor could host the image in JPEG format in its internal memory. Otherwise, exter-
nal RAM or FLASH memory would be needed.
Image transmission or relaying packet number, depend on the communication protocol, espe-
cially in the transport layer. By saving several packets, data can be retransmitted for reliability
after an error occurs. But this, in general, just affects the source node, not the relaying nodes,
except in the case where catching strategies were employed. It is always recommendable to
use internal memory to quickly react to the communication events.
For image processing, temporal images must be allocated. Some special algorithms can be
executed "on-the-y" without saving the overall image. In latter sections, we will describe
a basic model for that. But in general, for low-end platforms, an external RAM memory is
required, normally connected by an SPI port. Computing performance will decrease if an
intensive memory access is necessary. In that case, the algorithms must be released to support
data pre-catching strategies.
The most successful communication standard for WSN is the IEEE 802.15.4. But for high-end
nodes, Ultra Wide Band (UWB) is probably more suitable. Bandwidth for 802.15.4 compliant
devices in short distance transmissions ranges from 224kbps to 20kbps for 10 and 75 meters,
respectively.
Table 5 would clearly showthat video streaming in WSN is only possible for medium or high-
end platform, high compression, low frame rates and short distances. Fortunately, higher
compression is possible, but with very specic algorithms.
3.5 Processing and energy consumption
In traditional WSN, energy consumption (g. 8) comes mainly from radio transmissions. The
energy required to process data is considered negligible if, as usual, external or timed events
is quite feasible. Although, gray-scale images can not be compressed by such large factors
because the human eye is much more sensitive to brightness variations than to hue variations.
The threshold of visible loss using JPEGis often around 5:1 compression for gray-scale images.
Table 4 shows comparative results for JPEG compression.
Image resolution Format/Size Size in memory
Raw Low comp. High comp.
Low 128 X 96 (SQCIF) 36K 3K 369
176 x 144 (QCIF) 74K 7K 760
Med 320 x 240 (QVGA) 225K 22K 2K
352 x 288 (CIF-PAL) 297K 29K 2K
High 640 x 480 (VGA) 900K 90K 9K
704 x 576 (4CIF) 1M 118K 11K
Table 4. Size of images with several levels of compression
In many scientic papers, other compression schemes have been proposed, a great deal of
them based on wavelets transform (Lewis & Knowles, 1992; Villasenor et al., 1995). Haar
compression algorithms are very suitable for resource-limited devices like WSN nodes. More-
over, in image delivery, Haar compression allows variable data rates for adaptive transmission
schemes, by assigning different priorities to the packets. In this way, most relevant coefcients
will be reliably delivered and less relevant coefcients can be discarded if necessary.
Recently, compressions based on Wyner-Ziv encoder theory (Ziv & Lempel, 1978) are being
studied (Aaron et al., 2003), especially for collaborative compressions (Chia et al., 2009). The
main advantage is the simplicity and the light-weight computation of the encoder. For the
same reasons, but with a fairly different perspective, Compressive Sensing (CS) is another
promising technology. It exploits the sparse nature of an image to achieve very high compres-
sion rates.
A similar description can be done for audio data. Lossless compressions are more suitable
for generic applications. In this sense, lossy encoders should be correctly picked to optimize
audio data storage. Despite the fact that a vast amount of literature exists on lossless compres-
sion (Sayood, 2005), only fewalgorithms are shown to be feasible in WSNs due the strong HW
constraints. For example: Delta Coding, Human Coding, Lempel-Ziv SS, and their variants:
Delta Coding + Plain Huffman, Adaptive Huffman, Delta Coding+Adaptive Huffman, etc.
The best compression rate results are about 65% (Alippi et al., 2007).
The application of lossy techniques requires specic signal degradation studies, and they
should be picked and applied carefully. For example, many encoders take advantage of
the human psychoacoustic hearing or human speech properties like the popular (mp3), and
clearly, they are not directly applicable to compress vibrations or seismic data. Here again,
Compressive Sensing is a promising technology because no previous signal characteristic is
supposed, but sparsity is a very common property in nature.
3.4 Memory and bandwidth requirements
Memory supports compressed data storage and temporal raw data saving for multimedia
processing and data transmission/relay. In this context, raw data means images or audio data
without compression, with the same size or dimensions in which they have been captured, but
in a suitable format. For example, 8-bit unsigned integer RGB or YCbCr matrices for images,
and normalized oating-point (or 16-bit integer) arrays for audio. Memory occupation of
multimedia data is depicted in 6. In an image or video application, compressed images or
streams share memory with temporal images, packets for image delivery or relay, and local
and global variables.
IMAGE PROC. APPLICATION
COMPRESSED
IMAGES
RAW IMAGES
PACKETS
STORAGE
CAPTURED &
TEMP. IMAGES
DELIVERY/RELAY
VIDEO STREAMING
RAW FRAMES
&
COMPRESSED
CHUNKS
PACKETS
CAPTURED &
TEMP. IMAGES
DELIVERY/RELAY
Fig. 6. Memory occupation of multimedia data
Compressed images memory occupation depends on the application. In general, considering
the size of compressed images (table 4), it is not possible to save them into the microcontroller
internal memory. If no processing or different compression format was necessary, the CMOS
camera sensor could host the image in JPEG format in its internal memory. Otherwise, exter-
nal RAM or FLASH memory would be needed.
Image transmission or relaying packet number, depend on the communication protocol, espe-
cially in the transport layer. By saving several packets, data can be retransmitted for reliability
after an error occurs. But this, in general, just affects the source node, not the relaying nodes,
except in the case where catching strategies were employed. It is always recommendable to
use internal memory to quickly react to the communication events.
For image processing, temporal images must be allocated. Some special algorithms can be
executed "on-the-y" without saving the overall image. In latter sections, we will describe
a basic model for that. But in general, for low-end platforms, an external RAM memory is
required, normally connected by an SPI port. Computing performance will decrease if an
intensive memory access is necessary. In that case, the algorithms must be released to support
data pre-catching strategies.
The most successful communication standard for WSN is the IEEE 802.15.4. But for high-end
nodes, Ultra Wide Band (UWB) is probably more suitable. Bandwidth for 802.15.4 compliant
devices in short distance transmissions ranges from 224kbps to 20kbps for 10 and 75 meters,
respectively.
Table 5 would clearly showthat video streaming in WSN is only possible for medium or high-
end platform, high compression, low frame rates and short distances. Fortunately, higher
compression is possible, but with very specic algorithms.
3.5 Processing and energy consumption
In traditional WSN, energy consumption (g. 8) comes mainly from radio transmissions. The
energy required to process data is considered negligible if, as usual, external or timed events
Format/Size Uncompressed Low Compression High Compression
25 FPS 15 FPS 25 FPS 15 FPS 25 FPS 15 FPS
128 X 96 (SQCIF) 7Mbps 4Mbps 720Kbps 432Kbps 72Kbps 43Kbps
176 x 144 (QCIF) 14Mbps 8Mbps 1Mbps 891Kbps 148Kbps 89Kbps
320 x 240 (QVGA) 43Mbps 26Mbps 4Mbps 2Mbps 450Kbps 270Kbps
352 x 288 (CIF-PAL) 58Mbps 34Mbps 5Mbps 3Mbps 594Kbps 356Kbps
640 x 480 (VGA) 175Mbps 105Mbps 17Mbps 10Mbps 1Mbps 1Mbps
704 x 576 (4CIF) 232Mbps 139Mbps 23Mbps 13Mbps 2Mbps 1Mbps
Table 5. Required Bandwidth for Video Streaming
Fig. 7. Bandwidth usage in Streaming vs Image size and compression ratio
trigger the execution of an application, a task or a phase. Once completed, the node falls in a
low power state (standby, hibernation, sleep. . . ). For this reason, most research activities have
been addressed to reduce radio transmissions, optimizing the protocol stack, compressing
and processing data in the application level. Using these simple models, simulations allow
the estimation of results for large networks (Hu et al., 2007; Zhong et al., 2004). Recently,
different papers try to describe energy consumption more precisely. In (Kellner et al., 2008),
energy states of different devices and OS activities are modeled as nite state machines.
In Wireless Multimedia Sensors, intensive computing keeps the MCU in the active state for
a high percent of the working cycle. In this way, energy for signal processing is not negli-
gible. Although, a high-end processor wastes more energy, intensive computing takes less
time, and the overall energy can be lower. A newparadigm comes from this situation: nding
the optimal MCU performance/consumption ratio (PCR) to minimize the computing energy.
Inuencing factors are clock frequency, internal architecture, the set of instructions, and low-
Time
P
o
w
e
r

c
o
n
s
u
m
p
t
i
o
n
sleep
w
a
k
e
-
u
p
Communications
Data Acquisition
Processing
Fig. 8. Typical Power Consumption in WSN
power modes. There is no method to derive energy from these factors, so simulations and
empirical tests are needed. But a sensor node must execute intensive and non-intensive tasks.
Depending on the frequency of both, a high or low-performance processor will be necessary.
In the next section we propose a general and exible architecture that optimizes power con-
sumption made up of a High-End MCUor DSP acting as a co-processor, and a Low-End MCU
as the main processor. This architecture allows the adaptation of the performances and power
to the instantaneous activity, enabling or disabling the coprocessor.
4. Hardware architecture alternatives and Application Optimization
Devices described in standard WMSN architectures are clearly divided into categories associ-
ated to different tiers. All the devices in the same tier are almost identical in resources. But in
Single-tier Heterogeneous Architecture, high-end and low-end nodes work together to sup-
port multimedia streaming, still-image delivery, data relay, and multimedia processing. We
propose a set of simple, also practical, hardware architectures that cover the functionality of
all the categories. Moreover, we present multimedia manipulation schemes to increase acqui-
sition and delivery performances in low-end nodes.
4.1 Image data: Split On the Fly algorithm
Proposed Split on the Fly (SoF) algorithm is designed for low-end platforms to increase the
image resolution they can work with. These platforms consist of a resource limited MCU
connected to a CMOS image sensor, as shown in gure 9. Most of these image sensors allow
a command-based dialog using a serial interface (I2C or SPI). Typically, it is possible to select
the image format (CIF, GCIF, JPEG), the data format (e.g. YCbCr 4:2:2, GRB 4:2:2 or RGB Raw
Data), camera parameters (exposure, gain and white balance) and image quality parameters
(brightness, contrast, saturation, sharpness gamma, and windowing). Other operations or
algorithms must normally be executed in the microcontroller, and the original and processed
images have to be saved in the internal memory. Only very low image resolutions (e.g. 30
30) can be processed in the internal memory.
This platform can support data relay, image capture, transmission and video streaming for
very low resolutions. Compression algorithms, and transport protocols for reliable services,
are very limited, or not possible because of memory constraints.
Format/Size Uncompressed Low Compression High Compression
25 FPS 15 FPS 25 FPS 15 FPS 25 FPS 15 FPS
128 X 96 (SQCIF) 7Mbps 4Mbps 720Kbps 432Kbps 72Kbps 43Kbps
176 x 144 (QCIF) 14Mbps 8Mbps 1Mbps 891Kbps 148Kbps 89Kbps
320 x 240 (QVGA) 43Mbps 26Mbps 4Mbps 2Mbps 450Kbps 270Kbps
352 x 288 (CIF-PAL) 58Mbps 34Mbps 5Mbps 3Mbps 594Kbps 356Kbps
640 x 480 (VGA) 175Mbps 105Mbps 17Mbps 10Mbps 1Mbps 1Mbps
704 x 576 (4CIF) 232Mbps 139Mbps 23Mbps 13Mbps 2Mbps 1Mbps
Table 5. Required Bandwidth for Video Streaming
Fig. 7. Bandwidth usage in Streaming vs Image size and compression ratio
trigger the execution of an application, a task or a phase. Once completed, the node falls in a
low power state (standby, hibernation, sleep. . . ). For this reason, most research activities have
been addressed to reduce radio transmissions, optimizing the protocol stack, compressing
and processing data in the application level. Using these simple models, simulations allow
the estimation of results for large networks (Hu et al., 2007; Zhong et al., 2004). Recently,
different papers try to describe energy consumption more precisely. In (Kellner et al., 2008),
energy states of different devices and OS activities are modeled as nite state machines.
In Wireless Multimedia Sensors, intensive computing keeps the MCU in the active state for
a high percent of the working cycle. In this way, energy for signal processing is not negli-
gible. Although, a high-end processor wastes more energy, intensive computing takes less
time, and the overall energy can be lower. A newparadigm comes from this situation: nding
the optimal MCU performance/consumption ratio (PCR) to minimize the computing energy.
Inuencing factors are clock frequency, internal architecture, the set of instructions, and low-
Time
P
o
w
e
r

c
o
n
s
u
m
p
t
i
o
n
sleep
w
a
k
e
-
u
p
Communications
Data Acquisition
Processing
Fig. 8. Typical Power Consumption in WSN
power modes. There is no method to derive energy from these factors, so simulations and
empirical tests are needed. But a sensor node must execute intensive and non-intensive tasks.
Depending on the frequency of both, a high or low-performance processor will be necessary.
In the next section we propose a general and exible architecture that optimizes power con-
sumption made up of a High-End MCUor DSP acting as a co-processor, and a Low-End MCU
as the main processor. This architecture allows the adaptation of the performances and power
to the instantaneous activity, enabling or disabling the coprocessor.
4. Hardware architecture alternatives and Application Optimization
Devices described in standard WMSN architectures are clearly divided into categories associ-
ated to different tiers. All the devices in the same tier are almost identical in resources. But in
Single-tier Heterogeneous Architecture, high-end and low-end nodes work together to sup-
port multimedia streaming, still-image delivery, data relay, and multimedia processing. We
propose a set of simple, also practical, hardware architectures that cover the functionality of
all the categories. Moreover, we present multimedia manipulation schemes to increase acqui-
sition and delivery performances in low-end nodes.
4.1 Image data: Split On the Fly algorithm
Proposed Split on the Fly (SoF) algorithm is designed for low-end platforms to increase the
image resolution they can work with. These platforms consist of a resource limited MCU
connected to a CMOS image sensor, as shown in gure 9. Most of these image sensors allow
a command-based dialog using a serial interface (I2C or SPI). Typically, it is possible to select
the image format (CIF, GCIF, JPEG), the data format (e.g. YCbCr 4:2:2, GRB 4:2:2 or RGB Raw
Data), camera parameters (exposure, gain and white balance) and image quality parameters
(brightness, contrast, saturation, sharpness gamma, and windowing). Other operations or
algorithms must normally be executed in the microcontroller, and the original and processed
images have to be saved in the internal memory. Only very low image resolutions (e.g. 30
30) can be processed in the internal memory.
This platform can support data relay, image capture, transmission and video streaming for
very low resolutions. Compression algorithms, and transport protocols for reliable services,
are very limited, or not possible because of memory constraints.
CAM
SPI
Fig. 9. Low-End node architecture
Delivery of higher image resolutions without compression can be supported if the camera
sensor includes internal memory. In this case, nothing but performance limits the transmis-
sion. In any case, raw images can be delivered without processing if they are split on the y,
transmitting each row after being read from the camera. With this scheme, just one row needs
to be saved in the memory before delivering.
Without transport protocols, a packet loss can degrade the image at the sink node. In this
case a simple interpolation algorithm (e.g. bilinear) can be used to estimate the lost pixels and
reconstruct the image. In a burst of lost packets, a region of the image will be completely lost
and restoration will be impossible.
Reliability of this scheme and the quality of the restored image will increase if the rows are
also split into two packets with the odd and even pixels, respectively. Choosing different
delivery paths for both increases the probability of a packet loss, but one of them will also be
more likely to reach the sink node.
Again, lost pixels can be estimated using an interpolation algorithm. But in this case, the
real pixels are uniformly distributed throughout the image, so the quality of the image recon-
structed by interpolation increases. See gure 10
CAM
ROW 2
ROW 1
ROW 1 ROW 2
NETWORK PATH
NON-INTERLEAVING SCHEME
ROW 2
ROW 2
ROW 1
ROW 1
NETWORK PATH 2 ROW 2
ROW 1
NETWORK PATH 1
INTERLEAVING SCHEME
CORRECT PIXELS
LOST & INTERPOLED PIXEL
LOW-END ARCHITECTURE (SOURCE NODE)
IMAGE AT SINK NODE
Fig. 10. Split on the Fly scheme (with and without interleaving)
4.2 Medium-End architecture
SoF is a suitable algorithmfor MCUs with a lack of memory. But the energy required to deliver
the image through the network is high. Let N be the size of a square image. Then,
E = O(N
2
) (1)
A compression algorithm can reduce this energy signicantly, but no splitting on the y scheme
is possible, because the original and compressed images must be saved in the memory. An
external memory (i.e. an I2C RAM) can solve the problem. But, the chip increases the en-
ergy consumption in three ways: its power supply, the time needed to transfer data from the
camera (seconds), and a factor caused by the inefcient data access during the compression
algorithm.
CAM
SPI
I2C,
Flash, ..
External Memory
Fig. 11. MCU with external memory
As shown in (Molina et al., 2009) all of them are insignicant compared to the required pro-
cessing time. Using Split on the Fly scheme as the reference, the energy saving of this solution
can be approximated to:
E N
2
E
BB
(1 R
C
) + T
P
PWC (2)
where R
C
is the compression rate, E
BB
the energy to broadcast a byte, T
P
the processing time,
and PWC the CPU power consumption for the processing time. The rst factor, R
C
, is directly
related to the image quality. The Haar transform (Raviraj & Sanavullah, 2007) is recommend-
able because it reduces the image size with a reasonable computation cost and high quality.
4.3 High-End architecture
To reduce the processing time, we propose a dual processor architecture: A low performance
MCU managing the WSN services, and a high performance microcontroller acting as a multi-
media co-processor. Even though the power consumption of this co-processor is much higher,
the processing time is substantially lower. But co-processor selection is not trivial. If the option
is another MCU or a processing card, a detailed analysis is necessary because it is possible not
to achieve the desired energy reduction. This is the case of CMUCam (Carnegie Mellon Uni-
versity, 2007) . Multimedia co-processor should be optimized for data processing, and must
necessarily contain an internal structure far different from the main MCU (e.g. PIC32XMX-
460 with 32K RAM selected in (Molina et al., 2009)). A simple overclocked version of the main
CPU reduces the processing time proportionally, but the power consumption increases the
same rate, as shown following:
Using the general equation of power consumption of a CMOS integrated circuit (IC):
P
IC
= KV
2
f (3)
The energy in an overclocked CPU (E
OV
) for a period T
OV
is
E
OV
= P
OV
T
OV
= KV
2
f
OV
T
OV
= KV
2
f
OV
T
f
f
OV
= KV
2
f T = E (4)
CAM
SPI
Fig. 9. Low-End node architecture
Delivery of higher image resolutions without compression can be supported if the camera
sensor includes internal memory. In this case, nothing but performance limits the transmis-
sion. In any case, raw images can be delivered without processing if they are split on the y,
transmitting each row after being read from the camera. With this scheme, just one row needs
to be saved in the memory before delivering.
Without transport protocols, a packet loss can degrade the image at the sink node. In this
case a simple interpolation algorithm (e.g. bilinear) can be used to estimate the lost pixels and
reconstruct the image. In a burst of lost packets, a region of the image will be completely lost
and restoration will be impossible.
Reliability of this scheme and the quality of the restored image will increase if the rows are
also split into two packets with the odd and even pixels, respectively. Choosing different
delivery paths for both increases the probability of a packet loss, but one of them will also be
more likely to reach the sink node.
Again, lost pixels can be estimated using an interpolation algorithm. But in this case, the
real pixels are uniformly distributed throughout the image, so the quality of the image recon-
structed by interpolation increases. See gure 10
CAM
ROW 2
ROW 1
ROW 1 ROW 2
NETWORK PATH
NON-INTERLEAVING SCHEME
ROW 2
ROW 2
ROW 1
ROW 1
NETWORK PATH 2 ROW 2
ROW 1
NETWORK PATH 1
INTERLEAVING SCHEME
CORRECT PIXELS
LOST & INTERPOLED PIXEL
LOW-END ARCHITECTURE (SOURCE NODE)
IMAGE AT SINK NODE
Fig. 10. Split on the Fly scheme (with and without interleaving)
4.2 Medium-End architecture
SoF is a suitable algorithmfor MCUs with a lack of memory. But the energy required to deliver
the image through the network is high. Let N be the size of a square image. Then,
E = O(N
2
) (1)
A compression algorithm can reduce this energy signicantly, but no splitting on the y scheme
is possible, because the original and compressed images must be saved in the memory. An
external memory (i.e. an I2C RAM) can solve the problem. But, the chip increases the en-
ergy consumption in three ways: its power supply, the time needed to transfer data from the
camera (seconds), and a factor caused by the inefcient data access during the compression
algorithm.
CAM
SPI
I2C,
Flash, ..
External Memory
Fig. 11. MCU with external memory
As shown in (Molina et al., 2009) all of them are insignicant compared to the required pro-
cessing time. Using Split on the Fly scheme as the reference, the energy saving of this solution
can be approximated to:
E N
2
E
BB
(1 R
C
) + T
P
PWC (2)
where R
C
is the compression rate, E
BB
the energy to broadcast a byte, T
P
the processing time,
and PWC the CPU power consumption for the processing time. The rst factor, R
C
, is directly
related to the image quality. The Haar transform (Raviraj & Sanavullah, 2007) is recommend-
able because it reduces the image size with a reasonable computation cost and high quality.
4.3 High-End architecture
To reduce the processing time, we propose a dual processor architecture: A low performance
MCU managing the WSN services, and a high performance microcontroller acting as a multi-
media co-processor. Even though the power consumption of this co-processor is much higher,
the processing time is substantially lower. But co-processor selection is not trivial. If the option
is another MCU or a processing card, a detailed analysis is necessary because it is possible not
to achieve the desired energy reduction. This is the case of CMUCam (Carnegie Mellon Uni-
versity, 2007) . Multimedia co-processor should be optimized for data processing, and must
necessarily contain an internal structure far different from the main MCU (e.g. PIC32XMX-
460 with 32K RAM selected in (Molina et al., 2009)). A simple overclocked version of the main
CPU reduces the processing time proportionally, but the power consumption increases the
same rate, as shown following:
Using the general equation of power consumption of a CMOS integrated circuit (IC):
P
IC
= KV
2
f (3)
The energy in an overclocked CPU (E
OV
) for a period T
OV
is
E
OV
= P
OV
T
OV
= KV
2
f
OV
T
OV
= KV
2
f
OV
T
f
f
OV
= KV
2
f T = E (4)
The camera sensor is connected directly to the co-processor, so no external memory is needed,
and consequently, no delay time to transfer the pixels, no extra time to data access, etc.
NODE MICROCONTROLLER
IMAGE CO-PROCESSOR
Fig. 12. High-End node architecture
5. Conclusions and future work
Extending proposed concepts and optimizations on any WMSNs standard architectures is
obvious. Split on the Fly algorithm (SoF) is very easy to implement in any platform, and it can
serve as a reference to compare it with.
Balancing the computing load between a dual processor architecture as proposed in the high-
end one, allows the optimization of energy consumption, adapting the performance to any
instantaneous processing prole.
Using FPGAs to implement the image co-processor is a promising research line. The capac-
ity of a FPGA to execute parallel simple operations matches with the spatial redundancy of
many image processing algorithms. It is reasonable to assume that the optimization of these
algorithms for FPGAs, will reduce the processing time up to milliseconds, decreasing the
processing energy signicantly. Compared to ASICs, the power consumption of an FPGA
is considerably higher. However, an FPGA can be reprogrammed to update it to a specic
application.
6. References
Aaron, A., Rane, S., Zhang, R. & Girod, B. (2003). Wyner-ziv coding for video: Applications
to compression and error resilience, DCC 03: Proceedings of the Conference on Data
Compression, IEEE Computer Society, Washington, DC, USA, p. 93.
Abrach, H., Bhatti, S., Carlson, J., Dai, H., Rose, J., Sheth, A., Shucker, B., Deng, J. & Han,
R. (2003). Mantis: System support for multimodal networks of in-situ sensors, In
2nd ACM International Workshop on Wireless Sensor Networks and Applications (WSNA,
pp. 5059.
Akyildiz, I. F., Melodia, T. & Chowdhury, K. R. (2008). Wireless multimedia sensor networks:
Applications and testbeds, Proceedings of the IEEE 96(10): 15881605.
URL: http://dx.doi.org/10.1109/JPROC.2008.928756
Alippi, C., Camplani, R. & Galperti, C. (2007). Lossless compression techniques in wireless
sensor networks: Monitoring microacoustic emissions, Robotic and Sensors Environ-
ments.
Arora, A., Dutta, P., Bapat, S., Kulathumani, V., Zhang, H., Naik, V., Mittal, V., Cao, H.,
Demirbas, M., Gouda, M., Choi, Y., Herman, T., Kulkarni, S., Arumugam, U.,
Nesterenko, M., Vora, A. & Miyashita, M. (2004). A line in the sand: a wireless
sensor network for target detection, classication, and tracking, Computer Networks
46(5): 605 634. Military Communications Systems and Technologies.
URL: http://www.sciencedirect.com/science/article/B6VRG-4CXKNBT-
1/2/38791b700b4804d30e9df8aca23712e7
Bnzit, F. (2009). Distributed Average Consensus for Wireless Sensor Networks, PhD thesis, cole
Polytechnique Fdrale de Lausanne.
Cao, Q. & Abdelzaher, T. (n.d.). The liteos operating system: Towards unix-like abstractions
for wireless sensor networks.
Cao, Q., Yan, T., Stankovic, J. &Abdelzaher, T. (2005). Analysis of target detection performance
for wireless sensor networks, in V. K. Prasanna, S. Iyengar, P. G. Spirakis & M. Welsh
(eds), Distributed Computing in Sensor Systems, Vol. 3560 of Lecture Notes in Computer
Science, Springer Berlin / Heidelberg, pp. 276292.
Carnegie Mellon University (2007). Cmucam3 datasheet version 1.02. Pittsburgh, PA.
Cha, H., Choi, S., Jung, I., Kim, H., Shin, H., Yoo, J. & Yoon, C. (2007). Retos: resilient, expand-
able, and threaded operating system for wireless sensor networks, IPSN 07: Proceed-
ings of the 6th international conference on Information processing in sensor networks, ACM,
New York, NY, USA, pp. 148157.
Chia, W. C., Ang, L.-M. & Seng, K. P. (2009). Multiview image compression for wireless
multimedia sensor network using image stitching and spiht coding with ezw tree
structure, Intelligent Human-Machine Systems and Cybernetics, International Conference
on 2: 298301.
Dai, H. & Han, R. (2004). Tsync: a lightweight bidirectional time synchronization service for
wireless sensor networks, SIGMOBILE Mob. Comput. Commun. Rev. 8(1): 125139.
Dunkels, A., Grnvall, B. & Voigt, T. (2004). Contiki - a lightweight and exible operating
system for tiny networked sensors.
Elson, J. &Rmer, K. (2003). Wireless sensor networks: a newregime for time synchronization,
SIGCOMM Comput. Commun. Rev. 33(1): 149154.
Eswaran, A., Rowe, A. & Rajkumar, R. (2005). Nano-rk: An energy-aware resource-centric
rtos for sensor networks, RTSS 05: Proceedings of the 26th IEEE International Real-Time
Systems Symposium, IEEE Computer Society, Washington, DC, USA, pp. 256265.
Faheem, Y., Boudjit, S. & Chen, K. (2010). Wireless multimedia sensor networks:requirements
& current paradigm, L Update .
International Conference on Sensors.
Feng, W., Kaiser, E. & Le Baillif, M. (2005). Panoptes: scalable low-power video sensor net-
working technologies., ACM Trans. Multimedia Comput. Commun. Appl. 1: 151167.
Gay, D., Levis, P., von Behren, R., Welsh, M., Brewer, E. &Culler, D. (2003). The nesc language:
A holistic approach to networked embedded systems, PLDI 03: Proceedings of the
ACM SIGPLAN 2003 conference on Programming language design and implementation,
Girao, J., Westhoff, D., Mykletun, E. & Araki, T. (2007). Tinypeds: Tiny persistent encrypted
data storage in asynchronous wireless sensor networks, Ad Hoc Networks 5(7): 1073
1089.
Gu, L. (2006). t-kernel: Providing reliable os support to wireless sensor networks, In Proc. of the
4th ACM Conf. on Embedded Networked Sensor Systems (SenSys, ACM Press, pp. 114.
Han, C.-c., Kumar, R., Shea, R., Kohler, E. & Srivastava, M. (n.d.). Abstract a dynamic operat-
ing system for sensor nodes.
He, T., Huang, C., Blum, B. M., Stankovic, J. A. & Abdelzaher, T. (2003). Range-free local-
ization schemes for large scale sensor networks, MobiCom 03: Proceedings of the 9th
The camera sensor is connected directly to the co-processor, so no external memory is needed,
and consequently, no delay time to transfer the pixels, no extra time to data access, etc.
NODE MICROCONTROLLER
IMAGE CO-PROCESSOR
Fig. 12. High-End node architecture
5. Conclusions and future work
Extending proposed concepts and optimizations on any WMSNs standard architectures is
obvious. Split on the Fly algorithm (SoF) is very easy to implement in any platform, and it can
serve as a reference to compare it with.
Balancing the computing load between a dual processor architecture as proposed in the high-
end one, allows the optimization of energy consumption, adapting the performance to any
instantaneous processing prole.
Using FPGAs to implement the image co-processor is a promising research line. The capac-
ity of a FPGA to execute parallel simple operations matches with the spatial redundancy of
many image processing algorithms. It is reasonable to assume that the optimization of these
algorithms for FPGAs, will reduce the processing time up to milliseconds, decreasing the
processing energy signicantly. Compared to ASICs, the power consumption of an FPGA
is considerably higher. However, an FPGA can be reprogrammed to update it to a specic
application.
6. References
Aaron, A., Rane, S., Zhang, R. & Girod, B. (2003). Wyner-ziv coding for video: Applications
to compression and error resilience, DCC 03: Proceedings of the Conference on Data
Compression, IEEE Computer Society, Washington, DC, USA, p. 93.
Abrach, H., Bhatti, S., Carlson, J., Dai, H., Rose, J., Sheth, A., Shucker, B., Deng, J. & Han,
R. (2003). Mantis: System support for multimodal networks of in-situ sensors, In
2nd ACM International Workshop on Wireless Sensor Networks and Applications (WSNA,
pp. 5059.
Akyildiz, I. F., Melodia, T. & Chowdhury, K. R. (2008). Wireless multimedia sensor networks:
Applications and testbeds, Proceedings of the IEEE 96(10): 15881605.
URL: http://dx.doi.org/10.1109/JPROC.2008.928756
Alippi, C., Camplani, R. & Galperti, C. (2007). Lossless compression techniques in wireless
sensor networks: Monitoring microacoustic emissions, Robotic and Sensors Environ-
ments.
Arora, A., Dutta, P., Bapat, S., Kulathumani, V., Zhang, H., Naik, V., Mittal, V., Cao, H.,
Demirbas, M., Gouda, M., Choi, Y., Herman, T., Kulkarni, S., Arumugam, U.,
Nesterenko, M., Vora, A. & Miyashita, M. (2004). A line in the sand: a wireless
sensor network for target detection, classication, and tracking, Computer Networks
46(5): 605 634. Military Communications Systems and Technologies.
URL: http://www.sciencedirect.com/science/article/B6VRG-4CXKNBT-
1/2/38791b700b4804d30e9df8aca23712e7
Bnzit, F. (2009). Distributed Average Consensus for Wireless Sensor Networks, PhD thesis, cole
Polytechnique Fdrale de Lausanne.
Cao, Q. & Abdelzaher, T. (n.d.). The liteos operating system: Towards unix-like abstractions
for wireless sensor networks.
Cao, Q., Yan, T., Stankovic, J. &Abdelzaher, T. (2005). Analysis of target detection performance
for wireless sensor networks, in V. K. Prasanna, S. Iyengar, P. G. Spirakis & M. Welsh
(eds), Distributed Computing in Sensor Systems, Vol. 3560 of Lecture Notes in Computer
Science, Springer Berlin / Heidelberg, pp. 276292.
Carnegie Mellon University (2007). Cmucam3 datasheet version 1.02. Pittsburgh, PA.
Cha, H., Choi, S., Jung, I., Kim, H., Shin, H., Yoo, J. & Yoon, C. (2007). Retos: resilient, expand-
able, and threaded operating system for wireless sensor networks, IPSN 07: Proceed-
ings of the 6th international conference on Information processing in sensor networks, ACM,
Chia, W. C., Ang, L.-M. & Seng, K. P. (2009). Multiview image compression for wireless
multimedia sensor network using image stitching and spiht coding with ezw tree
structure, Intelligent Human-Machine Systems and Cybernetics, International Conference
on 2: 298301.
Dai, H. & Han, R. (2004). Tsync: a lightweight bidirectional time synchronization service for
wireless sensor networks, SIGMOBILE Mob. Comput. Commun. Rev. 8(1): 125139.
Dunkels, A., Grnvall, B. & Voigt, T. (2004). Contiki - a lightweight and exible operating
system for tiny networked sensors.
Elson, J. &Rmer, K. (2003). Wireless sensor networks: a newregime for time synchronization,
SIGCOMM Comput. Commun. Rev. 33(1): 149154.
Eswaran, A., Rowe, A. & Rajkumar, R. (2005). Nano-rk: An energy-aware resource-centric
rtos for sensor networks, RTSS 05: Proceedings of the 26th IEEE International Real-Time
Systems Symposium, IEEE Computer Society, Washington, DC, USA, pp. 256265.
Faheem, Y., Boudjit, S. & Chen, K. (2010). Wireless multimedia sensor networks:requirements
& current paradigm, L Update .
International Conference on Sensors.
Feng, W., Kaiser, E. & Le Baillif, M. (2005). Panoptes: scalable low-power video sensor net-
working technologies., ACM Trans. Multimedia Comput. Commun. Appl. 1: 151167.
Gay, D., Levis, P., von Behren, R., Welsh, M., Brewer, E. &Culler, D. (2003). The nesc language:
A holistic approach to networked embedded systems, PLDI 03: Proceedings of the
ACM SIGPLAN 2003 conference on Programming language design and implementation,
Girao, J., Westhoff, D., Mykletun, E. & Araki, T. (2007). Tinypeds: Tiny persistent encrypted
data storage in asynchronous wireless sensor networks, Ad Hoc Networks 5(7): 1073
1089.
Gu, L. (2006). t-kernel: Providing reliable os support to wireless sensor networks, In Proc. of the
4th ACM Conf. on Embedded Networked Sensor Systems (SenSys, ACM Press, pp. 114.
Han, C.-c., Kumar, R., Shea, R., Kohler, E. & Srivastava, M. (n.d.). Abstract a dynamic operat-
ing system for sensor nodes.
He, T., Huang, C., Blum, B. M., Stankovic, J. A. & Abdelzaher, T. (2003). Range-free local-
ization schemes for large scale sensor networks, MobiCom 03: Proceedings of the 9th
annual international conference on Mobile computing and networking, ACM, New York,
NY, USA, pp. 8195.
Hengstler, S., Prashanth, D., Fong, S. & Aghajan, H. (2007). Mesheye: A hybrid-resolution
smart camera mote for applications in distributed intelligent surveillance., Int. Conf.
Inf. Process. Sensor Netw. (IPSN), Cambridge, MA, USA., pp. 360369.
Hill, J., Szewczyk, R., Woo, A., Hollar, S., Culler, D. & Pister, K. (2000). System architecture di-
rections for networked sensors, ASPLOS-IX: Proceedings of the ninth international con-
ference on Architectural support for programming languages and operating systems, ACM,
Hu, P., Zhou, Z., Liu, Q. & Li, F. (2007). The hmm-based modeling for the energy level predic-
tion in wireless sensor networks, Industrial Electronics and Applications, 2007. ICIEA
2007. 2nd IEEE Conference on, pp. 2253 2258.
Ihler, A. T., Fisher, III, J. W., Moses, R. L. & Willsky, A. S. (2004). Nonparametric belief prop-
agation for self-calibration in sensor networks, IPSN 04: Proceedings of the 3rd inter-
national symposium on Information processing in sensor networks, ACM, New York, NY,
USA, pp. 225233.
IMOTE2 (n.d.). http://www.xbow.com/Products/Imote2.aspx (accessed August 16, 2009).
Kellner, S., Pink, M., Meier, D. &Blass, E.-O. (2008). Towards a realistic energy model for wire-
less sensor networks, Wireless on Demand Network Systems and Services, 2008. WONS
2008. Fifth Annual Conference on, pp. 97 100.
Kulkarni, P., Ganesan, D., Shenoy, P. & Lu, Q. (2005). Senseye: A multi-tier camera sensor
network, ACM Multimedia.
Lewis, A. & Knowles, G. (1992). Image compression using the 2-d wavelet transform, IEEE
Transactions on Image Processing 1: 244250.
Mao, G., Fridan, B. & Anderson, B. (2007). Wireless sensor network localization techniques,
Computer Networks. Elsevier .
Marti, M., Kusy, B., Simon, G. & Ldeczi, A. (2004). The ooding time synchronization pro-
tocol, SenSys 04: Proceedings of the 2nd international conference on Embedded networked
sensor systems, ACM, New York, NY, USA, pp. 3949.
Miluzzo, E., Lane, N., Campbell, A. & Olfati-Saber, R. (2008). Calibree: A self-calibration sys-
tem for mobile sensor networks, in S. Nikoletseas, B. Chlebus, D. Johnson & B. Krish-
namachari (eds), Distributed Computing in Sensor Systems, Vol. 5067 of Lecture Notes in
Computer Science, Springer Berlin / Heidelberg, pp. 314331.
Molina, F. J., Babancho, J., Mora, J. M. &Leon, C. (2009). Poster abstract: Practical issues in im-
age acquisition and transmission over wireless sensor network, IPSN 09: Proceedings
of the 2009 International Conference on Information Processing in Sensor Networks, IEEE
Mottola, L. & Picco, G. P. (2010). Programming wireless sensor networks: Fundamental con-
cepts and state of the art, ACM Computing Surveys .
Patwari, N., Ash, J., Iii, A. O. H., Hero, A. O., Moses, R. M., Correal, N. S., Labs, M., Kyper-
ountas, S., Correal, N. S. &(b, A. (2005). Locating the nodes: Cooperative localization
in wireless sensor networks.
Polastre, J., Szewczyk, R., Sharp, C. & Culler, D. (2004). The mote revolution: low power
wireless sensor network devices., Hot Chips 16: A Symposium on High Performance
Chips.
Rahimi, M., Baer, R., Iroezi, O., Garcia, J., Warrior, J., Estrin, D. & Srivastava, M. (2005). Cy-
clops: In situ image sensing and interpretation in wireless sensor networks., ACM
Conf. Embed. Netw: Sensor Syst. (SenSys), San Diego, CA, USA, pp. 192204.
Raviraj, P. & Sanavullah, M. (2007). The modied 2D-Haar transformation in image compres-
sion, Middle-East journal of Scientic Research 2(2): 7378.
Rubio, B., Diaz, M. &Troya, J. M. (2007). Programming approaches and challenges for wireless
sensor networks, ICSNC 07: Proceedings of the Second International Conference on Sys-
tems and Networks Communications, IEEE Computer Society, Washington, DC, USA,
p. 36.
Sayood, K. (2005). Introduction to Data Compression, Third Edition (Morgan Kaufmann Series in
Multimedia Information and Systems), Morgan Kaufmann Publishers Inc., San Fran-
cisco, CA, USA.
Sheng, B., Li, Q. & Mao, W. (2006). Data storage placement in sensor networks, MobiHoc
06: Proceedings of the 7th ACM international symposium on Mobile ad hoc networking and
computing, ACM, New York, NY, USA, pp. 344355.
Soro, S. & Heinzelman, W. (2009). A survey of visual sensor networks, Advances in Multimedia
pp. 122.
Stargate XScale Platform (n.d.). http://blog.xbox.com/xblog/Stargate_Xscale_Platform/ (ac-
cessed August 16, 2009).
Villasenor, J., Belzer, B. &Liao, J. (1995). Wavelet lter evaluation for image compression, IEEE
Zhong, L., Rabaey, J. & Wolisz, A. (2004). An integrated data-link energy model for wire-
less sensor networks, Communications, 2004 IEEE International Conference on, Vol. 7,
pp. 3777 3783 Vol.7.
Ziv, J. & Lempel, A. (1978). Compression of individual sequences via variable-rate coding.
Zoumboulakis, M. & Roussos, G. (n.d.). Integer-based optimisations for resource-constrained
sensor platforms.
annual international conference on Mobile computing and networking, ACM, New York,
NY, USA, pp. 8195.
Hengstler, S., Prashanth, D., Fong, S. & Aghajan, H. (2007). Mesheye: A hybrid-resolution
smart camera mote for applications in distributed intelligent surveillance., Int. Conf.
Inf. Process. Sensor Netw. (IPSN), Cambridge, MA, USA., pp. 360369.
Hill, J., Szewczyk, R., Woo, A., Hollar, S., Culler, D. & Pister, K. (2000). System architecture di-
rections for networked sensors, ASPLOS-IX: Proceedings of the ninth international con-
ference on Architectural support for programming languages and operating systems, ACM,
Hu, P., Zhou, Z., Liu, Q. & Li, F. (2007). The hmm-based modeling for the energy level predic-
tion in wireless sensor networks, Industrial Electronics and Applications, 2007. ICIEA
2007. 2nd IEEE Conference on, pp. 2253 2258.
Ihler, A. T., Fisher, III, J. W., Moses, R. L. & Willsky, A. S. (2004). Nonparametric belief prop-
agation for self-calibration in sensor networks, IPSN 04: Proceedings of the 3rd inter-
national symposium on Information processing in sensor networks, ACM, New York, NY,
USA, pp. 225233.
IMOTE2 (n.d.). http://www.xbow.com/Products/Imote2.aspx (accessed August 16, 2009).
Kellner, S., Pink, M., Meier, D. &Blass, E.-O. (2008). Towards a realistic energy model for wire-
less sensor networks, Wireless on Demand Network Systems and Services, 2008. WONS
2008. Fifth Annual Conference on, pp. 97 100.
Kulkarni, P., Ganesan, D., Shenoy, P. & Lu, Q. (2005). Senseye: A multi-tier camera sensor
network, ACM Multimedia.
Lewis, A. & Knowles, G. (1992). Image compression using the 2-d wavelet transform, IEEE
Mao, G., Fridan, B. & Anderson, B. (2007). Wireless sensor network localization techniques,
Computer Networks. Elsevier .
Marti, M., Kusy, B., Simon, G. & Ldeczi, A. (2004). The ooding time synchronization pro-
tocol, SenSys 04: Proceedings of the 2nd international conference on Embedded networked
sensor systems, ACM, New York, NY, USA, pp. 3949.
Miluzzo, E., Lane, N., Campbell, A. & Olfati-Saber, R. (2008). Calibree: A self-calibration sys-
tem for mobile sensor networks, in S. Nikoletseas, B. Chlebus, D. Johnson & B. Krish-
namachari (eds), Distributed Computing in Sensor Systems, Vol. 5067 of Lecture Notes in
Computer Science, Springer Berlin / Heidelberg, pp. 314331.
Molina, F. J., Babancho, J., Mora, J. M. &Leon, C. (2009). Poster abstract: Practical issues in im-
age acquisition and transmission over wireless sensor network, IPSN 09: Proceedings
of the 2009 International Conference on Information Processing in Sensor Networks, IEEE
Mottola, L. & Picco, G. P. (2010). Programming wireless sensor networks: Fundamental con-
cepts and state of the art, ACM Computing Surveys .
Patwari, N., Ash, J., Iii, A. O. H., Hero, A. O., Moses, R. M., Correal, N. S., Labs, M., Kyper-
ountas, S., Correal, N. S. &(b, A. (2005). Locating the nodes: Cooperative localization
in wireless sensor networks.
Polastre, J., Szewczyk, R., Sharp, C. & Culler, D. (2004). The mote revolution: low power
wireless sensor network devices., Hot Chips 16: A Symposium on High Performance
Chips.
Rahimi, M., Baer, R., Iroezi, O., Garcia, J., Warrior, J., Estrin, D. & Srivastava, M. (2005). Cy-
clops: In situ image sensing and interpretation in wireless sensor networks., ACM
Conf. Embed. Netw: Sensor Syst. (SenSys), San Diego, CA, USA, pp. 192204.
Raviraj, P. & Sanavullah, M. (2007). The modied 2D-Haar transformation in image compres-
sion, Middle-East journal of Scientic Research 2(2): 7378.
Rubio, B., Diaz, M. &Troya, J. M. (2007). Programming approaches and challenges for wireless
sensor networks, ICSNC 07: Proceedings of the Second International Conference on Sys-
tems and Networks Communications, IEEE Computer Society, Washington, DC, USA,
p. 36.
Sayood, K. (2005). Introduction to Data Compression, Third Edition (Morgan Kaufmann Series in
Multimedia Information and Systems), Morgan Kaufmann Publishers Inc., San Fran-
cisco, CA, USA.
Sheng, B., Li, Q. & Mao, W. (2006). Data storage placement in sensor networks, MobiHoc
06: Proceedings of the 7th ACM international symposium on Mobile ad hoc networking and
computing, ACM, New York, NY, USA, pp. 344355.
Soro, S. & Heinzelman, W. (2009). A survey of visual sensor networks, Advances in Multimedia
pp. 122.
Stargate XScale Platform (n.d.). http://blog.xbox.com/xblog/Stargate_Xscale_Platform/ (ac-
cessed August 16, 2009).
Villasenor, J., Belzer, B. &Liao, J. (1995). Wavelet lter evaluation for image compression, IEEE
Zhong, L., Rabaey, J. & Wolisz, A. (2004). An integrated data-link energy model for wire-
less sensor networks, Communications, 2004 IEEE International Conference on, Vol. 7,
pp. 3777 3783 Vol.7.
Ziv, J. & Lempel, A. (1978). Compression of individual sequences via variable-rate coding.
Zoumboulakis, M. & Roussos, G. (n.d.). Integer-based optimisations for resource-constrained
sensor platforms.
Imaging in UWB Sensor Networks
0
Imaging in UWB Sensor Networks
Technische Universitt Ilmenau
Germany
1. Introduction
Sensor networks consist of a number of spatially distributed nodes. These nodes perform
measurements and collect information about their surrounding. They transfer data to neigh-
boring nodes or to a data fusion center. Often measurements are performed in cooperation of
several nodes.
If the network consists of Ultra-Wideband (UWB) radar sensors, the network infrastructure
can be used for a rough imaging of the surrounding. In this way bigger objects (walls, pillars,
machines, furniture) can be detected and their position and shape can be estimated. These
are valuable information for the autonomous orientation of robots and for the inspection of
buildings, especially in case of dangerous environments (re, smoke, dust, dangerous gases).
Applications of UWB sensor networks are summarized in Thom (2007).
In this article basic aspects of imaging in UWB sensor networks are discussed. We start with
a brief description of two types of UWB radar devices: impulse radar and Noise/M-sequence
radar. Network imaging is based on principles of Synthetic Aperture Radar (SAR). Starting
from SAR some special aspects of imaging in networks are explained in section 3. Sections 4
and 5 form the main part of the article. Here two different imaging approaches are described
in more detail. The rst method is multistatic imaging, i.e. the measurements are performedin
cooperation of several sensor nodes at xed places and one mobile node. The second approach
is imaging by an autonomous mobile sensor, equipped with one Tx and two Rx units. This
sensor uses a network of xed nodes for its own orientation.
Part of the described methods have been practically realized in a laboratory environment.
Hence, practical examples support the presentation. Conclusions and references complete the
article.
2. Ultra Wideband (UWB) Radar
2.1 Main Characteristics
The main characteristic of UWB technology is the use of a very wide frequency range. A
system is referred to as UWB system if it operates in a frequency band of more than 500 MHz
width, or if the fractional bandwith bw
f
= 100% ( f
H
f
L
)/f
C
is larger than 25%. Here f
H
, f
L
,
and f
C
denote the upper and lower frequency limit and the centre frequency, respectively. For
imaging applications the large bandwidth is of interest because it guarantees a high resolution
in range direction, as explained in the next section. UWB systems always coexist with other
radio services, operating in the same frequency range. To avoid intereference, a number of
frequency masks and power restrictions have been agreedinternationally. Current regulations
are summarized in FCC (2002); Luediger & Kallenborn (2009).
24
A number of principles for UWB radar systems have been proposed (see Sachs et al. (2003)).
In the rest of this section we briey describe the two dominant methods Impulse Radar and
M-Sequence Radar.
2.2 Impulse Radar
An Impulse radar measures distances by transmission of single RF pulses and subsequent
reception of echoe signals. The frequency spectrum of an electric pulse covers a bandwidth
which is inversely proportional to its duration. To achieve ultra-wide bandwidth, the single
pulses generated in an impulse radar must have a duration of t
pulse
1ns or even less. They
can be generated by means of switching diodes, transistors or even laser-actuated semicon-
ductor switches, see Hussain (1998) for an overview. Pulse shaping is required to adapt the
frequency spectrum to common frequency masks. The principle of an impulse radar is shown
in Fig. 1.
Pulse
Gener.
Tx Rx
s
T
(t)
s
R
(t)
object
h
C
(t)
Fig. 1. Principle of an impulse radar.
The transmitter signal s
T
(t) is radiated by the Tx-antenna. The received signal s
R
(t) consists
of a small fraction of the transmitted energy that was scattered at the object. s
R
(t) can be
calculated by convolution of s
T
(t) with the channel impulse response h
C
(t):
s
R
(t) =
h
C
(t
) s
T
(t t
) dt
. (1)
Determination of the channel impulse reponse is possible via de-convolution, favourably per-
formed with the Fourier-transformed quantities S
R
(), S
T
() in the frequency domain:
h
C
(t) = F
1
S
R
()
S
T
()
F()
. (2)
F() is a bandpass lter that suppresses high amplitudes at the edges of the frequency band,
and F
1
symbolizes the inverse Fourier transform.
The minimum delay between two subsequent pulses (repetition time t
rep
) is given by t
rep
=
d
max
/c, where d
max
is the maximum propagation distance and c is the speed of light. For
smaller pulse distances no unique identication of pulse propagation times would be possi-
ble, especially in the case of more than one object. The necessity to introduce t
rep
limits the
average signal energy since only a fraction t
pulse
/t
rep
of the total measurement time is used for
transmission and the signal peak amplitude must not exceed the allowed power restrictions.
Advantageously, the temporal shift between transmission and reception of signals reduces the
problem of Tx/Rx crosstalk.
2.3 Noise Radar and M-Sequence Radar
Noise signals can possess a frequency spectrum which is as wide as the spectrum of a single
short pulse. Because of random phase relations between the single Fourier components the
signal energy of a noise signal is distributed over the entire time axis. Signals of this kind can
be used in radar systems and an example is shown in Fig. 2.
Noise
Gener.
Tx x
Rx

s
T
(t) s
R
(t) s
P
(t,)
s
T
(t-)
s
P
()
object
Fig. 2. Principle of a noise radar.
The relation between s
R
(t), h
C
(t), and s
T
(t) is of course the one already given in Equ. (1). In
this kind of radar device, information about the propagation channel is extracted by corre-
lation of the received signal with the transmitted s
T
(t). The correlator consists of a variable
delay element introducing delay , a multiplicator that produces the product signal s
P
(t, ):
s
P
(t, ) = s
R
(t) s
T
(t ) (3)
and an integrator that forms the signal average for one particular over all t. Introduction of
the convolution integral (1) into (3) and averaging over a time interval that is long in compar-
ison to usual signal variations gives the following expression for the averaged signal s
P
():
s
P
() = lim
T
1
2T
T
T
s
R
(t) s
T
(t ) dt = lim
T
1
2T
T
h
C
(t
) s
T
(t t
) s
T
(t ) dt
dt. (4)
In case of white noise the average value of the product s
T
(t t
) s
T
(t ) is always zero,
except for t
= . This means the autocorrelation function of white noise is a -function. So

we can perform the following substitution:
lim
T
1
2T
T
T
s
T
(t t
) s
T
(t ) dt = (t
) lim
T
1
2T
T
T
s
2
T
(t) dt = (t
) s
2
T
(t). (5)
Applying this result we see that the correlation of noise excitation s
T
(t) and receiver signal
s
R
(t) delivers the channel impulse response h
C
(t) multiplied by a constant factor. This factor
is the square of the effective value of s
T
(t):
s
P
() = s
2
T
(t)
h
C
(t
) (t
) dt
= s
2
T
(t) h
C
(). (6)
A number of principles for UWB radar systems have been proposed (see Sachs et al. (2003)).
In the rest of this section we briey describe the two dominant methods Impulse Radar and
M-Sequence Radar.
2.2 Impulse Radar
An Impulse radar measures distances by transmission of single RF pulses and subsequent
reception of echoe signals. The frequency spectrum of an electric pulse covers a bandwidth
which is inversely proportional to its duration. To achieve ultra-wide bandwidth, the single
pulses generated in an impulse radar must have a duration of t
pulse
1ns or even less. They
can be generated by means of switching diodes, transistors or even laser-actuated semicon-
ductor switches, see Hussain (1998) for an overview. Pulse shaping is required to adapt the
frequency spectrum to common frequency masks. The principle of an impulse radar is shown
in Fig. 1.
Pulse
Gener.
Tx Rx
s
T
(t)
s
R
(t)
object
h
C
(t)
Fig. 1. Principle of an impulse radar.
The transmitter signal s
T
(t) is radiated by the Tx-antenna. The received signal s
R
(t) consists
of a small fraction of the transmitted energy that was scattered at the object. s
R
(t) can be
calculated by convolution of s
T
(t) with the channel impulse response h
C
(t):
s
R
(t) =
h
C
(t
) s
T
(t t
) dt
. (1)
Determination of the channel impulse reponse is possible via de-convolution, favourably per-
formed with the Fourier-transformed quantities S
R
(), S
T
() in the frequency domain:
h
C
(t) = F
1
S
R
()
S
T
()
F()
. (2)
F() is a bandpass lter that suppresses high amplitudes at the edges of the frequency band,
and F
1
symbolizes the inverse Fourier transform.
The minimum delay between two subsequent pulses (repetition time t
rep
) is given by t
rep
=
d
max
/c, where d
max
is the maximum propagation distance and c is the speed of light. For
smaller pulse distances no unique identication of pulse propagation times would be possi-
ble, especially in the case of more than one object. The necessity to introduce t
rep
limits the
average signal energy since only a fraction t
pulse
/t
rep
of the total measurement time is used for
transmission and the signal peak amplitude must not exceed the allowed power restrictions.
Advantageously, the temporal shift between transmission and reception of signals reduces the
problem of Tx/Rx crosstalk.
2.3 Noise Radar and M-Sequence Radar
Noise signals can possess a frequency spectrum which is as wide as the spectrum of a single
short pulse. Because of random phase relations between the single Fourier components the
signal energy of a noise signal is distributed over the entire time axis. Signals of this kind can
be used in radar systems and an example is shown in Fig. 2.
Noise
Gener.
Tx x
Rx

s
T
(t) s
R
(t) s
P
(t,)
s
T
(t-)
s
P
()
object
Fig. 2. Principle of a noise radar.
The relation between s
R
(t), h
C
(t), and s
T
(t) is of course the one already given in Equ. (1). In
this kind of radar device, information about the propagation channel is extracted by corre-
lation of the received signal with the transmitted s
T
(t). The correlator consists of a variable
delay element introducing delay , a multiplicator that produces the product signal s
P
(t, ):
s
P
(t, ) = s
R
(t) s
T
(t ) (3)
and an integrator that forms the signal average for one particular over all t. Introduction of
the convolution integral (1) into (3) and averaging over a time interval that is long in compar-
ison to usual signal variations gives the following expression for the averaged signal s
P
():
s
P
() = lim
T
1
2T
T
T
s
R
(t) s
T
(t ) dt = lim
T
1
2T
T
h
C
(t
) s
T
(t t
) s
T
(t ) dt
dt. (4)
In case of white noise the average value of the product s
T
(t t
) s
T
(t ) is always zero,
except for t
= . This means the autocorrelation function of white noise is a -function. So

we can perform the following substitution:
lim
T
1
2T
T
T
s
T
(t t
) s
T
(t ) dt = (t
) lim
T
1
2T
T
T
s
2
T
(t) dt = (t
) s
2
T
(t). (5)
Applying this result we see that the correlation of noise excitation s
T
(t) and receiver signal
s
R
(t) delivers the channel impulse response h
C
(t) multiplied by a constant factor. This factor
is the square of the effective value of s
T
(t):
s
P
() = s
2
T
(t)
h
C
(t
) (t
) dt
= s
2
T
(t) h
C
(). (6)
An M-sequnce radar is a special form of a noise radar, where s
T
(t) consists of a maximum
length binary sequence, see Sachs (2004) for details. This pseudo-stochastic signal is gener-
ated in a shift register with feedback. Both noise radar and M-sequence radar use the full
measurement duration for transmission and reception of signals, maximizing the UWB signal
energy in this way. Decoupling between transmitter and receiver becomes more important
since Tx and Rx operate at the same time.
3. Specics of Imaging in Sensor Networks
3.1 Synthetic Aperture Radar (SAR)
Imaging in sensor networks is based on results of conventional microwave imaging, i.e. imag-
ing with only one single Tx/Rx antenna pair. Especially the principles of "Synthetic Aperture
Radar" (SAR) can be adapted to the special needs of sensor network imaging. Instead of using
an antenna with a large aperture, here the aperture is synthesized by movement of antennas
and sequential data acquisition from different positions. For an overview of SAR imaging see
Oliver (1989), and for a typical UWB-SAR application see Gu et al. (2004). In 3.3 relations be-
tween the length of the scan path (aperture) and image resolution are explained. To achieve
reasonable resolution, the antenna aperture of a radar imaging system must be signicantly
bigger than the wavelength . Processing of SAR data is explained in connection with general
processing in 4.1 .
3.2 Arrangement of Network Nodes and Scan Path
The network consists of a number of nodes. These are individual sensors with Rx and/or Tx
capabilities. Specialized nodes can collect data from several other nodes, and typically one
node forms the fusion center, where the image is computed from the totality of the acquired
data.
The network can be completed by so called anchor nodes. These are nodes at known, xed
positions. Primarily they support position estimation of the mobile nodes, but additionally
they can be employed in the imaging process.
The spatial arrangement of network nodes (network topology) strongly inuences the per-
formance of an imaging network. Together topology and scan path must guarantee that all
objects are illuminated by the Tx antennas and that a signicant part of the scattered radiation
can be collected by the Rx antennas.
A number of frequently choosen scan geometries (node positions and scan paths) are shown
in Fig. 3. At least one antenna must move during the measurement; or an array of antennas
has to be used, as in Fig. 3(b). Two main cases can be distinguished with respect to the scan
path selection:
1. The object positions are already known. In this case imaging shall give information on
the shape of objects and small modications of their position.
2. The object positions are entirely unknown. In this case a rough image of the entire
surrounding has to be created.
The optimum scan geometry is concave shaped in case 1, e.g. Fig. 3(b) and (c). This shape
guarantees that the antennas are always directed towards the objects, so that a signicant part
of the scattered radiation is received. If the region of interest is accessible from one side only,
then semi circle or linear scan geometries are appropriate choices.
(a) (b)
(c)
(d)
Fig. 3. Typical scan geometries in imaging sensor networks: (a) linear scan, (b) full circle, (c)
semi circle, (d) arbitrary scan path. Filled triangles and circles: antennas, hatched gures:
objects.
In entirely unknown environment a previous optimization of node positions is not possible.
In this case the nodes are placed at random positions. They should have similar mutual dis-
tances. Node positioning can be improved after initial measurements, if some nodes dont
receive sufcient signals. A network of randomly placed nodes requires the use of omnidirec-
tional antennas, which can cause a reduction of the signal to clutter ration in comparison to
directional antennas.
3.3 Resolution
Resolution is a measure of up to which distance two closely spaced objects are still imaged
separately. In radar technique we must distinguish between range resolution
z
(along the
direction of wave propagation) and cross range resolution
x
(perpendicular to the direction
of propagation). An approximation for the former is:
z
=
c
2bw
. (7)
It is immediately understandable that
z
improves with the bandwidth bw because the speed
of light c divided by bw is a measure for the width of the propagating wave packet in the
spatial domain. The 2 results from two times passage of the geometrical distance in radar
measurements.
A rough estimation of cross range resolution
x
can be derived by means of Fig. 4. d and d
1
are
the path lengths to the end points of
x
when the antenna is at one end position of the aperture
A. We assume the criterion that two neighbouring points can be resolved if a path difference
d = d d
1
of the order of half the wavelength appears during movement of the antenna
along the aperture A, resulting in a signal phase difference of 2 (two way propagation).
Typically the distances are related to each other as follows: A
x
, R
x
, R > A. Under
An M-sequnce radar is a special form of a noise radar, where s
T
(t) consists of a maximum
length binary sequence, see Sachs (2004) for details. This pseudo-stochastic signal is gener-
ated in a shift register with feedback. Both noise radar and M-sequence radar use the full
measurement duration for transmission and reception of signals, maximizing the UWB signal
energy in this way. Decoupling between transmitter and receiver becomes more important
since Tx and Rx operate at the same time.
3. Specics of Imaging in Sensor Networks
3.1 Synthetic Aperture Radar (SAR)
Imaging in sensor networks is based on results of conventional microwave imaging, i.e. imag-
ing with only one single Tx/Rx antenna pair. Especially the principles of "Synthetic Aperture
Radar" (SAR) can be adapted to the special needs of sensor network imaging. Instead of using
an antenna with a large aperture, here the aperture is synthesized by movement of antennas
and sequential data acquisition from different positions. For an overview of SAR imaging see
Oliver (1989), and for a typical UWB-SAR application see Gu et al. (2004). In 3.3 relations be-
tween the length of the scan path (aperture) and image resolution are explained. To achieve
reasonable resolution, the antenna aperture of a radar imaging system must be signicantly
bigger than the wavelength . Processing of SAR data is explained in connection with general
processing in 4.1 .
3.2 Arrangement of Network Nodes and Scan Path
The network consists of a number of nodes. These are individual sensors with Rx and/or Tx
capabilities. Specialized nodes can collect data from several other nodes, and typically one
node forms the fusion center, where the image is computed from the totality of the acquired
data.
The network can be completed by so called anchor nodes. These are nodes at known, xed
positions. Primarily they support position estimation of the mobile nodes, but additionally
they can be employed in the imaging process.
The spatial arrangement of network nodes (network topology) strongly inuences the per-
formance of an imaging network. Together topology and scan path must guarantee that all
objects are illuminated by the Tx antennas and that a signicant part of the scattered radiation
can be collected by the Rx antennas.
A number of frequently choosen scan geometries (node positions and scan paths) are shown
in Fig. 3. At least one antenna must move during the measurement; or an array of antennas
has to be used, as in Fig. 3(b). Two main cases can be distinguished with respect to the scan
path selection:
1. The object positions are already known. In this case imaging shall give information on
the shape of objects and small modications of their position.
2. The object positions are entirely unknown. In this case a rough image of the entire
surrounding has to be created.
The optimum scan geometry is concave shaped in case 1, e.g. Fig. 3(b) and (c). This shape
guarantees that the antennas are always directed towards the objects, so that a signicant part
of the scattered radiation is received. If the region of interest is accessible from one side only,
then semi circle or linear scan geometries are appropriate choices.
(a) (b)
(c)
(d)
Fig. 3. Typical scan geometries in imaging sensor networks: (a) linear scan, (b) full circle, (c)
semi circle, (d) arbitrary scan path. Filled triangles and circles: antennas, hatched gures:
objects.
In entirely unknown environment a previous optimization of node positions is not possible.
In this case the nodes are placed at random positions. They should have similar mutual dis-
tances. Node positioning can be improved after initial measurements, if some nodes dont
receive sufcient signals. A network of randomly placed nodes requires the use of omnidirec-
tional antennas, which can cause a reduction of the signal to clutter ration in comparison to
directional antennas.
3.3 Resolution
Resolution is a measure of up to which distance two closely spaced objects are still imaged
separately. In radar technique we must distinguish between range resolution
z
(along the
direction of wave propagation) and cross range resolution
x
(perpendicular to the direction
of propagation). An approximation for the former is:
z
=
c
2bw
. (7)
It is immediately understandable that
z
improves with the bandwidth bw because the speed
of light c divided by bw is a measure for the width of the propagating wave packet in the
spatial domain. The 2 results from two times passage of the geometrical distance in radar
measurements.
A rough estimation of cross range resolution
x
can be derived by means of Fig. 4. d and d
1
are
the path lengths to the end points of
x
when the antenna is at one end position of the aperture
A. We assume the criterion that two neighbouring points can be resolved if a path difference
d = d d
1
of the order of half the wavelength appears during movement of the antenna
along the aperture A, resulting in a signal phase difference of 2 (two way propagation).
Typically the distances are related to each other as follows: A
x
, R
x
, R > A. Under
R
d
d
1
x
A/2
x
d
(a)
(b)
A/2
Fig. 4. Approximation of cross range resolution
x
. (b) is a zoomed version of the encircled
section in (a).
these circumstances d and d
1
can be assumed as being parallel on short lengthscales. The
angle appears both in the small triangle with sides
x
and d, and in the big triangle with
half aperture A/2 and range R:
sin =
d
x
, tan =
A/2
R
. (8)
After rearranging of both equations, insertion into each other, and application of the relation
sin(arctan(x)) = x/
1 + x
2
we get an expression for
x
:
x
=
d 2R
1 + (A/(2R))
2
2A

R
2A
. (9)
Here d was replaced by /2. The extra 2 in the denominator results from the fact that
the calculation was performed with only half the actual aperture length. With the assumed
relation between A and R the square root expression can be set to 1 in this approximation.
While range resolution depends on the bandwidth, cross range resolution is mainly depen-
dent on the ratio between aperture and wavelength. In UWB systems resolution is estimated
with an average wavelength. In imaging networks the two cases range and cross range are
always mixed. For a proper resolution approximation the node arrangement and the signal
pulse shape must be taken into account.
3.4 Localization of Nodes and Temporal Synchronization
Imaging-algorithms need the distance Tx object Rx at each position of the mobile nodes.
This requires knowledge of all anchor node positions and continuous tracking of the mobile
nodes. Time-based node localization is possible only with exact temporal synchronization of
the singe nodes.
3.4.1 Localization of Nodes
Before we list the different localization tasks, we introduce abbreviations for the localization
methods:
TOA: Time of arrival ranging/localization
TDOA:Time difference of arrival localization
AOA: Angle of arrival localization
ADOA: Angle difference of arrival localization
RTT: Round trip time ranging
RSS: Received signal strength ranging
It is not necessary to explain these methods here, because this subject is covered in the liter-
ature extensively. Summaries can be found in Patwari et al. (2005) and in Sayed et al. (2005).
AOA and ADOA methods are explained in Rong & Sichitiu (2006), and TDOA methods are
discussed in Stoica & Li (2006).
The single tasks are:
1. The positions of the static nodes (anchor nodes) must be estimated. If the network is
a xed installation, then this task is already fullled. Otherwise anchor node positions
can be found by means of TOAlocalization (if synchronization is available) or by means
of RTT estimations (synchronization not required).
2. The positions of mobile nodes must be tracked continuously. If the sensors move along
predened paths, then their positions are known in advance. In case of synchronization
between mobile nodes and anchors, position estimation is possible with TOA meth-
ods. Without synchronization node positions may be found by TDOA, AOA, or ADOA
methods. RSS is not very precise; RTT could be used in principle but requires much
effort.
Methods that involve angle measurements (AOA and ADOA) can only be performed if the
sensor is equipped with directional antennas or with an antenna array. Time-based methods
require exact synchronization; in case of TDOA only on the individual sensor platform, for
TOA and RTT within the network. The large bandwidth and good temporal resolution of
UWB systems are huge advantages for time-based position measurements.
3.4.2 Temporal Synchronization of Network Nodes
Two main reasons exist for temporal synchronization of network nodes:
1. Application of time-based localization methods.
2. Use of correlation receivers in M-sequence systems.
Point 1 was alredy discussed. The necessity of synchronization in networks with correlation
receivers can be seen from Fig. 5.
Fig. 5. Mismatch between a received M-sequence signal and the reference signal because of
differing clock frequencies 1/t
C1
and 1/t
C2
of Tx and Rx. The total time shift is N
M
t
C
(N
M
: Number of chips; t
C
: time difference per cycle).
R
d
d
1
x
A/2
x
d
(a)
(b)
A/2
Fig. 4. Approximation of cross range resolution
x
. (b) is a zoomed version of the encircled
section in (a).
these circumstances d and d
1
can be assumed as being parallel on short lengthscales. The
angle appears both in the small triangle with sides
x
and d, and in the big triangle with
half aperture A/2 and range R:
sin =
d
x
, tan =
A/2
R
. (8)
After rearranging of both equations, insertion into each other, and application of the relation
sin(arctan(x)) = x/
1 + x
2
we get an expression for
x
:
x
=
d 2R
1 + (A/(2R))
2
2A

R
2A
. (9)
Here d was replaced by /2. The extra 2 in the denominator results from the fact that
the calculation was performed with only half the actual aperture length. With the assumed
relation between A and R the square root expression can be set to 1 in this approximation.
While range resolution depends on the bandwidth, cross range resolution is mainly depen-
dent on the ratio between aperture and wavelength. In UWB systems resolution is estimated
with an average wavelength. In imaging networks the two cases range and cross range are
always mixed. For a proper resolution approximation the node arrangement and the signal
pulse shape must be taken into account.
3.4 Localization of Nodes and Temporal Synchronization
Imaging-algorithms need the distance Tx object Rx at each position of the mobile nodes.
This requires knowledge of all anchor node positions and continuous tracking of the mobile
nodes. Time-based node localization is possible only with exact temporal synchronization of
the singe nodes.
3.4.1 Localization of Nodes
Before we list the different localization tasks, we introduce abbreviations for the localization
methods:
TOA: Time of arrival ranging/localization
TDOA:Time difference of arrival localization
AOA: Angle of arrival localization
ADOA: Angle difference of arrival localization
RTT: Round trip time ranging
RSS: Received signal strength ranging
It is not necessary to explain these methods here, because this subject is covered in the liter-
ature extensively. Summaries can be found in Patwari et al. (2005) and in Sayed et al. (2005).
AOA and ADOA methods are explained in Rong & Sichitiu (2006), and TDOA methods are
discussed in Stoica & Li (2006).
The single tasks are:
1. The positions of the static nodes (anchor nodes) must be estimated. If the network is
a xed installation, then this task is already fullled. Otherwise anchor node positions
can be found by means of TOAlocalization (if synchronization is available) or by means
of RTT estimations (synchronization not required).
2. The positions of mobile nodes must be tracked continuously. If the sensors move along
predened paths, then their positions are known in advance. In case of synchronization
between mobile nodes and anchors, position estimation is possible with TOA meth-
ods. Without synchronization node positions may be found by TDOA, AOA, or ADOA
methods. RSS is not very precise; RTT could be used in principle but requires much
effort.
Methods that involve angle measurements (AOA and ADOA) can only be performed if the
sensor is equipped with directional antennas or with an antenna array. Time-based methods
require exact synchronization; in case of TDOA only on the individual sensor platform, for
TOA and RTT within the network. The large bandwidth and good temporal resolution of
UWB systems are huge advantages for time-based position measurements.
3.4.2 Temporal Synchronization of Network Nodes
Two main reasons exist for temporal synchronization of network nodes:
1. Application of time-based localization methods.
2. Use of correlation receivers in M-sequence systems.
Point 1 was alredy discussed. The necessity of synchronization in networks with correlation
receivers can be seen from Fig. 5.
Fig. 5. Mismatch between a received M-sequence signal and the reference signal because of
differing clock frequencies 1/t
C1
and 1/t
C2
of Tx and Rx. The total time shift is N
M
t
C
(N
M
: Number of chips; t
C
: time difference per cycle).
Over the sequence duration of N
M
t
C1
a maximum shift of
1
2
t
C1
is tolerable. This corre-
sponds to a maximum clock frequency difference of
f
C
<
1
2 N
M
f
C1
. (10)
A comprehensive introduction into synchronization methods and protocols is given in Ser-
pedin & Chaudhari (2009). Originally, many of these methods were developed for commu-
nications networks. The good time resolution of UWB signals makes them a candidate for
synchronization tasks. An example is given in Yang & Yang (2006).
3.5 Data Fusion
Processing of data in an imaging sensor network is distributed across the nodes. Part of the
processing steps are performed at the individual sensors while, after a data transfer, nal
processing is done at the fusion center. An example ow chart is shown in Fig. 6.
Acqu. 1 Acqu. 2 Acqu. N
Pre-
Proc. 1
Pre-
Proc. 2
Pre-
Proc. N
Data Fusion
Image,
other information
raw Data
Tx position
Radargram,
calibrated Data
Fig. 6. Data processing in a network with one Tx and N Rx. The single steps are Data acquisi-
tion (Acqu.), Pre-processing (Pre-Proc.), and Data Fusion.
After transmitting a pulse or an M-sequence by the Tx, data are acquired by the Rx hardware.
Typically the sensor hardware performs some additional tasks: analog to digital conversion,
correlation with a known signal pattern (in case of M-sequence systems), and accumulation
of measurements to improve the signal to noise ratio.
The next step is pre-processing of the raw data, usually performed in a signal processor at the
sensor node. De-convolution of raw data with a measured calibration function can increase
the usable bandwidth and in this way it can increase range resolution. In M-sequence systems
data must be shifted to achieve coincidence between the moment of signal transmission and
receiver time zero (Sachs (2004)). The result of pre-processing can be visualized in a radar-
gram (Fig. 15). It displays the processed signals in form of vertical traces against the slow
time dimension of sensor movement. For some analyses only the TOA of the rst echo is of
importance. Then pre-processing includes a discrimination step, which reduces the informa-
tion to a single TOA value.
Data fusion is a generic term for methods that combine information from the single sensor
nodes and produce the image. While acquisition and pre-processing dont vary a lot between
the different imaging methods, data fusion is strongly dependent on network topology, sensor
pathways, and imaging method. Examples are described in section 4.
Additional information, required for imaging, are the positions of mobile nodes. As long as
the sensors follow predetermined pathways, this information is always available. In other
cases the mobile node positions must be estimated by means of mechanical sensors or the
position is extracted from radar signals.
Fusion is not always the last processing step. By application of image processing methods
supplementary information can be extracted from the radar image.
4. Imaging in Distributed Multistatic Networks
4.1 Multistatic SAR Imaging
The multitude of different propagation pathways in a distributed sensor network can be used
for rough imaging of the environment. A signal, transmitted by a Tx, is reected or scattered
at walls, furniture, and other objects. The individual Rx receive these scattered signals from
different perspectives. The information about the object position is contained in the signal
propagation times. The principle of this method is shown in Fig. 7. The propagation paths
are sketched for a signal scattered at an objects corner. In principle, the positions of Tx and Rx
could be swapped, but an arrangement with only one Tx and several Rx has the advantage of
simultaneous operation of all Rx.
Tx
Rx
1
Rx
2
Rx
3
Rx
4
(0,0)
y
x
d
TO
d
O1
d
O2
d
O3
d
O4
Fig. 7. Principle of imaging in a multistatic network. The four receivers (Rx
i
) are placed at
xed positions. The Transmitter (Tx) moves along the curved pathway.
Prerequisites for application of this method are knowledge of the Rx positions, temporal syn-
chronization of all nodes, and application of omnidirectional antennas. The synthetic aperture
has the shape of an arbitrary path through the environment. Data acquisition is carried out as
follows:
The Tx moves through the region. It transmits signals every few centimeters.
Over the sequence duration of N
M
t
C1
a maximum shift of
1
2
t
C1
is tolerable. This corre-
sponds to a maximum clock frequency difference of
f
C
<
1
2 N
M
f
C1
. (10)
A comprehensive introduction into synchronization methods and protocols is given in Ser-
pedin & Chaudhari (2009). Originally, many of these methods were developed for commu-
nications networks. The good time resolution of UWB signals makes them a candidate for
synchronization tasks. An example is given in Yang & Yang (2006).
3.5 Data Fusion
Processing of data in an imaging sensor network is distributed across the nodes. Part of the
processing steps are performed at the individual sensors while, after a data transfer, nal
processing is done at the fusion center. An example ow chart is shown in Fig. 6.
Acqu. 1 Acqu. 2 Acqu. N
Pre-
Proc. 1
Pre-
Proc. 2
Pre-
Proc. N
Data Fusion
Image,
other information
raw Data
Tx position
Radargram,
calibrated Data
Fig. 6. Data processing in a network with one Tx and N Rx. The single steps are Data acquisi-
tion (Acqu.), Pre-processing (Pre-Proc.), and Data Fusion.
After transmitting a pulse or an M-sequence by the Tx, data are acquired by the Rx hardware.
Typically the sensor hardware performs some additional tasks: analog to digital conversion,
correlation with a known signal pattern (in case of M-sequence systems), and accumulation
of measurements to improve the signal to noise ratio.
The next step is pre-processing of the raw data, usually performed in a signal processor at the
sensor node. De-convolution of raw data with a measured calibration function can increase
the usable bandwidth and in this way it can increase range resolution. In M-sequence systems
data must be shifted to achieve coincidence between the moment of signal transmission and
receiver time zero (Sachs (2004)). The result of pre-processing can be visualized in a radar-
gram (Fig. 15). It displays the processed signals in form of vertical traces against the slow
time dimension of sensor movement. For some analyses only the TOA of the rst echo is of
importance. Then pre-processing includes a discrimination step, which reduces the informa-
tion to a single TOA value.
Data fusion is a generic term for methods that combine information from the single sensor
nodes and produce the image. While acquisition and pre-processing dont vary a lot between
the different imaging methods, data fusion is strongly dependent on network topology, sensor
pathways, and imaging method. Examples are described in section 4.
Additional information, required for imaging, are the positions of mobile nodes. As long as
the sensors follow predetermined pathways, this information is always available. In other
cases the mobile node positions must be estimated by means of mechanical sensors or the
position is extracted from radar signals.
Fusion is not always the last processing step. By application of image processing methods
supplementary information can be extracted from the radar image.
4. Imaging in Distributed Multistatic Networks
4.1 Multistatic SAR Imaging
The multitude of different propagation pathways in a distributed sensor network can be used
for rough imaging of the environment. A signal, transmitted by a Tx, is reected or scattered
at walls, furniture, and other objects. The individual Rx receive these scattered signals from
different perspectives. The information about the object position is contained in the signal
propagation times. The principle of this method is shown in Fig. 7. The propagation paths
are sketched for a signal scattered at an objects corner. In principle, the positions of Tx and Rx
could be swapped, but an arrangement with only one Tx and several Rx has the advantage of
simultaneous operation of all Rx.
Tx
Rx
1
Rx
2
Rx
3
Rx
4
(0,0)
y
x
d
TO
d
O1
d
O2
d
O3
d
O4
Fig. 7. Principle of imaging in a multistatic network. The four receivers (Rx
i
) are placed at
xed positions. The Transmitter (Tx) moves along the curved pathway.
Prerequisites for application of this method are knowledge of the Rx positions, temporal syn-
chronization of all nodes, and application of omnidirectional antennas. The synthetic aperture
has the shape of an arbitrary path through the environment. Data acquisition is carried out as
follows:
The Tx moves through the region. It transmits signals every few centimeters.
All Rx receive the scattered signals. From the totality of received signals a radargram
can be drawn for each Rx.
The recorded data are processed in two ways:
The Tx position at the individual measurement points are reconstructed from the LOS
signals between Tx and Rx (dotted lines in Fig. 7).
The image is computed by means of a simple migration algorithm.
Separately for each receiver Rx
i
an image is computed. The brightness in one point B
i
(x, y) is
the coherent sum of all signals s
r
, originating from the scatterer at position (x, y), summarized
along the aperture (n is the number of the measurement along the Tx path).
B
i
(x, y) =
N
n=1
s
r
(
in
, n) . (11)
The delay
in
is the time required for wave propagation along the way Tx object point
(x, y) Rx
i
with speed c:
in
=
1
c
(d
TO
(n) +d
Oi
) =
1
c
(x x
Tx
(n))
2
+ (y y
Tx
(n))
2
+
(x
Rx
i
x)
2
+ (y
Rx
i
y)
2
.
(12)
The meaning of the used symbols can be seen from Fig. 7. This migration algorithm summa-
rizes signals along ellipses, which have their foci at the respective Tx and Rx position. The
ellipses for all possible Tx-Rx constellations have in common that they touch the considered
object point.
For improved performance, migration algorithms, based on wave equations must be applied,
see Margrave (2001). Stolt Migration, computed in the wavenumber-domain, is a fast migra-
tion method (Stolt (1978)). However, it requires an equally spaced net of sampling points.
Therefore it cannot be applied in case of an arbitrary-shaped scan path.
4.2 Cross-Correlated Imaging
The summation, mentioned in the previous section, cumulates intensity of the image at posi-
tions, where objects, which evoked echoes in measured impulse responses, are present. How-
ever, this simple addition of multiple snapshots also creates disturbing artefacts in the focused
image (see Fig. 9(a)). The elliptical traces do not only intersect at objects positions. They in-
tersect also at other positions and even the ellipsis themselves make the image interpretation
difcult or impossible.
In order to reduce these artefacts, a method based on cross-correlatedback projection was pro-
posed in Foo & Kashyap (2004). This method suggests a modication of the snapshots com-
putation. Instead of a simple remapping of an impulse response signal the modied snapshot
is created by a cross-correlation of two impulse responses:
B(x, y) =
1
N
N
n=1
T/2
T/2
s
r
d
TO
(n) + d
Oi
c
+
s
ref
d
TO
(n) + d
Oiref
c
+
d , (13)
where s
ref
is an impulse response measured by an auxiliary reference receiver at a suitable
measurement position. Since two different delay terms (d
TO
(n) + d
Oi
)/c and (d
TO
(n) +
d
Oiref
)/c have to match the actual scattering scenario in conjunction, the probability to add
"wrong" energy to an image pixel (x,y), which does not coincide with an object, will be re-
duced. The integration interval T is chosen to match the duration of the stimulation impulse.
Further improvement of this method was proposed in Zetik et al. (2005a); Zetik et al. (2005b);
and Zetik et al. (2008). The rst two references introduce modications that improve the per-
formance of the cross-correlated back projection from Foo & Kashyap (2004) by additional
reference nodes. This drastically reduces artefacts in the focused image. In Zetik et al. (2008),
a generalised form of the imaging algorithm, which is suitable for application in distributed
sensor networks, is proposed:
B(x, y) =
1
N
N
n=1
W
n
(x, y)A[s
rn
(x, y), s
ref1n
(x, y), ..., s
refMn
(x, y)] , (14)
d=0.6m
x
y
Rx
Turntable [0m,0m]
Tx
d
2
Ladder [-1.1m,1.5m]
Fig. 8. Cross-correlated imaging with a circular aperture. The data measured at a certain
position are multiplied with reference data, acquired at a position with 120
offset.
where A[.] is an operator, which averages N spatially distributed observations and W
n
(x, y)
are weighing coefcients. It is assumed that all Naveraged nodes can "see" an object at the po-
sition (x, y). In case of point-like objects, the incident EM waves are scattered in all directions.
Hence, sensor nodes can be arbitrarily situated around the position (x, y) and they will still
"see" an object (if there is one). However, extended objects, such as walls, reect EM waves
like a mirror. A sensor node can "see" only a small part of this object, which is observed under
the perpendicular viewing angle. Therefore, the selection of the additional reference nodes
must be done very carefully. A proper selection of sensor nodes for point like and also dis-
tributed objects is discussed in detail in Zetik et al. (2008). The weighting coefcients W
n
(x, y)
are inversely related to the number of nodes (measurement positions) that observe a specic
part in the focused image. This reduces over and under illumination of the focused image by
taking into account the topology of the network.
The following measured example demonstrates differences between images obtained by the
conventional SAR algorithm (11) and the cross-correlated algorithm (14). The measurement
constellation is shown in Fig. 8. The target - a metallic ladder - was observed by a sensor,
which was moving along a circular track in its vicinity. The sensor comprised two closely
All Rx receive the scattered signals. From the totality of received signals a radargram
can be drawn for each Rx.
The recorded data are processed in two ways:
The Tx position at the individual measurement points are reconstructed from the LOS
signals between Tx and Rx (dotted lines in Fig. 7).
The image is computed by means of a simple migration algorithm.
Separately for each receiver Rx
i
an image is computed. The brightness in one point B
i
(x, y) is
the coherent sum of all signals s
r
, originating from the scatterer at position (x, y), summarized
along the aperture (n is the number of the measurement along the Tx path).
B
i
(x, y) =
N
n=1
s
r
(
in
, n) . (11)
The delay
in
is the time required for wave propagation along the way Tx object point
(x, y) Rx
i
with speed c:
in
=
1
c
(d
TO
(n) +d
Oi
) =
1
c
(x x
Tx
(n))
2
+ (y y
Tx
(n))
2
+
(x
Rx
i
x)
2
+ (y
Rx
i
y)
2
.
(12)
The meaning of the used symbols can be seen from Fig. 7. This migration algorithm summa-
rizes signals along ellipses, which have their foci at the respective Tx and Rx position. The
ellipses for all possible Tx-Rx constellations have in common that they touch the considered
object point.
For improved performance, migration algorithms, based on wave equations must be applied,
see Margrave (2001). Stolt Migration, computed in the wavenumber-domain, is a fast migra-
tion method (Stolt (1978)). However, it requires an equally spaced net of sampling points.
Therefore it cannot be applied in case of an arbitrary-shaped scan path.
4.2 Cross-Correlated Imaging
The summation, mentioned in the previous section, cumulates intensity of the image at posi-
tions, where objects, which evoked echoes in measured impulse responses, are present. How-
ever, this simple addition of multiple snapshots also creates disturbing artefacts in the focused
image (see Fig. 9(a)). The elliptical traces do not only intersect at objects positions. They in-
tersect also at other positions and even the ellipsis themselves make the image interpretation
difcult or impossible.
In order to reduce these artefacts, a method based on cross-correlatedback projection was pro-
posed in Foo & Kashyap (2004). This method suggests a modication of the snapshots com-
putation. Instead of a simple remapping of an impulse response signal the modied snapshot
is created by a cross-correlation of two impulse responses:
B(x, y) =
1
N
N
n=1
T/2
T/2
s
r
d
TO
(n) + d
Oi
c
+
s
ref
d
TO
(n) + d
Oiref
c
+
d , (13)
where s
ref
is an impulse response measured by an auxiliary reference receiver at a suitable
measurement position. Since two different delay terms (d
TO
(n) + d
Oi
)/c and (d
TO
(n) +
d
Oiref
)/c have to match the actual scattering scenario in conjunction, the probability to add
"wrong" energy to an image pixel (x,y), which does not coincide with an object, will be re-
duced. The integration interval T is chosen to match the duration of the stimulation impulse.
Further improvement of this method was proposed in Zetik et al. (2005a); Zetik et al. (2005b);
and Zetik et al. (2008). The rst two references introduce modications that improve the per-
formance of the cross-correlated back projection from Foo & Kashyap (2004) by additional
reference nodes. This drastically reduces artefacts in the focused image. In Zetik et al. (2008),
a generalised form of the imaging algorithm, which is suitable for application in distributed
sensor networks, is proposed:
B(x, y) =
1
N
N
n=1
W
n
(x, y)A[s
rn
(x, y), s
ref1n
(x, y), ..., s
refMn
(x, y)] , (14)
d=0.6m
x
y
Rx
Turntable [0m,0m]
Tx
d
2
Ladder [-1.1m,1.5m]
Fig. 8. Cross-correlated imaging with a circular aperture. The data measured at a certain
position are multiplied with reference data, acquired at a position with 120
offset.
where A[.] is an operator, which averages N spatially distributed observations and W
n
(x, y)
are weighing coefcients. It is assumed that all Naveraged nodes can "see" an object at the po-
sition (x, y). In case of point-like objects, the incident EM waves are scattered in all directions.
Hence, sensor nodes can be arbitrarily situated around the position (x, y) and they will still
"see" an object (if there is one). However, extended objects, such as walls, reect EM waves
like a mirror. A sensor node can "see" only a small part of this object, which is observed under
the perpendicular viewing angle. Therefore, the selection of the additional reference nodes
must be done very carefully. A proper selection of sensor nodes for point like and also dis-
tributed objects is discussed in detail in Zetik et al. (2008). The weighting coefcients W
n
(x, y)
are inversely related to the number of nodes (measurement positions) that observe a specic
part in the focused image. This reduces over and under illumination of the focused image by
taking into account the topology of the network.
The following measured example demonstrates differences between images obtained by the
conventional SAR algorithm (11) and the cross-correlated algorithm (14). The measurement
constellation is shown in Fig. 8. The target - a metallic ladder - was observed by a sensor,
which was moving along a circular track in its vicinity. The sensor comprised two closely
X coordinate [m]
Y

c
o
o
r
d
i
n
a
t
e

[
m
]
2 1 0 1 2
2.5
2
1.5
1
0.5
0
0.5
1
100 95 90 85 80 75 70
(a)
X coordinate [m]
Y

c
o
o
r
d
i
n
a
t
e

[
m
]
2 1 0 1 2
2.5
2
1.5
1
0.5
0
0.5
1
100 95 90 85 80 75 70
(b)
Fig. 9. (a) Image taken with the arrangement shown in Fig. 8 and processed with conventional
migration algorithm. (b) Image processed with cross-correlated migration algorithm.
spaced antennas. One antenna was transmitting an UWB signal covering a bandwidth from
3.5 to 10.5 GHz. The second antenna was receiving signals reected from the surroundings.
Both antennas were mounted on the arm attached to the turntable. About 800 impulse re-
sponses were recorded. The originof the local coordinate systemwas selected to be the middle
of the turntable.
Firstly, the measured impulse responses were fused by the conventional SAR algorithm (11).
The result of this imaging on a logarithmical scale is depicted in Fig. 9(a). The whole image is
distorted by data fusion artefacts and is hard to interpret. The result can be improved by the
generalised imaging algorithm (14). Here, the operator A[.] was replaced by the minimum
value operator. It took the minimum magnitude from 3 observations s
rn
, s
ref1n
and s
ref2n
.
The position of two additional reference nodes Rref1n and Rref2n was computed adaptively
for each pixel (x,y) of the focused image and to each measured impulse response Rn. The
adaptation criterion was the 120
difference in the viewing angles of all 3 nodes. The reduction

of disturbing artefacts is evident in Fig. 9(b).
4.3 Indirect Imaging of Objects
The procedure explained in 4.1 can be reversed. Instead of measuring the signals reected at
objects, indirect imaging detects free LOS paths between objects within the area of interest.
An example is shown in Fig. 10. Generally a network of anchor nodes is required. First
these anchors estimate their respective positions and, later on, they operate as Rx nodes at
xed positions. A mobile Tx moves around the area of objects and anchor nodes. The Tx
emits UWB signals which are received at all Rx nodes. From the received signals two kinds
of information are extracted: the propagation time, which is a measure for the current Tx-Rx
distance, and the information about LOS or NLOS between Tx and Rx. The path of the Tx
can be reconstructed from the totality of distance estimates. The second information allows
creation of a map of LOS paths between the Tx path and the respective Rx position. An overlay
of all individual LOS path maps reveals positions and approximate contours of the objects.
Diffraction at the edges of objects limits the performance of the described procedure and
causes a underestimation of object dimensions. The method is explained in more detail in
Hirsch et al. (2010).
(a) LOS paths (b) Contours
Fig. 10. Indirect Imaging of objects. (a) LOS paths (dark regions) between the Tx pathway
(small circles) and Rx node 3. (b) Position of objects (lled boxes) and estimated object con-
tours (open boxes).
5. Imaging by Autonomous Rotating Sensors within a Network
5.1 Design
The networks presented in the previous sections consist of a number of nodes at xed po-
sitions and one mobile node that moves along the imaging aperture. The imaging process
requires cooperation of all nodes. Now we introduce a sensor that autonomously operates
within a network of anchor nodes. It consists of a mobile platform equipped with one Tx and
two Rx units and with the corresponding antennas. The sensor can move within the area of
the network, varying its perspective in this way; and it can rotate to acquire 360
panoramic
views. Because of the similarity to the ultrasound locating system of a bat we call it a bat-
type sensor. By means of the anchor nodes the bat sensor can estimate its own position and
its present orientation. Fig. 11 shows the geometry and a laboratory prototype. In principal
the anchor nodes could be used as additional illuminators or as additional receivers. These
aspects were not investigated in the frame of this work.
5.2 Orientation within the Network
An image of the environment is typically assembled from several individual measurements
performed with the bat-type sensor at different locations. For the correct assignment of these
images the position and orientation of the sensor within the room must be estimated. As
long as temporal synchronization exists between the network of anchor nodes and the bat-
type sensor, a variety of time of arrival (TOA) localization methods can be applied for this
purpose.
Here we present a method that neither requires temporal synchronization between mobile
sensor and anchor nodes nor synchronization within the network of anchor nodes. It is based
on angle measurements and can be classied as angle difference of arrival (ADOA) local-
ization. Line of sight from the mobile sensor to at least three anchor nodes is required. The
basic idea consists in establishment of a system of two equations, where the input parameters
X coordinate [m]
Y

c
o
o
r
d
i
n
a
t
e

[
m
]
2 1 0 1 2
2.5
2
1.5
1
0.5
0
0.5
1
100 95 90 85 80 75 70
(a)
X coordinate [m]
Y

c
o
o
r
d
i
n
a
t
e

[
m
]
2 1 0 1 2
2.5
2
1.5
1
0.5
0
0.5
1
100 95 90 85 80 75 70
(b)
Fig. 9. (a) Image taken with the arrangement shown in Fig. 8 and processed with conventional
migration algorithm. (b) Image processed with cross-correlated migration algorithm.
spaced antennas. One antenna was transmitting an UWB signal covering a bandwidth from
3.5 to 10.5 GHz. The second antenna was receiving signals reected from the surroundings.
Both antennas were mounted on the arm attached to the turntable. About 800 impulse re-
sponses were recorded. The originof the local coordinate systemwas selected to be the middle
of the turntable.
Firstly, the measured impulse responses were fused by the conventional SAR algorithm (11).
The result of this imaging on a logarithmical scale is depicted in Fig. 9(a). The whole image is
distorted by data fusion artefacts and is hard to interpret. The result can be improved by the
generalised imaging algorithm (14). Here, the operator A[.] was replaced by the minimum
value operator. It took the minimum magnitude from 3 observations s
rn
, s
ref1n
and s
ref2n
.
The position of two additional reference nodes Rref1n and Rref2n was computed adaptively
for each pixel (x,y) of the focused image and to each measured impulse response Rn. The
adaptation criterion was the 120
difference in the viewing angles of all 3 nodes. The reduction

of disturbing artefacts is evident in Fig. 9(b).
4.3 Indirect Imaging of Objects
The procedure explained in 4.1 can be reversed. Instead of measuring the signals reected at
objects, indirect imaging detects free LOS paths between objects within the area of interest.
An example is shown in Fig. 10. Generally a network of anchor nodes is required. First
these anchors estimate their respective positions and, later on, they operate as Rx nodes at
xed positions. A mobile Tx moves around the area of objects and anchor nodes. The Tx
emits UWB signals which are received at all Rx nodes. From the received signals two kinds
of information are extracted: the propagation time, which is a measure for the current Tx-Rx
distance, and the information about LOS or NLOS between Tx and Rx. The path of the Tx
can be reconstructed from the totality of distance estimates. The second information allows
creation of a map of LOS paths between the Tx path and the respective Rx position. An overlay
of all individual LOS path maps reveals positions and approximate contours of the objects.
Diffraction at the edges of objects limits the performance of the described procedure and
causes a underestimation of object dimensions. The method is explained in more detail in
Hirsch et al. (2010).
(a) LOS paths (b) Contours
Fig. 10. Indirect Imaging of objects. (a) LOS paths (dark regions) between the Tx pathway
(small circles) and Rx node 3. (b) Position of objects (lled boxes) and estimated object con-
tours (open boxes).
5. Imaging by Autonomous Rotating Sensors within a Network
5.1 Design
The networks presented in the previous sections consist of a number of nodes at xed po-
sitions and one mobile node that moves along the imaging aperture. The imaging process
requires cooperation of all nodes. Now we introduce a sensor that autonomously operates
within a network of anchor nodes. It consists of a mobile platform equipped with one Tx and
two Rx units and with the corresponding antennas. The sensor can move within the area of
the network, varying its perspective in this way; and it can rotate to acquire 360
panoramic
views. Because of the similarity to the ultrasound locating system of a bat we call it a bat-
type sensor. By means of the anchor nodes the bat sensor can estimate its own position and
its present orientation. Fig. 11 shows the geometry and a laboratory prototype. In principal
the anchor nodes could be used as additional illuminators or as additional receivers. These
aspects were not investigated in the frame of this work.
5.2 Orientation within the Network
An image of the environment is typically assembled from several individual measurements
performed with the bat-type sensor at different locations. For the correct assignment of these
images the position and orientation of the sensor within the room must be estimated. As
long as temporal synchronization exists between the network of anchor nodes and the bat-
type sensor, a variety of time of arrival (TOA) localization methods can be applied for this
purpose.
Here we present a method that neither requires temporal synchronization between mobile
sensor and anchor nodes nor synchronization within the network of anchor nodes. It is based
on angle measurements and can be classied as angle difference of arrival (ADOA) local-
ization. Line of sight from the mobile sensor to at least three anchor nodes is required. The
basic idea consists in establishment of a system of two equations, where the input parameters
r
d
1
a
x
Rx
1
Rx
2
Tx
y
d
2
a
(a) Geometry (b) Prototype
Fig. 11. Geometry and laboratory prototype of a bat-type sensor. The point, where r, d
1
, and
d
2
come together, is an object point.
are the known positions of the anchor nodes and the difference angles between three anchor
node directions, measured by the bat-type sensor. The solutions are the x- and y-position of
the mobile sensor. Afterwards the sensor orientation can be estimated.
The situation is sketched in Fig. 12. We distinguish between the global coordinate system (x
g
,
y
g
), dened by anchor nodes A
1
, A
2
, A
3
, and the coordinate system of the bat-type sensor,
which has its origin at coordinates (x
b
, y
b
) within the global system and which is rotated
against that system by
b
. The mathematical effort is reduced if the global system is arranged
in such a way that one node forms the origin and another node is placed directly on one
coordinate axis. This choice does not reduce the generality of the method.
All anchor nodes operate in Tx mode. Then the bat sensor can easily estimate the direction
angles to this nodes
1
,
2
,
3
within its own coordinate system. During a sensor rotation the
signal strength reaches a maximum at rotation angles where the sensor antennas are directed
towards an anchor node. These angles can be extracted from the radargram with good accu-
racy because the TOA traces of left and right Rx antenna intersect at the transmitter positions,
see Fig. 15. Because of the unknown
b
this information is not directly usable for position
estimation but the angle differences
12
=
2
1
and
31
=
1
3
can be used, since they
remain the same in both coordinate systems.
a
1
, a
2
, a
3
form the connection lines between sensor position (x
b
, y
b
) and the anchor nodes;
and
12
,
31
represent the cutting angles between these lines. Now we establish the system of
equations, which connects the cutting angles with the slopes m
i
of the connection lines:
tan
12
=
m
2
m
1
1 +m
2
m
1
, tan
31
=
m
1
m
3
1 +m
1
m
3
. (15)
The slopes follow from the anchor node coordinates and from the bat-type sensor position:
y
g
x
b
y
b
12
31
3
1
s
31
a
2 a
1
a
3
b
A
2
=(x
2
,0)
A
3
=(x
3
,y
3
)
x
g
s
12 A
1
=(0,0)
x
b
y
b
x
Fig. 12. Estimation of position (x
b
, y
b
) and orientation (
b
) of a bat-type sensor within the
global coordinate system (x
g
, y
g
). See text for details.
m
1
=
y
b
x
b
, m
2
=
y
b
x
2
x
b
, m
3
=
y
3
y
b
x
3
x
b
. (16)
Insertion of (16) in (15), expansion of the resulting expressions, and summary of the coef-
cients of x
b
and y
b
, respectively, delivers the following system of equations:
x
2
b
+ y
2
b
= x
2
x
b
+
x
2
tan
12
y
b
, x
2
b
+ y
2
b
=
x
3
+
y
3
tan
31
x
b
+
y
3
x
3
tan
31
y
b
. (17)
Equalization of the two equations (17) gives the following expression, where we introduce the
abbreviations S and T:
0 =
x
2
+ x
3
+
y
3
tan
31
x
b
+
y
3
x
3
tan
31
x
2
tan
12
y
b
= Sx
b
+ Ty
b
. (18)
In this way we have found expressions for y
b
and y
2
b
:
y
b
=
S
T
x
b
, y
2
b
=
S
T
2
x
2
b
. (19)
Introduction of (19) in the rst equation (17) gives a formula for x
b
:
x
b
= x
2
1
S
tan
12
T
1 +
S
T
. (20)
Finally the orientation of the bat-type sensor is calculated. The angle between a
1
and the x-axis
is
x
= arctan(y
b
/x
b
). From Fig. 12 one can see that the orientation angle
b
must be:
r
d
1
a
x
Rx
1
Rx
2
Tx
y
d
2
a
(a) Geometry (b) Prototype
Fig. 11. Geometry and laboratory prototype of a bat-type sensor. The point, where r, d
1
, and
d
2
come together, is an object point.
are the known positions of the anchor nodes and the difference angles between three anchor
node directions, measured by the bat-type sensor. The solutions are the x- and y-position of
the mobile sensor. Afterwards the sensor orientation can be estimated.
The situation is sketched in Fig. 12. We distinguish between the global coordinate system (x
g
,
y
g
), dened by anchor nodes A
1
, A
2
, A
3
, and the coordinate system of the bat-type sensor,
which has its origin at coordinates (x
b
, y
b
) within the global system and which is rotated
against that system by
b
. The mathematical effort is reduced if the global system is arranged
in such a way that one node forms the origin and another node is placed directly on one
coordinate axis. This choice does not reduce the generality of the method.
All anchor nodes operate in Tx mode. Then the bat sensor can easily estimate the direction
angles to this nodes
1
,
2
,
3
within its own coordinate system. During a sensor rotation the
signal strength reaches a maximum at rotation angles where the sensor antennas are directed
towards an anchor node. These angles can be extracted from the radargram with good accu-
racy because the TOA traces of left and right Rx antenna intersect at the transmitter positions,
see Fig. 15. Because of the unknown
b
this information is not directly usable for position
estimation but the angle differences
12
=
2
1
and
31
=
1
3
can be used, since they
remain the same in both coordinate systems.
a
1
, a
2
, a
3
form the connection lines between sensor position (x
b
, y
b
) and the anchor nodes;
and
12
,
31
represent the cutting angles between these lines. Now we establish the system of
equations, which connects the cutting angles with the slopes m
i
of the connection lines:
tan
12
=
m
2
m
1
1 +m
2
m
1
, tan
31
=
m
1
m
3
1 +m
1
m
3
. (15)
The slopes follow from the anchor node coordinates and from the bat-type sensor position:
y
g
x
b
y
b
12
31
3
1
s
31
a
2 a
1
a
3
b
A
2
=(x
2
,0)
A
3
=(x
3
,y
3
)
x
g
s
12 A
1
=(0,0)
x
b
y
b
x
Fig. 12. Estimation of position (x
b
, y
b
) and orientation (
b
) of a bat-type sensor within the
global coordinate system (x
g
, y
g
). See text for details.
m
1
=
y
b
x
b
, m
2
=
y
b
x
2
x
b
, m
3
=
y
3
y
b
x
3
x
b
. (16)
Insertion of (16) in (15), expansion of the resulting expressions, and summary of the coef-
cients of x
b
and y
b
, respectively, delivers the following system of equations:
x
2
b
+ y
2
b
= x
2
x
b
+
x
2
tan
12
y
b
, x
2
b
+ y
2
b
=
x
3
+
y
3
tan
31
x
b
+
y
3
x
3
tan
31
y
b
. (17)
Equalization of the two equations (17) gives the following expression, where we introduce the
abbreviations S and T:
0 =
x
2
+ x
3
+
y
3
tan
31
x
b
+
y
3
x
3
tan
31
x
2
tan
12
y
b
= Sx
b
+ Ty
b
. (18)
In this way we have found expressions for y
b
and y
2
b
:
y
b
=
S
T
x
b
, y
2
b
=
S
T
2
x
2
b
. (19)
Introduction of (19) in the rst equation (17) gives a formula for x
b
:
x
b
= x
2
1
S
tan
12
T
1 +
S
T
. (20)
Finally the orientation of the bat-type sensor is calculated. The angle between a
1
and the x-axis
is
x
= arctan(y
b
/x
b
). From Fig. 12 one can see that the orientation angle
b
must be:
b
=
x
1
+ . (21)
An alternative angle of arrival (AOA) localization method, employing pairs of angle measure-
ments between neighbor nodes, is described in Rong & Sichitiu (2006).
5.3 Calculation of Echo Proles
In this section we calculate the signal propagation time for the paths fromthe Tx via the object
to the Rx. The sensor acquires data during a full rotation. Thats why the radar propagation
time follows a systematic dependency. Knowledge of this dependency is required for the
design of signal processing and focus algorithms. Additionally, the calculation delivers the
angles of incidence both for the Tx and for the Rx antenna for each azimuth angle of the bat
sensor. This allows inclusion of the antenna characteristics in the calculation of proles of
signal intensity vs. azimuth angle. Such proles are valuable for classication of reecting
objects, for calculation of the eld of view, and for the estimation of cross-range resolution.
The use of polar coordinates is convenient because of the rotational symmetry of the arrange-
ment. The object coordinates are azimuth and radius r. The bat sensor azimuth angle is
. In the inital sensor position the bar with the 3 antennas is aligned along the y-axis. In
this position is 0 and the sensor is looking towards the positive x-axis. The sensor rotates
counter-clockwise (positive sense of rotation).
First we consider the simpler case of signal reection at a point-like object, shown in Fig.
11(a). Because the Tx antenna is placed in the origin of the local bat coordinate system, r (the
distance Tx-object) does not depend on . The length of d
1
and d
2
(distances object-Rx
1
, Rx
2
)
are determined by the oblique-angled triangles consisting of radius r, the arm of length a, and
d
1
or d
2
. The law of cosines is applied to nd d
1
and d
2
in these triangles (in or the
upper sign is always valid for index 1 while the lower sign is valid for index 2):
d
1/2
=
a
2
+ r
2
2 a r cos(/2 ( )) . (22)
Then the total propagation times are:
1/2
() = (1/c) (r + d
1/2
()) . (23)
In radar imaging objects are often treated as if they were composed of single point scatter-
ers. This is a simplication and fails in the case of bigger objects. In case of objects that are
signicantly bigger than the wavelength specular reection dominates. From this reason we
now consider the case of a reection at a wall, which represents the extreme case of a spacious
object and is more realistic than the point scatterer model in indoor imaging applications. A
sketch of a bat sensor in front of a wall is shown in Fig. 13.
For convenience we use an azimuth angle that is /2 . In this way the sensor orientation
is parallel to the wall at = 0. In case of the wall reection the point where the reection
occurs is not further xed. Instead it moves during the rotation of the sensor. By means of the
geometric construction in Fig. 13 those paths are to be found, which connect Tx and Rx
1/2
after a reection at the wall and which fulll the law of reection at the same time. This law
states that the angle of incidence is equal to the reection angle. Then in our construction the
perpendicular fromthe reection point separates the x-axis in two regions of equal lengths s
x1
and s
x2
, respectively.
To determine the propagation path lengths s
T1/T2
() and s
R1/R2
() we rst need to nd the
dependency of s
x1
and s
x2
on . We establish a linear equation of the general form y = mx + n
for the lines that contain s
R1/R2
:
Rx
2
Rx
1
Tx
s
x2
s
x1
s
x2
s
x1
s
y
2s
y
x
y
T2
T1
R2
R1
s
R1
s
R2
a
a
s
T2
s
T1
Fig. 13. Propagation paths in case of a bat sensor that rotates in front of a wall. The sensor is
symbolized by the bar Rx
1
-Tx-Rx
2
. The wall goes parallel to the x-axis at a distance s
y
. The
antenna main lobe directions are given by the dashed lines. See text for details.
y =
a sin 2 s
y
a cos
x + 2 s
y
. (24)
a cos and a sin are the projections of the bar a onto x- and y-axis. Setting y = 0 and re-
arranging Equ. (24) the intersection with the x-axis is found and in this way the length of
s
x1/x2
:
s
x1/x2
() =
s
y
a cos
a sin 2 s
y
. (25)
Using this result explicit expressions for the path lengths can now be given:
s
T1/T2
() =
s
2
x1/x2
() + s
2
y
. (26)
s
R1/R2
() =
(a cos s
x1/x2
())
2
+ (s
y
a sin )
2
. (27)
The total propagation times
1/2
() are the quotients of the total propagation distances and
the speed of light:
1/2
() =
s
T1/T2
() + s
R1/R2
()
c
. (28)
Fig. 15 illustrates this dependency on a practical example. The result of Equ. (28) is shown for
azimuth angles from 0
to 359
(remember: = /2 ). When the bat sensor is facing the

b
=
x
1
+ . (21)
An alternative angle of arrival (AOA) localization method, employing pairs of angle measure-
ments between neighbor nodes, is described in Rong & Sichitiu (2006).
5.3 Calculation of Echo Proles
In this section we calculate the signal propagation time for the paths fromthe Tx via the object
to the Rx. The sensor acquires data during a full rotation. Thats why the radar propagation
time follows a systematic dependency. Knowledge of this dependency is required for the
design of signal processing and focus algorithms. Additionally, the calculation delivers the
angles of incidence both for the Tx and for the Rx antenna for each azimuth angle of the bat
sensor. This allows inclusion of the antenna characteristics in the calculation of proles of
signal intensity vs. azimuth angle. Such proles are valuable for classication of reecting
objects, for calculation of the eld of view, and for the estimation of cross-range resolution.
The use of polar coordinates is convenient because of the rotational symmetry of the arrange-
ment. The object coordinates are azimuth and radius r. The bat sensor azimuth angle is
. In the inital sensor position the bar with the 3 antennas is aligned along the y-axis. In
this position is 0 and the sensor is looking towards the positive x-axis. The sensor rotates
counter-clockwise (positive sense of rotation).
First we consider the simpler case of signal reection at a point-like object, shown in Fig.
11(a). Because the Tx antenna is placed in the origin of the local bat coordinate system, r (the
distance Tx-object) does not depend on . The length of d
1
and d
2
(distances object-Rx
1
, Rx
2
)
are determined by the oblique-angled triangles consisting of radius r, the arm of length a, and
d
1
or d
2
. The law of cosines is applied to nd d
1
and d
2
in these triangles (in or the
upper sign is always valid for index 1 while the lower sign is valid for index 2):
d
1/2
=
a
2
+ r
2
2 a r cos(/2 ( )) . (22)
Then the total propagation times are:
1/2
() = (1/c) (r + d
1/2
()) . (23)
In radar imaging objects are often treated as if they were composed of single point scatter-
ers. This is a simplication and fails in the case of bigger objects. In case of objects that are
signicantly bigger than the wavelength specular reection dominates. From this reason we
now consider the case of a reection at a wall, which represents the extreme case of a spacious
object and is more realistic than the point scatterer model in indoor imaging applications. A
sketch of a bat sensor in front of a wall is shown in Fig. 13.
For convenience we use an azimuth angle that is /2 . In this way the sensor orientation
is parallel to the wall at = 0. In case of the wall reection the point where the reection
occurs is not further xed. Instead it moves during the rotation of the sensor. By means of the
geometric construction in Fig. 13 those paths are to be found, which connect Tx and Rx
1/2
after a reection at the wall and which fulll the law of reection at the same time. This law
states that the angle of incidence is equal to the reection angle. Then in our construction the
perpendicular fromthe reection point separates the x-axis in two regions of equal lengths s
x1
and s
x2
, respectively.
To determine the propagation path lengths s
T1/T2
() and s
R1/R2
() we rst need to nd the
dependency of s
x1
and s
x2
on . We establish a linear equation of the general form y = mx + n
for the lines that contain s
R1/R2
:
Rx
2
Rx
1
Tx
s
x2
s
x1
s
x2
s
x1
s
y
2s
y
x
y
T2
T1
R2
R1
s
R1
s
R2
a
a
s
T2
s
T1
Fig. 13. Propagation paths in case of a bat sensor that rotates in front of a wall. The sensor is
symbolized by the bar Rx
1
-Tx-Rx
2
. The wall goes parallel to the x-axis at a distance s
y
. The
antenna main lobe directions are given by the dashed lines. See text for details.
y =
a sin 2 s
y
a cos
x + 2 s
y
. (24)
a cos and a sin are the projections of the bar a onto x- and y-axis. Setting y = 0 and re-
arranging Equ. (24) the intersection with the x-axis is found and in this way the length of
s
x1/x2
:
s
x1/x2
() =
s
y
a cos
a sin 2 s
y
. (25)
Using this result explicit expressions for the path lengths can now be given:
s
T1/T2
() =
s
2
x1/x2
() + s
2
y
. (26)
s
R1/R2
() =
(a cos s
x1/x2
())
2
+ (s
y
a sin )
2
. (27)
The total propagation times
1/2
() are the quotients of the total propagation distances and
the speed of light:
1/2
() =
s
T1/T2
() + s
R1/R2
()
c
. (28)
Fig. 15 illustrates this dependency on a practical example. The result of Equ. (28) is shown for
azimuth angles from 0
to 359
(remember: = /2 ). When the bat sensor is facing the

90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
()
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
()
n
o
r
m
.

a
m
p
l
.
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
n
o
r
m
.

a
m
p
l
.
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
n
o
r
m
.

a
m
p
l
.
(a)
(b)
(c)
Fig. 14. Simulated amplitude vs. bat sensor rotation for point scatterer (left) and wall re-
ection (right). The length of the bat arm (variable a in Fig. 13) was 0.25 m. The distances
between the center of rotation of the bat and the reector were (a) 0.5 m, (b) 1.0 m, and (c) 5.0
m. Hatched line: signal at Rx
1
, solid line: signal at Rx
2
. The radiation pattern used in this
simulation is the measured pattern of a double-ridge horn antenna described in Schwarz et al.
(2010).
object both propagation times are equally long and the curves
1/2
() intersect. The objects
are visible only within a fraction of the rotation angle due to the antenna directivity.
For the calculation of the angles of incidence we rst need to determine the auxiliary angle
1/2
, which is part of the right-angled triangle with sides s
x1/x2
, s
R1/R2
, and s
y
. It is
sin
1/2
() =
s
y
s
T1/T2
()
. (29)
With this result we can give expressions for the angles of emergence
T1/T2
at the Tx antenna:
T1/T2
() =

2
arcsin
s
y
s
T1/T2
()
. (30)
In a similar way we determine the angles of incience
R1/R2
at the receiver antennas. Here we
have to use the auxiliary angle
1/2
that can be found from the sum of the internal angles of
()
t

(
n
s
)
0 45 90 135 180 225 270 315
0
5
10
15
20
25
Rx
2
Rx
1
R
2 a / c
Fig. 15. Radargram of receiver Rx
1
with the curves of time of arrival (TOA) vs. bat azimuth
angle for Rx
1
and Rx
2
. The Ellipse R marks the radar reex for which the TOA curves have
been calculated.
the triangles with corner points Tx, Rx
1/2
, and the intersection point of s
R1/R2
with the x-axis.
For Rx
1
and Rx
2
these sums are:
= (
1
) +
1
+ = (
2
) +
2
+ . (31)
By using these relations and Equ. (29) an expression for the angles of incidence can be given:
R1/R2
() =

2

1/2
() =

2
arcsin
s
y
s
T1/T2
()
. (32)
Now the amplitude A
1/2
of the received signal vs. the rotation angle can be computed.
Given an antenna radiation patter p() the amplitude can be computed by a multiplication
of weighting factors for the incidence at the Rx and Tx antennas and for the total propagation
pathlength:
A
1/2
() = p(
R1/R2
()) p(
T1/T2
())
s
ref
s
T1/T2
() + s
R1/R2
()
. (33)
The reference distance s
ref
is the total propagation distance at = 0. In Fig. 14 results are
plotted for the antenna DRH-90 (Schwarz et al. (2010)). At short distances (curves (a) and
(b)) the successive appearance of the object rst in the visibility range of Rx
2
and then of
Rx
1
can clearly be seen. This effect is reduced in case of the wall reection. The pattern at
the largest distance (5 m) are practically identical with the square of the measured antenna
radiation pattern. In this simulation the antennas are treated as point objects. This is still a
simplication.
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
()
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
()
n
o
r
m
.

a
m
p
l
.
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
n
o
r
m
.

a
m
p
l
.
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
90 45 0 45 90
0
0.2
0.4
0.6
0.8
1.0
n
o
r
m
.

a
m
p
l
.
(a)
(b)
(c)
Fig. 14. Simulated amplitude vs. bat sensor rotation for point scatterer (left) and wall re-
ection (right). The length of the bat arm (variable a in Fig. 13) was 0.25 m. The distances
between the center of rotation of the bat and the reector were (a) 0.5 m, (b) 1.0 m, and (c) 5.0
m. Hatched line: signal at Rx
1
, solid line: signal at Rx
2
. The radiation pattern used in this
simulation is the measured pattern of a double-ridge horn antenna described in Schwarz et al.
(2010).
object both propagation times are equally long and the curves
1/2
() intersect. The objects
are visible only within a fraction of the rotation angle due to the antenna directivity.
For the calculation of the angles of incidence we rst need to determine the auxiliary angle
1/2
, which is part of the right-angled triangle with sides s
x1/x2
, s
R1/R2
, and s
y
. It is
sin
1/2
() =
s
y
s
T1/T2
()
. (29)
With this result we can give expressions for the angles of emergence
T1/T2
at the Tx antenna:
T1/T2
() =

2
arcsin
s
y
s
T1/T2
()
. (30)
In a similar way we determine the angles of incience
R1/R2
at the receiver antennas. Here we
have to use the auxiliary angle
1/2
that can be found from the sum of the internal angles of
()
t

(
n
s
)
0 45 90 135 180 225 270 315
0
5
10
15
20
25
Rx
2
Rx
1
R
2 a / c
Fig. 15. Radargram of receiver Rx
1
with the curves of time of arrival (TOA) vs. bat azimuth
angle for Rx
1
and Rx
2
. The Ellipse R marks the radar reex for which the TOA curves have
been calculated.
the triangles with corner points Tx, Rx
1/2
, and the intersection point of s
R1/R2
with the x-axis.
For Rx
1
and Rx
2
these sums are:
= (
1
) +
1
+ = (
2
) +
2
+ . (31)
By using these relations and Equ. (29) an expression for the angles of incidence can be given:
R1/R2
() =

2

1/2
() =

2
arcsin
s
y
s
T1/T2
()
. (32)
Now the amplitude A
1/2
of the received signal vs. the rotation angle can be computed.
Given an antenna radiation patter p() the amplitude can be computed by a multiplication
of weighting factors for the incidence at the Rx and Tx antennas and for the total propagation
pathlength:
A
1/2
() = p(
R1/R2
()) p(
T1/T2
())
s
ref
s
T1/T2
() + s
R1/R2
()
. (33)
The reference distance s
ref
is the total propagation distance at = 0. In Fig. 14 results are
plotted for the antenna DRH-90 (Schwarz et al. (2010)). At short distances (curves (a) and
(b)) the successive appearance of the object rst in the visibility range of Rx
2
and then of
Rx
1
can clearly be seen. This effect is reduced in case of the wall reection. The pattern at
the largest distance (5 m) are practically identical with the square of the measured antenna
radiation pattern. In this simulation the antennas are treated as point objects. This is still a
simplication.
5.4 Detection of Individual Reectors
Typically, a radar image is created from the acquired data by application of a migration algo-
rithm. This procedure can as well be applied to the rotating bat-type sensor. In this case delays
have to be calculated (similar to Equ. (12)) for both Rx separately. An example is shown in
Fig. 17(b).
A disadvantage of this procedure is the appearance of parts of ellipses or circles in the nal
image, as they can be seen in the example. These are remains from the summation along
curves. Furthermore, almost all migration algorithms were derived for point-like scatterers,
but in indoor imaging most objects cause specular reections.
Remove constant signals Remove constant signals
Data Rx
1
Data Rx
2
Amplitude correction Amplitude correction
Multiplication
2D Correlation
2D Pulse shape
Creation of a table of single reflexes
Parameter extraction
Draw reflection point and reflecting surface
Image
Fig. 16. Creation of an image from individual reections at at surfaces.
An attempt to create a more realistic nal image is detection of individual at reectors from
the radar returns. This is based on the following consideration: The reection from a at
surface reaches a maximum when the irradiation occurs along the direction of the surface
normal; from this it follows that during sensor rotation, signal maxima will occur when the
sensor looks towards the surface normal of objects. Signal components that are reected into
directions far away from the surface normal cannot be detected from the bat-type sensor, be-
cause of the sensors small dimensions.
(a) Photograph
A
B
C
D
F
E
G
H
I
Path of bat sensor
(b) Conventional processing
-5 -4 -3 -2 -1 0 1
-3
-2
-1
0
1
2
3
4
5
6
x(m)
y
(
m
)
(c) Detection of single scatterers.
-5 -4 -3 -2 -1 0 1
-3
-2
-1
0
1
2
3
4
5
6
x(m)
y
(
m
)
(d) Schematic representation of reecting
surfaces.
Fig. 17. Imaging with bat-type sensor in a stairwell. The images show an overlay of 15 mea-
surement results, performed along the marked sensor path.
Now each reex in the radargram can be interpreted as a at surface fragment. The center
of gravity of the reex delivers the coordinates of the reection point and the signal intensity
is a measure for the reector dimensions. Using this information one can construct an image
consisting of individual at surface fragments.
The method requires special processing of the Rx signals, schematically described in Fig. 16.
Pre-processing includes the removal of constant signals (originating from cross-talk between
5.4 Detection of Individual Reectors
Typically, a radar image is created from the acquired data by application of a migration algo-
rithm. This procedure can as well be applied to the rotating bat-type sensor. In this case delays
have to be calculated (similar to Equ. (12)) for both Rx separately. An example is shown in
Fig. 17(b).
A disadvantage of this procedure is the appearance of parts of ellipses or circles in the nal
image, as they can be seen in the example. These are remains from the summation along
curves. Furthermore, almost all migration algorithms were derived for point-like scatterers,
but in indoor imaging most objects cause specular reections.
Remove constant signals Remove constant signals
Data Rx
1
Data Rx
2
Amplitude correction Amplitude correction
Multiplication
2D Correlation
2D Pulse shape
Creation of a table of single reflexes
Parameter extraction
Draw reflection point and reflecting surface
Image
Fig. 16. Creation of an image from individual reections at at surfaces.
An attempt to create a more realistic nal image is detection of individual at reectors from
the radar returns. This is based on the following consideration: The reection from a at
surface reaches a maximum when the irradiation occurs along the direction of the surface
normal; from this it follows that during sensor rotation, signal maxima will occur when the
sensor looks towards the surface normal of objects. Signal components that are reected into
directions far away from the surface normal cannot be detected from the bat-type sensor, be-
cause of the sensors small dimensions.
(a) Photograph
A
B
C
D
F
E
G
H
I
Path of bat sensor
(b) Conventional processing
-5 -4 -3 -2 -1 0 1
-3
-2
-1
0
1
2
3
4
5
6
x(m)
y
(
m
)
(c) Detection of single scatterers.
-5 -4 -3 -2 -1 0 1
-3
-2
-1
0
1
2
3
4
5
6
x(m)
y
(
m
)
(d) Schematic representation of reecting
surfaces.
Fig. 17. Imaging with bat-type sensor in a stairwell. The images show an overlay of 15 mea-
surement results, performed along the marked sensor path.
Now each reex in the radargram can be interpreted as a at surface fragment. The center
of gravity of the reex delivers the coordinates of the reection point and the signal intensity
is a measure for the reector dimensions. Using this information one can construct an image
consisting of individual at surface fragments.
The method requires special processing of the Rx signals, schematically described in Fig. 16.
Pre-processing includes the removal of constant signals (originating from cross-talk between
Tx and Rx antenna at the sensor platform) and distance-dependent amplitude correction of
the received signals. After these processing steps a radargramcan be drawn for each Rx sepa-
rately (Fig. 15). Pointwise multiplication of both radargrams increases resolution, because the
products of the reexes occupy a smaller area. All product reexes have a very similar shape.
This allows application of a 2Dcorrelation of the product-radargramwith an example pattern,
leading to signicant improvement of the signal to noise ratio. Experiments have shown that
a 2D Gauss-shaped function is a well suited pattern:
G(, r) = exp
exp
r r
0
r
. (34)
0
and r
0
are azimuth and radius of the reex, respectively. For the equipment used in the
practical tests (UWB radar 3.5-10.5 GHz, double-ridge horn antenna) the best parameters were
= 8.5
and r =55 mm. Afterwards a threshold detector isolates individual reexes. A

table of these reexes is created. Then 3 parameters are extracted fromeach reex :
0
, r
0
, and
the amplitude. Finally the image map is drawn (Fig. 17(c)). The detected reex centers are
marked with small points while the small bars are aligned along the surface orientation. The
lengths of the bars are proportional to the signal amplitude. The image generated in this way
comes already close to a schematic representation of the object surfaces (Fig. 17(d)).
6. Conclusions
UWB sensor networks can be used for rough imaging of the environment. Measurement
results, presented e.g. in Fig. 9 and 17, demonstrate that the shape of simple objects or the
structure of a room can be imaged with sufcient quality.
Methods which have been described elsewhere explicitely (e.g.4.3) were mentioned only
briey. New methods (especially chapter 5) were explained in more detail.
The methods presented in this article are especially useful for imaging of interiors. Special
applications, e.g. through-wall imaging, have not been mentioned. 3D imaging in a sensor
network doesnt seem to be a realistic issue. This would require time-consuming scanning of
a 2D surface.
The subject is still under development. Some of the tasks that must be solved by future re-
search are:
Completely wireless operation of the imaging network must be achieved by introduc-
tion of wireless node synchronization.
The time for data acquisition must be reduced, i.e. methods must be developed that
generate an image from a smaller amount of raw data.
Image artefacts must be reduced, e.g. by improved migration algorithms.
Imaging in UWB sensor networks can become part of surveillance systems with these im-
provements.
7. References
Federal Communications Commission, (2002). "Revision of part 15 of the Communications
Rules", FCC 02-48, April 2002, pp. 1-174. http://hraunfoss.fcc.gov
Foo, S. & Kashyap, S., "Cross-correlated back projection for UWB radar imaging", Proceedings
of IEEE Antennas and Propagation Society International Symposium, pp. 1275-1278, ISBN:
0-7803-8302-8, Monterey, CA, USA, September 2004, IEEE Conference Publishing,
Piscataway, NJ, USA.
Gu, K.; Wang, G. & Li, J., (2004). "Migration based SAR imaging for ground penetrating radar
systems", IEE Proceedings Radar Sonar & Navigation, Vol. 151, No. 5, November 2004,
pp. 317-325, ISSN: 1350-2395.
Hirsch, O.; Janson, M.; Wiesbeck, W. & Thom, R. S., (2010). "Indirect Localization and Imag-
ing of Objects in an UWB Sensor Network", IEEE Transactions on Instrumentation and
Measurement, Vol. 59, No. 11, Nov. 2010, pp. 2949-2957, ISSN: 0018-9456.
Hussain, M. G. M., (1998). "Ultra-wideband impulse radar - an overview of the principles",
IEEE Aerospace and Electronic Systems Magazine, Vol.13, No.9, September 1998, pp.9-
14, ISSN: 0885-8985.
Luediger, H. & Kallenborn, R., (2009). "Generic UWB Regulation in Europe", Frequenz, Journal
of RF-Engineering and Telecommunications, Vol.63, No.9-10, Sept./Oct. 2009, pp.172-
174, ISSN: 0016-1136.
Margrave, F. G., (2001). Numerical Methods of Exploration Seismology with algorithms in MATLAB,
http://www.crewes.org/ResearchLinks/FreeSoftware/EduSoftware/
NMES_Margrave.pdf.
Oliver, C. J., (1989). "Synthetic-aperture radar imaging", Journal of Physics D: Applied Physics,
Vol.22, No. 7, July 1989, pp. 871-890, doi: 10.1088/0022-3727/22/7/001.
Patwari, N.; Ash, J.N.; Kyperountas, S.; Hero, A.O., III; Moses, R.L.; Correal, N.S., (2005).
"Locating the nodes", IEEE Signal Processing Magazine, Vol. 22, No. 4, June 2005, pp.
54-69, ISSN: 1053-5888.
Rong, P. & Sichitiu, M. L., (2006). "Angle of Arrival Localization for Wireless Sensor Net-
works", Proceedings of SECON 2006, pp. 374-382, ISBN: 1-4244-0626-9, Reston, VA,
USA, September 2006, IEEE Catalog Number: 06EX1523.
Sachs, J.; Peyerl, P.; Zetik, R. & Crabbe, S., (2003). "M-Sequence Ultra-Wideband-Radar: State
of Development and Applications", Proceedings of Radar 2003, pp. 224-229, ISBN:
0-7803-7871-7, Adelaide, Australia, September 2003, Causal Productions, Adelaide,
Australia.
Sachs, J. (2004). "M-Sequence Radar", In: Ground Penetrating Radar, Daniels, D. J., (Ed.), 2nd
ed., pp. 225-236, Inst. of Electrical Engineers, ISBN:0-85296-862-0, Stevenage, U.K. .
Sayed, A. H.; Tarighat, A.; Khajehnouri, N., (2005). "Network-Based Wireless Location", IEEE
Signal Processing Magazine, Vol.22, No.4, July 2005, pp.24-40, ISSN: 1053-5888.
Schwarz, U.; Thiel, F,; Seifert, F.; Stephan, R. &Hein, M. A., (2010). "Ultra-Wideband Antennas
for Magnetic Resonance Imaging Navigator Techniques", IEEE Trans. on Antennas and
Propagation, vol. 58, No. 6, June 2010, pp.2107-2112, ISSN: 0018-926X.
Serpedin, E. & Chaudhari, Q. M. (2009). Synchronization in Wireless Sensor Networks: Parameter
Estimation, Performance Benchmarks and Protocols, Cambridge University Press, ISBN:
978-0-521-76442-1, Cambridge, UK.
Stoica, P. & Li, J., (2006). "Lecture Notes - Source Localization from Range-Difference Mea-
surements", IEEE Signal Processing Magazine, Vol.23, No.6, November 2006, pp.63-66,
ISSN: 1053-5888.
Tx and Rx antenna at the sensor platform) and distance-dependent amplitude correction of
the received signals. After these processing steps a radargramcan be drawn for each Rx sepa-
rately (Fig. 15). Pointwise multiplication of both radargrams increases resolution, because the
products of the reexes occupy a smaller area. All product reexes have a very similar shape.
This allows application of a 2Dcorrelation of the product-radargramwith an example pattern,
leading to signicant improvement of the signal to noise ratio. Experiments have shown that
a 2D Gauss-shaped function is a well suited pattern:
G(, r) = exp
exp
r r
0
r
. (34)
0
and r
0
are azimuth and radius of the reex, respectively. For the equipment used in the
practical tests (UWB radar 3.5-10.5 GHz, double-ridge horn antenna) the best parameters were
= 8.5
and r =55 mm. Afterwards a threshold detector isolates individual reexes. A

table of these reexes is created. Then 3 parameters are extracted fromeach reex :
0
, r
0
, and
the amplitude. Finally the image map is drawn (Fig. 17(c)). The detected reex centers are
marked with small points while the small bars are aligned along the surface orientation. The
lengths of the bars are proportional to the signal amplitude. The image generated in this way
comes already close to a schematic representation of the object surfaces (Fig. 17(d)).
6. Conclusions
UWB sensor networks can be used for rough imaging of the environment. Measurement
results, presented e.g. in Fig. 9 and 17, demonstrate that the shape of simple objects or the
structure of a room can be imaged with sufcient quality.
Methods which have been described elsewhere explicitely (e.g.4.3) were mentioned only
briey. New methods (especially chapter 5) were explained in more detail.
The methods presented in this article are especially useful for imaging of interiors. Special
applications, e.g. through-wall imaging, have not been mentioned. 3D imaging in a sensor
network doesnt seem to be a realistic issue. This would require time-consuming scanning of
a 2D surface.
The subject is still under development. Some of the tasks that must be solved by future re-
search are:
Completely wireless operation of the imaging network must be achieved by introduc-
tion of wireless node synchronization.
The time for data acquisition must be reduced, i.e. methods must be developed that
generate an image from a smaller amount of raw data.
Image artefacts must be reduced, e.g. by improved migration algorithms.
Imaging in UWB sensor networks can become part of surveillance systems with these im-
provements.
7. References
Federal Communications Commission, (2002). "Revision of part 15 of the Communications
Rules", FCC 02-48, April 2002, pp. 1-174. http://hraunfoss.fcc.gov
Foo, S. & Kashyap, S., "Cross-correlated back projection for UWB radar imaging", Proceedings
of IEEE Antennas and Propagation Society International Symposium, pp. 1275-1278, ISBN:
0-7803-8302-8, Monterey, CA, USA, September 2004, IEEE Conference Publishing,
Piscataway, NJ, USA.
Gu, K.; Wang, G. & Li, J., (2004). "Migration based SAR imaging for ground penetrating radar
systems", IEE Proceedings Radar Sonar & Navigation, Vol. 151, No. 5, November 2004,
pp. 317-325, ISSN: 1350-2395.
Hirsch, O.; Janson, M.; Wiesbeck, W. & Thom, R. S., (2010). "Indirect Localization and Imag-
ing of Objects in an UWB Sensor Network", IEEE Transactions on Instrumentation and
Measurement, Vol. 59, No. 11, Nov. 2010, pp. 2949-2957, ISSN: 0018-9456.
Hussain, M. G. M., (1998). "Ultra-wideband impulse radar - an overview of the principles",
IEEE Aerospace and Electronic Systems Magazine, Vol.13, No.9, September 1998, pp.9-
14, ISSN: 0885-8985.
Luediger, H. & Kallenborn, R., (2009). "Generic UWB Regulation in Europe", Frequenz, Journal
of RF-Engineering and Telecommunications, Vol.63, No.9-10, Sept./Oct. 2009, pp.172-
174, ISSN: 0016-1136.
Margrave, F. G., (2001). Numerical Methods of Exploration Seismology with algorithms in MATLAB,
http://www.crewes.org/ResearchLinks/FreeSoftware/EduSoftware/
NMES_Margrave.pdf.
Oliver, C. J., (1989). "Synthetic-aperture radar imaging", Journal of Physics D: Applied Physics,
Vol.22, No. 7, July 1989, pp. 871-890, doi: 10.1088/0022-3727/22/7/001.
Patwari, N.; Ash, J.N.; Kyperountas, S.; Hero, A.O., III; Moses, R.L.; Correal, N.S., (2005).
"Locating the nodes", IEEE Signal Processing Magazine, Vol. 22, No. 4, June 2005, pp.
54-69, ISSN: 1053-5888.
Rong, P. & Sichitiu, M. L., (2006). "Angle of Arrival Localization for Wireless Sensor Net-
works", Proceedings of SECON 2006, pp. 374-382, ISBN: 1-4244-0626-9, Reston, VA,
USA, September 2006, IEEE Catalog Number: 06EX1523.
Sachs, J.; Peyerl, P.; Zetik, R. & Crabbe, S., (2003). "M-Sequence Ultra-Wideband-Radar: State
of Development and Applications", Proceedings of Radar 2003, pp. 224-229, ISBN:
0-7803-7871-7, Adelaide, Australia, September 2003, Causal Productions, Adelaide,
Australia.
Sachs, J. (2004). "M-Sequence Radar", In: Ground Penetrating Radar, Daniels, D. J., (Ed.), 2nd
ed., pp. 225-236, Inst. of Electrical Engineers, ISBN:0-85296-862-0, Stevenage, U.K. .
Sayed, A. H.; Tarighat, A.; Khajehnouri, N., (2005). "Network-Based Wireless Location", IEEE
Signal Processing Magazine, Vol.22, No.4, July 2005, pp.24-40, ISSN: 1053-5888.
Schwarz, U.; Thiel, F,; Seifert, F.; Stephan, R. &Hein, M. A., (2010). "Ultra-Wideband Antennas
for Magnetic Resonance Imaging Navigator Techniques", IEEE Trans. on Antennas and
Propagation, vol. 58, No. 6, June 2010, pp.2107-2112, ISSN: 0018-926X.
Serpedin, E. & Chaudhari, Q. M. (2009). Synchronization in Wireless Sensor Networks: Parameter
Estimation, Performance Benchmarks and Protocols, Cambridge University Press, ISBN:
978-0-521-76442-1, Cambridge, UK.
Stoica, P. & Li, J., (2006). "Lecture Notes - Source Localization from Range-Difference Mea-
surements", IEEE Signal Processing Magazine, Vol.23, No.6, November 2006, pp.63-66,
ISSN: 1053-5888.
Stolt, R. H., (1978). "Migration by Fourier Transform", Geophysics, Vol.43, No.1, February 1978,
pp.23-48, ISSN: 0016-8033.
Thom, R. S.; Hirsch, O.; Sachs, J. & Zetik, R., (2007). "UWB Sensor Networks for Position
Location and Imaging of Objects and Environments", Proceedings of the 2nd European
Conference on Antennas and Propagation (EuCAP2007), pp. 1-9, ISBN: 978-0-86341-842-
6, Edinburgh, U.K., November 2007, IET Conferences, London, U.K.
Yang, Y. & Yang, K., (2006). "Time Synchronization for Wireless Sensor Networks using
the Principle of Radar Systems and UWB Signals", Proceedings of IEEE International
Conference on Information Acquisition 2006, pp. 160-165, ISBN: 1-4244-0528-9, Weihai,
China, August 2006, IEEE Conference Publishing, Piscataway, NJ, USA.
Zetik, R.; Sachs, J. & Thom, R., (2005) "Modied Cross-Correlation Back Projection for UWB
Imaging: Numerical Examples", Proceedings of IEEE International Conference on Ultra-
Wideband, pp. 650-654, ISBN: 0-7803-9397-X, Zrich, Switzerland, September 2005,
IEEE.
Zetik, R.; Sachs, J. & Thom, R., (2005) "Modied Cross-Correlation Back Projection for UWB
Imaging: Measurement Examples", Proceedings of the 6th International Scientic Con-
ference on Digital Signal Processing and Multimedia Communications - DSP-MCOM2005,
pp. 56-59, ISBN: 80-8073-323-6, Kosice, Slovakia, September 2005, Technicka uni-
verzita Kosice, Kosice, Slovakia.
Zetik, R.; Sachs, J. & Thom, R., (2010) "Imaging of distributed objects in UWB sensor net-
works", In: Short Pulse Electromagnetics 9, Sabath, F.; Giri, D. ; Rachidi, F. & Kaelinpp,
A. (Ed.), pp. 97-104, Springer-Verlag New York Inc, ISBN: 9780387778440, New York,
NY, USA.

Wireless Sensor Networks Application Centric Design

Uploaded by

Copyright:

Available Formats

Wireless Sensor Networks Application Centric Design

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wireless Sensor Networks Application Centric Design

Uploaded by

Copyright:

Available Formats

What are some of the applications discussed for wireless sensor networks?

What are some of the applications discussed for wireless sensor networks?

What are some of the challenges discussed for implementing wireless sensor networks?

What are some of the challenges discussed for implementing wireless sensor networks?

Wireless sensor netWorks:

encoded symbols, where k

k asymptotically, with encoding/decoding complexity of the order O(k logk).

encoded symbols, where k

k asymptotically, with encoding/decoding complexity of the order O(k logk).

) k encoded packets, where

usually is a small positive num-

) k encoded packets, where

usually is a small positive num-

. Several special sensing functions

. Several special sensing functions

> N, provided that the weighting array

> N, provided that the weighting array

. At last, the I/Qsignals are summed by an adder, amplied by a

. At last, the I/Qsignals are summed by an adder, amplied by a

offset in degree, is the I and

is the amplitude probability density function of

offset in degree, is the I and

is the amplitude probability density function of

delay of the input of the

delay of the output of the I path I

delay of the input of the

delay of the output of the I path I

(A/) are drawn from the estimated location of the node.

(A/) is defined as the one-hop

(A/) are drawn from the estimated location of the node.

(A/) is defined as the one-hop

are estimated, which can be achieved by the reference nodes mutually

are estimated, which can be achieved by the reference nodes mutually

. The noise component associated with Y

which indicates that

. The noise component associated with Y

which indicates that

(k is the potential number of neighbors) (k = 6 in Figure 1).

[Note: 0 and 1 are sensor readings (sample)].

) only indicate the relative locations of its neighbors

(k is the potential number of neighbors) (k = 6 in Figure 1).

[Note: 0 and 1 are sensor readings (sample)].

) only indicate the relative locations of its neighbors

is for running the transmitter or receiver circuitry and

), to receive the message, the radio expands (E

follows a k-component finite mixture distribution

), j = 1,2,,k; I = 1,2,,n. (1)

is the mixing weight or sometimes called the prior weight and

is to maximize the log-likelihood criterion as

) with its mean vector

and covariance matrix

) in the finite mixture model:

is for running the transmitter or receiver circuitry and

), to receive the message, the radio expands (E

follows a k-component finite mixture distribution

), j = 1,2,,k; I = 1,2,,n. (1)

is the mixing weight or sometimes called the prior weight and

is to maximize the log-likelihood criterion as

) with its mean vector

and covariance matrix