Trends in Telecommunications Technologies (2010)

I
Trends in Telecommunications
Technologies
Trends in Telecommunications
Technologies
Edited by
Christos J. Bouras
In-Tech
intechweb.org
Published by In-Teh
In-Teh
Olajnica 19/2, 32000 Vukovar, Croatia
Abstracting and non-profit use of the material is permitted with credit to the 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. Publisher assumes no responsibility liability for any damage or injury to persons or
property arising out of the use of any materials, instructions, methods or ideas contained inside. After
this work has been published by the In-Teh, authors have the right to republish it, in whole or part, in any
publication of which they are an author or editor, and the make other personal use of the work.
© 2010 In-teh
www.intechweb.org
Additional copies can be obtained from:
[email protected]
First published March 2010

Printed in India
Technical Editor: Goran Bajac

Cover designed by Dino Smrekar
Trends in Telecommunications Technologies,

Edited by Christos J. Bouras
p. cm.
ISBN 978-953-307-072-8
V
Preface
This book is the culmination of an effort to gather world-class scientists, engineers and
educators engaged in the fields of telecommunications to meet and present their latest
activities. Telecommunication is the assisted transmission of signals over a distance for
the purpose of communication. Having drastically transformed the human way of living,
telecommunications are considered the revolution of our times, and the catalyst for present
and future technological and scientific developments. Being a very active research field, new
advances in telecommunications are constantly changing the landscape and introduce new
capabilities and enhanced ways of communication. Literature in the field is extensive and
constantly enlarged. This book therefore intends to increase the dissemination level of the
latest research advances and breakthroughs by making them available to a wide audience in
a compact and friendly to the user way. It furthermore aims to provide a particularly good
way for experts in one aspect of the field to learn about advances made by their colleagues
with different research interests.
The main focus of the book is the advances in telecommunications modeling, policy, and
technology. In particular, several chapters of the book deal with low-level network layers
and present issues in optical communication technology and optical networks, including
the deployment of optical hardware devices and the design of optical network architecture.
Wireless networking is also covered, with a focus on WiFi and WiMAX technologies. The
book also contains chapters that deal with transport issues, and namely protocols and policies
for efficient and guaranteed transmission characteristics while transferring demanding data
applications such as video. Finally, the book includes chapters that focus on the delivery of
applications through common telecommunication channels such as the earth atmosphere.
This book is useful for researchers working in the telecommunications field, in order to read
a compact gathering of some of the latest efforts in related areas. It is also useful for educators
that wish to get an up-to-date glimpse of telecommunications research and present it in
an easily understandable and concise way. It is finally suitable for the engineers and other
interested people that would benefit from an overview of ideas, experiments, algorithms and
techniques that are presented throughout the book.
VI
Bringing this book to the publication stage is the reward of the effort put by the editors,
contributors and reviewers of all the presented chapters. We would therefore like to
acknowledge the valuable contributions of all the authors of the chapters contained in this
book, who decided to offer their insights and expertise in order to help assemble, in our
opinion, a highly useful and quality book. We would also like to thank the publishing house,
which supported this effort and helped make the result available to a wide audience all over
the world.
Patras, 17th March, 2010

Christos J. Bouras
Professor
University of Patras and RACTI
VII
Contents
Preface V
1. A Novel PFC Circuit for Three-phase utilizing Single Switching Device 001
Keiju Matsui and Masaru Hasegawa
2. Advanced Modulation Formats and Multiplexing Techniques for Optical

Telecommunication Systems 013
Ghafour Amouzad Mahdiraji and Ahmad Fauzi Abas
3. A Survey on the Design of Binary Pulse Compression Codes

with Low Autocorrelation 039
Maryam Amin Nasrabadi and Mohammad Hassan Bastani
4. Virtual Multicast 063

Petr Holub and Eva Hladká
5. The Asymmetrical Architecture of New Optical Switch Device 087

Mohammad Syuhaimi Ab-Rahman and Boonchuan Ng
6. Adaptive Active Queue Management for TCP Friendly Rate Control (TFRC)
Traffic in Heterogeneous Networks 109
Rahim Rahmani and Christer Åhlund
7. Queues with session arrivals as models for optimizing the traffic control
in telecommunication networks 123
Sergey Dudin and Moon Ho Lee
8. Telecommunication Power System: energy saving, renewable sources

and environmental monitoring 145
Carmine Lubritto
9. Propagation Models and their Applications in Digital Television Broadcast

Network Design and Implementation 165
Armoogum V., Soyjaudah K.M.S., Mohamudally N. and Fogarty T.
10. Interference Modeling for Wireless Ad Hoc Networks 185

Altenis V. Lima-e-Lima, Carlos E. B. Cruz Pimentel and Renato M. de Moraes
VIII
11. Energy Saving Drives New Approaches to Telecommunications Power System 201
Rais Miftakhutdinov
12. Directional Routing Protocol in Wireless Mobile Ad Hoc Network 235

L.A.Latiff, N. Fisal, S.A. Arifin and A. Ali Ahmed
13. Free Space Optical Technologies 257

Davide M. Forin, G. Incerti, G.M. Tosi Beleffi, A.L.J. Teixeira, L.N. Costa,
P.S. De Brito Andrè, B. Geiger, E. Leitgeb and F. Nadeem
14. Novel multiple access models and their probabilistic description 297
Dmitry Osipov
15. Performance analysis of multi-server queueing system operating under control

of a random environment 317
Che Soong Kim, Alexander Dudin, Valentina Klimenok and Valentina Khramova
16. Interdomain QoS paths finding based on overlay topology

and QoS negotiation approach 345
Şerban Georgică Obreja and Eugen Borcoci
17. Dual Linearly Polarized Microstrip Array Antenna 367

M. S. R Mohd Shah, M. Z. A Abdul Aziz, M. K. Suaidi and M. K. A Rahim
18. A Time-Delay Suppression Technique for Digital PWM Control Circuit 389
Yoichi Ishizuka
19. Layer 2 Quality of Service Architectures 399

Christos Bouras, Vaggelis Kapoulas, Vassilis Papapanagiotou,
Leonidas Poulopoulos, Dimitris Primpas and Kostas Stamos
20. Secrecy on the Physical Layer in Wireless Networks 413

Eduard A. Jorswieck, AnneWolf, and Sabrina Gerbracht
21. Performance Analysis of Time-of-Arrival Mobile Positioning

in Wireless Cellular CDMA Networks 437
M. A.Landolsi, A. H. Muqaibel, A. S. Al-Ahmari, H.-R. Khan and R. A. Al-Nimnim
22. Mobility and Handoff Management in Wireless Networks 457

Jaydip Sen
23. GPS Total Electron Content (TEC) Prediction at Ionosphere Layer

over the Equatorial Region 485
Norsuzila Ya’acob, Mardina Abdullah and Mahamod Ismail
24. Performance Evaluation Methods to Study IEEE 802.11 Broadband Wireless

Networks under the PCF Mode 509
Vladimir Vishnevsky and Olga Semenova
25. Next Generation Optical Access Networks: from TDM to WDM 537
Ll. Gutierrez, P. Garfias, M. De Andrade, C. Cervelló-Pastor and S. Sallent
IX
26. Building energy efficiency design for telecommunication base

stations in Guangzhou 561
Yi Chen, Yufeng Zhang and Qinglin Meng
27. Dynamic Space-Code Multiple Access (DSCMA) System: A Double Interference

Cancellation Multiple Access Scheme in Wireless Communications System 583
Chee Kyun Ng, Nor Kamariah Noordin, Borhanuddin Mohd Ali,
and Sudhanshu Shekhar Jamuar
28. Video Streaming in Evolving Networks under Fuzzy Logic Control 613
Martin Fleury, Emmanuel Jammeh, Rouzbeh Razavi, Sandro Moiron
and Mohammed Ghanbari
29. The Development of Crosstalk-Free Scheduling Algorithms for Routing

in Optical Multistage Interconnection Networks 643
Mohamed Othman, and Tg Dian Shahida Raja Mohd Auzar
30. Traditional float charges: are they suited to stationary

antimony-free lead acid batteries? 665
T. M. Phuong Nguyen, Guillaume Dillenseger Christian Glaize and Jean Alzieu
31. Neighbor Discovery: Security Challenges in Wireless Ad hoc

and Sensor Networks 693
Mohammad Sayad Haghighi and Kamal Mohamedpour
32. New Trends in Network Anomaly Detection 715

Yasser Yasami and Saadat Pourmozaffari
33. An Efficient Energy Aware Routing Protocol for Real Time Traffics
in Wireless Sensor Networks 735
Amir Hossein Mohajerzadeh and Mohammad Hossein Yaghmaee
34. Quality of Service Differentiation in WiMAX Networks 753

Pedro Neves, Susana Sargento, Francisco Fontes,
Thomas M. Bohnert and João Monteiro
A Novel PFC Circuit for Three-phase utilizing Single Switching Device 1
X1
A Novel PFC Circuit for Three-phase

utilizing Single Switching Device
Keiju Matsui and Masaru Hasegawa
Chubu University
Japan
1. Introduction
For consumer or industrial applications, electrical appliances use various types of rectifier,
which give rise to distorted input current due their non linear characteristics. Problems are
created by the various harmonics, generated in the power system. Under such
circumstances, IEC guideline was instituted ten and several years ago, and has recently been
superseded.(JISC.2005). With the spread of the use of such nonlinear equipments, it is
anticipated that we can not avoid the problems due to harmonics. With the relatively
increased capacity of industry applications, PWM rectifiers can be expected to be used in
three phase and single phase applications. (Takahashi. 1985, IEEJ Committee. 2000). Also in
office environments, OA equipments, inverter type fluorescent lamps and inverter type air-
conditioners are frequently used, surely bringing harmonic problems with them. Under
such conditions, various new type PFC schemes are presented and discussed.
(Takahashi.1900,Fujiwara.1991,Takeuchi2005). Methods intending to improve the current
towards a sinusoidal waveform by using switching devices will incur high cost performance
and yet troublesome noise problems. Certain applications require a switch-less scheme to
maintain the electromagnetic environmental standards. (Yamamoto. 2001, Takeuchi. 2007).
Also in the future, main stream methods will intend to achieve sinusoidal waveforms. From
thinking about research stream until now, more simplified method or low cost scheme
would be discussed and developed in a similar manner also in the future. On the basis of the
perceived requirements, in this paper, we propose and discuss a novel PFC circuit for three
phase, employing a single switch in such a manner as to render the waveform as sinusoidal
as possible.
2. Operational Principle
2.1 Prasad-Ziogas Circuit
Figure 1 shows a conventional circuit, comprising a three-phase circuit, using single
switching device. (Prasad & Ziogas. 1991). The principle of operation is such that the three
phase circuit is periodically shorted by a single switching device at a high frequency, so that
the input current waveform is created in proportion to input voltage waveform. The input
current waveform becomes synchronized with the input voltage, so that the circuit scheme
2 Trends in Telecommunications Technologies
is constructed as PFC circuit. In this paper, this circuit is named the PZ (Prasad-Ziogas)
circuit, one of these individuals being famous for contributions toward power electronics
development.
eu iu
Lu C1
O iv A
S1
B O' Ro Vd
iw
C
C1
Fig. 1. Three-phase single switch PFC circuit by Prasad-Ziogas.
iu iu Lu
Lu A
eu
eu eu>0 C1
eu>0
O O'
(a) S1 turn-on （b）S1 turn-off

Fig. 2. Equivalent circuit for Prasad-Ziogas.
Figure 2 shows the equivalent circuit of the PZ circuit. These characteristics may be
explained as follows; In Figure 2 (a), when S1 turns-on, the equivalent circuit is established
as shown, where the operation will be explained as a current-discontinuous mode for
simplicity of circuit analysis. In Figure 2 (b), the terminal voltage across O and O' of fictional
neutral point can be derived from Figure 1, the amplitude being E0/6 with an operational
frequency three times supply frequency, where E0 is the output dc link voltage. In Figure 2
(a), when S1 is turned on, circuit equation can be established as follows;
di u
eu  Lu (1)
dt
The analogous equations can be described also in phase v and phase w. From Eq. (1), the
input current is increasing in proportion to amplitude of eu at S1 turn-on. (see Figure 3 (b)
and (c)). When the switch is turned-off, the equivalent circuit is established as shown in
Figure 2 (b), where, by analogy with the other phases, the equations become as follows;
di u
e u  v AO  L u
dt
di v (2)
e v  v BO  Lv
dt
di
e w  v CO  Lw w
dt
From these equations, it is clear that each phase current is decreasing in proportion to eu-vAO
etc. These waveforms are shown for the S1-off period in Figure 3. If the current waveforms
are decreasing, as shown by the dashed lines, the resultant current values could be obtained
in proportion to the input voltage values. However, the terms for attenuation, such as eu-vAO,
are nonlinear. (see Figure 4 showing vAO). Actual waveforms are attenuated by means of the
terms like eu-vAO etc. (Murphy. 1985). If ev, for an example, has a small value, the degree of
attenuation may be small, so that a gently decaying dashed line would be obtained, as
shown. In this example case, however, the attenuation term is ev-vBO, so that the attenuation
degree becomes severe. As a result, the sharply decaying solid line can be obtained, because
of significant attenuation, producing nonlinear waveforms.
S1 on S1 off S1 on
eu
(a)
eu iu
0
0
(b) iv
e iv
ev
eu
(c) 0 eu
ew ev i iw
3. Inut current waveforms at S1 switching.
ee
Fig. 3. Input current waveforms at S1 switching.
vAO 2Vd
3 Vd
3
0
Fig. 4. Conceptual voltage waveform, vAO.

Figure 4 shows conceptual waveform as vAO. When S1 is turned-off, the corresponding diode
conducts. Depending on whether the amplitude of vAO=2Vd/3 or Vd/3, where Vd is the
output voltage, the degree of attenuation at S1 turn-off is varied.
(a) eu 200V
0 0
(b) iuiu 100A

0 0
(c) vL
vL 300V
0 0
Fig. 5. Explanation of distorted input current waveform in conventional method.
Figure 5 shows the operational waveforms for Figure 1 from circuit simulation. From these
figures, the reasons for waveform distortion in the conventional input current can be
explained to a certain extent. From the phase voltage, eu, in Figure 5 (a), the input current
waveform, iu, appears as in Figure 5 (b), using single device switching. It can be found that
the envelope of a six stepped waveform vAO appears and the distortion of iu is generated as
in Figure 5 (b). The term eu-vAO in (2) appears as an envelope of the applied voltage across
the input inductor in Figure 5 (c). From equation vL=Ludiu/dt, it can be seen that the integral
of vL becomes the input current, iu, so that the improvement scheme for input current
waveform can be determined from observing the inductor voltage wave, vL, to a certain
extent.
2.2 Operation Principle of the Proposed Circuit

Figure 6 shows one of the proposed types of, three-phase, single switch converter. In this
paper, we will discuss the boost type converter. In the future, however, it may be possible
that a buck type converter could be realized under adequate discussion. Thus, this paper
title does not restrict the concept to the boost type converter. The circuit configuration
originates from the above mentioned Prasad-Ziogas circuit. The notable feature is that
several electrolytic capacitors are parallel-connected to rectifying diodes. By means of this
configuration, the input voltage circuit is always connected to either dc output bus, so that
continuity and improvement of the input current can be realized. In such a way, a boosted
dc voltage, utilizing the PFC scheme, can be obtained in comparison to the conventional
circuit. The circuit operation can be roughly divided into six periods, where each period is
60 degrees. From the operation waveforms in Figure 7 and the operational periods shown in
Figure 8, the circuit operation can be discussed. To simplify the analysis of the operation, we
will assume a unity power factor of phase, u, where fundamental voltage and current
components are almost synchronized with each other.
Lsw
eu
Lu Cu R0
ew A C0
O ev S1
vd vs
Cx
Fig. 6. Proposed circuit configuration.
Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ Ⅵ
(a) eu t3 t4 t5 200V
0 t0 t1 t2 t6 0
(b) iu 100A
0 0
(c) iv 100A
0 0
(d) iw 100A
0 0
(e) vcu 400V
0 0
(f) vcx 400V
0 0
(g) vAO 400V
0 0
(h) iDu 100A
0 0
(i) icu 100A
0 0
(j) vs 1.0kV
0 0
Fig. 7. Waveforms for proposed circuit.

(a) period Ⅰ(t0<t<t1)

The voltage in phase u is gradually increasing from eu=0. In usual three phase circuit, the
current at small values of supply voltage can not be rising due to a large dc link voltage, so
that the current becomes zero for usual circuits, or suppressed to fairly reduced value, even
in the Prasad-Ziogas circuit. In the proposed circuit, however, the capacitor voltage, vcu, is
gradually discharged from being fully charged at the dc link voltage. (see Figure 7 (e)).
During this discharging period, diode current, iDu, does not flow. In the other phases, v and
w, diodes, Dv and Dw, conduct, although parallel-capacitor currents do not flow. The
capacitor charge and discharge currents, icu and icx, each of which are connected to the
constant dc link voltage, are equal, i.e., |icu|=|icx|=|iu/2|. These results can be derived
from the equation Cu×dvcu/dt=-Cx×dvcx/dt. In Figure 7 (h), (i) and (b), these results can be
seen as iDu+2iCu=iu. The diode current in phase w is decreasing toward zero as ew decreases.
When this voltage polarity is reversed, and the Dw current is commutated to Dz circuit, this
period comes to an end.
(b) period Ⅱ(t1<t<t2)

As the capacitor, Cz, in parallel with Dz, is charged to vcz=vd, this period starts from the
beginning of the discharge current icz. The current, Du, in phase u and the current, Dy, in
phase v continue to flow, supplying the dc bus. At the end of this period, the voltage, ev, is
reversed and iv is greater than zero.
(c) period Ⅲ(t2<t<t3)

The voltage, ev, begins to rise and the commutation from Dy to Dv commences. The capacitor
voltage, vcv , is varied in a similar manner to vcu in period Ⅰ. The current, iv, rises from the
zero point of ev. In phase u, the current, iu, is decreasing toward zero according to the
decrease in eu, when this period comes to an end.
In the subsequent period of negative eu, an analogous operation is repeated such as a
commutation of Du to Cx and Dx etc. A remarkable characteristic of this strategy is that there
is no discontinuity of the input current wave, as compared to the conventional three phase
diode circuit having 120 degree current wave. In the proposed method, on the other hand,
one terminal is always connected to either dc link circuit through a capacitor, achieving a
continuous and improved waveform. In this paper, the boost chopper strategy has been
considered and discussed, such that the stored charge is forced to flow from capacitors, so
that the functions of charge and discharge become more efficient and smoothing of the input
current becomes more effective. As an indication of the improvement of input current
waveform, the vAO waveform is shown in Figure 7 (g). This waveform can be derived from
the conventional six step inverter circuit by an analogous procedure. In the proposed circuit,
however, due to the intermediate capacitors, the vAO waveform becomes smoothed, as
shown in Figure 7 (g). Through such improved waveforms, instead of usual six step wave
vAO as in Figure 4, input current waveform can be improved, as shown in Figure 7 (b).
3. Operational Characteristics
By employing circuit constants in Table 1, various characteristics can be resolved. The
operational waveforms in Figure 7 can be resolved by using these circuit constants. Figure 9
shows the relationship between output power and THD. The characteristics are compared
between the conventional and
P P
P
vcu Du Du Dv vcv
Du icv
icu Dw
icx iu icy
iu
iw N
N eu
eu ev
eu ew v d
ev ew
P ew
ev
iv icw iw
iv
Dy Dy Dz icz vcz Dz
N N
N
（a） period Ⅰ （c） period Ⅲ
（b） period Ⅱ
P P P
Dv Dw
Dv Dw vcw
iv iw
iv
ev N ew
ev ew
eu ew eu ev
P P
eu
iw iu
iu
Dx vcx Dz Dx Dy vcy
Dx N
N N
（d） period Ⅳ （e） period Ⅴ （f） period Ⅵ
Fig. 8. Operating circuit.

proposed methods, where the input inductors are taken as parameter. In the reduced power
region of conventional circuit, the THD is deteriorated. In a region of increased power of the
circuit, the harmonics are relatively suppressed due to the function of the input inductor,
and the THD can be improved. For the proposed circuit, in this manner, the THD can be
entirely suppressed and improved.
LL: Line inductance 0.25 mH

Rs: Line resistance 0.2 Ω
Lin: Input filter inductance
proposed 5 mH
conventional 3 mH
Lsw: Switching inductance 0.2 mH
Ca: Auxiliary capacitance 150 μF
Co: Output capacitance 6000 μF
Ro: Load resistance
proposed 42 Ω
conventional 22 Ω
fsw: Switching frequency 20 kHz
vs: Supply line voltage 200 V
fs: Supply frequency 60 Hz
Table 1. Circuit constants
30
Conventional 2mH
3mH
Proposed 3mH, 200μF
20
THD [%]
4mH, 150μF
5mH, 150μF
10
0
0 4 8 12 16
Output power [kW]
Fig. 9. Relationship between output power and THD.
In a region of more increased power for the conventional method, we might expect that a
more improved THD could be obtained, but the actual result is to the contrary. Because
voltage drop across the input inductor is significant in the increased power region, a more
increased power can not be obtained. Due to an LC resonant operation, where the stored
electric charge in the C is charged and
discharged, a little increased power can easily be obtained, offering one remarkable feature
of this strategy.
1.0
0.8
0.6 Conventional 2mH

3mH
Power
0.4 Proposed 3mH, 200μF

4mH, 150μF
0.2 5mH, 150μF
00 8
4 12 16
Output power [kW]
Fig. 10. Relationship between output power and power factor.
Figure 10 shows the relationship between the output power and power factor using the
same circuit constants from Table 1. In the conventional circuit, as the output power is
increased, the power factor is reduced a little. The reason is that the voltage drop across the
input inductor is fairly significant. In the proposed method, however, though the THD
characteristic is much improved, the power factor characteristic is a little deteriorated. For
this reason, it could be said that this strategy is unsuitable for an application requiring the
avoidance of reduced power factor over a variable wide power range. Rather, it is suitable
for an application requiring constant output power or extended operation term with
constant power.
Figure 11 shows the relationship between auxiliary capacitance, Ca, and the THD. Results at
Ca=0 are represented for Prasad-Ziogas circuit, where THD is 12% of deteriorated value.
Employing the proposed auxiliary capacitance, however, as the value of capacitance is
increased, the THD characteristic is fairly improved, where output power is adjusted so as
to maintain unity power factor. Consequently, as the capacitance is increased, the input
current and the power are also increased, as shown by dotted line in Fig 11. Such increased
current can partly contribute an improvement in the THD.
15 40
Input filter inductance 3mH
Output power P0 [kW]

Convention 5mH
l 30
10
THD [%]
Output power 20
5
10
Proposed
0 0
0 100 200 300 400 500 600
Auxiliary capacitance [μF]
Fig. 11. Relationship between auxiliary capacitance and THD.
4. Development to Buck-converters
This paper discussed about boost type converters, but those type ones can be easily develop
toward any type of converters. In this section, we will discuss about the application for buck
type converter in single phase. Figure 12 shows such proposed circuit for buck converter,
where previously mentioned auxiliary intermediate capacitors are installed. In this circuit
configuration, the intended characteristics could be realized. The basic circuit is constructed
by the conventional buck-converter. The distinctive feature of the discussed circuit is to be
described as follows: Previously mentioned relatively large capacitors such as electrolytic
ones are parallel-connected to diodes in a similar way. By means of those connections, the
input circuit is always connected to either terminal of the dc link circuit, such as positive or
negative one, which makes possible the input current continuity and the improvement of
input current waveform. In general, for buck-type converter when the PFC circuits are
designed both for three-phase and single-phase, it might be difficult to construct the suitable
PFC ones, even with fairly large inductions or transforms. The distinctive feature of the
proposed strategy employs a very simple way, in which some electrolytic capacitors are
merely parallel-connected. By means of this parallel connection of capacitors, non-linearity
of input current waveform becomes linear one, which brings the waveform improvement. In
the usual buck chopper circuit having the constant dc link voltage, the output current does
not flow in a certain period. This feature brings non-sinusoidal waveforms. In a case of the
proposed circuit, the dc link voltage is constructed by the sum of series-connected double
capacitor voltage such as vC1 + vC2. As a result, due to assistance of each capacitor charge
and discharge operation, the input current always cntinues to flow, though the dc link
voltage vR becomes constant , so that each appearance should be in sinusoidally wave.
Ld
＋ S
ie D1 C1 D2＋ C
2
ve vR Co ＋ Ro
＋＋
D3 C3 D4 C4
Fig. 12. Development to buck-converter.
5. Conclusions
An improved circuit strategy has been proposed and discussed, based on an extension of the
Prasad-Ziogas circuit, offering significant improvement in the THD characteristic. The
results have been presented and compared. The proposed circuit uses single switching
device like the conventional one and the characteristics can be improved sufficiently by
using a simple auxiliary capacitor connection. In this way, a three phase PFC circuit can be
realized in a simple manner. Another feature in the proposed circuit is the ability to obtain a
fairly increased power capacity, making the circuit suitable for high capacity converter.
However, the circuit is employs several capacitors in the series current path. As a result, a
somewhat reduced power factor region is observed, particularly in the lower output power
region. Consequently, the circuit is unsuitable for an application requiring a wide variable
power range and a reduced power operation for a long period.
In the future, after further consideration and discussion, a novel buck-type converter using
proposed method might be realized.
6. References
Converter-circuit and Control-strategy Committee on IEEJ. (2000). Current Circumstances
and Trends of PFC Converter-circuit and Control Strategies. Technical Report of
Institute of Electrical Engineers in Japan, No.785
Fujiwara, K. & Nomura, H. (1995). A Power Factor Correction for Single-Phase Diode
Rectifiers without employing PWM Strategy. IPEC-Yokohama '95, pp.1501 -1506
JISC.(2005). Electromagnetic compatibility (EMC) - Part 3-2: Limits. JISC61000-3-20. PFC
Circuit Investigation Committee on IEEJ. (2000).
Murphy, J. M. D. & Turnbull, F. G. (1988). Power Electronic Control of AC Motors.
Pergamon Press, p.112
Prasad, A. R. & Ziogas, P. D. (1991). An Active Power Factor Correction Technique for
Three-Phase Diode Rectifiers, IEEE Trans. Power Electronics, Vol.6, No.1, pp.83-92
Takahashi, I. & Ikeshita, W. (1985). Improvement of Input Current Waveforms of a Single-
Phase Rectifier Circuit. IEEJ Trans. Vol.105-B, No.2, pp.82-90
Takahashi, I. Hori, K.(1997). Improvement of Input Current Waveforms of a Single Phase

Diode Rectifier by Passive Devices. IEEJ Trans. Vol.117-D, No.1, pp.13 - 18
Takeuchi, N. & Matsui, K. (2007). A Discussion on PFC Circuit Using Ladder Type Filter.
Industry Applications Society of the Institute of Electrical Engineers of Japan , Vo.I,
No.98, pp.519-522.
Yamamoto, I. & Matsui, K. (2001). A Power Factor Correction with Two-Input Current
Mode using Voltage Doubler Rectifier. IEEJ Trans. IA, Vol.121-D, No.2, pp.225 -
230
Advanced Modulation Formats
and Multiplexing Techniques for Optical Telecommunication Systems 13
X2
Advanced Modulation Formats and Multiplexing

Techniques for Optical Telecommunication
Systems
Ghafour Amouzad Mahdiraji1 and Ahmad Fauzi Abas2
1UCSI University & 2Universiti Putra Malaysia
Malaysia
1. Introduction
Since ancient times, one of the principal needs of people has been to communicate. This
need created interest in devising communication systems for sending messages from one
place to another. The advent of high performance computer processors brought many
advantages for digital communications over that of analog. These benefits include more
features, easy storage and faster processing. These caused huge amount of information,
which is increasing exponentially every year, to be carried over communication networks.
Various types of communication system appeared over the years. Among the basic
motivations behind each type are to improve the transmission fidelity, increase the data
rate, and increase the transmission distance between stations. All these facilities are
achievable utilizing optical fiber communications. Optical fiber offers several advantages
over the traditional media (e.g., twisted wire pair and coaxial cable). Its decisive advantages
are huge bandwidth and very low attenuation and noise (Arumugam, 2001). The first,
results in higher bit rate, and the second, results in longer transmission distance. These
potentials can be further pushed by utilizing multiplexing techniques and/or advanced
modulation formats.
The invention of wavelength division multiplexing (WDM) (G. E. Keiser, 1999) contributes
great benefit to the optical fiber communication systems especially after the introduction of
Erbium-doped fiber amplifier (EDFA). Using WDM, about forty channels can be
accommodated in the C-band at 100 GHz (0.8 nm) channel spacing. Based on this condition,
up to 1.6 Tb/s transmission capacity has been reported (Zhu et al., 2001). More channels can
be transmitted using ultra-dense WDM technique by considering channel spacing of as close
as 12.5 GHz (Ciaramella, 2002; Sang-Yuep, Sang-Hoon, Sang-Soo, & Jae-Seung, 2004). Using
such channel spacing, up to 2.5 Tb/s transmission is reported (Gyo-Sun et al., 2007) by
multiplexing 256 12.5 Gb/s channels, and transmitted over 2000 km standard single mode
fiber (SSMF). Larger transmission capacity can be achieved by utilizing the S- and L-bands
(Freund et al., 2005; Seo, Chung, & Ahn, 2005). Using triple bands (S + C + L), 10.92 Tb/s
transmission is experimentally reported (Fukuchi et al., 2001) by using 273 WDM channels
and 50 GHz spacing.
Higher transmission capacity can be achieved with higher bit rate per WDM channel. Time
division multiplexing (TDM) is the most commonly used for multiplexing high number of
lower bit rate channels to form a higher bit rate. For example, a 40 Gb/s data stream can be
achieved by multiplexing four 10 Gb/s data using electrical TDM (ETDM) (Dong-Soo, Man
Seop, Yang Jing, & Nirmalathas, 2003; Krummrich et al., 2002; Lach, Bulow, Kaiser, Veith, &
Bouchoule, 1998; Lee, Garthe, Pettitt, & Hadjifotiou, 1997; Miyamoto, Yoneyama, Otsuji,
Yonenaga, & Shimizu, 1999; Yoneyama et al., 1999). Using such system, 3.2 Tb/s (80 × 40
Gb/s) bidirectional WDM/ETDM transmission over 40 km SSMF is experimentally reported
(Scheerer et al., 1999). The feasibility to realize transmission systems, subsystems, and
electronic and optoelectronic components operating at bit rates beyond 40 Gb/s has been
demonstrated in numerous papers (Andre, Kauffmann, Desrousseaux, Godin, &
Konczykowska, 1999; Derksen, Moller, & Schubert, 2007; Elbers, 2002; Jansen et al., 2007;
Kauffmann et al., 2001; Lach & Schuh, 2006; Lach et al., 2007; Schuh et al., 2005). Recently,
up to 107 Gb/s full-ETDM transmission is experimented in laboratory and tested over
installed fiber in the field (Derksen et al., 2007; Jansen et al., 2007; Lach et al., 2007).
WDM channels capacity can be doubled using polarization division multiplexing (PDM)
technique (Hinz, Sandel, Noe, & Wust, 2000; Hinz, Sandel, Wust, & Noe, 2001; Martelli et al.,
2007; Sandel, Wust, Mirvoda, & Noe, 2002; Suzuki, Kubota, Kawanishi, Tanaka, & Fujita,
2001; Yan, Zhang, Belisle, Willner, & Yao, 2007). Combining PDM with WDM system, 10.2
Tb/s (256 × 42.7 Gb/s) transmission in C + L bands is experimentally demonstrated, which
offers 1.28 b/s/Hz SE (Bigo et al., 2001).
Further improvement in WDM network capacity can be realized by using advanced
modulation formats. Amongst different types of available modulation formats, differential
quaternary phase-shift keying (DQPSK) transmission is currently under serious
consideration for high-speed long-haul optical transmission systems due to its reduced
optical bandwidth and high tolerance to chromatic dispersion (CD) relative to traditional
binary systems (A. F. Abas, 2006; Cho, Grigoryan, Godin, Salamon, & Achiam, 2003;
Christen, Nuccio, Xiaoxia, & Willner, 2007; A. H. Gnauck, Winzer, Dorrer, &
Chandrasekhar, 2006; Morita & Yoshikane, 2005; Schubert et al., 2006; Weber et al., 2005;
Weber, Ferber et al., 2006; Yoshikane & Morita, 2004). Using the mentioned technique,
WDM channel capacity can be doubled with requiring transceivers operating at the same
baud rate. This improves the spectral efficiency (SE) of WDM system. Using WDM and
carrier-suppressed-return-to-zero (CS-RZ) DQPSK format, 4 Tb/s (50 × 85.4 Gb/s) with 70
GHz spacing has been experimentally tested (Yoshikane & Morita, 2004). Using that
configuration, 1.14 b/s/Hz spectral efficiency (SE) was achieved. Elsewhere, using RZ-
DQPSK, 5.12 Tb/s (64 × 85.4 Gb/s) with 50 GHz channel interval and 1.6 b/s/Hz SE was
experimentally demonstrated (Morita & Yoshikane, 2005).
Combining DQPSK with PDM, quadruples WDM channel capacity (Ahmad Fauzi Abas,
Hidayat, Sandel, Milivojevic, & Noe, 2007; Charlet et al., 2008; Pardo et al., 2008a; Renaudier
et al., 2008; Savory, Gavioli, Killey, & Bayvel, 2007; Wree et al., 2003). With this
configuration, Gnauck et al. (Alan H. Gnauck et al., 2007; A. H. Gnauck et al., 2008),
demonstrated a record of 25.6 Tb/s transmission over 240 km using 160 WDM channels
with 50 GHz grid in the C + L bands. In their experiment, they employed 85.4 Gb/s RZ-
DQPSK modulation and polarization multiplexing to attain 160 Gb/s in each WDM channel,
resulting in a SE of 3.2 b/s/Hz in each band (Alan H. Gnauck et al., 2007; A. H. Gnauck et
al., 2008). This was the record in optical communication systems in 2008.
Recently, Zhou et al. (Zhou et al., 22-26 March 2009) has reported 320 × 114 Gb/s PDM-RZ-8
quadrature amplitude modulation (QAM)dense WDM transmission with channel spacing
of 25 GHz over 580 km ultra-low-loss SMF-28. This is a record capacity of 32 Tb/s till 2009.
Duty cycle division multiplexing (DCDM) is another newly reported multiplexing technique
that can support multiple users per WDM channel (Mahdiraji et al., (In Press)). In this
technique, the multiplexed signals have a rising edge transition at the beginning of the
multiplexed symbol. This unique property has never been reported in other multiplexing
techniques and modulation formats. Considering that property, the technique allows
aggregate bit rate to be recovered at symbol/baud rate. Based on our knowledge, this is the
latest multiplexing technique reported to date.
In the following sections, details on principles, operation and implementation of various
modulation format and multiplexing techniques are presented.
2. Modulation Formats
Modulation is a process to form the baseband signal using high frequency carrier signal to
become more suitable for transmission over long communication link. Advanced
modulation formats improves the channel utilization and capacity. There are various types
of multiplexing techniques and modulation formats commonly used in optical fiber
communication system, which will be further discussed in the following Subsections.
2.1 Amplitude Shift Keying

In optical fiber communication systems, the baseband signals are modulated onto high
frequency optical carriers. Various types of modulation can be used for that purpose.
Amplitude modulation (AM) or amplitude-shift keying (ASK) or on-off keying (OOK) is the
simplest and commonly used technique in optical fiber communication systems, where AM
is referred to analog signals, and ASK and OOK referred to digital signals. In this technique,
the baseband signal is multiplied by a carrier frequency, thus (assuming binary signaling),
the binary 0 is transmitted with 0 W and binary 1 with A W. At the receiver, the
demodulation can be easily performed using a photodetector, which converts the optical
signal to the electrical signal, resulting in the original transmitted pattern. Figure 1 shows
example of a ASK modulation format.
Fig. 1. Example of ASK modulation foramt, (a) binary signal, and (b) ASK modulated signal
In advanced communication systems, instead of transmitting single bit per symbol, using
two level binary signals, more than one bit per symbol can be achieved, which it results in
higher transmission capacity. This technique is called multilevel signaling. The number of
signal level M, follows the rule of M  2 b where the b is the number of bits per symbol, thus
called M-ary signaling. In ASK, the value of M = 4 (4-ary ASK) is mostly used to double the
transmission capacity while maintaining the spectral width (Avlonitis, Yeatman, Jones, &
Hadjifotiou, 2006; Cimini & Foschini, 1993; Muoi & Hullett, 1975; Walklin & Conradi, 1999).
The 8-ary ASK is also studied over fiber optic communication for tripling the transmission
capacity (Walklin & Conradi, 1999). The improvement in channel capacity was obtained at
the cost of power penalty in the OSNR and system receiver sensitivity. For example, receiver
sensitivity of 4-ary ASK coded with NRZ and RZ signaling experienced around 3.8 dB and
6.6 dB penalty in comparison to binary NRZ and RZ respectively (Avlonitis et al., 2006). This
is due to the fragmentation of the main eye to the several smaller eyes for the 4-ary ASK.
2.2 Phase Shift keying

In phase modulation, binary data are modulated onto the optical carrier referring to the
phase difference between binary 0 and 1. This technique is called phase-shift keying (PSK)
or BPSK for binary PSK. Example of BPSK modulation is shown in Figure 2. In this example,
binary 1 is signed as sin(ωt) and binary 0 is signed as sin(ωt + π) or –sin(ωt).
Fig. 2. Example of BPSK modulation format, (a) binary signal, and (b) BPSK modulated
signal
In the early days, PSK did not receive much interest due to its demodulator’s complexity.
Instead, differential PSK (DPSK) had received more interests (Ho, 2005). In DPSK, the data
are first encoded differentially as the differential encoder shown in Figure 3(a). The encoded
data are then modulated onto optical carrier using a phase modulator (PM) or Mach-
Zehnder modulator (MZM), which externally changes the optical phase from its original
phase to a relatively π phase shift. In response to the driving baseband signal (Ho, 2005),
MZM is preferable to PM due to better chromatic dispersion tolerance. Figure 4 shows
example of DPSK, which Figure 4(a) shows the binary signal, and 4(b) is the DPSK
modulated signal.
Fig. 3. DPSK transceiver, (a) DPSK transmitter, (b) DPSK balanced receiver, and (c) DQPSK
transmitter
Fig. 4. Example of DPSK modulation format, (a) binary signal, and (b) DPSK modulated
signal
At the receiver, since DPSK can not directly be demodulated, a delay interferometer (DI) is
inserted in the optical path at the receiver to convert the differential phase modulation into
intensity modulation. As shown in Figure 3(b), a DI splits the received signals into two
paths, which experience one-bit delay to let two neighboring bits interfere at the DI output.
At port a (the destructive port), the two optical fields interfere destructively whenever there
is no phase change, and constructively whenever there is a phase change between
subsequent bits, thus converting phase modulation into intensity modulation (Winzer &
Essiambre, 2006).
Maintaining good interference is the most critical aspect in the design of DPSK receivers
(Ho, 2005; Winzer & Essiambre, 2006; Winzer & Hoon, 2003). Due to energy conservation
within the DI, the second DI output port b (the constructive port) yields the logically
inverted data pattern. In principle, one of the two DI output ports is sufficient to detect the
DPSK signal (single-ended detection). However, the 3-dB sensitivity advantage of DPSK is
only seen for balanced detections (Ho, 2005; Winzer & Essiambre, 2006). A balanced
detection (as shown in Figure 3(b) made with two photodetectors) considers the difference
between ports a and b signal providing a larger signal than that of a single-branch receiver
(Ho, 2005).
In advanced communication systems, similar to the M-ary ASK, M-ary DPSK are used
instead of binary DPSK. The most reports are on M = 4, which called 4-ary DPSK or
differential quadrature PSK (DQPSK). DQPSK is the only true multilevel modulation format
(more than one bit per symbol) that has received appreciable attention in optical
communications so far (A. F. Abas, 2006; Cho et al., 2003; Christen et al., 2007; A. H. Gnauck
et al., 2006; Kawanishi et al., 2007; Morita & Yoshikane, 2005; Nasu et al., 2008; Schubert et
al., 2006; van den Borne et al., 2008; Weber et al., 2005; Weber, Ferber et al., 2006; Yoshikane
& Morita, 2004). It experiences four phase shifts, 0, +π/2, –π/2, and π (sin(ωt), sin(ωt + π/2),
sin(ωt – π/2), and sin(ωt + π)), for data modulation, and operates at a symbol rate of half the
aggregate bit rate. Figure 3(c) shows a schematic of DQPSK based on (Winzer & Essiambre,
2006), consisting of a continuously operating laser source, a splitter to divide the light into
two paths of equal intensity, two nested MZMs operated as PMs, an optical π/2 phase
shifter in one of the paths, and a combiner to produce a single-output signal. Figure 5 shows
example of a QPSK/DQPSK modulated signal. In this example, the binaries 00, 01, 10, and
11 are signed with 0°, 90°, 270°, and 180° respectively. The QPSK and DQPSK modulated
signal are the same. The different is referred to the encoder before the modulator. If the
encoder is a differential encoder, then the modulated signal is DQPSK, otherwise it is QPSK.
Fig. 5. Example of QPSK/DQPSK modulated signal
At the receiver, DQPSK signal first splits into two equal parts, and detected by two balanced
receivers of the form depicted in Figure 3(b). The two balanced receivers are used in parallel
to simultaneously demodulate the two binary data streams contained in the DQPSK signal.
Note that, the delay produced by DI has to be equal to the symbol duration for DQPSK
demodulation, which is twice the bit duration. In general, feedback-controlled DI tuning
within the receiver is needed for both DPSK and DQPSK (Winzer & Essiambre, 2006). In
DQPSK, the reduction of bandwidth is beneficial for achieving high SE in WDM systems, as
well as for increasing tolerance to CD.
The higher bandwidth reduction or higher channel capacity can be achieved by increasing
the value of M. This is achieved by combination of the amplitude and phase modulation
which is called M-ary-quadrature amplitude modulation (QAM). In fact, the QAM (or 4-ary
QAM) is produced by two PSK/DPSK signals, thus, it is the same as QPSK/DQPSK. Higher
value of M, for example M = 16, which means 4 bits per symbol, is produced by utilizing 4
different amplitude levels combined with 8 different phase levels. In theory, this will lead to
32 states, allowing us to encode 5 bits per symbols (25 = 32). However, only 16 of these states
are used to encode 4 bits (log2(16) = 4) per symbol (Lathi, 1998; Zahedi, 2002). Using QAM
technique, 20-Msymbol/s using 128-QAM with coherent transmission over 500 km
(Nakazawa, Yoshida, Kasai, & Hongou, 2006), and 1-Gsymbol/s using 64-QAM coherent
transmission over 150 km optical fiber has been reported (Jumpei, Kasai, Yoshida, &
Nakazawa, 2007; Yoshida, Goto, Kasai, & Nakazawa, 2008). In M-ary signaling, since there
is b number of bit information per symbol, therefore, one symbol error produced b number
of errors. This is called error propagation, which is more serious at higher value of b and M.
2.3 Duobinary
Optical duobinary (DB) has attracted great attention in recent years. The two main
advantages attributed to this modulation format are increased tolerance to the effects of CD
and improved SE (Ibrahim, Bhandare, & Noe, 2006; Lender, 1964; Said, Sitch, & Elmasry,
2005; Wei et al., 2002; Yonenaga & Kuwano, 1997; Yonenaga, Kuwano, Norimatsu, &
Shibata, 1995). The fundamental idea of DB modulation (electrical or optical) that were first
described by Lender (Lender, 1964) is to deliberately introduce intersymbol interference (ISI)
by overlapping data from adjacent bits. This correlation between successive bits in a binary
signal leads the signal spectrum to be more concentrated around the optical carrier (Said et
al., 2005). This is accomplished by adding a data sequence to a 1-bit delayed version of itself,
which can be obtained by passing the binary signal through the delay-and-add filter as
shown in Figure 6(a) (Said et al., 2005). For example, if the (input) data sequence is x(nT) =
(0, 0, 1, 0, 1, 0, 0, 1, 1, 0), we would instead transmit the (output) data sequence y(nT) = (0, 0,
1, 0, 1, 0, 0, 1, 1, 0) + (*, 0, 0, 1, 0, 1, 0, 0, 1, 1) = (0, 0, 1, 1, 1, 1, 0, 1, 2, 1). Here the sign *
denotes the initial value (z(nT)) of the input sequence, which is assumed to be zero. Note
that while the input sequence is binary and consists of 0s and 1s, the output sequence is a
ternary sequence consisting of 0s, 1s, and 2s. Mathematically, DB results in y(nT) = x(nT) +
x(nT – T), where T is the bit period and n in the number of bit sequences (in above example
n is 10). At the receiver, the input sequence x(nT) can be recovered from the received y(nT)
based on z(nT) = y(nT) – z(nT – T) (detail refer to (Ramaswami & Sivarajan, 2002)).
There is one problem with this scheme, however; a single transmission error will cause all
further bits to be in error, until another transmission error occurs to correct the first one!
This phenomenon is known as error propagation (Ramaswami & Sivarajan, 2002). The
solution to the error propagation problem is to encode the actual data to be transmitted in a
differential form. For example, the x(nT) is encoded into d(nT) = (0, 0, 1, 1, 1, 1, 0, 1, 0, 1). To
see how differential encoding solves the problem, observe that if sequences of consecutive
bits are all in error, their differences will still be correct. However, such an approach would
eliminate the bandwidth advantage of DB signaling (Ramaswami & Sivarajan, 2002). The
bandwidth advantage of DB signaling can only be exploited by using a ternary signaling
scheme.
Fig. 6. Generating DB signal, (a) digital filter for electrical DB signal, (b) dual-derive MZM
and (c) MZM bias and derive conditions for optical DB signal
The primary version of DB, which used three levels signal, increases the sensitivity penalty
(Said et al., 2005). To avoid the penalty, the three level DB signals need to be encoded in
both the amplitude and the phase of the optical carrier (Yonenaga & Kuwano, 1997;
Yonenaga et al., 1995). Such a scheme is called optical AM-PSK (Ramaswami & Sivarajan,
2002) and most studies of optical DB signaling today are based on AM-PSK. If the data is
differentially encoded before the DB filter, the carrier phase information becomes
redundant, and hence, the received data can be decoded using a conventional binary direct-
detection receiver (Yonenaga & Kuwano, 1997; Yonenaga et al., 1995). This DB signal can be
generated by applying a baseband, three-level electrical DB signal to a dual-drive MZM
(Figure 6 (c)) that is biased at maximum extinction ratio, as shown in Figure 6(b) (Yonenaga
& Kuwano, 1997; Yonenaga et al., 1995). Conceptually, the carrier is a continuous wave
signal, a sinusoid denoted by a cos(ωt). The three levels of the ternary signal correspond to
–a cos(ωt) = a cos(ωt + π), 0 = 0 cos(ωt), and cos(ωt), which is denoted by –1, 0, and +1,
respectively. These are the three signal levels corresponding to 0, 1, and 2, respectively, in
y(nT). The AM-PSK signal retains the bandwidth advantage of DB signaling.
3. Multiplexing Techniques
Multipleixng is an essential part in a communcation system where multiple users transmit
data simultaneously through a single link, whether the link is a coaxial cable, a fiber, radio
or satellite. Multiplexing is widely employed in communication systems due to its capability
to increase the channel utilization or the transmission capacity and decrease system costs.
Figure 7 depicts the multipleixng function in its simplest form. There are n inputs to a
multiplexer. The multiplexer multiplex or combine these inputs in a way so that they are
separable. The demultiplexer performs opposite process as multiplexer to separate the

multiplexed data, and delivers them to the appropriate output lines. If each input to the
multiplexer carrying k bps digital data, the total data rate or the aggrigate rate of the link is
nk. There are various types of multiplexing techniques commonly used in optical fiber
communication system, which working principles are discussed in the following
Subsections.
Fig. 7. Multiplexing
3.1 Time Division Multiplexing

Several low bit rate signals can be multiplexed, or combined to form a high bit rate signal by
sharing the time. Because the medium is time shared by various incoming signals, this
technique is generally called time division multiplexing (TDM). For those implemented in
electrical domain, they are called electrical TDM (ETDM). Example of TDM system for
multiplexing two channels is shown in Figure 8. In TDM systems, if n number of users with
the same pulse width of T s (seconds) is multiplexed, the pulse width of the multiplexed
signals is T/n. In TDM, the multiplexer and demultiplexer needs to operate at frequency
equal to the total aggregate bitrate, which is n times faster than the bit rate of a single user.
1 0 0 T T/2 1 0 0
ch1 ch1
1 1 0 0 10
T T
1 0 1 1 0 1
ch2 ch2
Fig. 8. Example of a TDM system for multipleing two channels
The multiplexer typically interleaves the lower speed streams to obtain the higher speed
stream. The interleaving can be performed on a bit-by-bit (Figure 8) or packet-by-packet
basis. Framing is required for both cases because at the receiving terminal, the incoming
digital streams must be divided and distributed to the appropriate output channels. For this
purpose, the receiving terminal must be able to identify the timing of each bit correctly. This
requires the receiving system to uniquely synchronize in time with the beginning of each
frame, with each slot in a frame, and each bit within a slot. This is accomplished by adding
framing and synchronization bits to the data bits. These bits are part of the so-called
overhead bits.
Optical time division multiplexing (OTDM) has a similar concept to electrical TDM, only
that it is implemented in optical domain. Figure 9 illustrates the basic concept of point-to-
point transmission system using bit-interleaved OTDM. In this system, access nodes share
different channels that operate at a fraction of the media rate. For example, the channel rates
could vary from 100 Mb/s to 10 Gb/s, whereas the time-multiplexed media rate is around
100 Gb/s. In Figure 9, a laser source produces a regular stream of very narrow RZ optical
pulses at a repetition rate R. This rate typically ranges from 2.5 Gb/s to 10 Gb/s, which
corresponds to the bit rate of the electronic data tributaries feeding the system. An optical
splitter divides the pulse train into n separate streams. In Figure 9, the pulse stream is 10
Gb/s and n = 4. Each of these channels is then individually modulated by an electrical
tributary data source at a bit rate R. The modulated outputs are delayed individually by
different fractions of the clock period, and interleaved through an optical combiner to
produce an aggregate bit rate of nR. At the receiving end, the aggregate pulse stream is
demultiplexed into the original n independent data channels for further signal processing.
In this technique, a clock-recovery mechanism operating at the base bit rate R is required at
the receiver to drive and synchronize the demultiplexer (G. Keiser, 2000).
OTDM requires very narrow RZ pulses to be able to interleave data of different users within
a bit interval. These narrow pulses require higher spectral width. In addition, this system
becomes vulnerable to CD and polarization mode dispersion (PMD) as well as creating the
need for a higher optical signal-to-noise ratio (OSNR) in the wavelength channels due to the
very short pulses. A higher OSNR is obtained by employing a higher signal power and will
make the system more sensitive to fiber nonlinearity (Weber, Ludwig et al., 2006).
Fig. 9. Example of an ultrafast point-to-point transmission system using OTDM technique

(G. Keiser, 2000)
3.2 Wavelength Division Multiplexing

In wavelength-division multiplexing (WDM) systems, different independent users transmit
data over a single fiber using different wavelengths (G. E. Keiser, 1999; Palais, 2005).
Conceptually, WDM scheme, which is illustrated in Figure 10, is similar to frequency
division multiplexing (FDM) used in microwave radio and satellite systems. At the
transmitter side, n independent users’ data are modulated onto n high frequency carriers,
each with a unique wavelength (λ). These wavelengths can be spaced based on ITU-T
standards. A wavelength multiplexer combines these optical signals and couples them into a
single fiber. At the receiving end, a demultiplexer is required to separate the optical signals
into appropriate channels. This is done with n optical filters, whereby their cut-off frequency
is set based on the transmitted light source frequency. The total capacity of a WDM link
depends on how close the channels can be spaced in the available transmission window. In
late 1980s, with the advent of tunable lasers that have extremely narrow linewidth, one then
can have very closely spaced signal bands. This is the basis of dense WDM (DWDM) (G.
Keiser, 2000; G. E. Keiser, 1999). Figure 10 shows a typical WDM network containing
various types of optical filter such as post-amplifier or booster, in-line amplifier and
preamplifier.
Fig. 10. Implementation of a typical WDM network
The major disadvantage of WDM is the low channel utilization and spectral efficiency
because one wavelength is required per user. Therefore, for multiplexing n users, n
wavelengths or light sources with n filters are required, which increase the cost of the
system. The goal of all other multiplexing techniques and modulation formats are to
increase channel utilization and/or channel capacity of the WDM systems.
3.3 Orthogonal Frequency Division Multiplexing

Orthogonal frequency division multiplexing (OFDM) is a special form of a multi-carrier
modulation (MCM) or subcarrier multiplexing (SCM). In MCM, information of different
users is modulated with different waveforms, which are called subcarriers. The channel
spacing between subcarriers has to be multiple of symbol rate, which reduces the spectral
efficiency (Shieh, Bao, & Tang, 2008). A novel approach which overlaps between subcarriers
by reducing the channel spacing employing orthogonal signal set is called OFDM. A
fundamental challenge of OFDM is on the number of subcarriers, where a large number of
them are needed so that other channel treats sub-channels as a flat channel. This leads to an
extremely complex architecture involving many oscillators and filters at both transmitting
and receiving ends (Shieh, Bao et al., 2008). A family of OFDM was first proposed by
Weinsten and Ebert (Weinsten & Ebert, 1971), in which OFDM modulation/ demodulation
was implemented by using inverse discrete Fourier transform (IDFT)/discrete Fourier
transform (DFT) (Weinsten & Ebert, 1971). This made OFDM attractive to be investigated for
applications in the optical domain because of its resilience to the channel dispersion (Shieh
& Athaudage, 2006; Shieh, Bao et al., 2008). The most critical assumption for OFDM is the
linearity in modulation, transmission, and demodulation. Consequently, a linear
transformation is the key goal for the OFDM implementation. This is realized in coherent
optical OFDM (CO-OFDM) with challenges in designing a linear modulator (RF-to-optical
up-converter) and demodulator (optical-to-RF down-converter) (Shieh, Bao et al., 2008;
Shieh, Yi, Ma, & Yang, 2008). A generic CO-OFDM system can be divided into five
functional blocks including (Shieh, Bao et al., 2008; Shieh, Yi et al., 2008) (i) the RF OFDM
transmitter, (ii) the RF-to-optical (RTO) up-converter, (iii) the optical channel, (iv) the
optical-to-RF (OTR) down-converter, and (v) the RF OFDM receiver as shown in Figure 11.
In the RF OFDM transmitter, the input digital data is converted from serial to parallel into a
block of bits consisting of information symbols. This information symbol will be mapped
into a two-dimensional complex signal. The subscripts of the mapped complex information
symbol correspond to the sequence of the subcarriers and OFDM blocks. The time-domain
OFDM signal is obtained through IDFT and a guard interval is inserted to avoid the channel
dispersion. The digital signal is then converted into analog form through a digital-to-analog
converter (DAC) and filtered with a low-pass filter (LPF) to remove the alias sideband
signal. The subsequent RTO up-converter transforms the baseband signal into the optical
domain using an optical in-phase/quadrature (I/Q) modulator comprising a pair of Mach-
Zehnder modulators (MZMs) with a 90º phase offset. The baseband OFDM signal is directly
up-converted to the optical domain and propagates inside the optical medium. At the
receiver, the optical OFDM signal is then fed into the OTR down-converter where it is
converted to a RF OFDM signal. In the RF OFDM receiver, the down-converted signal is first
sampled with an analog-to-digital converter (ADC). Then the signal needs to go through
sophisticated three-level synchronization before the symbol decision can be made. The three
levels of synchronization are (i) DFT window synchronization where the OFDM symbol is
properly delineated to avoid intersymbol interference; (ii) frequency synchronization,
namely, frequency offset needs to be estimated, compensated, and preferably, adjusted to a
small value at the start; (iii) the subcarrier recovery, where each subcarrier channel is
estimated and recovered (Shieh, Yi et al., 2008). Assuming successful completion of DFT
window synchronization and frequency synchronization, the sampled value of RF OFDM
signal passed through the DFT. The third synchronization of the subcarrier recovery
involves estimation of the OFDM symbol phase (OSP), and the channel transfer function.
Once they are known, an estimated value, which is calculated by the zero-forcing method is
used for symbol decision or to recover the transmitted value, which is subsequently mapped
back to the original transmitted digital bits (Shieh, Bao et al., 2008; Shieh, Yi et al., 2008).
Fig. 11. Conceptual diagram for a generic CO-OFDM system with direct up-down
conversion architecture (Shieh, Yi et al., 2008)
CO-OFDM has advantages in mitigating CD effects (Shieh & Athaudage, 2006; Shieh, Yi et
al., 2008), as it transmits a high data rate divided into several low subcarrier channels
resulting in longer signal pulse width. Also, the spectra of OFDM subcarriers are partially
overlapped, resulting in high optical spectral efficiency. On the other hand, CO-OFDM
requires very accurate synchronizations (Shieh, Bao et al., 2008; Shieh, Yi et al., 2008), very
sensitive to nonlinear effects (Shieh, Bao et al., 2008), very complex and costly.
3.4 Polarization Division Multiplexing

Polarization division multiplexing (PDM) is a method for doubling the system capacity or
spectral efficiency, in which two independently modulated data channels with the same
wavelength, but orthogonal polarization states are simultaneously transmitted in a single
fiber (Hayee, Cardakli, Sahin, & Willner, 2001; Martelli et al., 2008; Nelson & Kogelnik, 2000;
Nelson, Nielsen, & Kogelnik, 2001; Yao, Yan, Zhang, Willner, & Jiang, 2007). At the receiver
end, the two polarization channels are separated and detected independently. Figure 12
shows a simple sketch of a polarization multiplexed system. As shown in this figure, the
multiplexer requires a polarization beam combiner (PBC) to combine two channels with
orthogonal polarizations. State of polarization is controlled accurately with very high speed
optical polarization controller (PC). At the receiver, the polarization demultiplexer operates
opposite of the multiplexer, in which a dynamic PC controls the polarization state before the
polarization beam splitter (PBS) separates the two data streams.
The main advantage of PDM is that, it can be applied on existing fiber system without
having to change any part of transmission hardware or software (Yao et al., 2007). It can also
be used together with modulation format like DQPSK (Alan H. Gnauck et al., 2007; A. H.
Gnauck et al., 2008) or QPSK (Charlet et al., 2008; Pardo et al., 2008a, 2008b) to quadruple
system capacity and increase SE. Even though polarization multiplexing is straightforward,
separating the two channels with acceptable crosstalk at the receiving end is not trivial
because the polarization state of the multiplexed channels changes rapidly with time.
Therefore, coherent crosstalk due to misaligned signal in reference to the input state of
polarization in the polarizers or polarization beam splitters arises. In addition to that, PMD
is another impairment for PDM (Nelson & Kogelnik, 2000; Nelson et al., 2001). Different
techniques have been proposed to mitigate this impairment such as monitoring of clock tone
or pilot tones (Chraplyvy et al., 1996; Hill, Olshansky, & Burns, 1992), multi-level electronic
detection (Han & Li, 2006; Hayee et al., 2001), cross-correlation detection of the two
demultiplexed channels (Noe, Hinz, Sandel, & Wust, 2001), and coherent detection (Charlet
et al., 2008; Jansen, Morita, Schenk, & Tanaka, 2008; Pardo et al., 2008a), but they add to the
system complexity and cost (Yao et al., 2007).
Fig. 12. Schematic of a simple polarization division multiplexing system (Nelson &
Kogelnik, 2000; Yao et al., 2007)
3.5 Duty-Cycle Division Multiplexing

Duty cycle division multiplexing (DCDM) is a new multiplexing technique introduced
recently by (Abdullah, Abdalla, F.Abas, & Mahdiraji, 2007). In this technique, different users
sign with different RZ duty cycles and then combine together synchronously to form a
multilevel step shape signal. The multiplexing process can be Performed either in electrical
domain (E-DCDM) or optical domain (O-DCDM). Figure 13(a) shows example of an E-
DCDM system for multiplexing three users. Data of the User 1 to 3 (U1, U2, and U3), each
with let say 10 Gb/s pulse at pseudo random binary signal (PRBS) 210–1, are curved with
three RZ modulators (RZ1, RZ2, and RZ3), which produces different duty cycles (DCs). The
three RZ modulators operate synchronously based on a central clock. The ith RZ modulator
has a DC of Ti = (i × Ts)/(n + 1), where Ts represents the symbol duration and n is the
number of user. Thus, data of U1 is curved with RZ1 which is set at 25% DC (U25); data of
U2 is curved with RZ2 set at 50% DC (U50); and finally, data of U3 is curved at 75% DC
(U75). The signals with different DCs are then multiplexed synchronously using an electrical
adder and then modulated with a laser diode (LD) signal, using an intensity modulator
(IM). Figure 13(b) shows the eye diagram of the modulated signal obtained from the
receiver.
Fig. 13. (a) Schematic of E-DCDM system, (b) eye diagram, and (c) demultiplexer
The main advantage of DCDM is the inherent self-symbol synchronized system. As

highlighted in Figure 13(a), there is one and only one rising edge transition in each
multiplexed symbol (except the case that all user send bit 0), which located at the beginning
of the symbol. Comparing these properties with RZ-TDM, they can only support the bit
synchronization and required external symbol synchronization scheme. For comparison
purpose, three-user RZ-TDM signal is shown in Figure 13(a). Another unique advantage of
DCDM is the impulse transitions in the multiplexed signal spectrum. Figure 14(a) shows
modulated spectra of 3 × 10 Gb/s DCDM, where the modulation is performed using a LD
operated at 1550 nm. It can be seen that DCDM has multiple impulse transitions in its
spectra. In general, the number of impulses are equal to the number of multiplexing users
with spacing equal to the single user bit rate or the symbol rate, which in Figure 14(a) the
transitions repeated every 10 GHz. Comparing this against RZ-TDM shows that it has only
one impulse transition , which is located at frequency equal to the channel aggregate bit
rate, which it is 30 GHz away from the carrier frequency (Figure 14(b)). DCDM provides
smaller spectra width in comparison to RZ-TDM.
0
-10
Power (dBm)
-20
-30
-40
-50
-60
1549.52 1549.76 1550 1550.24 1550.48
0
-10
Power (dBm)
-20
-30
-40
-50
-60
1549.52 1549.76 1550 1550.24 1550.48
Wavelength (nm)
Fig. 14. Modulation spectra of 3 × 10 Gb/s DCDM and 30 Gb/s RZ-TDM
DCDM demultiplexer operates in electrical domain. Considering Figure 13(a), at the receiver
side, the optical signal is first detected by a P-i-N photodiode (PD) and passed through a
low-pass filter (LPF) followed by the demultiplexer. In the demultiplexer (Figure 13(b)), a
clock recovery circuit (CRC) and edge detection circuit (EDC) is used to recover the clock
and detect the beginning of each multiplexed symbol. A 10 GHz clock is recovered referring
to the impulse transition available in the signal spectra (Figure 14(a)). On the other hand,
considering the 10 GHz recovered clock, the beginning of ach multiplexing symbol can be
detected using the EDC. Three sampling circuits are synchronized with the recovered clock
or the detected edges. By putting appropriate delay lines for each sampler as shown in
Figure 13(b), the first, second, and third sampler (S1, S2, S3) take samples at Ts/8, 3Ts/8, and
5Ts/8 s per symbol respectively. The frequency of all samplers is equal to the symbol rate
(10 GHz). Outputs of the samplers are fed into the decision and regeneration unit. In this
unit, the sampled values are compared against three threshold values thr1, thr2, and thr3, and
the decision is performed based on the operations shown in Table 1. Regarding to the rules
in that table, for U25, the decision is made based on the information taken from the two
consecutive sampling points, S1 and S2. If amplitudes of those two adjacent sampling points
are equal, bit 0 is regenerated (rules 1 to 3 from Table 1 and cases 1, 3, 5, and 7 from Figure
13(a)). On the other hand, when the amplitude at S1 is one level greater than S2, bit 1 is
regenerated (rules 4 to 6 from Table 1 and cases 2, 4, 6, and 8 from Figure 13(a)). The same
method is used for U50, which utilizes information extracted from S2 and S3. Finally, U75 is
recovered from only S3 by comparing amplitude of S3 against thr1.
Rules for U1 Cases

1 if (S1 < thr1) & (S2 < thr1), then U1 = 0 1
2 if (thr1 ≤ S1 < thr2) & (thr1 ≤ S2 < thr2), then U1 = 0 3, 5
3 if (thr2 ≤ S1 < thr3) & (S2 ≥ thr2), then U1 = 0 7
4 if (thr1 ≤ S1 < thr2) & (S2 < thr1), then U1 = 1 2
5 if (thr2 ≤ S1 < thr3) & (thr1 ≤ S2 < thr2), then U1 = 1 4, 6
6 if (S1 ≥ thr3) & (S2 ≥ thr2), then U1 = 1 8
Rules for U2 Cases
1 if (S2 < thr1) & (S3 < thr1), then U2 = 0 1, 2
2 if (thr1 ≤ S2 < thr2) & (S3 ≥ thr1), then U2 = 0 5, 6
3 if (thr1 ≤ S2 < thr2) & (S3 < thr1), then U2 = 1 3, 4
4 if (S2 ≥ thr2) & (S3 ≥ thr1), then U2 = 1 7, 8
Rules for U3 Cases
1 if (S3 < thr1), then U3 = 0 1, 2, 3, 4
2 if (S3 ≥ thr1), then U3 = 1 5, 6, 7, 8
Table 1. Data recovery rules for three-DCDM users
The main disadvantage of DCDM system is that it required high OSNR in comparison to the
binary signalling such as RZ or NRZ. This is due to the fragmentation of the main eye to
several smaller eyes (Figure 13(b)) similar to the multilevel amplitude signalling.
Performance of DCDM is improved using optical multiplexer by adding the cost in
multiplexer, which required one modulator for each multiplexing user. In terms of
transmitter and receiver complexity, DCDM has simple transmitter and receiver. At the
same time, DCDM allows high speed aggregate bit rate to be recovered at symbol or baud
rate, which made DCDM receiver very economic. Furthermore, this technique allow more
users to be allocated in a WDM channel, which contributes towards improvement in SE.
4. Comparison between Multiplexing Techniques and Modulation Formats

Table 2 shows a comparison between different modulation formats and multiplexing
techniques at 40 Gb/s. The comparison is made based on transmitter (Tx) and receiver (Rx)
complexity, optical signal-to-noise ratio (OSNR), chromatic dispersion (CD), null-to-null
modulated bandwidth (MBW) and clock recovery frequency (CRF).
Multiplexing techniques improve transmission capacity of optical networks. TDM with a
simple transmitter and receiver can improve channel utilization by expending spectral
bands or bandwidth. This technique is limited by the electronics technology (for example,
based on today technology, the maximum aggregated bit rate per single channel reported is
114 Gb/s). PDM can double the channel capacity with the same spectral band as required by
TDM systems. However, it suffers from channel crosstalk due to PMD and requiring very
high speed polarization controllers, which increase the system complexity and the cost.
OFDM improves channels capacity by allowing hundreds of low speed channels to form a
single channel with very complex transmitter and receiver and requiring 3 stages of precise
synchronizations. DCDM with simple transceivers can support multiple channels per WDM
channel with requiring smaller spectral bandwidth in comparison to the RZ-TDM. It also
facilitates symbol synchronization and allows high speed transmission to be recovered at
low speed clock. However, it suffers from the high OSNR requirement.
40Gb/s
CD
Modulation & Tx Rx OSNR MBW CRF
(ps/
multiplexing complexity complexity (dBm) (GHz) (GHz)
nm)
techniques
Sim: 16.5 (E-3),19.8 (E-9)
NRZ-TDM 1M 1 PD 54 80 40
Exp: ~23.3 (E-9)
Sim: 14.4 (E-3), 18.3 (E-9)
50% RZ-TDM 2 Ms 1 PD 48 160 40
Exp: ~21 (E-9)
Sim: 14.9 (E-3), 15.1 (E-3),
67% CS-RZ-TDM 2 Ms 1 PD 42 120 40
18.8 (E-9)
DB Sim: 22.4 (E-9) 40
Sim: 11.7 (E-3),13.5 (E-3),
NRZ-DPSK 1M 1 DI + 2 PDs 18.5 (E-9) 74 80 40
Exp: ~20 (E-9)
Sim: 11.1 (E-3), 15.6 (E-9)
50% RZ-DPSK 2 Ms 1 DI + 2 PDs 50 160 40
Exp: ~18 (E-9)
Sim: 13.2 (E-3), 15 (E-3),
NRZ-DQPSK 2 Ms 2 DIs + 4 PDs 13.4 (E-3), 20.5 (E-9) 168 40 20
Exp: ~24.5 (E-9)
Sim: 12.2 (E-3), 15 (E-3),
50% RZ-DQPSK 2 Ms 2 DIs + 4 PDs 17.7 (E-9), 20.2 (E-9) 161 80 20
Exp: ~23.3 (E-9)
3PCs, 2 Ms,
PS, PBS, 2DIs,
RZ-DQPSK-PDM 1PM, 1RZ-PG, Exp: 13.7 (E-9) 40 10
4PDs, PC
2PBS,1PBC
2 PDs, 3 PCs,
NRZ-16-QAM 3 PCs, 1 M Sim: 20.9 (E-9) 20 10
Pol, TFL
E-DCDM (2 × 20
1M 1 PD Sim: 17.8 (E-3), 21.74 (E-9) 62 120 20
Gb/s)
E-DCDM (4 × 10
1M 1 PD Sim: 21.6 (E-3), 26.4 (E-9) 58 100 10
Gb/s)
E-DCDM (7 × 5.71
1M 1 PD Sim: 27 (E-3), 31.4 (E-9) 52 91.4 5.71
Gb/s)
WDM: Wavelength division multiplexing NRZ: Non return-to-zero RZ: Return-to-zero M: Modulator
DQPSK: Differential quadrature phase shift keying CD: Chromatic dispersion DB: Duobinary Num: Numerical
QAM: Quadrature amplitude modulation RS: Receiver sensitivity Tx: Transmitter Rx: Receiver
PDM: Polarization division multiplexing PS: Polarization stabilizer Exp: Experimental Pol: Polarizer
CRF: Clock recovery frequency DI: Delay interferometer PD: Photodetector OOK: On off keying
PBC: Polarization beam combiner PC: Polarization controller PG: Pulse generator
PBS: Polarization beam splitter TFL: Tunable fiber laser MBW: Modulated bandwidth
Table 2. Performance and complexity comparison between different multiplexing
techniques and modulation formats at 40 Gb/s aggregation bit rate (Daikoku, Yoshikane, &
Morita, 2005; Essiambre, Winzer, & Grosz, 2006; Jumpei et al., 2007; Leibrich, Serbay,
Baumgart, Rosenkranz, & Schimmler, 24-28 September 2006; Martelli et al., 2008; Ohm &
Freckmann, 2004)
Advanced modulation formats such as QASK, QAM, QPSK, QDPSK and DB double channel
capacity of optical network with requiring the same spectral band as TDM. However, they
have complex and costly transmitter and receiver. Beside all these techniques, WDM allows
the use of available spectra in optical domain but only at one channel per wavelength.
Combining WDM with individual or combination of the techniques mentioned above

and/or other techniques can increase the transmission capacity tremendously and improve
spectral efficiency of the optical fiber communication systems. Overall, the goal of all
multiplexing techniques and modulation format are to increase spectral efficiency of WDM
networks. To date, PDM and DQPSK and combination of these two techniques achieved the
most attention from the communication society. At the same time, DCDM has some
attractive and unique properties that have not been discovered in other techniques. This
technique has the potential to become an alternative multiplexing technique but it required
more investigation on the practical and experimental systems.
5. Conclusion
The main objective in the communication systems is to transmit as much as possible
information in as low as possible bandwidth and cost. Different modulation formats such as
M-ary ASK, QPSK/DQPSK, QAM and DB are proposed to improve WDM channel capacity.
M-ary ASK improves the channel capacity by transmitting more than one bit per signal
element utilizing different signal levels or multilevel signals. QPSK/DQPSK utilizes
multiple phases for transmit more than one bit per signal element. QAM improve channel
capacity by transmitting more than one bit per signal element utilizing different phases and
different amplitudes. DB use either three amplitude levels or two amplitudes (like binary
signals) and one redundant phase element instead of the third level. All the advanced
modulation schemes are limited to double or triple the channel utilization but all of them
improve the channel capacity or the spectral efficiency. On the other hand, different
multiplexing techniques such as TDM, WDM, OFDM, PDM and DCDM are proposed to
improve the channel utilization. In TDM, different users share the same WDM channel by
allocating each user different time slot. Using WDM, the available optical spectrum can be to
utilize to support multiple numbers of users. In this technique, each user is assigned a
wavelength as the carrier signal. In OFDM, utilizing IDFT many different RF signals can be
assigned as the carrier for many low bit rate users. PDM allows two users to be carried over
two different polarizations, vertical and horizontal polarizations. In DCDM, different users
signed with different RZ duty cycles to share the same WDM channel. Except PDM, all other
multiplexing techniques can support multiple users. Eventhough multiplexing techniques
improve the channel utilization, but all of them except PDM, failed to improve the spectral
efficiency of the link. PDM act like modulation schemes that can double the channel capacity
or the spectral efficiency. Amongst all these multiplexing techniques and modulation
schemes, PDM and QPSK/DQPSK or combinations of them have obtained the most
attention in the communication systems. Finally, DCDM can reduce the clock recovery
frequency significantly.
6. References
Abas, A. F. (2006). Chromatic Dispersion Compensation in 40 Gbaud Optical Fiber WDM Phase-
Shift-Keyed Communication Systems. Universität Paderborn.
Abas, A. F., Hidayat, A., Sandel, D., Milivojevic, B., & Noe, R. (2007). 100 km fiber span in
292 km, 2.38 Tb/s (16×160 Gb/s) WDM DQPSK polarization division multiplex
transmission experiment without Raman amplification. Optical Fiber Technology,
13(1), 46-50.
Abdullah, M. K., Abdalla, M. F., F.Abas, A., & Mahdiraji, G. A. (2007). Duty-cycle Division
Multiplexing (DCDM): A Novel and Economical Optical Multiplexing and Electrical
Demultiplexing technique for High Speed Fiber Optics Networks. Paper presented at the
Wireless and Optical Communications Networks, 2007. WOCN '07. IFIP
International Conference on.
Andre, P., Kauffmann, N., Desrousseaux, P., Godin, J., & Konczykowska, A. (1999). InP HBT
circuits for high speed ETDM systems. Paper presented at the Circuits and Systems,
1999. ISCAS '99. Proceedings of the 1999 IEEE International Symposium on.
Arumugam, M. (2001). Optical fiber communication -- An overview. Pramana - Journal of
physics, 57(5 & 6), 849-869.
Avlonitis, N., Yeatman, E. M., Jones, M., & Hadjifotiou, A. (2006). Multilevel amplitude shift
keying in dispersion uncompensated optical systems. Optoelectronics, IEE
Proceedings -, 153(3), 101-108.
Bigo, S., Frignac, Y., Charlet, G., Idler, W., Borne, S., Gross, H., et al. (2001). 10.2 Tbit/s
(256x42.7 Gbit/s PDM/WDM) transmission over 100 km TeraLightTM fiber with 1.28
bit/s/Hz spectral efficiency. Paper presented at the Optical Fiber Communication
Conference and Exhibit, 2001. OFC 2001.
Charlet, G., Renaudier, J., Mardoyan, H., Tran, P., Pardo, O. B., Verluise, F., et al. (2008).
Transmission of 16.4Tbit/s Capacity over 2,550km Using PDM QPSK Modulation Format
and Coherent Receiver. Paper presented at the Proc. OFC/NFOEC, San Diego,
California.
Cho, P. S., Grigoryan, V. S., Godin, Y. A., Salamon, A., & Achiam, Y. (2003). Transmission of
25-Gb/s RZ-DQPSK signals with 25-GHz channel spacing over 1000 km of SMF-28
fiber. Photonics Technology Letters, IEEE, 15(3), 473-475.
Chraplyvy, A. R., Gnauck, A. H., Tkach, R. W., Zyskind, J. L., Sulhoff, J. W., Lucero, A. J., et
al. (1996). 1-Tb/s transmission experiment. IEEE Photonics Technology Letters, 8,
1264-1266.
Christen, L., Nuccio, S. R., Xiaoxia, W., & Willner, A. E. (2007). Polarization-Based 43 Gb/s RZ-
DQPSK Receiver Design Employing a Single Delay-Line Interferometer. Paper presented
at the Lasers and Electro-Optics, 2007. CLEO 2007. Conference on.
Ciaramella, E. (2002). Nonlinear impairments in extremely dense WDM systems. Photonics
Technology Letters, IEEE, 14(6), 804-806.
Cimini, L. J., Jr., & Foschini, G. J. (1993). Can multilevel signaling improve the spectral
efficiency of ASK optical FDM systems? Communications, IEEE Transactions on,
41(7), 1084-1090.
Daikoku, M., Yoshikane, N., & Morita, I. (2005). Performance comparison of modulation formats
for 40 Gbit/s DWDM transmission systems. Paper presented at the Optical Fiber
Communication Conference, 2005. Technical Digest. OFC/NFOEC.
Derksen, R. H., Moller, M., & Schubert, C. (2007). 100-Gbit/s Full-ETDM Transmission
Technologies. Paper presented at the Compound Semiconductor Integrated Circuit
Symposium, 2007. CSIC 2007. IEEE.
Dong-Soo, L., Man Seop, L., Yang Jing, W., & Nirmalathas, A. (2003). Electrically band-
limited CSRZ signal with simple generation and large dispersion tolerance for 40-
Gb/s WDM transmission systems. Photonics Technology Letters, IEEE, 15(7), 987-989.
Elbers, J. P. (2002). High-capacity DWDM/ETDM transmission. Paper presented at the Optical
Fiber Communication Conference and Exhibit, 2002. OFC 2002.
Essiambre, R.-J., Winzer, P. J., & Grosz, D. F. (2006). Impact of DCF properties on system
design. Journal of Optical and Fiber Communications Reports, 3, 221-291.
Freund, R., Molle, L., Raub, F., Caspar, C., Karkri, M., & Weber, C. (2005). Triple-(S/C/L)-band
WDM transmission using erbium-doped fibre amplifiers. Paper presented at the Optical
Communication, 2005. ECOC 2005. 31st European Conference on.
Fukuchi, K., Kasamatsu, T., Morie, M., Ohhira, R., Ito, T., Sekiya, K., et al. (2001). 10.92-Tb/s
(273 x 40-Gb/s) triple-band/ultra-dense WDM optical-repeatered transmission experiment.
Paper presented at the Optical Fiber Communication Conference and Exhibit, 2001.
OFC 2001.
Gnauck, A. H., Charlet, G., Tran, P., Winzer, P., Doerr, C., Centanni, J., et al. (2007). 25.6-Tb/s
C+L-Band Transmission of Polarization-Multiplexed RZ-DQPSK Signals. Paper
presented at the Proc. OFC/NFOEC, Anaheim, California.
Gnauck, A. H., Charlet, G., Tran, P., Winzer, P. J., Doerr, C. R., Centanni, J. C., et al. (2008).
25.6-Tb/s WDM Transmission of Polarization-Multiplexed RZ-DQPSK Signals. J.
Lightwave Technology, 26(1), 79-84.
Gnauck, A. H., Winzer, P. J., Dorrer, C., & Chandrasekhar, S. (2006). Linear and nonlinear
performance of 42.7-Gb/s single-polarization RZ-DQPSK format. Photonics
Gyo-Sun, H., Un-Seung, P., Sung-Ho, K., Youngchul, C., Jae-Seung, L., & Byoung-Whi, K.
(2007). 2.5-Tb/s (256 x 12.4 Gb/s) Transmission of 12.5-GHz-Spaced Ultra-dense WDM
Channels over a Standard Single-Mode Fiber of 2000 km. Paper presented at the Optical
Fiber Communication and the National Fiber Optic Engineers Conference, 2007.
OFC/NFOEC 2007. Conference on.
Han, Y., & Li, G. (2006). Experimental demonstration of direct-detection quaternary
differential polarization-phaseshift keying with electrical multilevel decision.
Electronics Letters, 42, 109-111.
Hayee, M. I., Cardakli, M. C., Sahin, A. B., & Willner, A. E. (2001). Doubling of Bandwidth
Utilization Using Two Orthogonal Polarizations and Power Unbalancing in a
Polarization-Division-Multiplexing Scheme. IEEE Photonics Technology Letters, 13(8),
881-883.
Hill, P. M., Olshansky, R., & Burns, W. K. (1992). Optical polarization division multiplexing
at 4 Gb/s,”. IEEE Photonics Technology Letters, 4(5), 500-502.
Hinz, S., Sandel, D., Noe, R., & Wust, F. (2000). Optical NRZ 2 x 10 Gbit/s polarisation
division multiplex transmission with endless polarisation control driven by
correlation signals. Electronics Letters, 36(16), 1402-1403.
Hinz, S., Sandel, D., Wust, F., & Noe, R. (2001). Interference detection enabling 2 x 20 Gbit/s
RZ polarisation division multiplex transmission. Electronics Letters, 37(8), 511-512.
Ho, K.-P. (2005). Phase-modulated optical communicattion systems: Springer, USA.
Ibrahim, S. K., Bhandare, S., & Noe, R. (2006). Performance of 20 Gb/s quaternary intensity
modulation based on binary or duobinary modulation in two quadratures with
unequal amplitudes. Selected Topics in Quantum Electronics, IEEE Journal of, 12(4),
596-602.
Jansen, S. L., Derksen, R. H., Schubert, C., Zhou, X., Birk, M., Weiske, C. J., et al. (2007). 107-
Gb/s full-ETDM transmission over field installed fiber using vestigial sideband
modulation. Paper presented at the Optical Fiber Communication and the National
Fiber Optic Engineers Conference, 2007. OFC/NFOEC 2007. Conference on.
Jansen, S. L., Morita, I., Schenk, T. C. W., & Tanaka, H. (2008). Long-haul transmission of
16×52.5 Gbits/s polarization-division- multiplexed OFDM enabled by MIMO
processing (Invited). Journal of Optical Networking, 7(2), 173-182.
Jumpei, H., Kasai, K., Yoshida, M., & Nakazawa, M. A. N. M. (2007). 1 Gsymbol/s, 64 QAM
coherent optical transmission over 150 km with a spectral efficiency of 3 bit/s/Hz. Paper
presented at the Proc. OFC/NFOEC, Anaheim, California.
Kauffmann, N., Blayac, S., Abboun, M., Andre, P., Aniel, F., Riet, M., et al. (2001). InP HBT
driver circuit optimization for high-speed ETDM transmission. Solid-State Circuits,
IEEE Journal of, 36(4), 639-647.
Kawanishi, T., Sakamoto, T., Miyazaki, T., Izutsu, M., Higuma, K., & Ichikawa, J. (2007). 80
Gb/s DQPSK modulator. Paper presented at the Optical Fiber Communication and
the National Fiber Optic Engineers Conference, 2007. OFC/NFOEC 2007.
Conference on.
Keiser, G. (2000). Optical fiber communications (3rd edition ed.): McGraw-Hill.
Keiser, G. E. (1999). A review of WDM technology and applications. Optical Fiber Technology,
5, 3-39.
Krummrich, P. M., Gottwald, E., Weiske, C. J., Schopflin, A., Kotten, I., & Lankl, B. (2002). 40
Gbit/s ETDM for long haul WDM transmission. Paper presented at the Lasers and
Electro-Optics Society, 2002. LEOS 2002. The 15th Annual Meeting of the IEEE.
Lach, E., Bulow, H., Kaiser, M., Veith, G., & Bouchoule, S. (1998). 40 Gbit/s TDM field test over
111 km installed G.652 fiber. Paper presented at the Optical Communication, 1998.
24th European Conference on.
Lach, E., & Schuh, K. (2006). Recent Advances in Ultrahigh Bit Rate ETDM Transmission
Systems. Lightwave Technology, Journal of, 24(12), 4455-4467.
Lach, E., Schuh, K., Junginger, B., Veith, G., Lutz, J., & Moller, M. (2007). Challenges for 100
Gbit/s ETDM Transmission and Implementation. Paper presented at the Optical Fiber
Communication and the National Fiber Optic Engineers Conference, 2007.
Lathi, B. P. (1998). Modern Digital and Analog Communication Systems: Oxford university
press.
Lee, W. S., Garthe, D., Pettitt, G. A., & Hadjifotiou, A. A. (1997). 40-Gbit/s TDM transmission
over 160 km (2 x 80 km) of standard nondispersion-shifted fiber. Paper presented at the
Optical Fiber Communication. OFC 97., Conference on.
Leibrich, J., Serbay, M., Baumgart, S., Rosenkranz, W., & Schimmler, M. (24-28 September
2006, 24-28 September 2006). Receiver sensitivity of advanced modulation formats for 40
Gb/s DWDM transmission with and without FEC. Paper presented at the 32nd
European Conference and Exhibition on Optical Communication (ECOC), Cannes,
France.
Lender, A. (1964). Correlative Digital Communication Techniques. Communication

Technology, IEEE Transactions on, 12(4), 128-135.
Mahdiraji, G. A., Abdullah, M. K., Mokhtar, M., Mohammadi, A. M., Abdullah, R. S. A. R., &
Abas, A. F. ((In Press)). Duty-Cycle-Division-Multiplexing: Bit Error Rate
Estimation and Performance Evaluation. Optical Review.
Martelli, P., Boffi, P., Ferrario, M., Marazzi, L., Parolari, P., Pietralunga, S. M., et al. (2008).
Endless Polarization Stabilizer for High Bit-rate Polarization-Division Multiplexed Optical
Systems. Paper presented at the OFC/NFOEC.
Martelli, P., Boffi, P., Ferrario, M., Marazzi, L., Parolari, P., Righetti, A., et al. (2007).
Uncompensated 20 Gb/s Duobinary Polarization Division Multiplexing Transmission over
200 km. Paper presented at the Lasers and Electro-Optics, 2007. CLEO 2007.
Conference on.
Miyamoto, Y., Yoneyama, M., Otsuji, T., Yonenaga, K., & Shimizu, N. (1999). 40-Gbit/s TDM
transmission technologies based on ultra-high-speed ICs. Solid-State Circuits, IEEE
Journal of, 34(9), 1246-1253.
Morita, I., & Yoshikane, N. (2005). Merits of DQPSK for ultrahigh capacity transmission. Paper
presented at the Lasers and Electro-Optics Society, 2005. LEOS 2005. The 18th
Annual Meeting of the IEEE.
Muoi, T., & Hullett, J. (1975). Receiver Design for Multilevel Digital Optical Fiber Systems.
Communications, IEEE Transactions on [legacy, pre - 1988], 23(9), 987-994.
Nakazawa, M., Yoshida, M., Kasai, K., & Hongou, J. (2006). 20Msymbol/s, 64 and 128 QAM
coherent optical transmission over 525km using heterodyne detection with
frequency-stabilised laser. Electronics Letters, 42(12), 710-712.
Nasu, Y., Oguma, M., Takahashi, H., Inoue, Y., Kawakami, H., & Yoshida, E. (2008).
Polarization insensitive MZI-based DQPSK demodulator with asymmetric half-wave plate
configuration. Paper presented at the Optical Fiber communication/National Fiber
Optic Engineers Conference, 2008. OFC/NFOEC 2008. Conference on.
Nelson, L. E., & Kogelnik, H. (2000). Coherent crosstalk impairments in polarization
multiplexed transmission due to polarization mode dispersion. Optical Express,
7(10), 350-361.
Nelson, L. E., Nielsen, T. N., & Kogelnik, H. (2001). Observation of PMD-Induced Coherent
Crosstalk in Polarization-Multiplexed Transmission. IEEE Photonics Technology
Letters, 13(7), 738-740.
Noe, R., Hinz, S., Sandel, D., & Wust, F. (2001). Crosstalk detection schemes for polarization
division multiplex transmission. Journal of Lightwave Technology, 19, 1469-1475.
Ohm, M., & Freckmann, T. (2004). Comparison of different DQPSK transmitters with NRZ and
RZ impulse shaping. Paper presented at the Advanced Modulation Formats, 2004
IEEE/LEOS Workshop on.
Palais, J. C. (2005). Fiber optic communications (5th edition ed.): Pearson Prentice Hall.
Pardo, O. B., Renaudier, J., Mardoyan, H., Tran, P., Charlet, G., & Bigo, S. (2008a).
Investigation of design options for overlaying 40Gb/s coherent PDM-QPSK channels over
a 10Gb/s system infrastructure. Paper presented at the Optical Fiber
communication/National Fiber Optic Engineers Conference, 2008. OFC/NFOEC
2008. Conference on.
Pardo, O. B., Renaudier, J., Mardoyan, H., Tran, P., Charlet, G., & Bigo, S. (2008b).
Investigation of design options for overlaying 40Gb/s coherent PDM-QPSK channels over
a 10Gb/s system infrastructure. Paper presented at the OFC/NFOEC.
Ramaswami, R., & Sivarajan, K. N. (2002). Optical Networks: A Practical Perspective: Morgan
Kaufmann
Renaudier, J., Charlet, G., Salsi, M., Pardo, O. B., Mardoyan, H., Tran, P., et al. (2008). Linear
Fiber Impairments Mitigation of 40-Gbit/s Polarization-Multiplexed QPSK by
Digital Processing in a Coherent Receiver. Lightwave Technology, Journal of, 26(1), 36-
42.
Said, M. M. E., Sitch, J., & Elmasry, M. I. (2005). An electrically pre-equalized 10-Gb/s
duobinary transmission system. Lightwave Technology, Journal of, 23(1), 388-400.
Sandel, D., Wust, F., Mirvoda, V., & Noe, R. (2002). Standard (NRZ 1 x 40 Gb/s, 210 km) and
polarization multiplex (CS-RZ, 2 x 40 Gb/s, 212 km) transmissions with PMD
compensation. Photonics Technology Letters, IEEE, 14(8), 1181-1183.
Sang-Yuep, K., Sang-Hoon, L., Sang-Soo, L., & Jae-Seung, L. (2004). Upgrading WDM
networks using ultradense WDM channel groups. Photonics Technology Letters,
IEEE, 16(8), 1966-1968.
Savory, S. J., Gavioli, G., Killey, R. I., & Bayvel, P. (2007). Transmission of 42.8Gbit/s
polarization multiplexed NRZ-QPSK over 6400km of standard fiber with no optical
dispersion compensation. Paper presented at the OFC/NFOEC 2007 - Optical Fiber
Communication and the National Fiber Optic Engineers Conference 2007.
Scheerer, C., Glingener, C., Farbert, A., Elbers, J. P., Schopflin, A., Gottwald, E., et al. (1999).
3.2 Tbit/s (80 x 40 Gbit/s) bidirectional WDM/ETDM transmission over 40 km
standard singlemode fibre. Electronics Letters, 35(20), 1752-1753.
Schubert, C., Ludwig, R., Ferber, S., Schmidt-Langhorst, C., Huettl, B., & Weber, H. G.
(2006). Single Channel Transmission Beyond 1 Tbit/s. Paper presented at the Lasers
and Electro-Optics Society, 2006. LEOS 2006. 19th Annual Meeting of the IEEE.
Schuh, K., Junginger, B., Rempp, H., Klose, P., Rosener, D., & Lach, E. (2005). 85.4 Gbit/s
ETDM transmission over 401 km SSMF applying UFEC. Paper presented at the
Optical Communication, 2005. ECOC 2005. 31st European Conference on.
Seo, H. S., Chung, W. J., & Ahn, J. T. (2005). A novel hybrid silica wide-band amplifier
covering S+C+L bands with 105-nm bandwidth. Photonics Technology Letters, IEEE,
17(9), 1830-1832.
Shieh, W., & Athaudage, C. (2006). Coherent optical orthogonal frequency division
multiplexing. Electronics Letters, 42(10).
Shieh, W., Bao, H., & Tang, Y. (2008). Coherent optical OFDM: theory and design. Optical
Express, 16(2), 841-859.
Shieh, W., Yi, X., Ma, Y., & Yang, Q. (2008). Coherent optical OFDM: has its time come?
[Invited]. Journal of Optical Networking, 7(3), 234-255.
Suzuki, K., Kubota, H., Kawanishi, S., Tanaka, M., & Fujita, M. (2001). High-speed bi-
directional polarisation division multiplexed optical transmission in ultra low-loss
(1.3 dB/km) polarisation-maintaining photonic crystal fibre. Electronics Letters,
37(23), 1399-1401.
van den Borne, D., Jansen, S. L., Gottwald, E., Schmidt, E. D., Khoe, G. D., & de Waardt, H.
(2008). DQPSK modulation for robust optical transmission. Paper presented at the
Optical Fiber communication/National Fiber Optic Engineers Conference, 2008.
Walklin, S., & Conradi, J. (1999). Multilevel signaling for increasing the reach of 10 Gb/s
lightwave systems. Lightwave Technology, Journal of, 17(11), 2235-2248.
Weber, H. G., Ferber, S., Kroh, M., Schmidt-Langhorst, C., Ludwig, R., Marembert, V., et al.
(2005). Single channel 1.28 Tbit/s and 2.56 Tbit/s DQPSK transmission. Paper presented
at the Optical Communication, 2005. ECOC 2005. 31st European Conference on.
Weber, H. G., Ferber, S., Kroh, M., Schmidt-Langhorst, C., Ludwig, R., Marembert, V., et al.
(2006). Single channel 1.28 Tbit/s and 2.56 Tbit/s DQPSK transmission. Electronics
Letters, 42(3), 178-179.
Weber, H. G., Ludwig, R., Ferber, S., Schmidt-Langhorst, C. A., Kroh, M. A., Marembert, V.
A., et al. (2006). Ultrahigh-Speed OTDM-Transmission Technology. Lightwave
Technology, Journal of, 24(12), 4616-4627.
Wei, X., Liu, X., Chandrasekhar, S., Gnauck, A. H., Raybon, G., Leuthold, J., et al. (2002). 40
Gb/s Duobinary and Modified Duobinary Transmitter Based on an Optical Delay
Interferometer. Paper presented at the Optical Communication, 2002. ECOC 2002.
Weinsten, S. B., & Ebert, P. M. (1971). Data transmission by frequency-division multiplexing
using the discrete frouer transform. IEEE Transactions on Communication, 19, 628-
634.
Winzer, P. J., & Essiambre, R. J. (2006). Advanced Modulation Formats for High-Capacity
Optical Transport Networks. Lightwave Technology, Journal of, 24(12), 4711-4728.
Winzer, P. J., & Hoon, K. (2003). Degradations in balanced DPSK receivers. Photonics
Wree, C., Hecker-Denschlag, N., Gottwald, E., Krummrich, P., Leibrich, J., Schmidt, E. D., et
al. (2003). High spectral efficiency 1.6-b/s/Hz transmission (8 x 40 Gb/s with a 25-
GHz grid) over 200-km SSMF using RZ-DQPSK and polarization multiplexing.
Photonics Technology Letters, IEEE, 15(9), 1303-1305.
Yan, L. S., Zhang, B., Belisle, A., Willner, A. E., & Yao, X. S. (2007). Automatic All-Optical
Detection in Polarization-Division-Multiplexing System Using Power Unbalanced
Transmission. Paper presented at the Lasers and Electro-Optics, 2007. CLEO 2007.
Conference on.
Yao, X. S., Yan, L.-S., Zhang, B., Willner, A. E., & Jiang, J. (2007). All-optic scheme for
automatic polarization division demultiplexing. Optical Express, 15(12), 7407-7414.
Yonenaga, K., & Kuwano, S. (1997). Dispersion-tolerant optical transmission system using
duobinary transmitter and binary receiver. Lightwave Technology, Journal of, 15(8),
1530-1537.
Yonenaga, K., Kuwano, S., Norimatsu, S., & Shibata, N. (1995). Optical duobinary
transmission system with no receiver sensitivity degradation. Electronics Letters,
31(4), 302-304.
Yoneyama, M., Miyamoto, Y., Otsuji, T., Hirano, A. A., Kikuchi, H. A., Ishibashi, T. A., et al.
(1999). Fully electrical 40-Gbit/s TDM system prototype and its application to 160-Gbit/s
WDM transmission. Paper presented at the Optical Fiber Communication
Conference, 1999, and the International Conference on Integrated Optics and
Optical Fiber Communication. OFC/IOOC '99. Technical Digest.
Yoshida, M., Goto, H., Kasai, K., & Nakazawa, M. (2008). 64 and 128 coherent QAM optical
transmission over 150 km using frequency-stabilized laser and heterodyne PLL
detection. Optics Express, 16(2), 829-840.
Yoshikane, N., & Morita, I. (2004). 1.14 b/s/Hz spectrally-efficient 50 x 85.4 Gb/s transmission
over 300 km using copolarized CS-RZ DQPSK signals. Paper presented at the Optical
Fiber Communication Conference, 2004. OFC 2004.
Zahedi, E. (2002). Digital Data Communication: Prentice Hall.
Zhou, X., Yu, J., Huang, M.-F., Shao, Y., Wang, T., Magill, P., et al. (22-26 March 2009). 32Tb/s
(320x114Gb/s) PDM-RZ-8QAM transmission over 580km of SMF-28 ultra-low-loss fiber.
Paper presented at the OFC/NFOEC.
Zhu, B., Leng, L., Nelson, L. E., Knudsen, S., Bromage, J., Peckham, D., et al. (2001). 1.6 Tb/s
(40 × 42.7 Gb/s) transmission over 2000 km of fiber with 100-km dispersion-
managed spans. Paper presented at the Optical Communication, 2001. ECOC '01.
A Survey on the Design of Binary Pulse Compression Codes with Low Autocorrelation 39
X3
A Survey on the Design of Binary Pulse

Compression Codes with Low Autocorrelation
Maryam Amin_Nasrabadi* and Mohammad_Hassan Bastani†
*BasamadAzma Co., †Sharif University of Technology
Iran
1. Introduction
Simple pulsed radar is limited in range sensitivity by the average radiation power and, in
range resolution by the pulse length. The design of any radar always involves a compromise
between the two constraints. Waveform design aims to seek an appropriate harmony that
best suits the relevant application. The pulse compression theory has been introduced in
order to get a high range resolution as well as a good detection probability.
One of the basic types of pulse compression is binary phase coding which encodes the
transmitted pulse with information that is compressed (decoded) in the receiver of the
radar.
The study of the peak sidelobe level (PSL) binary sequences occurs as a classical problem of
signal design for digital communication and, in equivalent guise, in analytic number theory.
It has also become a notorious problem of combinatorial optimization. For years
mathematicians, engineers, physicists and chemists have sought a systematic way to
construct long binary sequences with low PSL.
In this Chapter, we describe pulse compression technique in radar waveform design. In
order to make the presentation self-contained, we start by providing a short summary of
waveform design and an introduction to principle behind pulse compression by compiling
the basic tools required for analyzing and comparing different radar signals.
After that, we talk about binary sequence, its desired properties and general types of
methods for finding and generating such waveforms. We keep on by an overview and
introducing the existing methods and search routine done.
We conclude the chapter with a brief survey of the results exhibited yet for low
autocorrelation binary sequences. We mention a table of complete results presented and also
introduce a histogram to unscramble them visually and predict the future.
2. Why is pulse compression needed?

One of the most important usages of radar is range finding which is done through
measuring time delay, Δt; it takes a pulse to travel the two-way path between the radar and
the target.
ct
R (1)
2
Where, c is denoted as speed light.
In general, a pulsed radar transmits and receives a train of pulses, as illustrated by Figure 1.
Fig. 1. Illustrating range.
By definition, Range Resolution is the ability to detect targets close proximity to each other
as distinct objects only by measurement of their ranges (distances from radar) which usually
expressed in terms of the minimum distance by which two targets of equal strength at the
same azimuth and elevation angles must be spaced to be separately distinguishable.
Resolution in the range domain ΔR corresponds to the resolution τ in the time domain, and
is set by the pulsewidth according to ΔR = cτ/2 (for pulse-compression waveform, τ is the
pulsewidth after pulse compression).
Without pulse compression, the instantaneous bandwidth of radar receiver, B, is equal to
pulse bandwidth which is usually set to 1/τ; thus
c c
R   (2)
2 2B
In general, radar users and designers alike seek to minimize ΔR in order to enhance the
radar performance. As suggested by equation (2), in order to achieve fine range resolution
one must minimize the pulse width or maximize the bandwidth.
On the other hand, as illustrated in Figure 1., during each PRI1 the radar radiates energy
only for τ seconds and listens for target returns for the rest of PRI.
Let Pav , Pt and Ep stand for average transmitted power, peak transmitted power and pulse
energy, respectively. So,
1 Pulse Repetition Interval

 
Pav  Pt.  (3)
T 
E p  Pt .  Pav .T (4)
With regard to equations (3) and (4) and also Figure 1., above solutions will reduce the
average transmitted power. Furthermore, in accordance with Radar Equation (equation (5)),
maximum range and thus detection probability will decrease.
1
 Pt G 2 2  4
Rmax   

(5)
 (4 ) KTe BFL( SNR) o min
3

So, it seems that the only way to account for these problems and have good range resolution
is increasing the peak transmitted power, whereas there are technical limitations for the
maximum peak power, such as maximum high voltage or power from the output stage, or
waveguide breakdown. So, the only approach for achieving fine range resolution while
maintaining adequate average transmitted power is using pulse compression techniques
(Mahafza & Alabama, 2000; Skolnik, 2001) which is the main subject of this chapter and will
be expatiated.
For clarity, consider this example:
Example:
Desired resolution: R = 15 cm
Required bandwidth: B = 1 GHz
Required pulse energy: E = 1 mJ
By using equation (4), two solutions are as follows:
Brute force approach:
Raw pulse duration: τ = 1 ns
Required peak transmitted power: P = 1 MW!
Pulse compression approach:
Pulse duration: τ = 0.1 ms
Required peak transmitted power: P = 10 W
It is good to mention that, required range resolution for a given radar is dependent on its
performance (detection, recognition, identification, etc.). For example, see table R26 in
(Barton & Leonov, 1998) about resolution required for target interpretation tasks.
3. Pulse compression
Pulse compression allows radar to use long waveforms in order to obtain high energy and
simultaneously achieve the resolution of a short pulse by internal modulation of the long
pulse. This technique can increase signal bandwidth through frequency or phase coding.
Although, amplitude modulation is not forbidden but usually is not used. The received echo
is processed in the receiver matched filter to produce a short pulse with duration 1/B, where
B is bandwidth of compressed pulse. This technique is of interest when the radar is not able
to generate enough required power. So, a concise summary for pulse compression is
gathering two opposite benefits “High Range Resolution” and “high detection probability”
concurrently. It can be stated that “radar pulse compression” is a substitute for “short pulse
radar”, although, each one has its own advantages and difficulties.
Some privileges of short-pulse radar are as follows (Skolnik, 2001):
- range resolution
- range accuracy
- clutter reduction
- interclutter visibility
- glint reduction
- multipath resolution
- multipath height-finding
- target classification
- doppler tolerance
- ECCM
- Minimum range
and some defects of short-pulse are given below:
- Interference with other frequncy bands

- Limited maximum range
Pulse compression has all advantages related to short pulse except short minimum range.
Furthermore, this technique has obviated limitation in average transmitted power belonging
to short pulse. In other hand, it has two disadvantages:
- Increased complexity for generating, transmitting and processing which cause more
expense.
- Appearing sidelobes in compressed pulse which result in decreased range resolution.
A twin good and bad effect of pulse compression technique can be shown by Figure 2.
Fig. 2. Resolving targets in range (I) two resolved targets by short-pulse (II) two unresolved
targets by long-pulse (III) two resolved target by using pulse compression with long-pulse
Consider two targets which can receive and reflect radar pulse. If these two reflected pulses
are narrow enough, they will be separated; A-pulse and B-pulse are indicated reflected
pulse from target A and B respectively (Figure 2-I). But, if these pulses are wide, they may
overlap and may not be separable (Figure 2-II). If these wide pulses are passed through
compression filter, two narrow pulses will be generated which can be distinguished easily
(Figure 2-III). This is an efficacy of pulse compression but, one must tolerate a bad effect
along with this advantage which is appearing extra pulses around the main one at the
output of compression filter (Figure 2-III). This is obvious that if these side pulses have large
amplitude, the radar will mistake.
Another parameter needed to introduce is pulse compression ratio which is define here:
uncompressed pulse width

Pulse compression ratio 
compressed pulse width
And can be stated as follow:
pulse compressio n ratio  B (6)
In equation (6), B and τ are denoted as pulse bandwidth and compressed pulse width and
usually Bτ≫1.
4. Different types of pulse compression technique

There are various kinds of pulse compression technique which can be categorized in two
general sets. In order to be familiar to these categorizes, some common types of them are
considered and since these signals have been discussed in details in many literatures, only a
synoptic account of them is cited including its benefits and difficulties. For more
information, the reader is referred to (Barton & Leonov, 1998; Skolnik, 2001; Farnett &
Stevens, 1991; Levanon & Mozeson, 2004).
4.1 Frequency Modulation

These waveforms can entail linear (LFM) or nonlinear (NLFM) modulation of the frequency
of the transmitted waveform. The summary of general characteristics of them given here
(Nathanson & Cohen, 1999 ; Barton & Leonov, 1998; Farnett & Stevens, 1991):
4.1.1 Linear frequency modulation (Chirp)

Advantages:
- It is quite insensitive to doppler shifts.

- It is the easiest waveform to generate.
- There is a variety of hardware being avaiable to form and process it.
Disadvantages:
- It has range-doppler cross coupling, resulting in measurement errors unless one of the
coordinates (range or doppler) is determined.
- Range sidelobes are high, compared with nonlinear FM and phase-coded waveforms. To
reduce sidelobe level, weighting is usually required, resulting 1-to-2-db loss in signal to
noise ratio.
4.1.2 Nonlinear frequency modulation

- It has very low range sidelobes without necessitating the use of special weighting for their
supression, and hence has no signal-to-noise ratio loss as does the LFM waveform.
- It is more sensitive to doppler frequency shifts.
- It is complex and its generation techniques has been developed limitedly.
4.2 Phase coding

This waveform is one in which intrapulse modulation is obtained by subdividing the pulse
into subpulses of equal duration, each having a particular phase. The phase of each subpulse
is set in accordance with a given code or code sequence. Common characteristics of phase
coded radar are as follows (Barton & Leonov, 1998; Farnett & Stevens, 1991):
- In comparison with LFM waveforms, they have lower range sidelobes.

- They are preferred in jamming conditions, as the coding of the transmitted signal gives an
additional degree of protection against ECM.
- Their resolution performance in a dense target environment or in presence of distributed
clutter can be rather poor.
- The implementation of phase coded waveforms is more complex than that of the LFM
waveforms.
This modulation can be categorized in two subsets:
4.2.1 Polyphase coding

This waveform uses codes with the number of discrete phase values greater than two. Some
properties of this modulation are:
- The range sidelobes of polyphase coded waveforms can be lower than those of the binary
coded ones of the same length, but the performance of these waveforms deteriorates
rapildly in the presence of doppler frequency shift.
- Generation and processing of polyphase-coded waveforms use technique similar to those
of frequency-coded waveforms, but their range sidelobe parameters are much better than
for unweighted FM waveforms.
4.2.2 Binary coding

It is the most widely used phase coded waveform which employs two phase. In this type of
pulse compression method, a long pulse with duration τ is divided into N “subpulse”, each
with duration τn, where

n  (7)
N
The phase of each segment is set to 0 and 180 in accordance with the sequence of element
in the phase code, as indicated in Figure 3.
Fig. 3. Binary coded waveform.

(A) binary sequence (modulator). (B) phase coded signal.
If these pahses are selected randomly, the resulted waveform will be a noise modulated one
and if they are chosen in accordance with some special patterns, the generated binary coded
signal will have better function and the goal of this chapter is finding such sequences
(Barton & Leonov, 1998; Farnett & Stevens, 1991; Skolnik, 2001).
By using bianry coded waveform, the effective width of matched filter output pulse is τn and
its amplitude is N times greater than primary one. Thus, pulse compression ratio is

pulse compressio n ratio  B  N (8)
n
Where, B is banwidth of modulated pulse and can be calculated by
1
B (9)
n
The duration of matched filter output is indeed 2τ. i.e.; in addition to main peak with width
τn the output of this filter spreads over a 2τ period in both sides of the main lobe. These extra
pulses are named time sidelobes.
Table 1. is shown a brief comparison between LFM and Biphas-coded signals (Skolnik 2001).
Also, the reader can refer to a table including summary of performance of various pulse
compression implementation in (Farnett & Stevens, 1991) and a good and depth overview of
HRR1 radar and comparison between several modulations mentioned in previous sections
in (Cohen, 1991; Levanon & Getz, 1994).
Property LFM Binary phase coded pulses
Good when weighting on Can be equal to 1/2N, and are not

Time sidelobes receive, and when a loss of easy to improve; poor doppler
about 1 dB can be tolerant sidelobes
Dopller Doppler tolerant Requires filter bank
Ambiguity Thumbtack (but with high sidelobes

Ridge
diagram in plateau)
Pulse Single filter can be used for Single filter can be used for transmit
compression transmit and receive; usually and receive, but with input at
filter analog for high resolution opposite end; usually digital
Less complex, especially if

Complexity More complex, (requires filter bank)
Strech can be used
High resolution (wide
Application Long pulses
bandwidth)
Range-doppler coupling; has Bandwidth limited by availability of

Other been more widely used than A/D converter; erroneously thought
other pulse compression to be less susceptible to ECM spoofing
Table 1. Comparison of linear FM and Binary phase-coded pulse compression waveforms
1 High Range Resolution

5. Matched filter
Matched filter have a principle position in pulse compression technique, so before starting to
talk about main question, its properties and characteristics are considered briefly.
The most unique characteristic of the matched filter is that it produces the maximum
achievable instantaneous SNR1 at its output when a signal plus additive white noise is
present at the input. The peak instantaneous SNR at the receiver output can be achieved by
matching the radar receiver transfer function to the received signal and this peak value can
be calculated by
2E
SNR  (10)
No
Where, E and No are denoted as input signal energy and input noise power respectively.
Thus, we can draw the conclusion that the peak instantaneous SNR depends only on the
signal energy and input noise power, and is independent of the waveform utilized by the
radar.
For this peak instantaneous SNR, matched filter impulse response is:
*
h(t )  si (  t ) (11)
Where, si(t) is radar transmitted signal. Equation (11) indicates that the peak occurs at τ
second after entering signal to matched filter.
Now, consider a radar system that uses signal si(t), and assume that a matched filter receiver
is utilized. The matched filter input signal can then be represented by
x(t )  Csi (t  t1 )  ni (t ) (12)
Where C is constant, t1 is an unknown time delay proportional to target range, and ni(t) is
input white noise.
The matched filter output y(t) can be expressed by the convolution integral between the
filter’s impulse response and x(t),

y (t )   x(u )h(t  u )du

(13)
And by using equation (11), the matched filter output signal can be written

 x(u ) s
*
y (t )  i (  t  u ) du  R xsi (t   ) (14)

1 Signal to Noise Ratio

Where R xsi (t   ) is cross-correlation between x(t) and si(t-τ).

Therefore, the matched filter output can be computed from the cross-correlation between the
radar received signal and a delayed replica of the transmitted waveform. If the input signal
is the same as the transmitted signal, the output of the matched filter would be the
autocorrelation function of the received (or transmitted) signal.
In practice, replicas of the transmitted waveforms are normally computed and stored in
memory for use by the radar signal processor when needed (Mahafza & Alabama, 2000;
Skolnik, 2001).
In pulse compression technique, initially a long pulse is generated and modulated in
transmitter and in receiver, a matched filter is used to compressed signal. The matched filter
output is compressed by factor equal to Bτ which is proportional to bandwidth and
pulsewidth. i.e.; by utilizing long pulse and wide band modulation, it is possible to gain
high compression ratio. Therefore, in using pulse compression, it is good to apply a
modulation which can maximize compression ratio while having low sidelobe in
compressed signal.
Sidelobe suppression technique can be used on the compressed pulse spectrum in order to
reduce the side lobe levels. Usually, the cost associated with such an approach is a loss in
the main lobe resolution, and a reduction in the peak value (i.e., loss in the SNR). For more
information, the reader can see ( Mahafza & Alabama, 2000; Baden & Cohen, 1990; Ackroyd
& Ghani, 1973; Rihaczek & Golden, 1971).
6. Autocorrelation function
Since, this function have a critical position in this matter, before continuing to talk about the
problem, it is good to review its definition and some important characteristics which
researchers are used.
Consider a real binary sequence of length N,
ak kN01 , ak  1 (15)
Some definitions related to these codes are as follow (Levanon & Mozeson, 2004; Golay,
1977):
1. Its ACF1,
N |n| 1
R (n)  a a
i 0
i i  | n| , n  0,1,,( N  1)
(16)
2. ML2 which is defined as the absolute maximum value of ACF,
1 AutoCorrelation Function
2 MainLobe
ML  max ( R ( n)) (17)

n
3. Sidelobes which are described as maximum values of ACF except absolute one.
4. PSL which is denoted as maximum of sidelobes,
PSL  max (| R ( n) |) (18)

n 0
5. E which means energy of a sequence,
N 1
E  2 R 2 ( n) (19)
n 1
6. ISL1 which is another parameter for measuring sidelobe levels,
 E 
ISL  10 log10  2  (20)
N 
7. MF2 which is first defined by Golay,
N2
MF  (21)
E
8. ACF is an even function,
R (n)  R ( n) (22)
9. ACF is a finite length sequence,
R (n)  0 , | n | N (23)
10. The absolute maximum value of ACF is at the origin,
R ( n )  R ( 0 ) , n (24)
11. ACF’s value for all N-bit sequences at the origion is equal and independent on codes’
elements,
1 Integrated Sidelobe Level

2 Merit Factor
R ( 0)  N (25)
12. ACF has a limitation on its values,
| R (n) | N  | n | , n (26)
And
1  PSL  ( N  1) (27)
Another concept which is introduced is Allomorphic forms and psl-preserving operations.

Each binary sequence can be stated in 4 forms in terms of autocorrelation function. If Ra(n) is
referred to ACF of sequence ak, these three codes have ACF equal to Ra(n):
- Inverse-amplitude or complement:
bk  , bk  ak , 0  k  ( N  1) (28)
- Inverse-time or reverse:
ck  , ck  a N  k 1 , 0  k  ( N  1) (29)
- Complement of reverse:
d k  , d k  a N  k 1 , 0  k  ( N  1) (30)
Then,
Ra (n)  Rb (n)  Rc (n)  Rd (n) , n (31)
As PSL is related to the absolute value of ACF, there are some other forms that can have the
PSL equal to ak:
ek  , ek  ( 1) k a k , 0  k  ( N  1) (32)
fk  , f k  (1) k 1 a k , 0  k  ( N  1) (33)
g k  , g k  ( 1) k a N  k 1 , 0  k  ( N  1) (34)
hk  , hk  ( 1) k 1 a N k 1 , 0  k  ( N  1) (35)

Such that,
Re (n)  R f (n)  Rg (n)  Rh (n)  (1) n Ra (n) , n (36)
7. Problem definition
Now, everything is ready in order to introduce main problem and its offered solutions. As
proved before, the output of the matched filter is the autocorrelation function of the input
signal (without any doppler shift frequencies and considering noise). So, a good criterion for
choosing biphase codes is that their autocorrelation have sidelobes as minimum as possible.
Barker sequences are one of these optimum codes whose peak sidelobe levels are equal to 1.
One of these codes is shown in Figure 4.A and its autocorrelation is drawn in Figure 4.B. As
indicated, level of these sidelobes is -22.3 dB below the main peak.
Fig. 4. Complex envelope of transmitted signal which is modulated by Barker sequence (A)
Barker sequence (B) complex envelope of matched filter output
Up to now, a few numbers of these codes have been discovered as the longest found one has
only 13 elements which is not appropriate for practical usage in radar. It has been shown
that there is not any odd-length Barker code longer than 13. It has remained an open
question for even-length Barker codes, but it is assumed that there is not any even-length
one longer than 4.
Since, most practical applications require peak-to-sidelobe ratio much greater than 13, a
compilation of sequences with the lowest possible sidelobes at the longer length is needed.
Finding optimal sets of M phases (or codes) for different radar applications has kept radar
engineers busy from the early days of radar. The number of possibilities of generating phase
codes of length M is unlimited. The criteria for selecting a specific code are the resolution
properties of the resulting waveform, frequency spectrum, and the ease with which the
system can be implemented. Sometimes the design is even more complicated by using
different phase codes for the transmitted pulse and the reference pulse used at the receiver
(possibly even with different lengths). This can improve resolution at the expense of a
suboptimal signal-to-noise ratio.
The problem of finding a code that leads to a predetermined range–Doppler resolution is very
complicated. A manageable problem is finding a code with a good correlation function. So, it is
needed to search for codes whose autocorrelation functions have sidelobes as low as possible.
There are several parameters for measuring the sidelobe levels which are used in different
conditions. The matched-filter peak sidelobe level ratio is often used to characterize the level
of interference expected from point targets. For volume or surface clutter the interference
level is characterized by the matched-filter integrated sidelobe level ratio. The science (or
art) of designing radar signals is based on finding signals that yield a matched-filter
response that matches a given application. For example, if closely separated targets are to be
detected and distinguished in a low-SNR scenario, a radar signal having a matched-filter
response that exhibits a narrow mainlobe (the peak) and low sidelobes is required. The
mainlobe width and sidelobe level requirements are a function of the expected target
separation and expected target RCS difference.
Now, the problem can be written in mathematical format:
It is desired to find N-bit binary sequences whose PSLs or ISLs have the minimum value among
all 2N existing codes.
These codes are often called MPS1 codes. Finding such these codes is classified in
optimization problems and so far, no accurate and analytical solution has been found for it.
- General solutions for this optimization problem are not known (Lindner 1975).
- The search for the least autocorrelated binary sequence resembles the search of the needle
in the haystack (Militzer et al.,1998).
- There is no known analytical technique to construct sequences with minimum PSL (Deng
& Fan, 1999).
- Although one can identify minimum PSL sequences by conducting an exhaustive search,
no general-setting solution for identifying least autocorrelated binary sequences of
arbitrary bit length have been described in the literature (Ferrara, 2006).
8. A survey of the methodologies and inquiries

The search methods of finding binary sequences of desired length and PSLs are categorized
in two general classes:
- Exhaustive or Global
- Partial or Local
which have their own advantages and disadvantages and are used in accordance with
designer’s goal.
1 Minimum Peak Sidelobe

8.1 Global search

Finding MPS codes involves exhaustive computer search. The only disadvantage of this
method is that it takes long time, but it associates with these benefits:
1. It can reach to the absolute minimum, despite existing local ones.

2. It is able to find all optimum solutions.
3. It does not need to define any intermediate criteria and can directly search on the
base of main one.
So, if all optimum solutions are required, the only key is global search. Of course, there are
some limitations for this method too. Calculating of autocorrelation function for each
sequence needs N2/2 binary multiplications and N2/2 normal summations. So, total amount
of calculations needed to compute ACFs for all N-bit binary sequences are 2N.N2/2 binary
multiplications and the same number for normal summations. i.e.; if one bit is added to a
code, the needed search time will be doubled minimally. Therefore, for large N, elapsing
cpu time becomes very huge, unless special computers are used. Of course, this kind of
computers has its own limitations.
The above computations are related to simple full search. i.e., at first, all N-bit sequences
whose number is 2N are generated and their ACF are computed. Then, optimum codes are
selected.
8.2 Local search

The only advantage of this method is that it requires relatively short time. But, it suffers
from these difficulties:
1. Although it finds rather reasonable answers, it can not guarantee that it is able to
reach optimum ones. Since, it involves in local minimums.
2. Even if it can find some optimal answers, it is not determined whether it has found
all optimum ones or not.
3. The approximation methods usually consider a primary code, then determine next
codes by using an intermediate criterion. They find new code by using previous one
in this manner. The intermediate criterion must satisfy the main one. In different
approach, these criteria are referred to as different names like “evaluation functions”,
“fitness functions”, “error functions”, etc. In such methods, the intermediate criteria
lead to a better solution, but it can be told that up to now, no such criterion have
been found so that is able to navigate the search routine to the best solution. So,
definition of the intermediate criterion has important position in these methods.
However, even with the most powerful computers, enumeration algorithms are only able to
globally search for the best sequences with rather small length within a reasonable amount
of time. Therefore, for longer length effective optimization method should be adopted to
search sequences with good rather than the best aperiodic ACF properties.
8.3 History of scientific endeavors

In 1975, Lindner searched all binary sequences up to length 40 in simple-full search method
and by utilizing a fast special minicomputer constructed for extensive investigation of the
correlation functions of binary sequences. It took about 50 days. For error detection
purposes, inverse-amplitude and inverse-time sequences were not excluded. He
summarized his results in a table and introduced several good parameters for each code
length which can be used as selection criteria (Lindner, 1975). Although, he has searched all
N-bit codes till length 40, he has discarded inverse-time and inverse-amplitude sequences
in stating the number of optimum codes. Later on, he published his detailed results in a
restored version in 2006 which included almost all optimum found sequences. The obvious
important benefits of such this table can be stated in two items:
1. Before starting to search for optimum codes, by referring to this table, one can be
informed the optimum PSL of each length.
2. The number of optimum reported codes can be a good criterion to check the validity and
accuracy of full search algorithms.
In none-simple exhaustive method, by considering some concepts and characteristics of

autocorrelation function (usually allomorphic forms), only a portion of N-element code
configuration space is searched.
In 1986, Kerdoc et al. searched sequences of length 51 and found that their minimum
attainable PSL is equal to 3. As Lindner, they utilized a special-purpose digital hardware
designed for the task. Also, they tried to found longest binary codes which have PSLs equal
to 3, 4 and 5. They claimed that there is not any code longer than 51 with PSL equal to 3 and
so far, it has remained correct (Kerdoc et al., 1986).
In 1990, Cohen et al. searched all binary sequences from bit length 41 through 48. They
employed psl-preserver concept to reduce the search space. Also, they introduced an
innovative and recursive algorithm to search a smaller number of codes. They used PSL as
selection criterion and enumerated all MPSL codes in these lengths. They completed the
efforts of Kerdoc and his co-authors and noted that there are no length 49 or 50 biphase
codes with peak sidelobes of three or less. As Lindner, they have excluded inverse-time and
inverse-amplitude sequences in stating the number of optimum codes, but there is a
difference between these two routines. Cohen has not searched them at all (Cohen et al.,
1990).
In 1996, Mertens searched all binary sequences up to length 48 again, but his criterion was
minimum possible energy (Emin) or maximizing MF. He used psl-preserver concept to
reduce search time. He compiled a table of sequences with minimum energy and suggested
an asymptotic value for MF in large code length (Mertens, 1996).
In 1998, Militzer et al. introduced an evolutionary algorithm and tried to determine the most
suitable values for the optimization parameters of the strategy. They used MF as a criterion
and showed their highest values for some lengths of skew-symmetric sequences. They
compiled a table for comparison their found MF with the highest ones which others have
reported (Militzer et al., 1998).
In 1999, Deng and Fan presented another new evolutionary algorithm to generate sequences
with low PSL. They obtained a list of sequences of length 49-100 which were better than the
other letters in most lengths at that time (Deng & Fan, 1999).
In 2001, Coxson et al. searched binary sequences up to length 69. They exploited all psl-
preserver operations and therefore, were able to reduce search space more than before. They
introduced a new algorithm which is induced from Cohen’s one (Cohen et al., 1990) and
exhibited a new version for Lindner’s and Cohen’s tables. They enumerated the MPSL codes
of each length up to 48 again after excluding all allomorphic forms of a code. Also, in
accordance with Kerdoc claim, they tried to provide examples of PSL=4 codes for each
length between 49 and 69 but their examples are correct only up to 60 (Coxson et al., 2001).
1n 2004, Coxson et al. exhibited an efficient exhaustive algorithm which exploited all psl-
preserver operations too. Also, they introduced a fast method for computing the aperiodic
autocorrelation function. They established its ability by finding examples of PSL=4 codes for
each length from 61 through 70. Also they searched all 64-bit sequences and found all MPSL
codes and exhibited all balanced ones in a table. It is the longest power of two codes that
have been fully searched (Coxson & Russo, 2004).
Next, Levanon and Mozeson provided a summary of optimal PSLs for lengths up to 69
(Levanon & Mozeson, 2004).
In 2006, Ferrara described an integer programming method for generating low
autocorrelation binary codes at arbitrary bit lengths. He compared PSL values and MFs (for
bit length 71 through 100) of the sequences obtained with this method to the best literature-
based minimal-PSL sequences and compiled a table of best minimum-PSL binary sequences
for bit lengths 71 through 100. His record of length 74 was better than the other found codes
(Ferrara, 2006).
In 2008, Nunn and Coxson updated table of best minimum-PSL binary sequences from bit
lengths 71 through 105. For bit lengths 71 to 82, codes with PSL 4 were found. Under the
generally accepted assumption that no PSL-3 binary codes exist for lengths greater than 51,
they established, with near certainty, that the optimal PSL for lengths 71-82 is 4 by searching
until a single PSL-4 code is discovered for each of these lengths. PSL-5 codes were produced
for all lengths from 83 to 105 (Nunn & Coxson, 2008).
8.4 The authors’ efforts

The problem of finding best possible PSLs for binary sequences has triggered the authors’
interest form year 2005.
The first exhibited method combined several contents and gained its efficiency from Genetic
algorithm. It used some other orders of allomorphic forms which reduced search spaces
more than the ordinary algorithms which only use three psl-preserver concepts. Although it
was a partial search method, it does not involve in local minimum. Also, it could be
implemented by a simple scheme for partitioning and parallelizing the search by the fixed
upper bound on PSL. Since, it used genetic algorithm, it was possible to optimize found
codes by several factors simultaneously included in fitness function. Although the presented
result for this algorithm is not good (a 126-bit code with PSL 11), it seems to be able to find
better codes by using the better simulation (Amin & Bastani, 2006).
The other suggested approach belonged to Global search. It utilized a branch-and-bound
search strategy and PSL-preserver concept. Also, it used some rules and properties of
autocorrelations which reduces the configuration space more. A fast recursive method for
computing ACFs of binary sequences was presented. In addition, this algorithm could be
implemented in parallel mode. All these items lead to less elapsed cpu time and faster
execution. For example, it was fast enough to search all codes up to length 50 without
requiring any special computer or workstation. Also, this method could be easily modified
for local search (Amin & Bastani, 2007).
The recent published method was based on Local one and introduced a new innovative
evolutionary algorithm inspired by the Genetic algorithm (Amin & Bastani, 2006) and
Length-Increment one (Amin & Bastani, 2007). Same as other evolutionary algorithm, at first
it starts by an initial population and the better the initial codes, the faster the execution is.
But, it uses some rules and lemmas which results in not depending on primary population.
This proposed algorithm is fast enough to yield optimum or near optimal codes, especially
in long length codes. This algorithm was used to generate optimum codes longer than 200,
but in a test execution on length shorter than 100 in a matter of hours, the authors were
surprised by the results which were very better than others mentioned in literatures. It was
able to improve 11 records of previous best mentioned PSLs. So, the authors decided to
publish these found records without trying to improve them more. At that time, the paper of
Dr. Coxson (Nunn & Coxson, 2008) had been accepted but the authors could not gain access
to their results. Later on, I saw mentioned paper and its results. I think it is one of our
proposed method excellences that some of its fast found PSLs still have remained as
minimum as accessible ones. Also, a histogram for MPSL have been represented which help
to visualize the results and predict longer ones. By the histogram, it is expected, not proved,
that longer codes have better mainlobe to sidelobe ratio, thus better compression ratio
(Amin & Bastani, 2008).
9. Results
The summary of best found PSLs is exhibited in Table 2. These results are accurate for
lengths up to 82 and for upper ones, all reported values are relative and it is the Nunn’s and
Coxson’s opinion that many of these codes (of length 83-105) are themselves optimal for
PSL.
Codes with a peak sidelobe of 2 were reported for N ≤ 28 except ones which Barker codes
have been found. The MPS codes reported for 28 <N ≤ 48 and N = 51 have a sidelobe level of
3, and the MPS codes of length N = 50 and 52 ≤ N ≤ 82 have a sidelobe level of 4 and the best
found PSL for all upper length to 105 is 5.
As all codes of length 1-48 and 64 have been searched exhaustively and all MPSL codes of
these lengths have been found, it is possible to mention the number of best optimal existing
codes. These numbers are stated after excluding all allomorphic forms of codes.
Table 2. gives a single MPS code for each length. For M ≤ 48 the listed codes are those that
have, from all those with minimum peak sidelobe, the minimum integrated sidelobe
(Levanon & Mozeson, 2004).
Table 2. lists codes in Hexadecimal. Each hexadecimal digit represents four binary bits, and
the convention is made that, upon base conversion, any unnecessary binary digits are
removed from the left side of the sequence.
Length PSL Example Number

2 1 3 1
3 1 6 1
4 1 E 1
5 1 1D 1
6 2 34 4
7 1 72 1
8 2 97 8
9 2 0D7 10
10 2 167 5
11 1 712 1
12 2 977 16
13 1 1F35 1
14 2 1483 9
15 2 182B 13
16 2 6877 10
17 2 0774B 4
18 2 190F5 2
19 2 5BB8F 1
20 2 5181B 3
21 2 16BB83 3
22 3 0E6D5F 378
23 3 38FD49 515
24 3 64AFE3 858
25 2 12540E7 1
26 3 2380AD9 242
27 3 25BBB87 388
28 2 8F1112D 2
29 3 164A80E7 283
30 3 2315240F 86
31 3 2A498C0F 251
32 3 01E5AACC 422
33 3 0CCAA587F 139
34 3 333FE1A55 51
35 3 00796AB33 111
36 3 3314A083E 161
37 3 0574276F9E 52
38 3 003C34AA66 17
39 3 13350BEF3C 30
40 3 2223DC3A5A 57
41 3 038EA520364 15
42 3 04447B874B4 4
43 3 005B2ACCE1C 12
44 3 0FECECB2AD7 15
45 3 02AF0CC6DBF6 4
46 3 03C0CF7B6556 1
47 3 069A7E851988 1
48 3 156B61E64FF3 4
49 4 012ABEC79E46F -
50 4 025863ABC266F -
51 3 0E3F88C89524B -
52 4 0945AE0F3246F -
53 4 0132AA7F8D2C6F -
54 4 0266A2814B3C6F -
55 4 04C26AA1E3246F -
56 4 099BAACB47BC6F -
57 4 01268A8ED623C6F -
58 4 023CE545C9ED66F -
59 4 049D38128A1DC6F -
60 4 0AB8DF0C973252F -
61 4 005B44C4C79EA350 -
62 4 002D66634CB07450 -
63 4 04CF5A2471657C6F -
64 4 4090A2E9E63237C2 1859
65 4 002DC0B0D9BCE5450 -
66 4 0069B454739F12B42 -
67 4 20506C9AB1E909CC2 -
68 4 009E49E3662A8EA50 -
69 4 026FDB09A83A118E15 -
70 4 1A133B4E3093EDD57E -
71 4 63383AB6B452ED93FE -
72 4 E4CD5AF0D054433D82 -
73 4 1B66B26359C3E2BC00A -
74 4 36DDBED681F98C70EAE -
75 4 6399C983D03EFDB556D -
76 4 DB69891118E2C2A1FA0 -
77 4 1961AE251DC950FDDBF4 -
78 4 328B457F0461E4ED7B73 -
79 4 76CF68F327438AC6FA80 -
80 4 CE43C8D986ED429F7D75 -
81 4 0E3C32FA1FEFD2519AB32 -
82 4 3CB25D380CE3B7765695F -
83 5 711763AE7DBB8482D3A5A -
84 5 CE79CCCDB6003C1E95AAA -
85 5 19900199463E51E8B4B574 -
86 5 3603FB659181A2A52A38C7 -
87 5 7F7184F04F4E5E4D9B56AA -
88 5 D54A9326C2C686F86F3880 -
89 5 180E09434E1BBC44ACDAC8A -
90 5 3326D87C3A91DA8AFA84211 -
91 5 77F80E632661C3459492A55 -
92 5 CC6181859D9244A5EAA87F0 -
93 5 187B2ECB802FB4F56BCCECE5 -
94 5 319D9676CAFEADD68825F878 -
95 5 69566B2ACCC8BC3CE0DE0005 -
96 5 CF963FD09B1381657A8A098E -
97 5 1A843DC410898B2D3AE8FC362 -
98 5 30E05C18A1525596DCCE600DF -
99 5 72E6DB6A75E6A9E81F0846777 -
100 5 DF490FFB1F8390A54E3CD9AAE -
101 5 DF490FFB1F8390A54E3CD9AAE -
102 5 2945A4F11CE44FF664850D182A -
103 5 77FAAB2C6E065AC4BE18F274CB -
104 5 E568ED4982F9660EBA2F611184 -
105 5 1C6387FF5DA4FA325C895958DC5 -
Table 2. Best-known binary codes.
It seems that for any peak sidelobe level there is a limit of the maximal value of N for which
a binary sequence with that sidelobe level exists. Now, it is possible to update Kerdoc’s table
in Table 3.
PSL Code Length (N) Example

1 13 1F35
2 28 DA44478
3 51 71C077376ADB4
4 82 3CB25D380CE3B7765695F
5 105 1C6387FF5DA4FA325C895958DC5
Table 3. Longest-known binary codes for PSLs from one to five.
MPSL histogram for binary sequences up to length 105 is drawn in Figure 5. This histogram
shows that, the longer the code, the smaller the PSL ratio and the best record is related to
length. So, it is expected, not proved, that longer codes have better mainlobe to peak
sidelobe ratio, thus better compression characteristic.
Fig. 5. MPSL Histogram in dB (absolute and relative values)

10. References
Amin_Nasrabadi, M.; Bastani, M. H. (2006), A new approach for long low autocorrelation
binary sequence problem using genetic algorithm, Proceedings of CIE International
Conference on Radar, Vol. II, pp. 1898-1900, 0-7803-9582-4, China, October 2006,
Shanghai.
Amin_Nasrabadi, M.; Bastani, M. H. (2007), Exhaustive search for long low autocorrelation
binary codes using Length-Increment algorithm, Proceedings of IET International
Conference on Radar Systems, pp. 1-4, 978-0-86341-848-8, Edinburgh, UK, October
2007, Edinburgh.
Amin_Nasrabadi, M.; Bastani, M. H. (2008), Best known PSLs for binary sequences from bit
length 71 through 100. Proceedings of International Symposium on Telecommunications,
pp. 697-700, 978-1-4244-2751-2, Iran, August 2008, Tehran.
Ackroyd, M. H.; Ghani, F. (1973). Optimum mismatched filters for sidelobe suppression.
IEEE Transactions on Aerospace and Electronic Systems, Vol. AES-9, No. 2, (March
1973) pp. 214-218, 0018-9251.
Baden, J. M.; Cohen, M. N. (1990). Optimal peak sidelobe filters for biphase pulse
compression, Proceedings of IEEE International Radar Conference, pp. 249-252,
Arlington, VA, USA, May 1990, Arlington.
Barton, D. K.; Leonov, S. A. (1998). Radar technology encyclopedia, Artech House, 0-89006-893-
3, Boston.
Cohen, M. N.; Fox, M. R. & Baden, J. M. (1990). Minimum peak sidelobe pulse compression
codes, Proceedings of the IEEE International Radar Conference, pp. 633-638, Arlington,
VA, USA, May 1990, Arlington.
Cohen, M. N. (1991). An overview of high range resolution radar techniques, Proceedings of
National Telesystems Conference, pp. 107-115, 0-7803-0062-9, Atlanta, GA, USA,
March 1991, Atlanta.
Coxson, G. E.; Hirschel, A. & Cohen, M. N. (2001). New results on minimum-PSL binary
codes, Proceedings of IEEE National Radar Conference, pp. 153-156, 0-7803-6707-3,
Atlanta, GA, USA, May 2001, Atlanta.
Coxson, G. E.; Russo, J. (2004). Efficient exhaustive search for optimal-peak-sidelobe binary
codes, Proceedings of the IEEE Radar Conference, pp. 438-443, 0-7803-8234-X,
Philadelphia, PA, April 2004, Philadelphia.
Deng, X.; Fan, P. (1999). New binary sequences with good aperiodic autocorrelations
obtained by evolutionary algorithm. IEEE Communications Letters, Vol. 3, No. 10,
(October 1999) pp. 288-290, 1089-7798.
Farnett, E. C.; Stevens, G. H. (1991). Pulse compression radar, In: Radar Handbook, Skolnik,
M. I., (Ed.), pp. 10.1-10.39, McGraw-Hill, 0-07-057913-X, Singapore.
Ferrara, M. (2006). Near-optimal peak-sidelobe binary codes. Proceedings of IEEE National
Radar Conference, pp. 400-403, 0-7803-9496-8, Syracuse, NY, Aprill 2006, Syracuse.
Golay, M. J. E. (1977). Sieves for low autocorrelation binary sequences. IEEE Transactions on
Information Theory, Vol. IT-23, No. 1, (January 1977) pp. 43-51, 0018-9448.
Kerdoc, A. M.; Mayer, R. & Bass, D. (1986). Longest binary pulse compression codes with
given peak sidelobe levels, Proceedings of the IEEE, Vol. 74, No. 2, pp. 366-366,
February 1986.
Levanon, N.; Getz, B. (1994). Comparison between linear FM and phase-coded CW radars.
IEE Proceedings Radar, Sonar & Navigation, Vol. 141, No. 4, (Augest 1994) pp. 230-
240, 1350-2395.
Levanon, N.; Mozeson, E. (2004). Radar signals, John Wiley & Sons, 0-471-47378-2, New
Jersey.
Lindner, J. (1975). Binary sequences up to length 40 with best possible autocorrelation
function. Electronics Letters, Vol. 11, No. 21, (October 1975) pp. 507-507, 0013-5194.
Mahafza, B. R.; Alabama, H. (2000). Radar systems & analysis and design using Matlab,
Chapman & Hall/CRC, 1-58488-182-8, USA.
Mertens, S. (1996). Exhaustive search for low-autocorrelation binary sequences. Journal of
Physics A: Mathematical and General, Vol. 29, No. 18, (September 1996) pp. L473-
L481, 0305-4470.
Militzer, B.; Zamparelli, M. & Beule, D. (1998). Evolutionary search for low autocorrelated
binary sequences. IEEE Transactions on Evolutionary Computation, Vol. 2, No. 1,
(Aprill 1998) pp. 34-39, 1089-778X.
Nathanson, F. E.; Cohen, M. N. (1999). Frequency-modulated pulse compression waveforms,
In: Radar design principles: signal processing and the environment, Nathanson, F. E.;
Reilly, J. P. & Cohen, M. N., (Ed.), pp. 583-634, SciTech, 9781891121098, New Jersey.
Nunn, C. J.; Coxson, G. E. (2008). Best-known autocorrelation peak sidelobe levels for binary
codes of length 71 to 105. IEEE Transactions on Aerospace and Electronic Systems, Vol.
44, No. 1, (January 2008) pp. 392-395, 0018-9251.
Rihaczek, A.; Golden, R. (1971). Range sidelobe suppression for Barker codes. IEEE
Transactions on Aerospace and Electronic systems, Vol. AES-7, No. 6, (November 1971)
pp.1087-1092, 0018-9251.
Skolnik, M. I. (2001). Introduction to radar systems, Tata McGraw-Hill, 0-07-044533-8, New
Dehli.
Virtual Multicast 63
0
4
Virtual Multicast
Petr Holub and Eva Hladká
Masaryk University and CESNET z. s. p. o.
Czech Republic
1. Introduction
The ability of the Internet to facilitate collaboration leads to widespread use of various video-
conferencing and more advanced collaborative environments. As a result, synchronous multi-
media transmissions have become more common. Various communication patterns emerged:
from many-to-many low-bandwidth streams for large scale collaboration over slow links
to few-to-few extreme-bandwidth streams as seen in collaboration based on high-definition
(HD) Holub et al. (2006), Jo et al. (2006) or even post-HD video Shimizu et al. (2006). These
applications require Internet to become more active, the classical passive transmission service
is no longer sufficient.
Multimedia streams are processed within the network, allowing, e.g., to establish a collaborat-
ing group where most members are connected to the high-bandwidth network links while a
minority has rather limited connection. If the network is capable of processing—compressing,
down-sampling, etc.—the data at the appropriate nodes (where the high and low throughput
network links interconnect), the communication quality should not be reduced to the lowest
common throughput denominator. The network must be able to support complex commu-
nication patterns and to process data internally. Robustness and failure resilience is another
area, where more support at the network level is expected. While classical transport proto-
cols like TCP support reliable data transmission, they are not appropriate for synchronous
multimedia environment, where delays are unacceptable. It may be undesirable to wait for a
timeout and then ask for a datagram retransmission, the network and applications themselves
must be able to detect and immediately mitigate any data corruption or loss. Up to now, new
requirements were served by different infrastructures tailored for a specific purpose. Nowa-
days, we need to merge them together in a network that uses packet transmission as its basis
protocol—to do this successfully, new models, approaches, and techniques are necessary.
The theoretical model of the virtual multicast naturally follows from graph-based model of
computer networks. The graph model of computer network can easily be extended to multi-
graphs, which allow multiple line to connect any individual nodes. Although most computer
networks are bi-directional, working a semi-duplex or full duplex regime, orientation can be
added for explicit description of direction of flows (multiple edges used to represent the bi-
directionality). As another step, we can add labels to the edges, representing some important
properties like throughput or latency of each link. Labels on nodes can denote their properties,
like different capabilities, latency of passing (bridging) data between edges of the node (the
internal latency), size of internal buffers, etc. We can also speak about internal network, which
is a part of the graph without any leaf node. It is also easy to identify end—i.e. leaf—elements.
Such a model is appropriate to study most usual flow patterns in contemporary computer
networks, namely the sender–receiver one. In this case, we have one node sending and ex-
actly one node receiving a particular data flow. The basic network problem is finding a route
between the communicating nodes, additional constraint is to guarantee available bandwidth
and eventually other properties like overall latency or jitter. The route is usually the one com-
posed from the smallest number of edges—the so called shortest path—but in some case any
path could fit—this is the case, e.g., in the interdomain routing. The mechanism for creat-
ing a route can work on a flow basis—we speak about connection oriented networks—or on
a datagram basis—the case of IP network. In the later case stability of the route is becom-
ing additional important parameter that could influence the behavior of the whole flow (e.g.,
there is no reordering of datagrams within a flow in the connection oriented networks). Each
path has one sending, one receiving, and zero or several internal nodes that are responsible for
forwarding data.
However, as the networks were exposed to larger number of more sophisticated applications,
more complex communicating patters emerged. The first one is a multicast, with still one
sender but multitude of receivers. A simple extension is a communicating mesh, where ev-
ery member of such a communicating group (the multicast group) could become a sender.
Yet more complex communication patterns are seen in the peer to peer networks, where we
may have multiple partially overlaid multicast groups communicating in parallel, it may use
flooding, different cases of wave communication patterns, etc.
All the more complex communication patterns can still be expressed in our simple graph
model using the sender–receiver paradigm. Multicast can be modeled by a set of sender to
receiveri flows, but to express it correctly some kind of coordination (synchronicity) must be
added to the model (data delivery to all receivers is expected to happen at the same time).
Also, even in networks with unlimited bandwidth the simultaneous sending of all streams by
just one element stress it above the optimal level (reducing efficiency of the communication
scenario).
To deal with such complex communicating patterns more effectively, we have to extend our
routing algorithm to find not paths, but whole subgraphs of the original graph. Flows going
through such subgraph are more efficient than collection of individual send–receiver flows.
The subgraphs represent overlay networks, that are specialized to transfer the particular flow
pattern in the most efficient way.
When mapped back to the underlying network, the subgraphs extend the requirements on
the internal path nodes. Simple forwarding (taking data from one link and sending them
to another) is no longer sufficient, data must be duplicated and further processed to fit the
communicating subgraph (overlay network) requirements. At the theoretical level this is just
a simple extension, but propagating it back to the network proved to be very difficult, if not
impossible work.
As an example, let’s briefly discuss the IP multicast. It has been established as a family of
protocols at the beginning of 80s in the last century Cheriton & Deering (1985); Deering &
Cheriton (1990). IP multicast is based on a family of multicast routing protocols (how to cre-
ate the appropriate subgraph of the network) and its implementation requires support at each
network element both for routing and also for multicast forwarding. The IP multicast includes
nodes that do datagram duplication—they must be able to forward incoming data to two or
even more output links. IP multicast does not guarantee delivery of datagrams, does not pro-
vide any feedback to sender, it is in fact very simple extended forwarding scheme. All vendors
of routers officially support multicast, yet it is not available on large parts of the Internet and
the situation is not expected to change in the future. Although simple, multicast still can in-
terfere with the basic sender–receiver communication patterns, imposes more load on routers
(duplication is more complicated than simple forwarding) and the multicast routing protocols
can introduce instability into the basic routing. As the result, multicast may not work prop-
erly or could be switched off by network administrators if they suspect it to be the cause of
a problem they have with the network1 Diot et al. (2000); Dressler (2003a;b); El-Sayed et al.
(2003).
If the IP multicast situation is far from satisfactory, what we can anticipate with more complex
extensions, where data have to be not only transmitted but also processed during transmis-
sion?
We must change the paradigm—instead of expecting the underlying network to provide all
the advanced functionality and increasing complexity above sustainable levels, more isola-
tion and independent deployment of support for complex communication (and data processing)
patterns is the possible answer. The isolation is provided by the overlay networks, that take
care of all the new functionality by themselves. The independence of deployment is achieved
through the user empowered approach. The overlay networks are constructed and managed (of-
ten just temporarily) by their own users, without any need for specific support from network
and its administrators.
Several years ago we started to build a network environment based on the user-empowered
approach for transport and processing data in IP networks. We used the concept of active
networks and designed and developed an Active Element—a programmable network node
designed for synchronous data distribution and processing, configurable without administra-
tor’s right—and used it as the basic building block for construction of complex communica-
tion patterns.
The initial phase of our research was influenced by the network-centric view. We designed
an active router Hladká & Salvet (2001a), an extension of the classical router that allows users
to define their own processing over individual data streams. The active network paradigm
which introduced the active network elements, opened also the door to more user oriented
approach. The active routers (and similar active network elements) are expected to be setup
and operated by system administrators, with users “only” injecting smaller or larger programs
to process their data within the network. Although the concept of active networks has been
proved to provide the new functionality necessary to fulfill new requirements of data trans-
mission and processing within the network, the whole idea collided with the conservative
approach of network vendors and administrators. As the multicast experience demonstrated,
it is very difficult to introduce new properties as they can interact in unpredictable way with
the simpler, previously introduced protocols. Also, security concerns could not be overem-
phasized. A network programmable by end users is ripe for being taken completely by a
hacker; this risk seen too high to be outweighted by the potential of new features.
At the same time as the active networks were developed, another paradigm that proved the
value in giving control to end users emerged—the peer to peer networks. They completely
1 To further illustrate this problem, we have performed a quick survey of Internet2 Bigvideo group mail-
ing list archive (https://mail.internet2.edu/wws/arc/bigvideo/). This list was in operation
from May 2003 to May 2006. It focused on education and problem solving for users of high-end video
technologies in advanced academical networks like the one operated by Internet2. The list was not lim-
ited to Internet2 community and there was a significant international contribution. As a majority of the
advanced video tools use multicast, 212 of total 625 messages, i.e., 34% was spent on multicast testing
and debugging.
abandon the network-centric view, implementing in fact many already available network pro-
tocols once again, providing complete orthogonality (and independence) on the underlying
network. The peer to peer networks are classical overlay networks, taking as granted only lim-
ited number of very simple properties of the underlying network and providing all the higher
level functionality—searching, routing, etc.—by themselves.
However, the complete independence on the underlying network leads to inefficiency. The
classical peer to peer networks could place their nodes only on the periphery of the network,
where the users’ stations are connected. The data distribution pattern required by the content
(which the peer to peer network understood) may fit very poorly into the actual underlying
network topology, overloading some lines while leaving other unused. Also, reliability of
the peer to peer network is usually based on an overwhelming redundancy, when the same
data is distributed, processed, and stored by many nodes—again a clear contradiction to the
network-centric approach where the efficiency (the cost of the infrastructure) is one of the
ruling paradigms.
We can see that the network-centric approach is highly efficient, but very slow in adopting
new features and rather unfriendly to users. On the other hand, pure user-centric overlay
approach, as represented by basic peer to peer networks is very inefficient (consuming more
resources than needed in the optimal case), but it is able to introduce new features fast and can
provide exactly the services the users are looking for. Another reason for that huge success
is also their single purpose—the peer to peer networks are not trying to solve all the users’
requirements, they focus on one service or just a small set of similar interconnected services.
Is it possible to take the positive from both approaches and leave out their negatives? Several
years ago we decided to try this combination, moving from the network-centric to the user-
centric approach, but not abandoning the network orientation completely. We extended the
active router model to fit into the user-centric paradigm. The original active router and its
implementation was based on Unix operating system and exploited both the kernel and user
components. Its installation and deployment thus required system administrator’s privileges
that ordinary user may not have. As the next step, we completely redesigned the active router
to become Active Element (AE), working in the user space of any operating system only. We
obtained a fully user controlled element, that can be installed on any machine user has access
to, without any specific privileges (e.g., on a server that is more strategically placed within
the network than end user desktop machine). However, the AE design still followed basic
network-centric pattern, being an evolutionary successor of active router, and thus became
a keystone for the distribution and processing infrastructure, not a node in a peer to peer
network. We still differentiate between an infrastructure and clients, but we put both into
users’ hands.
The user controlled Active Element is a very strong and flexible component to build different
distribution schemes. We started with an infrastructure for virtual multicast. We used this
infrastructure to study properties of the serial communication schema for group synchronous
communication instead of the parallel communication model of the native multicast. While
we had clearly demonstrated its advantages, especially in the area of security and reliability,
the limited scalability remained the major disadvantage and it became our natural next re-
search target. Instead of using just a single AE to do all the processing and distribution, we
designed a network of AEs with distinct control and data planes. This separation allowed us
to use the peer to peer principles at the control plane, taking advantage of the properties of
peer to peer networks like robustness and very high scalability. The inherent low efficiency
of peer to peer networks does not play significant role, as the amount of control data is al-
ways limited. The result is an easily configurable and fault tolerant network of AEs with a
reasonably high throughput capabilities.
However, the scalability is not one dimensional issue. While the network of AEs addressed the
scalability in terms of number of clients supported, very high quality video (e.g., that used in
the cinema theaters) generates so huge amount of data that may not be processed by a single
AE. Therefore, we extended our work on scalability to increase the AE processing capacity
through their internal parallelization. The parallelized AE runs on a cluster with fast internal
interconnect and is capable of processing in near real time even 10 Gbps data stream.
All this research and development would not be complete without an actual deployment.
Putting the AEs and their networks into production use provided a very valuable continuous
feedback on their design while experimentally testing their properties. The AEs were used
to build an infrastructure for collaborative environment used by several geographically dis-
tributed groups of researchers. Requirements from these groups initiated further research into
support of advanced communication and collaboration features like moderating or subgroup-
ing. The AEs started to play a role of directly controlled user tool to support these advanced
properties. This confirmed the strength of the general concept of user empowered building
blocks for data processing and distribution networks.
In another environment we used the idea of overlay network with AEs capable to provide new
functionality for the stereoscopic video streams synchronization. A simple software imple-
mentation running on commodity hardware is able to synchronize two streams of stereoscopic
digital video (DV, 25 Mbps) format successfully even when the original streams are highly de-
synchronized. The penalty of the synchronization is increased latency, as the “faster” stream
must wait for data in the slower stream, plus some processing latency is added to the fi-
nal perceived delay. While this delay may be problematic in interactive implementation, we
demonstrated that the AE-based synchronization element can be easily used for synchronized
unidirectional stereoscopic streaming to multiple end users even in highly adverse and desyn-
chronizing network conditions Hladká et al. (2005). While the stereoscopic streaming may not
be too common, this concept is usable for synchronization of stereo or multichannel (e.g., 5.1)
audio streams or for synchronization of separately sent audio and video streams.
The real strength of the AEs and the whole concept of controllable overlay networks is demon-
strated in the multi-point High Definition (HD) video distribution. If the HD video is to be
used for a synchronous collaborative environment, uncompressed streams must be sent over
the network. However, the required throughput of 1.5 Gbps per each stream was too high to
be sent reliably over a native multicast in heterogeneous network over multiple administrative
domains (even if it was available). The optimized AEs are able to replicate even such high de-
manding streams in near real time and were used to build infrastructure that supported one
of the world first multipoint videoconferences using uncompressed HD video Holub et al.
(2006). Later, improved AEs grouped into a network became key infrastructure for a virtual
classroom that ran full semester and connected 6 sites on two continents Matyska, Hladká &
Holub (2007). The Active Element network processed up to 18 Gbps bi-directional bandwidth,
fully confirming the usability of the AE design.
In this chapter, we present several classes of solutions following our long term research in this
field. The simplest solution to user-empowered data distribution and processing is a central
Active Element (AE) described briefly in Section 2, which is a programmable modular active
element, that can be run in the network easily without requiring any administrative privi-
leges. The AE distributes and optionally also processes the incoming data, which allows for
unique per-user processing capabilities—something that is impossible to do with traditional
data distribution systems like multicast. As any centralized solution, it has its positives and
shortcomings: while it is easy to setup and deploy, it has limited robustness and scalability,
both with respect to number of streams and the bandwidth of a single stream. When more
clients are collaborating or when higher robustness is needed, the AEs may be deployed as
static or dynamic self-organizing AE networks shown in Section 2.1. This field has been stud-
ied thoroughly from the data distribution efficiency and robustness point of view by many
groups previously and the most relevant body of work is referenced in Section 2.1. Our view
here is, however, more general, focusing not only on mere multicast-like data distribution,
but also on the possibilities enabled by additional data processing, operation in adverse net-
working environments, self-organization, etc. Another step forward needs to be taken when
bandwidth of a single stream exceeds capacity of any single AE in the AE network. Utiliz-
ing properties of real-time multimedia applications and data distribution protocols, we have
designed a distributed AE described in Section 2.2 that can be deployed on tightly coupled
clusters—but this solution becomes very complex when not only the data distribution but
also data processing is required. We demonstrate applications which have been built on top
of these technologies for synchronous data distribution and processing in Section 3. Related
work is summarized in Section 4 and we conclude with some remarks on directions for future
research in Section 5.
2. Active Elements
The Active Element (AE) Hladká et al. (2004) is a programmable element designed for syn-
chronous data distribution and processing while minimizing the latency of the distribution.
The word “reflector” is also being used in this context, which only refers to data distribution
capabilities. Since our approach is far more general and close to idea of active networks, we
have resorted to using the Active Element name. The architecture of the AE is flexible enough
to allow implementation of required features while leaving space for easy extensions. If the
data is sent to all the listening clients and all the clients are also actively sending, which is a
standard scenario for collaborative group of participants, the number of data copies is equal
to the number of the clients, and the limiting outbound traffic grows with n(n − 1), where n
is the number of sending clients.
From general point of view the AE is a user-controlled modular programmable router working
on the application layer. It runs entirely in user-space of the underlying operating system and
thus it works without the need for administrative privileges on the host computer. AEs are
based on our active router concept described in Hladká & Salvet (2001b), building on the same
principles of modularity, but adding the user-empowered approach. The AE architecture is is
shown in the Figure 1.
Data processing architecture.
Data routing and processing part of the AE comprises network listeners, shared memory, a packet
classifier, a processor scheduler, number of processors, and a packet scheduler/sender.
The network listeners are bound to one UDP port each. When a packet arrives to the listener
it places the packet into the shared memory and adds reference to a to-be-processed queue. The
packet classifier then reads the packets from that queue and determines a path of the data
through the processor modules. It also checks with routing AAA module whether the packet
is allowed or not (in the later case it simply drops that packet and creates event that may be
logged). Zero-copy processing is used in all simple processors (packet filters), minimizing
processing overhead (and thus packet delay). E. g. for simple data multiplication, the data
Reflector Kernel Processor 1
resource management packet

processor
messaging administrative AAA
interface 1 session
management management
session
management
Processor n
messaging
packet
interface n
routing processor processor
AAA scheduler
session
management
network packet
listener 1 classifier
packet
shared
network scheduler
memory /sender
listener n
data flow control information

Fig. 1. Architecture of Active Element with its individual modules and interactions.
are only referenced multiple times in the packet scheduler/sender queue before they are ac-
tually being sent. Only the more complex modules may require processing that is impossible
without use of packet copies.
The session management module follows the processors and fills the distribution list of the
target addresses. The filling step can be omitted if data passed through a special processor
that filled the distribution list structure and marked data attribute appropriately (this allows
client-specific processing). Processor can also poll session management module to obtain up to
date list of clients for specified session. Session management module also takes care of adding
new clients to the session as well as removing inactive (stale) ones. There are two ways of
adding clients for a session at the AE: implicit mode and explicit mode. In the implicit mode,
when new client sends packets for the first time, session management module adds client to
the distribution list (data from forbidden client has already been dropped by packet classifier).
This mechanism is designed to work with the multimedia systems like MBone Tools suite. The
explicit mode requires some specific action to be taken by the user or application to register for
the session at the AE, be it RTSP protocol Schulzrinne et al. (1998) or direct interaction through
one of native messaging interfaces of the AE. Information about the last activity of a client is
also maintained by the session module and is used for pruning stale clients periodically in the
implicit mode. Even when distribution list is not filled by the session management module,
packets must pass through it to allow addition of new clients and removal of stale ones.
When the packet targets are determined by the router processor a reference to the packet is
put into the to-be-sent queue. Then the packet scheduler/sender picks up packets from that
queue, schedules them for transmission, and finally sends them to the network. Per client
packet scheduling can also be used for e. g. client specific traffic shaping.
The processor scheduler is not only responsible for the processors scheduling but it also takes
care of start-up and (possibly forced) shutdown of processors which can be controlled via
administrative interface of the AE. It checks resource limits with routing AAA module while
scheduling and provides back some statistics for accounting purposes.
Architecture of management.
Communication with the AE from the administrative point of view is provided using messag-
ing interfaces, management module, and administrative AAA module of the AE. Commands for the
management module are written in a specific message language.
The administrative part of the AE can be accessed via secure messaging channels such as
HTTP with SSL/TLS encrypted transport or SOAP with GSI support. The user can authenti-
cate using various authentication procedures, e. g., combination of login and password, Ker-
beros ticket, or X.509 certificate. Authorization uses access control lists (ACLs) and is per-
formed on a per-command basis. Authentication, authorization, and accounting for the ad-
ministrative section of the AE is provided by an administrative AAA module. Each of these
interfaces unwraps the message if necessary and passes it to the management module. A
message language for communication with the management module is called Reflector Ad-
ministration Protocol (RAP) described in Denemark et al. (2003).
Prototype implementation and performance evaluation.
In order to evaluate the behavior of AE on recent high-performance infrastructure, we have
set up a testbed comprising sender and receiver machines (each 2× AMD Opteron 2.4 GHz,
2 GB memory, Linux 2.6.9 SMP kernel) and a machine running the AE (2× dual core Intel
Xeon 3.0 GHz, 8 GB memory, Linux 2.6.19 SMP kernel). The sender machine was equipped
with Chelsio T110 and both the receiver and the AE machine with Myricom Myri-10GE NICs.
All the three machines were connected to a 10GE Cisco 6506 switch.
The performance was measured using two implementations of the AE: the full featured com-
plex version described above2 (denoted as AE) and a high-performance simplified version
including only one receiving and one sending thread, which was designed for HD video dis-
tribution Holub et al. (2006) (thus denoted as HD-AE). The performance is summarized in
Fig. 2. The results indicate that even the more complex version is capable of distributing the
uncompressed HD video for up to 4 participants when Jumbo frames are used, which is nec-
essary for this application anyway.
2.1 AE Networks
As the scalability of AE is limited especially with respect to the number of data streams
(clients), the concept of single AE has been extended to a network of AE Holub et al.
(2005) while preserving its processing capabilities through modularity and retaining the user-
empowered approach to maximum extent. Its architecture features separated data distribu-
tion plane and control plane: while the data distribution is optimized for maximum perfor-
mance and minimum latency, the control plane has to provide maximum robustness even at
2 The AE concept has been implemented in C language for Unix-like operating systems under code name
RUM2. http://miro.cesnet.cz/software/software.cz.html.
20
forwarding
2 replicas
3 replicas
4 replicas
15
packet loss [%]
10
0
0 1000 2000 3000 4000 5000
bandwidth [Mbps]
20
forwarding
2 replicas
3 replicas
15 4 replicas
packet loss [%]
10
0
0 1500 3000 4500 6000 7500 9000
bandwidth [Mbps]
AE HD-AE
replica(s)
1500 B 8500 B 1500 B 8500 B
1 1450 4820 2700 6100
2 1000 2975 1660 3490
3 775 2170 1140 2510
4 600 1725 825 1960
Fig. 2. Performance of the modular full-featured AE compared to highly simplified version
optimized for HD video distribution (HD-AE). Stabilization in the upper right graph is be-
cause of sender card saturation above 6.5 Gbps. The table below the graphs shows maximum
stream bandwidth [Mbps] distributed with less than 0.1 % packet loss for both standard and
Jumbo frame sizes.
Network Management Network Information Service
Messaging
Kernel Processors
Modules
Network Packet
Shared Memory
Listeners Scheduler/Sender
data flow control information

Fig. 3. Architecture of Active Element with Network Management and Network Information
Service modules.
cost of performance. The control plane is responsible for management actions of the AE net-
work like monitoring, reconstruction of the network after node or network link failure and has
to survive all the perturbations. Thus we have chosen a P2P architecture of the control plane
which exhibits very strong resilience. The data plane on the other hand may be dynamically
rebuilt based on the information from the control plane; even the data distribution model may
change.
The AEs has networking capability, i.e., inter-element communication. The network manage-
ment is implemented via two modules dynamically linked to the AE: Network Management
(NM) and Network Information Service (NIS) as shown in Fig. 3. The NM takes care of build-
ing and managing the network of AEs, joining new content groups and leaving old ones, and
reorganizing the network in case of link failure. The NIS gathers and publishes information
about the specific AE (e.g., available network and processing capacity), about the network
of AEs, about properties important for synchronous multimedia distribution (e.g., pairwise
one-way delay, RTT, estimated link capacity), and also information on content and available
formats distributed by the network.
The data distribution plane is designed using loadable plug-ins to enable incorporating vari-
ous distribution models. A number of suitable models has been proposed previously by many
independent groups in the past, most of which fall into one of the two categories: (1) mesh
first distribution models like Narada Chu et al. (2000), Delaunay triangulation Liebeherr &
Nahas (2001), Bayeux Zhuang et al. (2001), and (2) tree first models like YOID Francis (2000),
TBCP Mathy et al. (2001), HMTP Zhang et al. (2002), SHDC Mathy et al. (2002), NICE Banerjee
et al. (2002), Overcast Jannotti et al. (2000), ZIGZAG Tran et al. (2003). Some other models may
also be found in El-Sayed et al. (2003); Li & Shin (2002).
Given the data processing capabilities of the AE, the usefulness of AE networks goes beyond
pure data distribution models. AEs in the network can be specialized in performing various
transformation of the data based on user request (e.g., AE running on host with enough CPU
power and sufficient network capacity can perform transformation of the data from high-
bitrate to low-bitrate). However, combinations of data distribution and data processing makes
makes scheduling problem particularly hard and first approaches have only been studied
recently using self-organizing CoUniverse platform Liška & Holub (2009).
Prototype implementation of the AE networks with P2P control plane based on JXTA-C3 has
been demonstrated in Procházka et al. (2005). A few simple optimizations to default JXTA
settings improved the performance significantly for synchronous applications with a limited
number of participants where down-time minimization is required despite increasing com-
munication overhead, thus making it suitable control-plane middleware.
The AE network is also designed to facilitate communication in adverse networking environ-
ments, i.e., environments where the network communication is obstructed by firewalls, net-
work address translators (NATs) and proxy servers. The data may be tunneled over TCP in-
stead of usual UDP, it may even mimic using HTTP and tunnel the data over HTTP proxy. The
AE may also be augmented by employing a VPN Holub et al. (2007) such as OpenVPN4 , which
boosts pervasivity, as it allows even tunneling through HTTP and SOCKS proxy servers. VPN
also enables deployment of strong authentication and very secure data encryption protocols.
Similar approaches have also been described in Alchaal et al. (2002). The solution that in-
tegrates these features directly into AE modules Bouček (2002); Salvet (2001) has significant
advantages despite having a more demanding implementation: it allows for dynamic failure
recovery properties in case of AE node failure or network link failure, as the client may join
the AE network using another AE node that is still available and reachable.
2.2 Distributed Active Element

Another scalability issue regarding both single AE and AE networks is scalability with respect
to the bandwidth of each individual data stream. In oder to improve on this, we have designed
a distributed AE Holub (2005); Holub & Hladká (2006), intended to be run on tightly coupled
clusters with low latency network interconnection for the control plane and high-bandwidth
interfaces for the data plane. The distributed AE splits a single stream into multiple sub-
streams, which are processed in parallel—thus possibly introducing packet reordering. This is
significantly different from general purpose load distribution systems like LACP IEEE 802.3ad
protocol, which have to avoid the packet reordering and therefore a single data stream is
processed sequentially5 . The distributed AE includes distribution unit to distribute the data
to the parallel processing units, and aggregation unit, which aggregates the data from the
parallel units.
Limited synchronization and FCT. The basic idea behind distributed AE utilizes the fact,
that most of the synchronous multimedia applications use non-guaranteed data transport like
UDP and thus they need to adapt to some packet reordering. However, significant data re-
ordering may either not be adapted upon or it results in latency increase as substantial buffer-
3 http://www.jxta.org/
4 http://openvpn.net/
5 This is done by using data flow identifiers hash to assign each data flow to a specific link of the of the
aggregated link group. Thus each single data flow must not exceed capacity of the single link.
Definition 5.4 (Ideal distributed AE) The ideal AE has processing capacity equal or higher
than stream bandwidth and it has an input queue size of qiAE . All the parallel units of the
ideal AE have the same parameters and performance and the total bandwidth of the traffic
is divided into streams with the same parameters. The ideal AE introduces no losses, nor
data corruption, nor data reordering in the data stream.

sAE
i sAE
o sSW
i
bj
distribution AE AE AE aggregating aggregating

unit input output input unit
buffer buffer buffer
Fig. 4. Model of the ideal distributed AE with ideal aggregation unit.
F IGURE 5.2: Model of the ideal distributed AE with ideal aggregation unit.
ing is necessary to reorder the packets. Without any explicit synchronization, the maximum
packet reordering can be
5.2.1 Ingress Distribution n(s AE + s AE + sSW + 1)
i o i
The ingress
where data distribution
n is number of paralleltakes
paths care
andofsdistributing
AE AE SWincoming are bufferdata across different paths
i , s o , si sizes on input of the AE,
inside the
output distributed
of AE, and inputAE. Foraggregation
of the the ideal distributed AE, it isassuitable
unit respectively, shown in to Figure
use simple
4. round-
robin
In distribution
order to decreaseaspacket
all thereordering
parallel AE units are by
introduced equivalent in theirAE,
the distributed performance.
we have introduced
a distributed algorithm for achieving less packet reordering compared to no explicit synchro-
Definition 5.5 (Ideal distribution unit) The ideal distribution unit distributes packets in
nization. The nodes are ordered in a ring with one node elected as a master node and they
round-robin fashion. In each round, it distributes n packets, one to each of the parallel
circulate a token which serves as a barrier so that no node can run too much ahead with
units. The distribution unit marks round number into each packet.
sending data. After reception of the token containing the current “active” round number,
each non-master node passes on the token immediately and may send only the data from the
Such an ideal distribution might not be suitable in the following cases:
round marked in the token until it receives to token again. When the master node receives
the •token
When from the lastAE
parallel node in the
units are ring, it finishes
of unequal sending the current
performance. In this round, increments
case, load balancingthe
rounddescribed
number inbelowthe token a passes
is useful. on the token. The mechanism is called Fast Circulating
Token (FCT) since the token is not held for the entire time period of data sending as usual in
When
the •token ringdata stream packets are not independent and the processing needs to have all
networks.
Becausetheofinter-dependent packets through
real world implementation of data thepacket
same path.
sending This
in might
common be for example
operating when
systems,
some that
we assume datasending
processing is donefor
procedure and somepacket
a single state inside the AE needsFurther
is non-preemptive. to be created and
we assume
maintained.
that token reception event processing has precedence over any other event processing in the
distributed
In thisAE. However,
case, as the
the packet data sending
distribution needs is non-preemptive,
to follow the packet if theinter-dependencies.
token arrives in the
middle of data
When packet sending,
distribution unit isit implemented
will be handled asjust
a partafterofthat packetapplication
sending sending is finished.
(e. g. user-
After more
spacedetailed analysis HolubUDP
library encapsulating (2005), it can be shown
sendto() function),the itmaximum
is possible reordering
to utilize induced
knowl-
by an edge
ideal distributed AE with
of data directly andFCT egress synchronization
distribute it correspondingly. and ideal
If theaggregating
distribution unit is is
unit
SW
n ( s + 3 ) ,
implemented as separate stand-alone network unit, the application can mark groups
i
when all queues operate in FIFO tail-drop mode. Thus the receiving application can adapt
its buffer size to this upper bound. On custom hardware, the FCT protocol can be adapted to
provide no packet reordering at all (called Exact Order Sending. More in-depth analysis can
be found in Holub (2005); Holub & Hladká (2006).
Prototype implementation and experimental evaluation. Prototype implementation of the
distributed AE is implemented in ANSI C language for portability and performance reasons.
The implementation comprises two parts: a load distribution library and the distributed AE
itself.
Because of lack of flexible enough load distribution hardware unit, we have implemented it
as a library, which allows simple replacement of standard UDP related sending functions in
existing applications and allows developers to have defined type of load distribution—either
pure round robin or load balancing.
Each parallel AE uses threaded modular implementation based on architecture described
above. Internal buffering capacity of each AE node has been set to 500 packets. Explicit
synchronization using FCT protocol has been implemented using MPICH implementation6
of MPI built with low-latency Myrinet GM 2.0 API7 (so called MPICH–GM). Prototype imple-
mentation has been tested on Linux.
For cost-effective prototype implementation, the aggregation unit was a implemented as com-
modity switch with sufficient capacity of internal switching fabrics.
The experimental results obtained on 10GE infrastructure, revealing that the distributed AE is
capable of completely saturating sender machine in a testbed similar to the one used for AE
performance evaluation above. Up to 8 parallel units were used for the measurement, con-
nected using Gigabit Ethernet NICs into GE ports of the Cisco 6506. Myrinet-2000 NICs and
switch were used for the low-latency interconnection. Packet distribution was implemented
as user-land UDP library and the aggregation was performed by the Cisco 6506 switch. When
the FCT protocol is used, the experimental evaluation showed the maximum packet reorder-
ing is below 15 for 8 parallel units, which makes it comparable to long-haul networks of good
quality. Without the FCT, the maximum reordering was up to 111 for the same setup, i.e., one
magnitude worse. Typical results can be seen in Figure 5.
The distributed AEs can also be incorporated into an AE network using the same approach
described in the previous chapter. However, because of running on more complicated infras-
tructure, the setup and start is more complex than for a single AE and thus the system has
worse fail-over behavior compared to the network of simple AEs. Another complication of
the distributed AE is in the processing of the passing data, which requires development of
parallel programming paradigms similar to MIT StreamIt Thies et al. (2002). The processing
may follow one of two possible approaches: (1) a context is maintained within one parallel
unit only (requires either that all the data requiring the same context to be processed in are
processed with one parallel unit only, or per-packet processing without a context is used), or
(2) the context is maintained within a subset of parallel nodes using the low-latency intercon-
nection of the cluster. These approaches will be further investigated in the future.
6 http://www-unix.mcs.anl.gov/mpi/mpich/
7 http://www.myri.com/scs/GM-2/doc/html/
Fig. 5. Packet reordering distribution with FCT and without synchronization, for 8 parallel
units and 3.4 Gbps per data flow.
3. Applications
Applications of virtual multicast range from simple user-empowered data distribution to com-
plex data parallel data processing tasks and per-user data processing. Overlay network cre-
ating virtual multicast can be also used to distribute data strongly protected environments.
These use cases are further discussed in this chapter.
3.1 Data Distribution

The AEs have been used routinely by different groups for collaboration, mostly with MBone
Tools8 , DVTS9 , and uncompressed HD video based on UltraGrid Holub et al. (2006). A recent
demonstration of uncompressed HD video with bandwidth usage of 1.5 Gbps per data stream
at SuperComputing|06 conference10 used a network with 3 optimized HD AEs in StarLight
(Chicago, USA) and achieved sustained aggregated data rate of 18 Gbps without any packet
8 http://www-mice.cs.ucl.ac.uk/multimedia/software/
9 http://www.sfc.wide.ad.jp/DVTS/
10 https://sitola.fi.muni.cz/igrid/index.php/SuperComputing_2006
loss. As an alternative setup, we have also used a combination of an AE with multiplication

on optical layer (optical multicast), which is, however, far from user-empowered as it requires
both direct access to Layer 1 network and installation of specialized hardware directly into
the network. The high-performance static AE network has also been used in production for
uncompressed HD video distribution for a distributed class on high-performance computing
taught by prof. Sterling at Louisiana State University Matyska, Holub & Hladká (2007). In
this case, dedicated λ-circuits spanning 5 institutions across the USA and one in the Czech
Republic were used and, therefore, a static configuration was the most appropriate as the
circuit topology was also statically configured. The 1.5 Gbps streams were distributed up to 7
locations as shown in Figure 6.
SL AE MU -- storage
UARK
MU -- live
LATECH
LSU AE SL AE
LSU – Frey LSU – Johnston MCNC NCSU
Fig. 6. Data distribution for LSU HPC Class by prof. Sterling based on uncompressed HD
video with 1.5 Gbps per stream.
With much lower bandwidth per stream but many more clients served, another AE network
is also used for streaming data distribution using VLC at the Masaryk University to get the
live video feeds from the lecturing halls even over networks without multicast support. Fur-
thermore, it is used for tunneling the data to the student dormitories which have very adverse
networking environments. This AE network also supports transcoding as described below.
3.2 Stream transcoding

Typical application of processing on an AE is stream transcoding. For live video stream distri-
bution from several lecturing halls at the Masaryk University, a transcoding processor for the
video and audio streams has been implemented as an AE module Liška & Denemark (2006). It
uses VideoLAN Client11 (VLC) as the actual transcoding back-end, thus giving us a large va-
riety of supported formats for both input and output. The transcoding module communicates
with VLC in three ways: the source data is delivered using Unix standard I/O, the transcoded
data is received from VLC using a local UDP socket in order to receive the data appropriately
packetized, while a local telnet interface is used for remote control of VLC.
For the specific application, the distribution schema is shown in Fig. 7. There are basically
two types of video stream sources: an MPEG-2 hardware encoder such as Teracue ENC-100
or a regular MPEG-4 streaming PC with video capture card and VideoLAN Client installed.
11 http://www.videolan.org
In both cases the video stream is generated as a standard MPEG Transport Stream (MPEG-
TS) at 2 Mbps and sent using unicast to the AE for further processing. The original data
is available to the students either using unicast (AE) or multicast from the AE, or they can
watch transcoded video from the gateway AE. Students then use VLC again for rendering the
streams at their computers. This allows us to provide students at the high-speed networks
with the maximum quality video, while students with slower networks (e.g., in dormitories)
are also supported and may participate in the class using transcoded streams with a lower bi-
trate. Depending on the settings, the transcoding can consume a considerable amount of pro-
cessing power and therefore the transcoding AE has significantly lower distribution capacity.
As a result it is set up at the beginning of the low bandwidth link working as a gateway or
bridge only, while another AE is use to actually distribute the transcoded data at large.
unicast unicast
MPEG-4 MPEG-4
multicast MPEG-2
MPEG-2 TS AE multicast MPEG-4 AE
unicast
MPEG-2 unicast MPEG-4
VLC
dormitory
Fig. 7. Video stream distribution schema for the live streaming from lecturing halls.
Performance evaluation
In order to evaluate efficiency and scalability of this solution, we have performed a series of
performance and latency measurements.
For performance measurements, we have used the following testbed: the AE was running on
a computer with the dual-core Pentium D at 3 GHz, 1 GB RAM, and a Gigabit Ethernet (GE)
NIC Broadcom NetXtreme BCM5721. Client computers were furnished with two Intel Xeon
3 GHz processors, 8 GB RAM, and a GE NIC BCM 5708. The testbed was interconnected using
two HP Procurve GE switches (2824 and 5406zl). All the computers were running Linux kernel
version 2.6. We have optimized buffer settings on NIC to 1 MB to improve the performance.
For transcoding, VLC 0.8.6b with ffmpeg library using libavutil 49.4.0, libavcodec 51.40.2, and
libavformat 51.11.0 was used. Source video for transcoding was in MPEG-2 format with full
PAL resolution (768×576) at 6 Mbps bitrate. MPGA audio accompanied the video and both
streams were encapsulated in MPEG Transport Stream format. The output stream was MPEG-
4 with 576×384 resolution at 1 Mbps bitrate, audio bitrate 128 kbps, all encapsulated again in
MPEG TS. Scalability and resource utilization is shown in Figure 8.
We have also performed a similar experiment for H.264 output video with full PAL resolution
at 2 Mbps bitrate using x264 library12 , but this didn’t work as the conversion used 100 % CPU
capacity which resulted in visible packet loss in the image.
12 http://www.videolan.org/developers/x264.html
Fig. 8. Performance characteristics of transcoding AE (1 MB buffer or NIC).

Output encoding Measure latency [ms]

MPEG-4 1220 ± 20
MPEG-4, keyint = 1 1240 ± 20
MJPEG, keyint = 1 1200 ± 20
H.263, keyint = 1, res. 704×576 1180 ± 20
Table 1. Transcoding AE latency measurements results.
Using the same setup, we have measured also the latency of the transcoding AE. The results
are summarized in Table 1. Obviously, this implementation of transcoding provides too much
latency for interactive video communication, but it is perfectly valid for streaming purposes
described above. The latency can be decreased to tens of milliseconds when implemented
directly as an AE module, not dependent on external transcoding tool—the latency added by
a simple AE module that just passes on the raw data in zero copy mode is 0.238 ms on given
infrastructure.
3.3 Video stream composition

Large group collaboration may easily result in too many windows at client sites (typical sce-
nario for AccessGrid13 ), and clients may not have sufficient power or desktop space to render
them all. In cases like this, it may be advantageous to down-sample the video streams and
compose several of them into a single stream directly on the AE. The same technique is im-
plemented in MCUs for H.323/SIP, but it was unavailable for MBone Tools. The first version
of video compositor Holer (2003) has been adapted to fit into the modular AE architecture as
a processor. This processor is based on the VIC tool McCanne & Jacobson (1995) and thus it
supports exactly the same set of video formats and the result is seen in Fig. 9. Up to four video
streams can be composed into one output stream. Input video formats are auto-detected, the
processor is able to work with different formats simultaneously. The output video format is
configurable by the end user.
Fig. 9. Example of video stream composition at AE using VIC video clients.
13 http://www.accessgrid.org/
3.4 Operation in Adverse Networking Environments and Security

The real-time communication for healthcare purposes is unique because of the two classes of
interconnected requirements: security and ability to operate even in heavily protected net-
working environments. The security is necessary as the specialist often need to communicate
very sensitive patients data. Because of the security requirements, the healthcare institutions
are usually trying to implement the most restrictive networking scenarios. E.g., we have been
collaborating with a hospital that has its network protected by a firewall and hidden behind a
NAT, that allows only HTTP traffic, which has to pass through two tiers of proxies. However,
even the specialists from this hospital need to communicate with their colleagues. The AE ap-
proach combined with VPNs have been deployed successfully for several healthcare related
projects and we were able to include even the institute mentioned above. As shown in Holub
et al. (2007), the OpenVPN approach only has a minimal impact on the performance of col-
laborative tools. Another important feature that we are developing in this field is efficient
aggregation of individual media streams—not only the video streams as discussed above—as
some of the institutions, especially in developing countries, have only very limited Internet
connection capacity.
4. Related work
Distribution of multimedia data over IP network leads to a multicast schema Almeroth (2000).
However, as the native multicast solution is not always appropriate (e.g., for many small
groups which is characteristic for interactive collaboration as it has been designed for small
number of large groups), reliable, or even available, other distribution schemes were de-
veloped following the approach of multicast virtualization El-Sayed et al. (2003); Li & Shin
(2002), e.g., Mtunnel Parnes et al. (1998) and UMTP Finlayson (2003). While many theo-
retical concepts for data distribution were developed namely during 1998 – 2003 period (see
the data distribution models referenced in Section 2.1), the practical approaches are still usu-
ally based on a central distribution unit or static topologies like the H.323 MCUs or reflec-
tors provided in the Virtual Room Videoconferencing System (VRVS)14 . The successor of
VRVS called Enabling Virtual Organizations (EVO)Galvez (2006) is based on a self-organizing
system of reflectors, again not empowering the end-user with tools to change the distribu-
tion topology. High-perormance dynamic data distribution system used for distribution of
200 Mbps compressed 4K video streams designed by NTT is called Flexcast Shimizu et al.
(2006). Another application-level multicast called Host Based Multicast (HBM) has been pro-
posed in El-Sayed (2004). The HBM author also investigated a combination of an IPsec based
VPN environment—while useful for data protection, it doesn’t improve on adaptability of
HBM for adverse networking environments. Other simpler UDP packet reflectors include
rcbridge Buchhorn (2005), reflector15 , and Alkit Reflex16 . However, all these systems are pri-
marily focused on pure data distribution and most of them even neglect the user-empowered
view, thus differing significantly from our highly modular and user-empowered AE based on
active network concept.
Another relevant field of work is parallel stream-oriented processing and programming of
such systems, which is of high importance for the distributed AE. A parallel programming
paradigm, that might be suitable for distributed AE programming, has been proposed in MIT
14 http://www.vrvs.org/
15 http://www.cs.ucl.ac.uk/staff/s.bhatti/teaching/z02/reflector.html
16 http://w2.alkit.se/reflex/
StreamIt system Thies et al. (2002). It enables efficient parallelization of the data processing
based on sent data structure and processing dependencies. Its suitability and possible adap-
tation will be further investigated.
5. Future work
In this chapter, we have explained basic principles of multicast virtualization, presented a
framework of Active Elements, designed for user-empowered synchronous data distribution
and processing. Depending on target environment and the streams that are being distributed,
the AEs may be deployed as a single central entity, or as a network of AEs for increased
scalability with respect to number of clients and increased failure resiliency, or as a distributed
AE to improve scalability with respect to the bandwidth of individual data stream. We have
demonstrated a number of applications both for data distribution and processing.
In the future, there are at least several areas to focus on. Utilizing a single AE, we would
like to introduce multi-level QoS approach to provide strict user and stream separation. This
is especially important when an AE is used for data processing. We would like to use a
virtualization-based approach to achieve this, and the virtual machines may also be used for
“programming” the AE, as the user may “inject” the whole virtual machine into the AE. For
the AE network, we would like to develop more complex signaling protocols to improve di-
agnostics (e.g., failure information needs to be distributed not only inside the AE network, but
also to the influenced users in some way). The virtual machine approach will also be used to
simplify migration of the processing modules in the AE network. Last but not least, we will
further investigate programming paradigms suitable for the distributed AE to enable truly
parallel stream processing.
This field of data distribution in overlay networks has been thoroughly examined by several
research groups between 1998 – 2003; some were examining the data distribution perspective,
while others were also looking at security issues. We provide a much broader view of the field
extending it with active network and user-empowered approaches. We have demonstrated
that while the research interest in this field dropped since 2003, new useful techniques can
still be invented and there are many practical applications worth analyzing.
Larger networks of AEs that are specialized in their functionality for data distribution as well
as processing goes beyond human capacity to manage such system. Thus we are researching
application of self-organization principles to application orchestration Liška & Holub (2009),
that could include not only AE and their networks, but also other components ranging from
individual applications running at users’ computers to allocation of network circuits.
Acknowledgments
This project has been kindly supported by the research intent “Optical Network of National
Research and Its New Applications” (MŠM 6383917201) and “Parallel and Distributed Sys-
tems” (MŠM 0021622419). We would like to appreciate various colleagues that worked in var-
ious stages of mutlicast virtualization projects, namely Jiří Denemark, Luděk Matyska, and
Tomáš Rebok.
6. References
Alchaal, L., Roca, V. & Habert, M. (2002). Offering a multicast delivery service in a pro-
grammable secure IP VPN environment., Networked Group Communication, Fourth In-
ternational COST264 Workshop, NGC 2002, Boston, MA, USA, October 23-25, 2002, Pro-
ceedings, ACM, pp. 5–10.
Almeroth, K. C. (2000). The evolution of multicast: From the MBone to inter-domain multicast
to Internet2 deployment, IEEE Network 14(1): 10–20.
Banerjee, S., Bhattacharjee, B. & Kommareddy, C. (2002). Scalable application layer multicast.,
Proceedings of the ACM SIGCOMM 2002 Conference on Applications, Technologies, Archi-
tectures, and Protocols for Computer Communication, August 19-23, 2002, Pittsburgh, PA,
USA, ACM, pp. 205–217.
Bouček, T. (2002). Kryptografické zabezpečení videokonferencí (Securing videoconferences using cryp-
tography), Master’s thesis, Military Academy of Brno, Czech Republic.
Buchhorn, M. (2005). Designing a multi-channel-video campus delivery and archive service,
The 7th Annual SURA/ViDe Conference, Atlanta, GA, USA.
Cheriton, D. R. & Deering, S. E. (1985). Host groups: a multicast extension for datagram
internetworks, Technical report, Stanford University, Stanford, CA, USA.
Chu, Y.-H., Rao, S. G. & Zhang, H. (2000). A case for end system multicast, ACM SIGMETRICS:
Measurement and Modeling of Computer Systems, pp. 1–12.
Deering, S. E. & Cheriton, D. R. (1990). Multicast routing in datagram internetworks and
extended LANs, ACM Transactions on Computer Systems 8: 85–110.
Denemark, J., Holub, P. & Hladká, E. (2003). RAP – Reflector Administration Protocol, Techni-
cal Report 9/2003, CESNET.
URL: http://www.cesnet.cz/doc/techzpravy/
Diot, C., Levine, B. N., Lyles, B., Kassem, H. & Balensiefen, D. (2000). Deployment issues for
the IP multicast service and architecture, IEEE Network 14(1): 78–88.
Dressler, F. (2003a). Availability analysis in large scale multicast networks, 15th IASTED In-
ternational Conference on Parallel and Distributed Computing and Systems (PDCS2003),
Vol. I, pp. 399–403.
Dressler, F. (2003b). Monitoring of Multicast Networks for Time-Synchronous Communication, Ph.d.
thesis, University of Erlangen-Nuremberg.
URL: http://www7.informatik.uni-erlangen.de/ dressler/publications/dissertation.pdf
El-Sayed, A. (2004). Application-Level Multicast Transmission Techniques Over The Internet, PhD
thesis, INRIA Rhône-Alpes, France.
URL: http://www.inrialpes.fr/planete/people/elsayed/phd/phd_final_080304.pdf
El-Sayed, A., Roca, V. & Mathy, L. (2003). A survey of proposals for an alternative group
communication service, IEEE Network 17(1): 46–51.
Finlayson, R. (2003). The UDP multicast tunneling protocol. IETF Draft <draft-finlayson-
umtp-09.txt>.
URL: http://tools.ietf.org/id/draft-finlayson-umtp-09.txt
Francis, P. (2000). YOID: extending the internet multicast architecture. Unpublished report for
ACIRI, http://www.aciri.org/yoid/docs/index.html.
Galvez, P. (2006). From VRVS to EVO (Enabling Virtual Organizations), TERENA Networking
Conference 2006, Catania, Italy.
Hladká, E., Holub, P. & Denemark, J. (2004). An active network architecture: Distributed
computer or transport medium, 3rd International Conference on Networking (ICN’04),
Gosier, Guadeloupe, pp. 338–343.
Hladká, E., Liška, M. & Rebok, T. (2005). Stereoscopic video over ip networks, in P. Dini
& P. Lorenz (eds), International Conference on Networking and Services 2005 (ICNS’05),
IEEE, p. 6.
Hladká, E. & Salvet, Z. (2001a). An active network architecture: Distributed computer or

transport medium, in P. Lorenz (ed.), Networking – ICN 2001: First International Con-
ference Colmar, France, July 9-13, 2001, Proceedings, Part II, Vol. 2094 of Lecture Notes in
Computer Science, Springer-Verlag, Heidelberg, pp. 612–619.
Hladká, E. & Salvet, Z. (2001b). An active network architecture: Distributed computer or
transport medium, in P. Lorenz (ed.), Networking – ICN 2001: First International Con-
ference Colmar, France, July 9-13, 2001, Proceedings, Part II, Vol. 2094 of Lecture Notes in
Computer Science, Springer-Verlag, Heidelberg, pp. 612–619.
Holer, V. (2003). Slučování videostreamů (Videostream Merging), Bachelor thesis, Faculty of In-
formatics, Masaryk University Brno, Brno, Czech Republic.
Holub, P. (2005). Network and Grid Support for Mulitmedia Distribution and Processing, PhD
thesis, Faculty of Informatics, Masaryk University Brno, Czech Republic.
Holub, P. & Hladká, E. (2006). Distributed active element for high-performance data distribu-
tion, NPC 2006: Network and Parallel Computing, Tokio, Japan, pp. 27–36.
Holub, P., Hladká, E. & Matyska, L. (2005). Scalability and robustness of virtual multicast
for synchronous multimedia distribution, Networking - ICN 2005: 4th International
Conference on Networking, Reunion Island, France, April 17-21, 2005, Proceedings, Part II,
Vol. 3421/2005 of Lecture Notes in Computer Science, Springer-Verlag Heidelberg, La
Réunion, France, pp. 876–883.
Holub, P., Hladká, E., Procházka, M. & Liška, M. (2007). Secure and pervasive collabora-
tive platform for medical applications, Studies in Health Technology and Informatics
126: 229–238. IOS Press, Amsterdam, The Netherlands.
Holub, P., Matyska, L., Liška, M., Hejtmánek, L., Denemark, J., Rebok, T., Hutanu, A.,
Paruchuri, R., Radil, J. & Hladká, E. (2006). High-definition multimedia for mul-
tiparty low-latency interactive communication, Future Generation Computer Systems
22(8): 856–861.
Jannotti, J., Gifford, D. K., Johnson, K. L., Kaashoek, M. F. & O’Toole, J. J. W. (2000). Overcast:
reliable multicasting with on overlay network, OSDI’00: Proceedings of the 4th confer-
ence on Symposium on Operating System Design & Implementation, USENIX Association,
Berkeley, CA, USA, pp. 197–212.
Jo, J., Hong, W., Lee, S., Kim, D., Kim, J. & Byeon, O. (2006). Interactive 3D HD video transport
for e-science collaboration over UCLP-enabled GLORIAD lightpath, Future Genera-
tion Computer Systems 22(8): 884–891.
Li, Z. & Shin, Y. (2002). Survey of overlay multicast technology. Unpublished report for Net-
works Lab, UC Davis, http://networks.cs.ucdavis.edu/~lizhi/papers/
ECS289.pdf.
Liebeherr, J. & Nahas, M. (2001). Application-layer multicast with delaunay triangulations,
Global Telecommunications Conference, 2001. GLOBECOM ’01. IEEE, pp. 1651–1655
vol.3.
Liška, M. & Denemark, J. (2006). Real-time transcoding and scalable data distribution for
video streaming, Technical Report 30/2006, CESNET, Praha, Czech Republic.
URL: http://www.cesnet.cz/doc/techzpravy/2006/rum-streaming/
Liška, M. & Holub, P. (2009). Couniverse: Framework for building self-organizing collab-
orative environments using extreme-bandwidth media applications, Euro-Par 2008
Workshops - Parallel Processing. Las Palmas de Gran Canaria, Spain, Vol. 5415 of Lecture
Notes in Computer Science, Springer-Verlag, Berlin/Heidelberg, pp. 339–351.
Mathy, L., Canonico, R. & Hutchison, D. (2001). An overlay tree building control protocol,
Lecture Notes in Computer Science 2233: 76–87.
Mathy, L., Canonico, R., Simpson, S. & Hutchison, D. (2002). Scalable adaptive hierarchi-
cal clustering, NETWORKING ’02: Proceedings of the Second International IFIP-TC6
Networking Conference on Networking Technologies, Services, and Protocols; Performance
of Computer and Communication Networks; and Mobile and Wireless Communications,
Springer-Verlag, London, UK, pp. 1172–1177.
Matyska, L., Hladká, E. & Holub, P. (2007). Collaborative framework for distributed distance
learning, Proceedings of the 8th International Conference on Information Technology Based
Higher Education and Training (ITHET ’07), Kumamoto, Japan.
Matyska, L., Holub, P. & Hladká, E. (2007). Collaborative framework for distributed dis-
tance learning, Proceedings of the 4th High-End Visualization Workshop, Obergurgl, Ty-
rol, Austria, pp. 40–45.
McCanne, S. & Jacobson, V. (1995). vic: A flexible framework for packet video, Proceedings of
ACM Multimedia’95, San Francisco, CA, USA, pp. 511–512.
Parnes, P., Synnes, K. & Schefstrom, D. (1998). Lightweight application level multicast tunnel-
ing using mtunnel, Computer Communication 21(15): 1295–1301.
Procházka, M., Holub, P. & Hladká, E. (2005). Active element network with P2P control plane,
IWSOS 2006: 1st International Workshop on Self-Organizing Systems, Vol. 4124/2006 of
Lecture Notes in Computer Science, Springer-Verlag Heidelberg, University of Passau,
Germany, p. 257.
Salvet, Z. (2001). Enhanced UDP packet reflector for unfriendly environments, Technical Report
16/2001, CESNET, Praha, Czech Republic.
URL: http://www.cesnet.cz/doc/techzpravy/2001/16/
Schulzrinne, H., Rao, A. & Lanphier, R. (1998). Real Time Streaming Protocol (RTSP). RFC
2326.
Shimizu, T., Shirai, D., Takahashi, H., Murooka, T., Obana, K., Tonomura, Y., Inoue, T., Yam-
aguchi, T., Fujii, T., Ohta, N., Ono, S., Aoyama, T., Herr, L., van Osdol, N., Wang, X.,
Brown, M. D., DeFanti, T. A., Feld, R., Balser, J., Morris, S., Henthorn, T., Dawe, G.,
Otto, P. & Smarr, L. (2006). International real-time streaming of 4K digital cinema,
Future Generation Computer Systems 22(8): 929–939.
Thies, W., Karczmarek, M. & Amarasinghe, S. (2002). StreamIt: A language for streaming ap-
plications, Proceedings of the 11th International Conference on Compiler Construction, Vol.
2304/2002 of Lecture Notes in Computer Science, Springer-Verlag Heidelberg, Greno-
ble, France, pp. 179–196.
Tran, D. A., Hua, K. A. & Do, T. T. (2003). ZIGZAG: An efficient peer-to-peer scheme for media
streaming., INFOCOM.
Zhang, B., Jamin, S. & Zhang, L. (2002). Host multicast: A framework for delivering multicast
to end users., IEEE INFOCOM ’02, New York, NY.
Zhuang, S. Q., Zhao, B. Y., Joseph, A. D., Katz, R. H. & Kubiatowicz, J. D. (2001). Bayeux:
an architecture for scalable and fault-tolerant wide-area data dissemination, NOSS-
DAV ’01: Proceedings of the 11th international workshop on Network and operating systems
support for digital audio and video, ACM Press, New York, NY, USA, pp. 11–20.
The Asymmetrical Architecture of New Optical Switch Device 87
X5
The Asymmetrical Architecture

of New Optical Switch Device
Mohammad Syuhaimi Ab-Rahman and Boonchuan Ng
Universiti Kebangsaan Malaysia (UKM)
Malaysia
1. Introduction
The explosive growth of data and combusting traffic have led to a demanding of flexible,
efficient, survivable, and multifunctional device to support all of network functions,
management and testing. The sophisticated technologies today had led to realize any
complex design in optical architecture that has become impossible for some years ago. The
waveguide technologies become the first player in developing the new architecture of
optical switch device. The most optical devices today are built by symmetrical architecture.
Therefore, they can perform bi-directionally with the same function to both side of incoming
signal.
The introducing of new optical switching device named optical cross add and drop
multiplexing (OXADM) which has an asymmetrical architecture and can perform bi-
directionally function as the previous using the combination concept of optical cross connect
(OXC) and optical add and drop multiplexing (OADM). Its enable the operating wavelength
on two different optical trunks to be switched to each other while implementing add and
drop function simultaneously. With the asymmetrical architecture, its enable the operating
wavelength on two different optical trunks to be switched to each other and implementing
accumulating function simultaneously. Here, the operating wavelengths can be reused
again as a carrier of new data stream. The wavelength transfer between two different cores
of fiber will increase the flexibility, survivability, and also efficiency of the network
structure. To make device operational more efficient, micro-electromechanical system
(MEMS) switches are used to control the mechanism of operation such as wavelength
add/drop and wavelength routing operation. As a result, the switching performed within
the optical layer will be able to achieve high-speed restoration against failure/degradation
of cables, fibers and optical amplifiers.
The OXADM architecture consists of 3 parts; selective port, add/drop operation, and path
routing. Selective port permits only the interest wavelength going through and acts as filter.
With the switch configuration, add and drop function can be activated in second part of
OXADM architecture. The ‘accumulation’ feature is found in the third part. The functions of
OXADM include node termination, drop and add, routing, multiplexing, and also providing
mechanism of restoration for point-to-point (P2P), ring, and mesh metropolitan as well as
fiber-to-the-home (FTTH) access network.
The OXADM also provide survivability through restoration against failure by means of
dedicated and shared protection that can be applied in wavelength division multiplexing
(WDM) ring metropolitan network. In other application, OXADM can also work as any
single device such as multiplexer, demultiplexer, OXC, OADM, wavelength selective
coupler (WSC), and wavelength roundabout (WRB). With such the excellent features, the
OXADM is expected to be unique, universal, and with a high reliability that is used to
overcome the various functions in WDM communication network today.
The experimental results showed the value of crosstalk and return loss for OXADM is
bigger than 60 dB and 40 dB respectively. The results also showed the value of insertion loss
was less than 0.06 dB under ideal condition, the maximum length that can be achieved is 94
km. In the transmission using SMF-28 fiber, with the transmitter power of 0 dBm and
sensitivity –22.8 dBm at a P2P configuration with safety margin, the required transmission is
71 km with OXADM. The optical signal-noise-ratio (OSNR) measurements for every
function of OXADM are also proposed with the values are bigger than 20 dB.
2. Wavelength Routed Networks Switching

Optical networks are high-capacity telecommunications networks based on optical
technologies and components that provide routing, grooming, and restoration at the
wavelength level as well as wavelength-based services. As networks face increasing
bandwidth demand and diminishing fiber availability, network providers are moving
towards a crucial milestone in network evolution: the optical network. Optical networks
based on the emergence of the optical layer in transport networks, provide higher capacity
and reduced costs for new applications such as the Internet, video and multimedia
interaction, and advanced digital services (IEC, 2007).
The explosive growth of data, particularly internet traffic, has led to a dramatic increase in
demand for transmission bandwidth imposing an immediate requirement for broadband
transport networks. Currently telecommunications networks widely employ WDM in
single-mode (SM) optical fibers to interconnect discrete network locations and offer high
capacity, high-speed and long reach transmission capabilities. The information transmitted
in the optical domain is transferred through simple P2P links terminated by Synchronous
Digital Hierarchy (SDH) or Synchronous Optical Network (SONET) equipment forming
ring and mesh network topologies. This solution requires a number of intermediate service
layers introducing complex network architectures. Such a scenario provides unnecessarily
high switching granularity, numerous optoelectronic conversions and complicated network
management resulting in poor scalability for data services and slow service turn up with
high installation, operation and maintenance cost. With the recent technology evolution in
the area of optical communications, the WDM transport layer is migrating from simple
transmission links into elaborate networks providing similar functionality to that of the
SDH/SONET layer, with improved features, higher manageability, lower complexity, and
cost-effective. Integrated WDM networks performing switching (Ramaswami, 2001) and
routing are deployed in order to overcome the need for multi-layer network architectures
(Tzanakaki et al., 2002). In such network scenarios, high capacity optical routes are set in the
transport layer forming connections between discrete points of the network topology. In
wavelength routed networks switching is performed through OADM and OXC nodes.
2.1 Optical Add/Drop Multiplexers (OADM)

OADM is element that provides capability to add and drop traffic in optical network. The
function of an OADM is to insert (add) or extract (drop) one or more selected wavelengths
at a designated point in an optical network (Keiser, 2003). An OADM can be passive or
active device that located at sites supporting one or two (bi-directional) fiber pairs and
enable a number of wavelength channels to be dropped and added reducing the number of
unnecessary optoelectronic conversions, without affecting the traffic that is transmitted
transparently through the node (Mukherjee, 2006).
The features that an OADM should ideally support are listed below:
 Drop capability of any channel and insertion capability o f a channel on any unused
wavelength
 Dynamic reconfiguration supporting fast switching speed (- ms)
 Variable add/drop percentage up to 100%
 Scalable architecture in a modular fashion
 Drop and continue functionality enabling network architectures such as dual homing
rings
 Span and ring protection capabilities
 Minimum performance degradation in terms of noise, crosstalk, filtering, etc for add,
drop, and through paths
The OADM enhances the WDM terminals by adding several significant features (see Figure
1). The OADM systems have the capacity of up to 40 optical wavelengths. They efficiently
drop and add various wavelengths at intermediate sites along the network for solving a
significant challenge for existing WDM (IEC, 2007).
Fig. 1. OADM functionality (IEC, 2007).
Most important, OADM technology introduces asynchronous transponders to allow the

optical-network element to interface directly to high revenue generating services. It is now
possible for asynchronous transfer mode (ATM), frame relay (FR), native local area
networks (LANs), high-bandwidth Internet protocol (IP), and others to connect directly to
the network via a wavelength in the optical layer. Transponder technology also extends the
life of older lightwave systems by accepting its bandwidth directly into the optical layer,
converting its frequency to an acceptable standard, and providing protection, and
restoration. The OADM also is the foundation of optical bidirectional line switched rings
(OBLSRs), which are described in the next module (IEC, 2007).
An OADM can be used in both linear and ring network architectures and in practice operate
in either fixed or reconfigurable mode. In fixed OADMs the add/drop and express
(through) channels are predetermined and can only be manually rearranged after
installation. In reconfigurable OADMs the channels that are added/dropped or pass
transparently through the node can be dynamically reconfigured as required by the
network. These are more complicated structures but more flexible as they provide
provisioning on demand without manual intervention, therefore they can be set up on the
fly to allow adding or dropping of a percentage of the overall traffic. The reduction of
unnecessary optoelectronic conversions through the use of OADMs introduces significant
cost savings in the network (Tzanakaki et al., 2003).
An OADM allows the insertion or extraction of a wavelength from a fiber at a point between
terminals. An OADM can operate either statically or dynamically. Some vendors call a
dynamic device a reconfigurable OADM (R-OADM). A static version obviously is not as
flexible and may require a hardware change if a different wavelength needs to be dropped
or added. For example, a static OADM might use two optical circulators in conjunction with
a series of fixed-wavelength fiber Bragg gratings (FBG). A dynamic or R-OADM results if
the gratings are tunable. Although a dynamic feature adds greater flexibility to a network,
this versatility also requires more careful system design. In particular, tunable (wavelength-
selectable) optical filters may be needed at the drop receivers, and the OSNR for each
wavelength must be analyzed more exactly (Keiser, 2003).
The technology of choice is determined by the functional and the performance requirements
of the node. R-OADM can be divided into two categories partly and fully reconfigurable. In
partly reconfigurable architectures there is capability to select the channels to be
added/dropped, but there is a predetermined connectivity matrix between add/drop and
through ports restricting the wavelength assignment function. Fully reconfigurable OADMs
also provide the ability to select the channels to be added/dropped, but also offer flexible
connectivity between add/drop and through ports, which enables flexible wavelength
assignment with the use of tunable transmitters and receivers (Tzanakaki et al., 2003).
R-OADM can be divided into two main generations. The first is mainly applied in linear
network configurations and does not support optical path protection while the second is
applied in ring configurations and provide optical layer protection to support network
survivability. The two most common examples of fully R-OADM: Wavelength selective
(WC) and broadcast select (BS) architectures are illustrated in Figure 2b & 2c. The WS
architecture utilizes wavelength (de)multiplexing and a switch fabric interconnecting all
express and add/drop ports; while the BS is based on passive splitters/couplers and tunable
filters (Mukherjee, 2006).
The tunable filtering, present in the through path, should provide selective blocking of any
dropped channels and can be achieved using technologies such as or acousto-optic filters or
dynamic channel equalizers (DCEs) based on diffraction grating and liquid crystal or MEMS
technologies (Tzanakaki et al., 2003). The overall loss introduced by the through path of the
BS solution is noticeably lower than the loss of the WS approach, significantly improving the
OSNR of the node and therefore its concatenation performance in a practical transmission
link or ring network. In addition, the BS design offers superior filter concatenation
performance, advanced features such as drop and continue, and good scalability in terms off
add/drop percentage (Mukherjee, 2006).
(a)
(b) (c)
Fig. 2. (a) Generic OADM architecture, (b) WS OADM architecture, and (c) BS OADM
architecture
Figure 3 showed a simple OADM configuration that has four input and four output ports.
Here the add and drop functions are controlled by MEMS-based miniature mirrors that are
activated selectively to connect the desired fiber paths. When no mirrors are activated, each
incoming channel passes through the switch to the output port. Incoming signals can be
dropped from the traffic flow by activating the appropriate mirror pair. For example, to
have the signal carried on wavelength λ3 entering port 3 dropped to port 2D; the mirrors are
activated as shown in Figure 3. When an optical signal is dropped, another path is
established simultaneously, allowing a new signal to be added from port 2A to the traffic
flow. There are many variations on optical add/drop device configurations depending on
the switching technology used. However, in each case the operation is independent of
wavelength, data rate, and signal format (Keiser, 2003).
Depending on whether an engineer is designing a metropolitan area network (MAN) or a
long-haul network, different performance specifications need to be addressed when
implementing an OADM capability in the network. In general, because of the nature of the
services provided, changes in the add/drop configuration for a long-haul network tend to
occur less frequently than in a MAN. In addition, the channel spacing is much narrower in a
long-haul network, and the optical amplifiers which are used must cover a wider spectral
band. For an interesting analysis of the Erbium Doped Fiber Amplifier (EDFA) performance
requirements and the link power budgets used for an OADM capability in a MAN
environment (Keiser, 2003).
Wideband long-haul networks are essentially a collection of P2P trunk lines with one or
more OADMs for inserting and extracting traffic at intermediate points. Figure 4 depicted a
generic long-haul Dense WDM (DWDM) network. Such networks typically are configured
as large rings in order to offer reliability and survivability features. For example, if there is
cable cut somewhere, the traffic that was supposed to pass through that fault can be routed
in the opposite direction on the ring and still reach its intended destination. As shown in
Figure 4 are three 10 Gbps DWDM rings and the major switching centers where
wavelengths can be regenerated, routed, added, or dropped. The links between DWDM
nodes have optical amplifiers every 80 km to boost the optical signal amplitude and
regenerators every 600 km to overcome degradation in the quality of the optical signals.
Extended-reach long-haul networks allow path lengths without regenerators of several
thousand kilometers. Also illustrated are typical services between two end-users, such as
SONET/SDH, Gigabit Ethernet, or IP traffic (Keiser, 2003).
Fig. 3. Example of adding and dropping wavelengths with a 4x4 OADM device that uses
miniature switching mirrors
Fig. 4. A generic long-haul DWDM network which is configured as a set of large rings
2.2 Optical Cross-Connect (OXC)
OXC switches optical signals from input ports to output ports. These type of elements are
usually considered to be wavelength insensitive, i.e., incapable of demultiplexing different
wavelength signals on a given input fiber. OXC is located at nodes cross-connecting a
number of fiber pairs and also support add and drop of local traffic providing the interface
with the service layer. A basic cross-connect element is the 2x2 cross point element. A 2x2
cross point element routes optical signals from two input ports to two output ports and has
two states: cross states and bar states (see Figure 5). In the cross state, the signal from the
upper input port is routed to the lower output port, and the signal from the lower input port
is routed to the upper output port. In the bar state, the signal from the upper input port is
routed to the upper output port, and the signal from the lower input port is routed to the
lower output port (Mukherjee, 2006).
To support flexible path provisioning and network resilience, OXCs normally utilize a
switch fabric to enable routing of any incoming channels to the appropriate output ports
and access to the local client traffic. The features that an OXC should ideally support are
similar to these of an OADM, but additionally OXCs need to provide:
 Strictly non-blocking connectivity between input and output ports
 Span and ring protection as well as mesh restoration capabilities
Efficient use of fiber facilities at the optical level obviously becomes critical as service
providers begin to move wavelengths around the world. Routing and grooming are key
areas that must be addressed. This is the function of the OXC, as shown in Figure 6.
(a) Cross state
(b) Bar state
Fig. 5. OXC state Fig. 6. OXC block diagram (IEC, 2007)
In the optical domain, where 40 optical channels can be transported on a single fiber, a
network element is needed that can accept various wavelengths on input ports and route
them to appropriate output ports in the network. To accomplish this, the OXC needs three
building blocks as illustrated in Figure 7:
1. Fiber switching - the ability to route all of the wavelengths on an incoming fiber to a
different outgoing fiber
2. Wavelength switching - the ability to switch specific wavelengths from an incoming
fiber to multiple outgoing fibers
3. Wavelength conversion - the ability to take incoming wavelengths and convert them
(on the fly) to another optical frequency on the outgoing port; this is necessary to
achieve strictly non-blocking architectures when using wavelength switching (IEC,
2007).
Fig. 7. Switching and conversion with OXC (IEC, 2007).
3. Optical Cross and Add/Drop Multiplexers (OXADM)

OXADMs are elements which provide the capabilities of add and drop function and cross
connecting traffic in the network, similar to OADM and OXC (Caraglia, 2003; Mutafungwa,
2000; Tsushima, 1998; Tzanakaki, 2003). OXADM consists of three main subsystem; a
wavelength selective demultiplexer, a switching subsystem, and a wavelength multiplexer.
Each OXADM is expected to handle at least two distinct wavelength channels each with a
coarse granularity of 2.5 Gbps of higher (signals with finer granularities are handled by
logical switch node such as SDH/SONET digital cross connects or ATM switches. There are
eight ports for add and drop functions, which are controlled by four lines of MEMs-optical
switch. The other four lines of MEMs switches are used to control the wavelength routing
function between two different paths (see Figure 8). The functions of OXADM include node
termination, drop and add, routing, multiplexing and also providing mechanism of
restoration for P2P, point-to-multipoint (P2MP), ring, and mesh metropolitan. With the
setting of the MEMs optical switch configuration, the device can be programmed to function
as another optical device such as multiplexer, demultiplexer, coupler, wavelength converter
(with fiber grating filter configuration), OADM, WRB, and etc for the single application.
Fig. 8. The block diagram of OXADM architecture Fig. 9. The classification of OXADM
3.1 OXADM Optical Hybrid Device
The designed 4-channels OXADM device is expected to have maximum operational loss of
0.06 dB for each channel when device components are in ideal condition. The maximum
insertion loss when considering the component loss at every channel is less than 6 dB. In the
transmission using SMF-28 fiber, with the transmitter power of 0 dBm and sensitivity –22.8
dBm at a point-to-point configuration with safety margin, the required transmission is 71
km with OXADM without regeneration (Rahman et al., 2006).
The OXADM architecture consists of 3 parts; selective port, add/drop operation, and path
routing. Selective port permits only the interest wavelength going through and acts as filter.
With the switch configuration, add and drop function can be activated in second part of
OXADM architecture. The signals are then re-routing to any port of output. The signals can
also be accumulated on one path and exit at any output port. The partition of OXADM
architecture is depicted in Figure 9.
3.2 OXADM Optical Switch
The control switch for OXADM optical switch is used to change the path of incoming from
the input port. When the control switches in ‘off’ state, no switching occurs and the signal
pass through the device as seen from the Figure 10a. But when the control switch B is ‘on’
state, the signal from the Input 1 will be switch to Output 2 (see Figure 10b). In contrast to
the control switch B is in reverse state, the accumulation function occurs which multiplex all
the signals from the inputs together and exit at the Output 2. This will be the same if the
control switch A is in ‘on’ state but the output is at Output 1. If both switches are in ‘on’
state, the signal will be switched to exchange their output port and works as an OXC device.
The functional of an OXADM optical switch can be summarized through the truth
functional table shown by Table 1. The incoming signals from the back will be switched to
neighbor output port or pass through the device. This is shown in Figure 11.
(a) Normal condition
(b) Switch B activated - Accumulation (c) Switch B activated – Accumulation

signal on output 2 signal on output 1
Fig. 10. OXADM optical switch works as re-configurable output port multiplexer
Fig. 11. OXADM optical switch works as 2x2 demultiplexer. When the switch is activated,
the signal will be switched to any output port.
Table 1. The truth table of OXADM optical switch
3.3 OXADM Multifunctional Switch
OXADM can also work as any single device such as multiplexer, demultiplexer, OXC,
OADM, WSC, and WRB.
1. Demultiplexer - There are two configuration of demultiplexer using OXADM, with
interleaver and without interleaver. The function of interleaver is to separate the
incoming signal before entering the OXADM ports (Figure 12ai). In contrast, the signal
will enter the back port and has automatically routed to their respective path (Figure
12aii).
2. Multiplexer - The input signals enter the add ports of OXADM and the multiplexed
signal out of the signal via output port (Figure 12b).
3. OADM - The separated wavelengths out of the interleaver have a capability to
add/drop function before they are combined and exit one of the output ports of
OXADM (Figure 12c).
4. WSC - Two WSC devices will be produced by OXADM. The signal will enter the input
port and will be separated out through the drop port of OXADM (Figure 12d).
5. OXC - The concept is similar to OADM but the OXADM also offer the function of cross-
connect the signals. The function is almost the same with OXC device (Figure 12e).
6. WRB - The new invented device that offer the management function of wavelengths.
Different with circulator in which the inputs and outputs are built separately (Figure
12f).
3.4 Other Application
FTTH is a simple, inexpensive, ideal, and attractive many parties in optical communication
today. A number of factors are increasing the interest among network service providers in
offering the triple play services of high-speed data access, voice, and video. Most
importantly, subscribers are finding a growing number of applications that drive their
desire for higher bandwidth, including Internet access, interactive games, and video
delivery. Since the fiber offers a vast amount of bandwidth that can be utilized for
communication, one of utilizing this is signal multiplexing. Due to the large bandwidth and
the associated high bit rates, the multiplexing process is beyond the capabilities of pure
electronic methods and has to be implemented optically as well. As the reach of fiber is
being extended to the access network it is economically attractive to share fibers between
different end-users without adding active components in the network (Menif and Fathallah,
2007).
(a) The functional diagram of OXADM:DEMUX (b) The functional diagram of OXADM:MUX
1x4, (i) Configuration 1 (ii) Configuration 2 4x1
(c) The functional diagram of OXADM:OADM 1x4 (d) The block diagram of OXADM:WSC
(e) The functional diagram of OXADM:OXC (f) The functional diagram of OXADM:WRB
Fig. 12. OXADM multifunctional switch
It involves the full installment of fiber from central office (CO) till to customer houses which
is called premises. Recently, installation of FTTH technology has started to accelerate as a
strong alternative to the existing broadband access technologies based on copper pairs
(Digital Subscriber Line, DSL) and coax (cable modems). This worldwide acceleration is
largely due to both the considerable decrease in capital expenses (CAPEX) of introducing
FTTH connectivity and its “future proof” nature in providing ever increasing user
bandwidth requirements (Yuksel et al., 2008). This technology ensures low operational
expenditures (OPEX) due to all elements in used are the passive optical device with small
number used. The maximum distance achievable is 20 km with gigabit of transmission rate.
FTTH consists of 3 significant elements: optical line terminal (OLT), optical splitter, and
optical network unit (ONU). The designed OXADM is can also used as the wavelength
management in OLT architecture for excellently FTTH. It has 4 inputs terminal which
represent the 4 different signal carriers to be multiplexed and exit at Output 1. Three
wavelengths are used typically 1310 nm for data/voice upstream transmission and 1490 nm
for the downstream data/voice transmission. Meanwhile 1550 nm is used to transmit
downstream video signal.
Since the possible wavelength for communication is 1200-1600 nm, 1625-1650 nm
wavelength regions is commonly considered for in-service monitoring purpose without
affecting the services to other end-users. Additional wavelength 1625 nm will be adding up
to the system to carries the troubleshooting signal for line status monitoring purposes. In the
downstream direction (OLT to users), packets are broadcast by the OLT and extracted by
their destination ONUs based on their media access control (MAC) address. In the upstream
direction (users to OLT), each ONU will use a time shared channel (TDM) arbitrated by the
OLT. OLT is function to aggregates Ethernet traffic from remote ONU devices through
passive optical splitters. The signals then are sent to the customer premises. This can be
defined in Figure 13 with the wavelength allocation highlighted.
Fig. 13. OXADM function as the wavelength management element in OLT in FTTH
3.5 Device Comparison
The OXADM device will be compared with two existing switching devices; directional
coupler (DC) switch and OXC. The non-selective DC switch has two states and one control
element. It has fixed number of input and output port that is two. The wide bandwidth
signal comes from the input port with switch to either one of output ports. It works bi-
directional with symmetrical function (Palais, 2005). Figure 14 showed the mechanism of
switching for directional coupler switch in normal (a) and active condition (b) & (c). The
application of DC switch is to control the signal path in WDM network and optical storage;
and can also perform the function of OADM in optical distributed network
OXC is the directional coupler witch but with many ports. The functional of OXC is cross-
connecting between output and input port (see Figure 15) (Mutafungwa, 2001). Same with
OXADM, the OXC is selective device but it does not have ‘accumulation’ feature. In contrast
with OXADM, OXC works bi-directional with symmetrical function. The application of
OXC is as a switching device in mesh network configuration and also in optical storage.
Table 2 summarized the differences of OXADM as compared with DC and OXC.
Table 2. Comparison between OXADM, DC, and OXC
Fig. 14. Switching mechanism of DC optical Fig. 15. Switching mechanism of OXC,
switch, a) normal (b) activate a) normal (b) activate
4. Experimental Results and Discussions

4.1 Experimental Setup
Two parameters have been studied experimentally to ensure the interference of uninterested
signal is minimized. Figure 16 presented the experimental set up to measure the crosstalk at
two ports of OXADM and the results have been redrawn in Figure 17 and 18 respectively.
The crosstalk value is bigger than 60 dB means the interested wavelength is in safety level
and the transmitted data can be interpreted at any receiver end.
(a) Configuration 1
(b) Configuration 2
Fig. 16. Crosstalk measurement set up
(a) Output 1 (b) Output 2
Fig. 17. Redrawing of measured output port at every port for configuration 1
(a) Output 1 (b) Output 2
Fig. 18. Redrawing of measured output port at every port for configuration 2
The other parameter should be considered for bi-directional device is return loss. Return
loss is the disturbance of uninterested signal against the direction of interested signal. This
can be explained using Figure 19. The return loss is measured by using set up in Figure 20
and the result is shown in Figure 21. The return signal coming out from the input port is
routing to OSA 3 using optical circulator. The value is 40 dB which is higher than minimum
safety value. Both experimental have shown that the OXADM optical switch has a good and
acceptable value of crosstalk and return loss.
Fig. 19. Return loss or leak reflected signal which contributes to crosstalk phenomena in bi-
directional device.
Fig. 20. Return loss measurement set up

Fig. 21. Redrawing of measured output port at every port for configuration 2 for return loss
measurement
4.2 Experimental Results

The OXADM device is characterized by using two tunable light sources and two optical
spectrum analyzers. The designed 4-channel OXADM device is expected to have maximum
operational loss of 0.6 dB for each channel when device components are in ideal condition.
The maximum insertion loss when considering the component loss at every channel is 6 dB.
The testing is carried out for every single function of OXADM. The function includes
bypass, path exchange and accumulation. The single operating wavelength test (wavelength
is 1510 nm), the results show the OSNR value for bypass function is 20 dB and path
exchange is also 20 dB. Each measurement result is indicated in Figure 22 till 24.
Fig. 22. The decrement of kilometers occurs by increasing the attenuation of OXADM which
represent the device losses
Fig. 23. The measured output power at two operating wavelength for bypass operation
Fig. 24. The measured output power for path exchange operation
The path splitting function (accumulation function in reverse mode as Figure 10b and 10c) is
also applied and the result shown in Figure 25 with OSNR > 24 dB. For backward operation
as depicted in Figure 26, the OSNR values for cross-connecting function (as Figure 11) are
bigger than 22 dB. This can be defined that the level of signal is 20 dB higher than noise level
for all single functions of OXADM optical switch. The 20 dB reference indicates the
acceptable value for the signal to noise ratio in data communication.
Fig. 25. The measured output power two operating wavelength for path splitting operation
(accumulation function in reverse mode)
Fig. 26. The measured output power at two operating wavelengths for path exchange
operation
5. The Developed Prototype

A prototype was designed and developed to enable us to evaluate the performance in an
actual propagation environment as illustrated in Figure 27 and 28.
Fig. 27. Photographic view of OXADM device set up
Fig. 28. Photographic view of two OXADM device prototypes set up in P2P network
6. Conclusion
A new switching device which utilizes the combined concepts of OADM and OXC
operation is presented through the development of OXADM. The experimental results show
the value of crosstalk and return loss is bigger than 60 dB and 40 dB respectively. In our
previous results have also shown the value of insertion loss was less than 0.06 dB under
ideal condition, the maximum length that can be achieved is 94 km. While when considering
the loss, with the transmitter power of 0 dBm and sensitivity –22.8 dBm at a P2P
configuration with safety margin, the required transmission is 71 km with OXADM. The
OXADM switching mechanism has been explained and compare with other existing optical
switch; DC optical switch and OXC. The OXADM optical device is particularly designed for
WDM metro application. It can be used as restoration switch in FTTH network. OXADM
can also work as any single device such as demultiplexer, multiplexer, OADM, OXC, WSC,
and WRB. In other application, the OXADM can also provide survivability through
restoration against failure by means of dedicated and shared protection that can be applied
in WDM ring metropolitan network.
Acknowledgement
This project is supported by Ministry of Science, Technology and Innovation (MOSTI),
Government of Malaysia, through the National Top-Down Project fund and National
Science Fund (NSF). The authors would like to thank the Photonic Technology Laboratory in
Institute of Micro Engineering and Nanoelectronics (IMEN), Universiti Kebangsaan
Malaysia (UKM), Malaysia, for providing the facilities to conduct the experiments. The
OXADM had firstly been exhibited in 19th International Invention, Innovation and
Technology Exhibition (ITEX 2008), Malaysia, and was awarded with Bronze medal in
telecommunication category.
7. References
Caraglia, C.; Haffaouz, S.; Zantvoort, J.V.; Leijtens, X. & Smit, M. (2003). A Photonic
Integrated Add-Drop Multiplexing for a 1.6 Tbits/s Optical Ring Network,
Proceeding Symposium IEEE/LEOS Benelux Chapter, Enshede, pp. 69-72, 2003
International Engineering Consortium. (2007). Optical Networks, White Paper, 2007,
Available: http://www.iec.org/online/tutorials/opt_net/index.asp
Keiser, G. (2003). Optical Communications Essentials, McGraw Hill, New York, US, ISBN: 0-07-
143353-8
Menif, M. & Fathallah, H. (2007). An Encoder/decoder Device Including a Single Reflective
Element for Optical Code Division Multiple Access System. Journal of Optical
Communications, Vol. 28, No. 3 (2007) 172-174, ISSN: 0173-4911
Mukherjee, B. (2006). Optical WDM Networks (Optical Networks), Springer, ISBN: 978-0-387-
29055-3
Mutafungwa, E. (2001). An Improved Wavelength-selective All Fiber Cross-connect Node,
Proceedings of 2001 IEEE Workshop on High Performance Switching and Routing, pp. 93-
96, ISBN: 0-7803-6711-1, Dallas, TX, USA, 2001
Tsushima, H.; Hanatani, S.; Kanetake, T.; Fee, J.A. & Liu, S.A. (1998). Optical Cross-connect
System for Survivable Optical Layer Networks, Hitachi Review, Vol. 47, No. 2, (1998)
85-90, ISSN: 0018-277X
Palais, J.C. (2005). Fiber Optic Communication, Prentice Hall, New Jersey, ISBN: 978-
0130085108
Rahman, M.S.A.; Husin, H.; Ehsan, A.A. & Shaari, S. (2006). Analytical Modeling of Optical
Cross Add and Drop Multiplexing Switch, Proceedings of 2006 IEEE International
Conference on Semiconductor Electronics, pp. 290-293, ISSN: 0-7803-9731-2, Kuala
Lumpur, Malaysia, 2006
Ramaswami, R. (2001). All-Optical Crossconnects in the Transport Network, Proceedings of
2001 Optical Fiber Communication Conference and Exhibit (OFC 2001), Vol. 3, pp. WZ1-
WZ1, ISBN: 1-55752-655-9, Anaheim, CA, USA, March 2001
Tzanakaki, A.; Wright, I. & Sian, S.S. (2002). Wavelength Routed Networks: Benefits and
Design Limitations, Proceedings of Cybernetics and Informatics (SCI2002), Orlando,
Florida, July 2002
Tzanakaki, A.; Zacharopoulus, I. & Tomkos, I. (2003). Optical Add/drop Multiplexers and
Optical Cross-connects for Wavelength Routed Network, Proceedings of 5th
International Conference on Transparent Optical Networks (ICTON 2003), pp. 41-46,
ISBN: 0-7803-7816-4, Warsaw, Poland, June 2003
Yuksel, K.; Moeyaert, V.; Wuilpart, M. & Mégret, P. 2008. Optical layer monitoring in
passive optical networks (PONs): a review. Proceedings of 10th Anniversary
International Conference on Transparent Optical Networks (ICTON 2008), pp. 92-98,
ISSN: 978-1-4244-2626-3, Athens, Greece, June 2008
Adaptive Active Queue Management
for TCP Friendly Rate Control (TFRC) Traffic in Heterogeneous Networks 109
X6
Adaptive Active Queue Management for TCP

Friendly Rate Control (TFRC) Traffic in
Heterogeneous Networks
Rahim Rahmania and Christer Åhlundb
a Mid Sweden University, b Luleå University of Technology
SWEDEN
1. Introduction
Proposals to handle differentiated and guaranteed services in Internet have not provided
the benefits that users and operators are expecting. Its complexity, with a large number of
interconnected networks, is difficult to handle in an efficient way. This is due to the resource
heterogeneity in terms of technologies and the inconsistent implementation of quality of
services (QoS) in different networks. Despite several research activities in the area of QoS,
Internet is still basically a best-effort network, which it is likely to stay, also in the far future.
Streaming-based servers utilizing UDP for the underlying transport need to use some form
of congestion control [1] to ensure the stability of Internet as well as the fairness to other
flows, like those using TCP. The TCP-Friendly Rate Control (TFRC) is such a congestion
control scheme appropriate for UDP. In this paper we target the question of how to
optimize network and user-perceived performance in a best-effort network. In particular,
we focus on the impact of end-to-end performance of the queue management scheme
utilized in a wireless network.
The first of the basic assumptions in this study is that the wireless last hop constitutes the
bottleneck of the end-to-end (E2E) path. TFRC is intended for applications such as streaming
media, where a relatively smooth sending rate is of importance. TFRC measures loss rate by
estimating the loss event ratio [2], and uses this measured rate to determine the sending rate
in packets per RTT. When a bottleneck is shared with large-packet TCP flows, this has
consequences for the rate achievable by TFRC. In particular a low bandwidth small-packet
TFRC flow, sharing a bottleneck with high-bandwidth large-packet TCP flows, may be
forced to slow down, even though the nominal rate of the TFRC application in bytes per
second is less than the total rate of the TCP flows. This is fair only if the network limitation
is defined by the number of packets per second, instead of bytes per second. In the TFRC
protocol the small-packets are intended for flows that need to send frequent small quantities
of information. It intends to support applications better, which should not have their
sending rates in bytes per second decreased because of the use of small packets. This is
restricted to applications, which do not send packets more often than every 10 ms. The
TFRC Small-Packet (TFRC-SP) variant is motivated partly by the approach in Ref. [3], with
the argument that it is reasonable for voice over IP (VoIP) flows to assume that the network
limitation is in bytes rather in packets per second, that the sending rate is the same in bytes
per second as for a TCP flow with 1500 byte packets, and that the packet drop rate is the
same. An application using TFRC-SP can have a fixed packet size or may vary its packet size
in response to congestion.
Wireless channels suffer from bursty error losses reducing TFRC throughput, because TFRC
incorrectly interprets packet loss as a sign of congestion. The maximum increase of TFRC
rate, given fixed RTT, is estimated to be 0.14 packets per RTT and 0.22 packets per RTT with
history discounting [4]. It takes four to eight RTTs for TFRC to halve its sending rate in the
presence of persistent congestion.
Active Queue Management (AQM) intends to achieve high link utilization without
introducing an excessive delay into the E2E path. For good link utilization it is necessary for
queues to adapt to varying traffic load. The AQM has been subject to extensive research in
the Internet community lately, and a number of methods to control the queue size have been
proposed. An increase in RTT not only degrades the control performance of an AQM
algorithm, but also leads to instability in the network.
The work described in this chapter examines throughput, aggressiveness, and smoothness
of TFRC with varying bandwidth. A dynamic model of TFRC enables applications to
address the basic feedback nature of AQM.
To demonstrate the generality of the proposed method an analytic model is described and
verified by extensive simulation of different AQM.
2. Adaptive AQM (AAQM) Algorithm

AAQM is a light weight algorithm aimed at maximizing the flow of packets through the
router, by continuously computing the quotient between the number of arriving and
departing packets. The algorithm applies probabilistic marking of incoming packets to keep
the quotient between arriving and departing packets just below 1 to minimize the queue
length and maximize the throughput. Minimizing the queue length means minimizing the
delay and packet loss.
2.1 Formal description

This part contains a more formal description regarding how the AAQM algorithm works .
The goal of the AAQM algorithm is to keep the packet arrival rate as close to the packet
departure rate as possible in order to maximize the flow of packets through the queue. The
packet arrival rate is determined by the end systems sending packets and can for this reason
be controlled by probabilistic marking of arriving packets. The packet departure rate is
determined by the packet size and bandwidth of the outgoing link. The packet arrival rate is
denoted by A and the packet departure rate is denoted by B, both of which are measured
over T seconds. The utilization U of the flow of packets through the queue is defined by
equation (1).
A (1)
U= , B≠0
B
If in the equation above , U>1 the queue grows, which can result in a queue overflow and
packet loss. Similarly, if, in the equation above , U<1 the queue shrinks, which could result
in a link underutilization and wasted bandwidth. Therefore to maximize the flow, U should
be as close to 1 as possible. The required benefit is that when U=1 a packet arrives for each
departing packet causing the link to be fully utilized, and if U is just below 1 the queue will
simultaneously shrink minimizing both loss and delay. Now, if the total running time of the
algorithm is divided into N non-overlapping time slots of size T seconds then the utilization
of each time slot n is given by equation (2).
An (2)
Un = , Bn ≠ 0, n = 1...N
Bn
Now the packet marking probability P will be adjusted depending on the value of U during
each time slot n . If U>1 then P can be increased to reduce the arrival rate and thus decrease
U. Likewise, if U<1 then P can be decreased to increase the arrival rate and thus increase U.
Equation (3) shows how the packet marking probability is adjusted.
Pn +1 = Pn + f (U n ), n = 1...N (3)
P0 =0
The function f is given by equation (4) and, depending on U, increases or decreases the
packet marking probability P.
− P , x < 1 (4)
f ( x) =  1
 P2 , x > 1
T, P1 and P2 are constants that need to be determined beforehand.
2.2 Parameter Tuning

Of the variables described above only P1, P2 and T are user adjustable. Two sets of
recommended values for these parameters are presented in Table 1. The first row of
parameters in Table 1 gives the algorithm a little better utilization at the cost of higher loss
and delay. In the second row of parameters gives the algorithm a little lower utilization but
improved loss and delay characteristics. The recommended parameters have been found by
performing several simulations using NS[13] of the network using different parameter
settings.
P1 P2 T
0.700 0.900 0.010
0.700 1.000 0.010
Table 1. Recommended AAQM Parameter.
When tuning the algorithm for use in a network some basic aspects must be kept in mind.
A larger T will give a larger time interval over which the quotient U is computed. This will
in turn lead to a better prediction of the flow through the queue but on the downside it will
also lead to a slower reaction to changes in the flow. Slow reactions to changes in the flow
can lead to high delay and packet loss. When choosing P1 and P2 it should be kept in mind
that P1 < P2 leads to low utilization, delay and packet loss. Similarly P1 > P2 gives high
utilization, delay and packet loss.
3. Simulation setup
We use the methods presented in [9] for the study of the average queue time to have a well-
tuned AQM
Wl l q
Pwin N / R0 Pque
-
P
e-sR
0
Caqm(S)
Fig. 1. Linearization of AQM with long-lived TCP traffic l and unresponsive traffic u.
with a queue averaging time less than RTTs. Figure 1 presents a block diagram of the
linearized AQM feedback system, where C aqm (s) denotes the transfer function of the AQM
controller, Wl the congestion window size in packets, R0 the round-trip time, N the number
of long-lived TCP connections, Pwin the window size, and Pque the queue transfer function.
The symbol l is used for long-lived TCP traffic (FTP) and u for unresponsive flows when
describing how these flows impact on the queue length q, loss probability p, and arrival
rates of both l and u. In this work we focus on randomly created short and long-lived flows
as well as their impact on AQM. For applications needing to maintain a slowly changing
sending rate, the equation-based congestion control is the most appropriate. This kind of
application is the source of short-lived flows. It is assumed that they never escape the slow
start phase of TCP, so their window sizes are increasing exponentially rather than linearly as
during TCP’s congestion-control phase. As a model of a short-lived flow we used the
sending rate of TFRC [4] which is a rate based congestion control mechanism ensuring
fairness when it co-exists with TCP flows [4]. Its throughput equation is a version of the
throughput equation for conformant TCP Reno [4]:
B
X= ,
2bp  3bp  (5)
RTT
3
t
+ RTO (3
 8
) p(1+ 32 p )
2

where X is the transmission rate in byte/sec, as a function of the packet size B in bytes, RTT
is the round-trip time in seconds, p the steady-state loss event rate, tRTO is the TCP
retransmission timeout value in seconds, and b the number of packets acknowledged by a
single TCP acknowledgement. TFRC uses this equation to adjust the sender rate to achieve
the goal of TCP friendliness.
The sender can calculate the retransmit timeout value tRTO using the usual TCP algorithm:
tRTO = SRTT + 4 * RTTvar , (6)
where RTTvar is the variance of RTT, and SRTT is the round trip time estimate. Different TCP
congestion control mechanisms use different clock granularities to calculate retransmit
timeout values, so it is not sure that TFRC accurately can model a typical TCP flow. Unlike
TCP, TFRC does not use this value to determine whether to retransmit or not. Hence the
consequences of this inaccuracy are less serious. In practice the simple empirical heuristic of
t RTO = 4 RTT works reasonably well to provide fairness with TCP [4]. For the dynamics of
the model for TFRC behavior, we use the simplified and modified model presented in [10].
This model ignores the TCP timeout mechanism and is described by the nonlinear
differential equation:
1 X(t) X(t − R(t))

X(t) = − . p(t − R(t)), (7)
R(t) 2 R(t − R(t))
 N(t)
−C + X(t), .q > 0
 R(t)
Q=  (8)
max 0,−C + N(t) X(t), q=0
  R(t) 
where X is the average packet rate (packets/s). The RTT is calculated as R(t) = q(t) C + T p ,
where q is the average queue length (packets), C is the link capacity (packets/s), and Tp is
the propagation delay. The parameter p is the probability of a packet mark, the 1 R(t) term
models additive increase of the packet rate, while the X(t) 2 term models multiplicative
decrease of the packet rates in response to a packet marking p. Eq. (8) models the bottleneck
queue length using the accumulated differences between packet arrival rates N(t)X(t) R(t) and
the link capacity C, where N (t ) is the load factor (number of traffic sessions).
For linearization of Eqs. (7) and (8) and the estimation of some of the parameters Ref. [10]
defines plant dynamics of the AQM feedback control as:
C 2 2N
P(s) = , (9)
2N 1
(s + )(s + )
R 2C R
In Eq. (5) C 2 2N is the high-frequency plant gain, which is an important parameter in the
design of AQM control schemes, since it affects the stability, transient response, and steady-
state performance. The variation of the queue length depends on the input rate, output rate,
and the AQM controller.
The design of adaptive AQM algorithm should be based on the mean RTT [11] since realistic
network conditions have heterogeneous flows with different RTTs. The link capacity C is
measured by keeping track of departed packets, and an estimate of the TFRC load N/R is
inferred from measurements of the dropping probability p. The TFRC throughput
N RC = p 2 provides a mean for estimating N / R .
For the parameterization of AQM dynamics the generalized fluid description of AQM
dynamics relating instantaneous queue length q and loss probability p from [12] is used
which is described as follow:
χ am = f θ (χ am ,q),
(10)
p = g(χ am ,q),
Where χ am denotes AQM state and f and g describe the AQM dynamic behavior. In RED [5],
χ am is the average queue length (RED bases its decision whether to mark a packet or not on
the average queue length). In REM [6] χ am is the marking probability and the link price
and in BLUE [7] χ am is the packet loss and link utilization history used to manage
congestion.
In AAQM [8] χ am is the quote between the numbers of arrived and departed packets
within a timer event. This kind of event occurs at fixed, evenly spaced, time intervals, see
[8]. The quote is used as the utilization factor of the queue. For tuning of the AAQM the
control parameters of the AAQM (P1, P2, T) are linked to the network parameters N, R and
C. In the absence of knowledge of these parameters we can develop an on-line estimation of
the link capacity C and the TFRC load N.
The link capacity Cˆ is estimated using the model in [12] extended with computation of the
packet arrival rate. The degree of TFRC traffic utilization of the link changes depending on
the competing traffic, and to smooth the TFRC capacity Cˆ a low-pass filter (LPF) and the
fluid mechanism is used:
θ C = −K Cθ C + K C Cˆ , (11)
where θ c denotes the estimated capacity and Kc the filter time constant. To estimate the
TFRC load we use the TFRC fluid model presented by Eqs. (7) and (8) and derive the
following steady-state TFRC relationship:
2
1 X0
0= − P0 , (12)
R 2R
N
0= X 0 − C, (13)
R
Where X0 is the equilibrium congestion packet rate and P0 is the equilibrium drop
probability. From formulas (12) and (13) we can obtain p0 2 = N RC . In the same way as we
estimated the link capacity, we can smooth the estimate of the N/RC by using the LPF:
n = −K n θ n + K n p / 2,
θ rc (14)
rc rc rc
Where θ rcn is the smoothed estimate of the N/RC and K rcn the filter time constant.
For self-tuning AQM uses the parameters estimated in Eqs. (11) and (14) described as:
χ am = fθ (χ am ,q),
(15)
p = gθ (χ am ,q),
Where fθ and gθ show the explicit dependency of AQM dynamics using the estimate
 n  .
variables θ = θ C ,θ rc
 
When the 802.11 MAC layer approaches saturation, contention delays induced by deferred
count-down timers, an increased contention window and retransmissions affecting the
performance of TFRC and TCP [16]. TFRC is unaware of the MAC layer congestion,
rendering in that the TFRC sender may overestimate the maximum sending rate. This
congests the MAC layer, and the wireless network consequently reaches a sub-optimal
stable state with respect to the throughput and round-trip time [17][18]. Because of the lack
of interaction between TFRC and the 802.11 MAC layer, a Rate Estimator (RE) is added to
TFRC [17]. The RE approximates the saturation capacity of the MAC layer, and by limiting
the sending rate, MAC layer congestion can be avoided.
0.1.0 12 1.0.1
1
100Mb
5ms 11Mb
13 1.0.2
11Mb
100Mb 10Mb
0.2.0 2 0 11
10ms 10ms
0.0.0 1.0.0
11Mb
100Mb 21 1.0.10
25ms
0.10.0
10
Fig. 2. Configuration of the simulated network
4. Simulations
The simulation results presented obtained using the NS-2 simulation tool [13]. Figure 2
depicts the topology used. The solid lines symbolize wired links and the dashed lines
wireless links. Nodes numbered 1-10 are fixed; those numbered 12 - 21 are mobile, and the
node numbered 11 is the 802.11b base station (infrastructure mode). The link between the
nodes number 0 and 11 represents the virtual bottleneck link running the AQM algorithms.
An access point has two interfaces an 802.11 wireless interface to transmit/receive frames on
the air and a wired interface. The disparity in channel capacity of these two interfaces makes
the access point a significant potential bottleneck link. The virtual bottleneck represents an
access point.
Data are originating at the nodes numbered 1 - 10 and received by the nodes numbered 12 -
21. Each link carries both a TFRC and a TCP flow using a Pareto traffic generator (to
generate aggregate traffic that exhibits a long-range dependency). The traffic generators
start randomly after one second of the simulation to avoid a deterministic behavior and lasts
for 100 seconds. Each simulation was run 30 times with different seeds for the random
number generator. TCP-SACK [14] is used with Selective Acknowledgments (SACK),
allowing a receiver to acknowledge out-of-order segments selectively. The TFRC flows are
modeled as short-lived small packets web flows, and the TCP flows as a mix of short-lived
flows and long-lived FTP flows. The sources of the short-lived web flows are modeled
according to Ref. [15].
In table 2 the different parameters settings used are listed. The TFRC and TCP timer
granularity used, i.e. the tick value, is set to 500 ms, and the TCP minimum retransmission
timeout to 1 s. The throughput at the bottleneck link, the queue size (in packets), and the
drop probability are used for performance evaluation throughout the simulations.
Queues Queue sizes Numbers of

(packets) web sessions
BLUE 5 400
Drop-Tail 10 800
RED 50 1600
REM 100
AAQM
Table 2. List of parameter settings used in different scenarios and packetsize 14 bytes.
The parameter settings used in the AQM algorithms are shown in Tab. III. For Blue the
values were obtained from Ref. [7], for RED from Ref. [5], for REM from Ref. [6], and for
AAQM from Ref. [8].
BLUE freeze_time= D1=0.001 d2=0.0002

10 ms
RED Wq=0.002 minth=20% of maxth=80% maxp=1
the queue size of the queue
size
REM Gamma=1 Phi=1.001 Bo=20 Updtime=0.0
AAQM P1=0.700 P2=1.000 T=0.010
Table 3. Parameter setting for the simulated aqm
5. Perfomance evaluation
To achieve satisfactory control performance, the design goals of AQM algorithms are
responsiveness and stability. Bursty sources are used for the performance evaluation with
varying queue sizes managed by the AQM algorithms. The bursts generated require an
AQM algorithm to efficiently and quickly adapt to the current situation to maintain a high
overall throughput and to avoid dropping more packets than necessary. As a result the drop
rate and throughput are compared for the different algorithms.
Figures 3 and 4 depict the results for the drop rate and overall throughput respectively. Four
queue sizes are used 5, 10, 50 and 100 packets. The curves are plotted using a 95%
confidence interval
Figures 3 shows that Drop-Tail (DT) exhibits a high drop rate. This is due to the fact that all
packets are dropped when the queue is full so it affects all flows. All sources will then
decrease its sending rate at approximately the same time. This also means that all sources
will increase the sending rate about the same time rendering in full queues once more and
this will create oscillation. and beyond a high drop rate low throughput is also exhibited.
Figures 3(a) and 3(b) show AAQM and RED have similar drop rate but in figures 3(c) and
3(d) RED has higher drop rate than AAQM.
The instantaneous queue length of RED is controlled in range of minth and maxth. In order to
be effective a RED queue must be configured with a sufficient amount of buffer space to
accommodate an applied load greater than the link capacity from the instant in time when
the applied load decreases at the bottleneck link in response to congestion notification. RED
must ensure that congestion notification is given at a rate which sufficiently suppresses the
transmitting sources without underutilizing the link. When a large number of TCP flows
are active the aggregate traffic generated is extremely bursty. Bursty traffic often defeats the
active queue management techniques used by RED since queue length grow and shrink
rapidly before RED can react. Exactly that happens in figures 3(c), 3(d), 4(c) and 4(d).
On the other hand the queues controlled by REM and BLUE are often empty. When the link
capacity is low AAQM requlates the queue length, where REM and BLUE oscillate between
an empty buffer and its limit of queue size. As a result REM and BLUE show poor
performance under a wide range of traffic environments.
Drop-Tail and REM must be configured with a sufficient amount of buffer space in order to
accommodate an applied load greater than the link capacity from the instant that congestion
is detected, using the queue length trigger, to the instant when the applied load decreases in
response to congestion notification.
In general, the bias against bursty traffic and the tendency towards global synchronization
can be eliminated by maintaining a stable packet loss over time. The steady-state control
performance of each AQM algorithm was evaluated and the packet loss rate studied at three
different traffic loads.
As shown in figure 3 and figure 4 the flow dynamics are severely oscillating and it takes a
long time to stabilize to a steady-state. The AAQM controller is able to compensate the
oscillatory of the flow dynamics and given satisfactory control performance such as a fast
and stable control dynamics. AAQM show the most robust steady-state control
performance, independent of traffic loads, in terms of relatively small mean value of the
packet loss rate as well as its variance.
In the experiments we allowed the AQM schemes to converge to steady state when there
were 400 web sessions, then we increased the web session number to 800 and 1600 to study
the performance of the AQM schemes when the number of short flows increases. From
figures 4(a) and 4(b) it can seen that AAQM has 1% better throughput than RED when the
number of sessions is 1600, and much higher throughput than BLUE, REM and DropTail In
figure 4(c) AAQM has 3% better throughput than RED, and from figure 4(d), can seen that
AAQM has 36% better throughput than RED.
The results from Figures 3 and 4 show that AAQM has the best responsiveness to congestion
as well as the most robust steady-state control.
6. Related work
The work presented in [9] studies the effects of unresponsive flows on AQM. It shows that
the queue averaging time is a result of a trade-off between AQM responsiveness and the
robustness of the uncontrolled flows. The average queue time results in a smooth or stable
congestion feedback, which introduces jitter in the queuing delay due to variation in the
unresponsive flows. Three types of flow types were considered: short-lived TCP, Markov
on-off UDP, and traffic with long-range dependencies (e.g ftp). Our work instead focuses on
short-lived flows and uses a more realistic model for VoIP traffic by using TFRC–SP. Also
[9] does not study the impact of unresponsive flows on the AQM algorithms, while we do.
Our focus are in responsiveness of UDP flows with co-existent TCP flows.
The work presented in [19] surveys two adaptive and proactive AQM algorithms using a
classical proportional-integral-derivative feedback control to achieve stability and
responsiveness. The TCP flows are modeled as long-lived FTP flows. In our work the flows
are modeled as the mix of long-live flows and short-live flows to fulfill the design goal of an
adaptive AQM to interact with a realistic flow composition.
In reference [22] the authors argues in favor of rate-based AQM for high-speed links. Also in
that work a proportional-integral controller for the AQM scheme is used. The design goal
was to match the aggregate rate of TCP flows to the available capacity while minimizing the
queue size. We study the integration of TFRC-SP and UDP in co-existence of with TCP in
heterogeneous networks.
The work presented in [23] uses a token-bucket model as a virtual queue (VQ) with a link
capacity less than the actual link capacity. If a packet arrives, it is placed in a queue in the
VQ if there is space available. Otherwise the packet is dropped. Accordingly the algorithm is
able to react at an earlier stage, even before the queue grows, making it very sensitive to the
traffic load and round trip time. However, the utility functions are much different from ours
due to the AQM control parameters. AAQM uses control law and link utilization in order to
manage congestion. The action of the control law in AAQM is to mark incoming packets in
order to maintain the quotient between arriving and departing packet as close to one as
possible.
The study in reference [24] focused just on the RED and the parameter setting of RED was
based on heuristics. It also studied RED against disturbances on the wireless access network.
Only one type of flow types was considered: short-lived TCP. We study UDP in co-existence
of with TCP and their impact on DT, RED, REM, BLUE and AAQM.
The work presented in [25] by using of proxy AQM between access point for WLAN and
wired network. The proxy reduces the overhead of the access point by implementing the
AQM functionality at the gateway. In the work they extended the RED/ARED scheme to a
proxy mode by calculating the average queue length and updating pmaof ARED. They
measured only a number of TCP flows.
In reference [26] a channel-aware AQM scheme is presented. This new approach provides
congestion signals for flow control not only based on the queue length but also the channel
condition and the transmission bit rate. For the performance evaluation of the new AQM in
multi-rate WLAN the bit rate of the wireless node in manuals fixed at different levels ( in
sequence of 2M, 1M, 11M, and 5.5 Mbps). Two TCP flows were considered. The main idea in
the work was to design an AQM for flow control in multi-rate WLAN.
a) b)
c) d)
Fig. 3. Packet loss rate for a packet size of 14 bytes; (a) queue size 5 packets, (b) queue size 10
packets, (c) queue size 50 packets, (d) queue size 100 packets.
a) b)
c) d)
Fig. 4. Throughput of the link for a packet size of 14 bytes; (a) queue size 5 packets, (b)
queue size 10 packets, (c) queue size 50 packets, (d) queue size 100 packets.
7. Conclusions and future work

The present work studies AQM algorithms for competing small-packet TFRC flows and
TCP flows in heterogeneous networks. We investigate the effects caused by unresponsive
flows using TFRC on the AQM performance that is measured using responsiveness and
stability. Through simulations it is shown that with suitable design of the AQM scheme the
end-to-end performance can be maintained for TFRC flows consisting of small packets. It is
shown that the control performance of Drop-tail, RED, and BLUE are very sensitive to the
traffic load and round trip time. With REM the links suffer in utilization as the buffer size
increases. In particular AAQM shows a stable queue length with low and smooth packet
loss rates independent of the traffic load.
Future works will particularly investigate the integration of the Guaranteed TFRC (GTFRC)
[20] with the DCCP protocol [21] and co-existence of DCCP with the Stream Control
Transmission Protocol (SCTP). The impact of these protocols on the AQM performance and
QoS will be studied in heterogeneous networks. An additionally study will be focused on
optimization of the buffer space requirements.
8. References
[1] S. Floyd, “Congestion Control Principles” RFC2914, September 2000.
[2] M. Handley, J.Pahdye, S. Floyd, and J. widmer, “ TCP Firendly Rate Control (TFRC):
Protocol Specification “. RFC3448, January 2003.
[3] RFC 3714.
[4] S.Floyd, M. Handley, J. Padhye, and J. Widmer. “ Equation-based congestion control for
unicast application”. In proc. of ACM SIGCOMM’00, Aug. 2000.
[5] Sally Floyd and Van Jacobson, “Random Early Detection Gateways for Congestion
Avoidance”, IEEE/ACM Transactions on Networking, vol. 1, Number 4, August
1993.
[6] Sanjeewa Athuraliya, “A Note on Parameter Values of REM with Reno-like Algorithms”,
Networking Laboratory, Caltech, March 2002.
[7] Wu-chang Feng, Dilip Kandlur, Debanjan Saha and Kang Shin, “BLUE: A New Class of
Active Queue Management Algorithms”, Department of EECS, University of
Michigan, Ann Arbor, MI 48105.
[8] R.Rahmani, “Adaptive Active Queue Management in Heterogeneous Networks ”in 26th
ITI 2004, Croatia, June 2004.
[9] C.V. Hollot, Y. Liu, V. Misra , D. Towsley “ Unresponsive flows and AQM Performance
“, In Proceedings IEEE, 2003.
[10] V. Misra, W.Gong, and D. Towsley, “Fluid based analysis of a network of AQM routers
supporting TCP flows with an application to RED, “ in Proceedings of
ACM/SIGCOMM, 2000.
[11] C.Hollot, V. Misra, D. Towsley, and Gong, “ Analysis and design of controllers for
AQM routers supporing TCP flows,” IEEE Transactions on Automatic Control,
vol. 47, no. 6, 2002.
[12] H.Zhang, C. Hollot, D. Towsley, and V. Misra, “ A self-tunning structure for adaptation
in TCP/AQM networks, “ Department of computer sciences, UMass, Amherst,
Tech. Rep., July 2002.
[13] The NS-2 Simulator available at http://www.isi.edu/nsnam/ns/.
[14] RFC 2018.
[15] Anja Feldmann , Anna C. , P. Huang and W. Willinger “Dynamics of IP traffic: A study
of the role of variability and the impact of control “sigcomm99 , pub acm , pages
301-313 , http://www.acm.org/sigcomm/sigcomm99/papers/session8-3.html.
[16] Saikat Ray , Jeffrey Carruthers, and David Starobinski, “ RTS/CTS-induced Congestion
in Ad-Hoc Wireless LANs” in Prcceedings of IEEE WCNC, Mar.2003, pp. 1516-
1521.
[17] Mingzhe Li, Choong-Soo Lee, Emmanuel Agu, Mark Claypool, and Robert Kinicki, “
Performance Enhancement of TFRC in Wireless Ad Hoc Networks, Distributed
Multimedia Systems (DMS)”, September 2004.
[18] henghua Fu, Xiaoqiao Meng, and Songwu Lu, “A Transport Protocol For Supporting
Multimedia Streaming in Mobile Ad Hoc Networks”, IEEE JSAC, December 2004.
[19] S.Ryu, C.Rump, C.Qiao “ Advances in Active Queue Management (AQM) Based TCP
Congestion Control” Telecommunication Systems Journal Kluwer publisher 25:3,4,
p 317-351, 2004.
[20] E. Lochin, L. Dairaine, G. Jourjon “gTFRC: a QoS-aware congestion control algorithm”,

The 5th International Conference on Networking (ICN'2006) Mauritius - April,
2006.
[21] S. Floyd, E. Kohler, J. Padhye :” Profile for DCCP Congestion Control ID 3: TFRC
Congestion Control “, RFC 4342, IETF (2006).
[22] J. Aweya, M. Ouellette, D. Y. Montuno, K. Felske, “Rate-based proportional-integral
control scheme for active queue management“ International Journal of Network
Management John wiley & Sons Publisher 16: p 203-231, 2006.
[23] S. Kunniyur, R. Srikant, “ An Adaptive Queue(AVQ) Algorithm for Active Queue
Management” ACM Transactions on Networking Vol.12 No.2 April 2004.
[24] F. Zheng, J. Nelson, “ An H∞ approach to cognestion control design fro AQM routers
supporting TCP flows in wireless access networks” Computer Networks vol. 51 p
1684-1704, 2007.
[25] S. Yi, M. Kappes, S. Grag, X. Deng, G. Kesidis, C. R. Das, ” Proxy-RED: An AQM
Scheme for Wireless Local Area Networks” 0-7803-8784-8/$20.00 2004, IEEE.
[26] Y. Xue, H. V. Nguyen, K. Nahrstedt “ CA-AQM: ChannelAware Active Queue
Management for wireless Networks” ICC 2007, IEEE.
Queues with
session arrivals as models for optimizing the traffic control in telecommunication networks 123
X7
Queues with session arrivals as models for

optimizing the traffic control in
telecommunication networks
Sergey Dudin 1 and Moon Ho Lee 2
1Belarusian
State University
Belarus
2Chonbuk National University
Korea
1. Introduction
Many problems in routing, scheduling, flow control, resources allocation and capacity
management in telecommunication, production, and transportation networks can be solved
with help of queueing theory. Typically, a user of a network generates not a single item
(packet, job, pallet, etc) but a whole bunch of items and service of this user assumes
sequential transmission of all these items. This is why the batch arrivals are often assumed
in analysis of queueing systems. It is usually assumed that, at a batch arrival epoch, all
requests of this batch arrive into the system simultaneously. However, the typical feature of
many nowadays networks is that requests arrive in batch, while arrival of requests
belonging to a batch is not instantaneous but is distributed in time. We call such batches as
sessions. The first request of a session arrives at the session arrival epoch while the rest of
requests arrive one by one in random intervals. The session size is random and it may be not
known a priori at the session arrival epoch. Such a situation is typical, e.g., in modeling
transmission of video and multimedia information. This situation is also typical in IP
networks, e.g., in World Wide Web with Hypertext Transfer Protocol (HTTP) where a
session can be interpreted as a HTTP connection and a request as a HTTP request. This
situation is also discussed in literature with respect to the modeling the Scheme of
Alternative Packet Overflow Routing ( SAPOR ) in IP networks.
In this scheme, the session is called as flow and represents a set of packets that should be
sequentially routed in the same channel. When a packet arrives, it is determined (e.g. by
means of IP address) if the packet is a part of a flow, already tracked. If the packet belongs
to an existing flow, the packet is marked for transmission. If the flow is not yet tracked and
the channel capacity is still available, the packet is admitted into the system and flow count
is increased. Otherwise the flow is routed on the overflow link (or is dropped at all) and the
packet is rejected in the considered channel. Tracked flows are cleared after they are
finished. Clearing of an inactive flow is done if no more packets belonging to this flow are
received within a certain time interval. Tracking and clearing of flows is performed by
means of a token mechanism. The number of tokens, which defines the maximal number of
flows that can be admitted into the system simultaneously, is very important control
parameter. If this number is small, the channel may be underutilized. If this number is too
large, the channel may become congested. Average delivering time and jitter may increase
essentially and Grade of Service becomes bad. So, the problem of defining the optimal
number of tokens is of practical importance and non-trivial. In (Kist et al., 2005),
performance measures of the SAPOR scheme of routing in IP networks are evaluated by
means of computer simulation.
Analogous situation also naturally arises in modeling information retrieving in relational
data bases where, besides the CPU and disc memory, some additional "threads" or
"connections" should be provided to start the user’s application processing. In this
interpretation, session means application while requests are queries to be processed within
this application and tokens are threads or connections.
In the paper (Lee et al., 2007), the Markovian queueing model with a finite buffer that suits
for analytical performance evaluation and capacity planning of the SAPOR routing scheme
as well as for modelling the other real-world systems with time distributed arrival of
requests in a session is considered. To the best of our knowledge, such kind of queueing
models was not considered and investigated in literature previously. In (Lee et al., 2007), the
problem of the system throughput maximization subject to restriction of the loss probability
for requests from accepted sessions is solved. In the paper (Kim et al., 2009), the analysis
given in (Lee et al., 2007) is extended in three directions. Instead of the stationary Poisson
arrival process of sessions, the Markov Arrival Process (MAP) is considered. It allows catching
the effect of correlation of flow of sessions. The presented numerical results show that the
correlation has profound effect on the system performance measures. The second direction
is consideration of the Phase type (PH) service process instead of an exponential service time
distribution assumed in (Lee et al., 2007). Because PH type distributions are suitable for
fitting an arbitrary distribution, this allows to take into account the service time distribution
and variance of this time in particular, carefully. The third direction of extension is the
following one. It is assumed in (Lee et al., 2007), that the loss (due to a buffer overflow) of
the request from the accepted session never causes loss of a whole session itself. More
realistic assumption in some situations is that the session might be lost (terminates
connection ahead of schedule). E.g., it can happen if the percentage of lost voice or video
packets (and quality of speech or movie) becomes unacceptable for the user. To take such a
possibility into account in some extent, it is assumed in this paper that the loss of a request
from the admitted session, with fixed probability, leads to the loss of a session to which this
request belongs. Influence of this probability is numerically investigated in the paper (Kim
et al., 2009).
In the present paper, the modification of model from (Kim et al., 2009) to the case of an
infinite buffer is under study. In contrast to the model with a finite buffer considered in
(Kim et al., 2009) where the problem of the throughput maximization was solved under
constraint on the probability of the loss of a request from an accepted session, here we do
not have such a loss. So, the problem of the throughput maximization is solved under
constraint on the average sojourn time of requests from the accepted sessions. In section 2,
the mathematical model is described in detail. Stability condition, which is not required in
the model (Kim et al., 2009) with a finite state space but is very important in the model with
an infinite buffer space, is derived in a simple form. This condition creates an additional
Queues with
constraint in maximization problem. The steady state joint distribution of the number of
sessions and requests in the system is analyzed by means of the matrix analytical technique
and expressions for the main performance measures of the system are given in section 3.
Section 4 is devoted to consideration of the request and the session sojourn time
distribution. Section 5 contains numerical illustrations and their short discussion and section
6 concludes the paper.
2. Mathematical model
We consider a single server queueing system with a buffer of an infinite capacity. The
requests arrive to the system in sessions. Sessions arrive according to the Markov Arrival
Process. Sessions arrival in the MAP is directed by an irreducible continuous time Markov
chain  t , t  0 with the finite state space {0,..., W }. The sojourn time of the Markov chain
 t , t  0 in the state  has an exponential distribution with the parameter    0 W 
After this sojourn time expires, with probability p ( , ), the process  t , t  0 transits to
k
the state   , and k sessions, k  0 1 arrive into the system. The intensities of jumps from
one state into another, which are accompanied by an arrival of k sessions, are combined
into the matrices Dk  k  0 1 of size ( W  1)  ( W  1) . The matrix generating function of
these matrices is D( z)  D0  D1 z  z  1 . The matrix D(1) is the infinitesimal generator of the
process  t  t  0 The stationary distribution vector  of this process satisfies the equations
 D(1)  0  e  1 Here and in the sequel 0 is the zero row vector and e is the column
vector of appropriate size consisting of 1’s. In case the dimensionality of the vector is not
clear from the context, it is indicated as a lower index, e.g. e W denotes the unit column
vector of dimensionality W  W  1 .
The average intensity  (fundamental rate) of the MAP is defined as
   D1e
The variance v of intervals between session arrivals is calculated as
v  2  1 ( D0 )1 e   2 
the squared coefficient c var of variation is calculated by
c var  2  ( D0 )1 e  1
while the correlation coefficient ccor of intervals between successive group arrivals is given
by
1
ccor  (  ( D0 )1 D1 ( D0 )1 e   2 )v
For more information about the MAP , its special cases and properties and related research
see (Fisher & Meier-Hellstern, 1993), (Lucantoni, 1991) and the survey paper by S.
Chakravarthy (Chakravarthy, 2001). Usefulness of the MAP in modeling

telecommunication systems is mentioned in (Heyman & Lucantoni, 2003), (Klemm et al.,
2003). Note, that the problem of constructing the MAP which fits well a real arrival
process, is not very simple. However, this problem has practical importance and is
intensively solving. For relevant references and the fitting algorithms see, e.g., (Heyman &
Lucantoni, 2003), (Klemm et al., 2003), (Asmussen et al., 1996) and (Panchenko & Buchholz,
2007).
Following (Kist et al., 2005) , we assume that admission of sessions (they are called flows in
(Kist et al., 2005) and called threads, connections, sessions, exchanges, windows, etc. in different
real-world applications) is restricted by means of tokens. The total number of available
tokens is assumed to be K  K  1 Further we consider the number K as a control parameter
and solve the corresponding optimization problem.
If there is no token available at a session arrival epoch the session is rejected. It leaves the
system forever. If the number of available tokens at the session arrival epoch is positive this
session is admitted into the system and the number of available tokens decreases by one. We
assume that one request of a session arrives at the session arrival epoch and if it meets free
server, it occupies the server and is processed. If the server is busy, the request moves to the
buffer and later it is picked up for the service according to the First Came - First Served
discipline.
After admission of the session, the next request of this session can arrive into the system in
an exponentially distributed with the parameter  time. The number of requests in the
session has the geometrical distribution with the parameter   0    1 i.e., probability that
the session consists of k requests is equal to  k  1 (1   ) k  1 The average size of the
session is equal to (1   )1 
If the exponentially distributed with the parameter  time since arrival of the previous
request of a session expires and new request does not arrive, it means that the arrival of the
session is finished. The token, which was obtained by this session upon arrival, is returned
to the pool of available tokens. The requests of this session, which stay in the system at the
epoch of returning the token, must be completely processed by the system. When the last
request is served, the sojourn time of the session in the system is considered finished.
The service time of a request is assumed having PH distribution. It means the following.
Request’s service time is governed by the directing process t , t  0, which is the continuous
time Markov chain with the state space { 1… M} The initial state of the process t  t  0 at
the epoch of starting the service is determined by the probabilistic row-vector
  (  1 …  M ) . The transitions of the process t  t  0 that do not lead to the service
completion, are defined by the irreducible matrix S of size M  M . The intensities of
transitions, which lead to the service completion, are defined by the column vector
S0  Se . The service time distribution function has the form B( x )  1   eSx e . Laplace-

e dB( x ) of this distribution function is  ( sI  S )1 S0  The average

 sx
Stieltjes transform
0
service time is given by b1   ( S )1 e . The matrix S  S0  is assumed to be irreducible. The

more detailed description of the PH -type distribution and its partial cases can be found,
e.g., in the book (Neuts, 1981). Usefulness of PH distribution in description of service
Queues with
process in telecommunication networks is stated, e.g., in (Pattavina & Parini, 2005) and
(Riska et al., 2002).
It is intuitively clear that the described mechanism of arrivals restriction by means of tokens
is reasonable. At the expense of some sessions rejection, it allows to decrease the sojourn
time and jitter for admitted sessions. This is important in modeling real-world systems
because the quality of transmission of accepted information units should satisfy imposed
requirements of Quality of Service. Quantitative analysis of advantages and shortcomings of
this mechanism as well as optimal choice of the number of tokens requires calculation of the
main performance measures of the system under the fixed value K of tokens in the system.
These measures can be calculated based on the knowledge of stationary distribution of the
random process describing dynamics of the system under study.
3. Stationary distribution of the system states

Let us assume that the number K  K  1 of tokens is fixed and let
 it be the total number of requests in the system, it  0
 kt be the number of sessions having token for admission to the system,
kt  0 K 
  t and t be the states of the directing processes of the MAP arrival
process and PH service process,  t  0 W  t  1 M
at the epoch t t  0
Note that when it  0 i.e., requests are absent in the system, the value of the component t 
which describes the state of the service directing process, is not defined. To avoid special
treatment of this situation, without loss of generality, we assume that if the server becomes
idle the state of the component t is chosen randomly according to the probabilistic vector
 and is kept until the next service beginning moment.
It is obvious that the four-dimensional process t  {it  kt  t t } t  0 is the irreducible
regular continuous time Markov chain.
Let us enumerate the states of this Markov chain in lexicographic order and refer to (i k ) as
macro-state consisting of M 1  ( W  1)M states (i k  )   0 W    1 M
For the use in the sequel, introduce the following notation:
     (1   )           I M1       I M1     I M1 
 C K  diag{ 0 1… K} is the diagonal matrix with the diagonal entries
{ 0 1… K} C K  C K  I M1 
 RK  diag{ 1… K}  I M1  I is an identity matrix, O is a zero matrix;

O O O … O O  0 1 0 … 0
  O … O O  0 0 1 … 0 
  
AO 2 2  … O O  E    

   
   
        0 0 0 … 1
  0 0 0 … 0 
O O O … K  K   

  O … O O  0 0 0 … 0
   2  … O O  0 0 0 … 0 
  
A1   O 2   … O O 
 E        
   
       0 0 0 … 0
O O O… ( K  1)  K   0 0 0 … 1 
 
  i  j is Kronecker delta,  i  j is equal to 1, if i  j and equal to 0 otherwise;
  is the symbol of Kronecker product of matrices;
  is the symbol of Kronecker sum of matrices;
 bT denotes transposed vector b .
Let Q be the generator of the Markov chain t  t  0 with blocks Qi  j consisting of

intensities (Qi  j )k  k  of the Markov chain t  t  0 transitions from the macro-state (i k ) to
the macro-state ( j k) k k  0 K The diagonal entries of the matrix Qi  i are negative and the
modulus of the diagonal entry of (Qi  i )k  k defines the total intensity of leaving the
corresponding state (i k  ) of the Markov chain. The block Qi  j  i j  0 has dimension
K 1  K 1  where K 1  (K  1)M 1 
Lemma 1. Generator Q has the three block diagonal structure:
 Q0 0 Q0 O O …
Q Q1 Q0 O … 
Q
2
 O Q2 Q1 Q0 …
 
      
where non-zero blocks Qi  j are defined by
Q0  0  A  I K  1  D0  I M  E  D1  I M 
Q1  A  I K  1  (D0  S )  E  D1  I M 
Q0   C K  E  D1  I M 
Q2  I K  1  I W  1  S 0  
Proof of the lemma consists of analysis of the Markov chain t  t  0 transitions during the
infinitesimal interval of time and further assembling the corresponding transition intensities
into the matrix blocks. Value   is the intensity of a token releasing due to the finish of the
session arrival,   is the intensity of a new request in the session arrival.
Let us investigate the Markov chain t  t  0 defined by the generator Q To this end, at first
we should derive conditions under which this Markov chain is ergodic (positive recurrent).
Queues with
Theorem 1. Markov chain t  {it  kt  t t } t  0 is ergodic if and only if the following
inequality is fulfilled:
K K 1
       kx k e W  1   x k D1e W  1  (1)
k 0 k 0
where  is the average service rate defined by
 1  b1   ( S )1 e
and x  ( x 0 … x K ) is the vector of the stationary distribution of the system MAPM K  0
with the MAP arrival process, defined by the matrices D0 and D1 and the average service
rate   .
Proof. It follows from (Neuts, 1981) that the ergodicity condition of the Markov chain
t  {it  kt  t t } t  0 is the fulfillment of inequality
yQ2 e  yQ0 e (2)
where the row vector y is solution to the system of linear algebraic equations of form
y(Q0  Q1  Q2 )  0 ye  1 (3)
It is easy to verify that
Q0  Q1  Q2  B  I M  I( K  1)( W  1)  (S  S0  )
where B is the generator of the Markov chain, which describes behavior of the
MAP  M  K  0 system with the MAP arrival process defined by matrices D0 and D1 and
average service rate   :
 D0 D1 O … O O 
  I D   I D … O O 
 0 1 
B 0 2  I D0  2  I … O O 
 
       
O O O … K  I D(1)  K  I 

According to the definition, vector x satisfies equations
xB  0 xe  1 (4)
By direct substitution into (3), it can be verified that the vector y which is solution to the
system (3), can be represented in the form y  x   , where  is the unique solution of the
system of linear algebraic equations
 (S  S0  )  0 e  1 (5)
By substituting vector y  x   into inequality (2), after some transformations we get

inequality (1). Theorem 1 is proven.
In what follows we assume that condition (1) is fulfilled. Then the following limits
(stationary probabilities) exist:
 (i k  )  lim P{it  i kt  k  t    t   } i  0 k  0 K    0 W    1 M

t 
Let us combine these probabilities into the row-vectors
 (i k )  ( ( i k  1)  ( i k  2)…  (i k  M ))

 (i k )  ( ( i k 0)  (i k 1)…  ( i k W ))
 i  ( (i 0)  (i 1)…  (i K )) i  0
The following statement directly stems from the results in (Neuts, 1981).
Theorem 2. The stationary probability vectors  i  i  0 are calculated by
 i   0 R i  i  0
where the matrix R is the minimal non-negative solution to the equation
R 2Q2  RQ1  Q0  O
and the vector  0 is the unique solution to the system of linear algebraic equations
 0 (Q0 0  RQ2 )  0  0 ( I  R )1 e  1
Having stationary probability vectors  i  i  0 been computed, we can calculate different

performance measures of the system. Some of them are given in the following statements.
Corollary 1. The probability distribution of the number of requests in the system is
computed by
lim P{it  i}   i e i  0
t 
The average number L of requests in the system is computed by

L   i i e   0 R( I  R )2 e
i 0
Queues with
The probability distribution of the number of sessions in the system is computed by

lim P{kt  k}   (i k )e   0 ( I  R )1 (e( k )  e ) k  0 K 
t 
i 0
where the column vector e( k ) has all zero entries except the k th one, which is equal to 1,
k  0 K 
The average number Z of sessions in the system is computed by
K  K
Z    k (i k )e   0 ( I  R )1  k( e( k )  e )
k 1 i 0 k 1
The distribution function R(t ) of a time, during which arrivals from an arbitrary session
occur, is computed by
t

 ( y )l  1
R(t )  (1   )   l  1 e  y dy  1  e  (1  )t 
l1 0 (l  1)
The average number T of requests processed by the system at unit of time (throughput) is
computed by
 K W M
T       (i k  )(S0 )   0 R( I  R )1 ( e( K  1)( W  1)  S0 )
i 1 k 0  0  1
Remark 1. In contrast to the model with a finite buffer, see (Lee et al., 2007) and (Kim et al.,
2009), where the arriving session can be rejected not only due to the tokens absence but also
due to the buffer overloading, distribution of the number of sessions in the model under
study does not depend on the number of requests in the system. It is defined by formula
lim P{kt  k}  x k e k  0 K 
t 
where the vectors x k  k  0 K  are the entries of the vector x  ( x 0 … x K ) which satisfies the
system (5). However, distribution  (i k ) i  0 k  0 K  does not have multiplicative form
because the number of requests in the system depends on the number of sessions currently
presenting in the system.
Remark 2. It can be verified that the considered model with the infinite buffer has the steady
state distribution of the process t  {it  kt  t t } t  0 coinciding with the steady state
distribution of the queueing model of the MAPPH  1 type with the phase service time
distribution having irreducible representation (   S ) and the MAP arrival process defined
by the matrices D and D  having the form
0 1
 D0 O O … O O   O D1 O … O O 
 I D   I O … O O  O  I D1 … O O 
 0  
  0
D 2  
I D0  2 I … O O  0 O
 D 2  I … O O 
0 1
   
              
O O O … K  I D(1)  K I  O O O … O K  I 
 
It is easy to verify that the fundamental rate of this MAP is equal to  which is defined in
(1). So, stability condition (1) is intuitively clear: the average service rate should exceed the
average arrival rate. Note that the first summand in expression
K K 1
     kx k e W  1   x k D1e W  1  for the rate  represents the rate of requests from already
k 0 k 0
accepted sessions, i.e., the rate of requests who are not the first in a session. The second
summand is the rate of the sessions arrival.
Theorem 2. The probability Pb( loss ) of an arbitrary session rejection upon arrival is computed
by

(D1  I M ) D1
Pb( loss )    (i K ) e  xK e
i 0  
The probability Pc( loss ) of an arbitrary request rejection upon arrival is computed by

(D1  I M ) D
Pc( loss )    ( i K ) e  xK 1 e
i 0

 
    x D e
where  K 1
Proof of formula for probability Pb( loss ) accounts that the session is rejected upon arrival if
and only if the number of sessions in the system at this epoch is equal to K . So
  (i K )(D 1  I M )e 
(D1  I M )
Pb( loss )  i 0
 K
   ( i K ) e

   (i k )(D
i 0 k 0
1  I M )e i 0
Rejection of a request can occur only if this request is the first in a session and the number of
sessions in the system at this session arrival epoch is equal to K . So
  (i K )(D 1  I M )e 
D1  I M
Pc( loss )  i 0
   ( i K ) e
 K 

   (i k )(D
i 0 k 0
1  k  I )  I M e i 0
Queues with
4. Distribution of the sojourn times

Let Vb ( x ) Vc ( x ) and Vc( a ) ( x ) be distribution functions of the sojourn time of an arbitrary
session, an arbitrary request, which is the first in a session, and an arbitrary request from the
admitted session, which is not the first in this session, in the system under study, and vb (s )
vc (s ) and vc( a ) (s ) Re s  0 be their Laplace-Stieltjes transforms (LSTs):
  
vb ( s )   e  sx dVb ( x ) vc ( s )   e  sx dVc ( x ) vc( a ) (s )   e  sx dVc( a ) ( x )
0 0 0
Formulae for calculation of the LSTs vc (s ) and vc( a ) (s) are the following:
1 
vc (s )  [ 0 (D1  I M )e (sI  S )1 S0    i (D1  I M )(e( K  1)( W  1)  ((sI  S )1 S0 ))(  ( sI  S )1 S0 )i ] 
 i 1
1
 [ 0 (D1  I M )e (sI  S )1 S0   0 R(  (sI  S )1 S0 ) 

( I  R(  (sI  S )1 S0 ))1 ( D1  I M )(e( K  1)( W  1)  ((sI  S )1 S0 ))]
1 K
vc( a ) ( s )  K   k 
[ (0 k )e (sI  S )1 S0 
  k
k 1 i 0

 (i k )e k 1

   (i k )(e W  1  ((sI  S )1 S0 ))(  (sI  S )1 S0 )i ]
i 1
Formulae for the average sojourn time Vc of an arbitrary request, which is the first in a
session, the average sojourn time Vc of an arbitrary non-rejected request, which is the first
in a session, and the average sojourn time Vc( a ) of an arbitrary request from the admitted
session, which is not the first in this session, are as follows:
1 
Vc  [ 0 (D1  I M )eb1    i (D1  I M )((e( K  1)( W  1)  ( S )1 e )  eib1 )] 
 i 1

 0 [( I  R( I  R )2 )(D1  I M )eb1  R( I  R )1 (D1  I M )(e( K  1)( W  1)  ( S )1 e ]

Vc
Vc  
1  Pb( loss )
K 
 k 
[ (0 k )eb1    (i k )(( e W  1  ( S )1 e )  eib1 )]
Vc( a )  k 1
K
i 1


  k
k 1 i 0

 ( i k )e
If the service time distribution is exponential, expression for the average sojourn time Vc of
an arbitrary arriving request, which is the first in a session, becomes simpler:
0
Vc  b1 ( I  R )2 D1e

Derivation of formula for calculation of the LST vb (s ) is more involved. Recall that the
sojourn time of an arbitrary session in the system lasts since the epoch of the session arrival
into the system until the moment when the arrival of a session is finished and all requests,
which belong to this session, leave the system. We will derive expression for the LST vb (s )
by means of the method of collective marks (method of additional event, method of
catastrophes), for references see, e.g., (Kasten & Runnenburg, 1956) and (Danzig, 1955). To
this end, we interpret the variable s as the intensity of some virtual stationary Poisson flow
of catastrophes. So, vb (s ) has meaning of probability that no one catastrophe arrives during
the sojourn time of an arbitrary session.
We will tag an arbitrary session and will keep track of its staying in the system. Let
v( s i l k  ) be the probability that catastrophe will not arrive during the rest of the tagged
session sojourn time in the system conditional that, at the given moment, the number of
sessions processed in the system is equal to k k  1 K  the number of requests is equal to
i i  0 the last (in the order of arrival) request of a tagged session has position number
l l  0 i in the system, and the states of the processes  t  t  t  0 are    . Position number
0 means that currently there is no one request of the tagged session in the system.
It follows from the formula of total probability that if we will have functions v( s i l k  )
been calculated the Laplace-Stieltjes transform vb (s ) can be computed by
1  K 1 W M W
vb (s )  Pb( loss ) 

      (i k  ) p 
i 0 k 0 0 1 0
(1)
   v(s i  1 i  1 k  1  ) (6)
The system of linear algebraic equations for functions v( s i l k  ) is derived by means of
formula of total probability in the following form:
W
v(s i l k  )  [   p(1)  ((1   k  K )v(s i  1 l k  1  )  (7)
  0
W
 k  K v(s i l k  ))    p(0)  v(s i l k  ) 
  0
M
(1   i  0 )(S0 )   [ v(s i  1 l  1 k  )(1   l 0 )  v(s i  1 0 k  ) l 0 ] 
  1
M
(1   i  0 )  (S )   v(s i l k  )    v(s i  1 i  1 k  ) 
  1
  ( k  1)v(s i  1 l k  )    ( k  1)v(s i l k  1  ) 

Queues with
  [(( sI  S )1 S0 ) (  (sI  S )1 S0 )l  1 (1   l 0 )   l 0 ]] 

(s    S   k )1  l  0 i i  0 k  1 K    0 W    1 M 
Let us explain formula (7) in brief. The denominator of the right hand side of (7) is equal to
the total intensity of the events which can happen after the arbitrary time moment:
catastrophe arrival, transition of the directing process of the MAP , transition of the directing
process of the PH service process, and expiring the time till the moment of possible request
arrival from sessions already admitted into the system. The first term in the square brackets
in (7) corresponds to the case when a new session arrives. The second term corresponds to
the case when transition of the directing process of the MAP occurs without new session
generation. The third term corresponds to the case when service completion takes place. The
fourth term corresponds to the case when the transition of the directing process of the PH
service process occurs without the service completion. The fifth term corresponds to the case
when the new request of the tagged session arrives into the system. In this case, the position
of the last request of the tagged session in the system is reinstalled from l to i  1 The sixth
term corresponds to the case when the new request from another session, which was
already admitted to the system, arrives. The seventh term corresponds to the case when
some non-tagged session terminates arrivals. The eighth term corresponds to the case when
the expected new request of the tagged session does not arrive into the system and arrival of
requests of the tagged session is stopped. This session will not more counted as arriving into
the system and the tagged request finishes its sojourn time when the last request, who is
currently the l th in the system, will leave the system. Number ((sI  S )1 S0 ) defines the
probability that catastrophe will not arrive during the residual service time conditional that
the directing process of the PH service is currently in the state   The number β (sI  S )1 S0
defines probability that catastrophe will not arrive during the service time of an arbitrary
request.
Let us introduce column vectors
v(s i l k )  ( v(s i l k  1)… v(s i l k  M ))T 

v(s i l k )  ( v(s i l k 0)… v(s i l k W ))T 
v(s i l )  ( v(s i l 1)… v(s i l K ))T 
v(s i )  ( v(s i 0)… v(s i i ))T  v(s )  ( v(s 0) v(s 1)…)T 
System (7) of linear algebraic equations can be rewritten to the matrix form as
(sI  Qˆ i  i )v(s i l )  Qˆ i  i  1v(s i  1 l )  Qˆ i  i  1v(s i  1 l  1)(1   0  l )  Qˆ i  i  1v( s i  1 0) 0 l  (8)
 I K    v(s i  1 i  1)    e K ( W  1)  ((sI  S )1 S0 )(  (sI  S )1 S0 )l  1  0TKM1  l  0 i i  0
where
Qˆ i  i  A1  I K  ( D0  S )(1   i  0 )  E  ((D1  I M ))  I K  ( D0  I M ) i  0  i  0
Qˆ i  i  1   C K  1  EK  (D1  I M ) i  0
 
Qˆ i  i  1  I K ( W  1)  S0   i  0 Qˆ 0 1  O
Let us introduce notation:

(s ) is three block diagonal matrix with non-zero blocks
i  j (s ) j  max{ 0 i  1 } i i  1 i  0
defined by
i  i (s )   I i  1  (sI  Qˆ i  i ) i  i  1  D3( i )  Qˆ i  i  1
 i  i  1  D1( i )  Qˆ i  i  1  D2( i )  I K    
Here the matrix D1( i ) of size (i  1)  ( i  2) is obtained from the identity matrix I i  1 by
means of supplementing from the right by the column 0Ti 1 The matrix D2( i ) of the same size
has the last column consisting of 1’s and other columns consisting of 0’s. The matrix D3( i ) of
size (i  1)  i is obtained from the identity matrix I i by means of supplementing from
above by the row (1 0… 0)
Vector B(s ) is defined by

B( s )  ( B0 (s )… BN (s )…)T
where
Bi (s )    (e K  e M1  e K  e W  1  (sI  S )1 S0  eK  e W  1  ((sI  S )1 S0 ) (sI  S )1 S0 …

eK  e W  1  ((sI  S )1 S0 )(  ( sI  S )1 S0 )i  1 )T  i  0
Using this notation we can rewrite the system (7) to the form
( s)v( s)  B( s)  0T  (9)
It can be verified that the diagonal entries of the matrix (s ) dominate in all rows of this
matrix. So the inverse matrix exists. Thus we proved the following assertion.
Theorem 3. The vector v(s ) consisting of conditional Laplace-Stieltjes transforms LST
v(s i l k  ) l  0 i i  0 k  1 K    0 W    1 M is calculated by
v(s )   1 (s )B(s ) (10)
Corollary 2. The average sojourn time Vb of an arbitrary session is calculated by
 K 1
(D1  I M ) v(s i  1 i  1 k  1)
Vb     (i k ) s  0 
i 0 k 0  s
Queues with
v ( s i 1i 1 k 1) dv ( s )

where column vectors s  s 0 are calculated as the blocks of the vector ds  s 0
defined by
dv(s ) d B( s )
s  0   1 (0)(  s  0  v(0))
ds ds
where v(0)     1 (0)e
Corollary 3. The average sojourn time Vb( accept ) of an arbitrary admitted session is calculated
by
Vb
Vb( accept )  
1  Pb( loss )
where Pb( loss ) is probability of an arbitrary session rejection upon arrival.
5. Optimization problem and numerical examples

It is obvious that the most important from economical point of view characteristic of the
considered model is the throughput T of the system because it defines the profit earned by
information transmission. If the number K that restricts the number of sessions, which can
be served in the system simultaneously, increases the throughput T of the system increases
and the probability Pb( loss ) of an arbitrary session rejection upon arrival decreases. So, it
seems to be reasonable to increase the number K as much as possible until stability
condition (1) is violated. However, such performance measures as the average sojourn time
of an arbitrary request and jitter are also very important because they should fit
requirements of Quality of Service. These performance measures become worse if the
number K grows. Evidently, it does not make sense to admit too many sessions into the
system simultaneously and provide bad Quality of Service (average sojourn time and jitter)
for them. So, the system manager should decide how many sessions can be allowed to enter
the system simultaneously to fit requirements of Quality of Service and to reach the
maximally possible throughput.
Thus, one should solve, e.g., the following non-trivial optimization problem:
T  T (K )  max (11)
subject to constraints (1) and
Vc  V  (12)
where V is the maximal admissible value of the sojourn time of the first request from non-
rejected session and is assumed to be fixed in advance.
This optimization problem can be easy solved by means of computer, based on presented
above expressions for the main performance measures of the system, by means of
enumeration, i.e., increasing the value K until constraints (1) and (12) are violated. The
optimal value of K in the optimization problem (1), (11), (12) will be denoted by K  
Corresponding computer program allows to validate the feasibility of such an optimization
algorithm and to illustrate the dependencies of the system characteristics on the system
parameters and the value of K In what follows several illustrative examples are presented.
Before to start description of these examples, let us mention that numerous experiments
show that the famous Little’s formula holds good for the system under study in the form
 L  Vc  where L is the average number of requests in the system and Vc is the average
sojourn time of an arbitrary request which is the first in a session.
5.1. Dependence of probabilities Pbloss of an arbitrary session loss and Pcloss of an

arbitrary request loss on the number K of tokens and correlation in the sessions
arrival process
The experiment has two goals. One is to illustrate quantitatively the dependence of
probabilities Pbloss of an arbitrary session loss and Pcloss of an arbitrary request loss on the
number K of tokens. The second goal is to show that for several different arrival processes
having the same average rate but different correlation this dependence is quite different.
This explains the importance of consideration of the model with the MAP arrival process of
sessions, which can be essentially correlated in real telecommunication networks, instead of
analysis of simpler model with the stationary Poisson arrival process of sessions.
We consider six different MAPs having the same fundamental rate   1 The first MAP is
the stationary Poisson arrival process. Variation coefficient of inter-arrival times is equal to
1. Four other MAPs have the variation coefficient equal to 2 but different coefficients of
correlation of successive intervals between sessions arrival. These four MAPs are described
as follows.
 MAP ( IPP  Interrupted Poisson Process ) flow with correlation coefficient equal
to 0 is defined by the matrices
 04 016 024   0 0 0
   
D0   13 694 681   D1   0 0 0 
 13 
13 270   
  1002 1672 0 
 MAP flow with correlation coefficient equal to 0.1 is defined by the matrices
 266 012 012   23 008 004 
   
D0   013 05 008   D1   009 018 002  
 014 008 032   05 001 004 
   
 316 012 012   284 006 002 
   
D0   01 045 009   D1   002 021 003  
 011 039   
 012  002 004 01 
 511 008 007   485 009 002 
   
D0   0029 0446 004   D1   0007 0333 0037  
 006 008 035   0 005 016 
 
Queues with
 The sixth MAP has correlation coefficient equal to -0.16 and the squared
correlation coefficient equal to 1.896. It is defined by the matrices
 3607 0   0347 326 
D0     D1   
 0 0617   0478 0139 
The service time distribution is Erlangian of order 2 with intensity of the phase equal to 16.
The rest of the parameters are the following:   2   09
Figures 1 and 2 illustrate the dependencies of probability Pbloss of an arbitrary session loss
and Pcloss of an arbitrary request loss on the number K of tokens for the listed above
different MAP s with the same fundamental rate but the different correlation.
Fig. 1. Dependence of probability Pbloss of arbitrary session loss on the number of tokens K
Fig. 2. Dependence of probability of an arbitrary request loss on the number of tokens K
One can pay attention that the curves corresponding to the different MAP s terminate at the
different points, e.g., the curve corresponding to the stationary Poisson process terminates at
the point K  5 the curve corresponding to the MAP s having correlation coefficient 0.3
terminates at the point K  12 The reason of termination is that the stationary distribution
existence condition violates for K larger than 5 and 12 correspondingly.
It is worth to mention, that the previous analysis of different queues with the Batch
Markovian Arrival Process given in many papers shows that usually the stability condition
depends on the average arrival rate, but does not depend on correlation. It the model under
study, stability condition (1) depends on correlation as well. This has the clear explanation:
stability condition includes the stationary distribution of the corresponding MAPM K  0
queueing system which describes the behavior of the number of busy tokens. As it is
illustrated in (Klimenok et al., 2005), this distribution essentially depends on the correlation
in the arrival process.
Conclusion that can be made based on these numerical results is the following: higher
correlation of the session’s arrival process implies higher value of Pbloss and Pcloss but larger
number of sessions which can be simultaneously processed in the system without
overloading the system. IPP process violates this rule a bit. This is well known very special
kind of arrival process. It has zero correlation. Intervals where arrivals occur more or less
intensively alternate with time periods when no arrivals are possible. Such irregular arrivals
make the IPP violating the conclusion made above. Note that the system with the negative
correlation in the arrival process has characteristics close to characteristics of the system
with the stationary Poisson process. While the more or less strong positive correlation
changes these characteristics essentially.
5.2. Dependence of the throughput of the system on the number of tokens and
correlation in the sessions arrival process
Let us consider the same system as in the previous experiment and consider optimization
problem (11), (12) where the limiting value of the average sojourn time for the first request
in non-rejected session is assumed to be V  40 Figure 3 illustrates the dependence of the
throughput T of the system on the number of tokens K As it is expected, the throughput
T is the increasing function of K for all arrival processes. However, the shape of this
function depends on the correlation in the sessions arrival process. The lines corresponding
to the different MAP s terminate when condition (12) is not hold true. So, as it is seen from
Figure 3 the optimal value K  of tokens is equal to 5 when the arrival process is the
stationary Poisson or has the negative correlation or is equal to 0.1 and is equal to 6 for the
rest of the arrival processes.
It is seen from Figures 1-3 that positive correlation has the negative impact on the system
performance. Although the number of simultaneously processed requests can be larger, loss
probability is higher and the throughput of the system is lesser.
Dependence of the average sojourn time Vc for the first request in non-rejected sessions on
the number of tokens in these examples is presented on Figure 4.
Fig. 3. Dependence of the throughput T of the system on the number of tokens under
restriction Vc  40
Queues with
Fig. 4. Dependence of Vc on K

It is seen that the average sojourn time Vc sharply increases when the number of tokens K
approaches the value K  5 or K  6 depending on correlation in the arrival process. For
the model with the stationary Poisson arrival process stationary distribution does not exist
for K  6
5.3. Dependence of the optimal number of tokens on the session size, arrival and
service rates
The goal of this experiment is to illustrate the dependence of the optimal number of tokens
on the session size, average arrival and average service rates.
Firstly, let as clarify the impact of the session size. We assume that the MAP process of
sessions is defined by the matrices
 688 00008   68 00792 

D0     D1   0016 02032  
 0 0008 0 22   
This MAP has the average rate equal to 1.37, correlation coefficient 0.4 and the squared
variation coefficient 9.4. As in the previous examples, the service time distribution is
assumed to be Erlangian of order 2 with the intensity of the phase equal to 16.
On Figure 5, we vary the parameter   which characterizes the distribution of the number
of requests in a session, in the interval [01 08] This implies that the average session size
varies in the interval [1111 5] Parameter V defining the limiting value of the average
sojourn time for the first request in non-rejected sessions is assumed to be equal to 0.8.
On Figure 6, we vary the parameter  in the interval [08 098] This implies that the
average session size varies in the interval [5 50] Parameter V is assumed to be equal to 8.
As it is expected, the optimal number K  is non-increasing function of   When 
increases from 0.1 to 0.8 the number K  decreases from 8 to 1 under restriction Vc  08. If
we take  greater than 0.8, restriction Vc  08 is not fulfilled even only 1 session is allowed
to enter the system. If the weaken this restriction to the restriction Vc  8 four sessions can
be processed in the system simultaneously for  equal to 0.8. Situation when restriction
Vc  8 is not fulfilled even only 1 session is allowed to enter the system occurs for  greater
than 0.98.
Fig. 5. Dependence of the optimal number K  of tokens on the parameter  under

restriction Vc  08
Fig. 6. Dependence of the optimal number K  of tokens on the parameter  under

restriction Vc  8
In the next example, we illustrate the impact of the average arrival rate. We consider the
IPP process defined above and vary the average arrival rate in the interval [1 11] by means
of multiplication of the matrices D0 and D1 by the corresponding factor. The service time
distribution is assumed to be Erlangian of order 2 with the intensity of the phase equal to 30.
Parameter V defining the limiting value of the average sojourn time is assumed to be equal
to 4. Figure 7 shows the dependence of the optimal number K  of tokens on the average
arrival rate  
Fig. 7. Dependence of the optimal number K  of tokens on the average arrival rate 
Queues with
As it is expectable, the optimal number K  of tokens decreases when  is increasing. The

same dependence takes place for other MAP s, only the points of the jumps of the lines are
different.
In the last example, we illustrate the impact of the average service rate. We consider the IPP
process defined above having the average arrival rate   1 The service time distribution is
assumed to be Erlangian of order 2 with intensity of the phase varied to get the average
service rate in the interval [35 20] Parameter V defining the limiting value of the average
sojourn time is assumed to be equal to 5.
Figure 8 shows the dependence of the optimal number K  of tokens on the average service
rate  
Fig. 8. Dependence of the optimal number K  of tokens on the average service rate 
As it is expectable, the optimal number K  of tokens increases when  is increasing. The

same dependence takes place for other MAP s, only again the points of the jumps of the
lines are different.
6. Conclusion
In this paper, the novel infinite buffer queueing model with session arrivals distributed in
time is analyzed. Ergodicity condition is derived. Joint distribution of the number of
requests in the system and number of currently admitted sessions is computed. The sojourn
time distribution of an arbitrary request and arbitrary session is given in terms of the
Laplace-Stieltjes Transform. Usefulness of the presented results is illustrated numerically.
Validity of Little’s formulas is checked by means of numerical experiment.
Results are planned to be extended to the systems with many servers, non-geometrical
session size distribution, and heterogeneous arrival flow.
Acknowledgement
This work was supported by World Class Univ. R32-2008-000-20014-0 NRF and KRF-2007-
521-D00330, Korea.
7. References
Asmussen, S.; Nerman, O. & Olsson, M. (1996) Fitting phase-type distributions via the EM
algorithm. Scandinavian Journal of Statistics, Vol. 23, pp. 419-441.
Chakravarthy, S.R. (2001) The session Markovian arrival process: a review and future
work, in: A. Krishnamoorthy, et al. (Eds.). Advances in Probability Theory and
Stochastic Process: Proc., Notable Publications, pp. 21-49.
Danzig, D. (1955) Chaines de Markof dans les ensembles abstraits et applications aux
processus avec regions absorbantes et au probleme des boucles. Ann. de l’Inst. H.
Pioncare, Vol. 14, pp. 145-199.
Fisher, W. & Meier-Hellstern, K.S. (1993) The Markov-modulated Poisson process
(MMPP) cookbook. Performance Evaluation, Vol. 18, pp. 149-171.
Heyman, D.P. & Lucantoni, D. (2003) Modelling multiple IP traffic streams with rate
limits, IEEE/ACM Transactions on Networking, Vol. 11, pp. 948-958.
Kasten, H. & Runnenburg, J. Th. (1956) Priority in waiting line problems, Mathematisch
Centrum, Amsterdam, Holland.
Klemm, A.; Lindermann, C.; & Lohmann, M. (2003) Modelling IP traffic using the
session Markovian arrival process. Performance Evaluation, Vol. 54, pp. 149-173.
Klimenok, V.; Kim, C.S.; Orlovsky, D. & Dudin, A. (2005) Lack of invariant property of
Erlang loss model in case of the MAP input. Queueing Systems, Vol. 49, pp. 187-213.
Kim, C.S.; Dudin, S. & Klimenok, V. (2009) The MAPPH  1N queue with time phased
arrivals as model for traffic control in telecommunication networks. Performance
Evaluation, Vol. 66, pp. 564-579.
Kist, A.A.; Lloyd-Smith, B. & Harris, R.J. (2005) A simple IP flow blocking model.
Performance Challenges for Efficient Next Generation Networks. Proceedings of 19-
th International Teletraffic Congress, 29 August – 2 September 2005, Beijing, pp. 355-
364.
Lee, M.H.; Dudin, S. & Klimenok, V. (2007) Queueing Model with Time-Phased Session
Arrivals. In: Managing Traffic Performance in Converged Networks Proceedings 20th
International Teletraffic Congress, Ottawa, Canada, June 2007. Berlin: Springer.
Lecture Notes in Computer Science, Vol. 4516. pp. 716-730.
Lucantoni, D.M. (1991) New results on the single server queue with a session Markovian
arrival process. Communications in Statistics-Stochastic Models, Vol. 7, pp. 1-46.
Neuts M.F., (1981) Matrix-geometric solutions in stochastic models. Baltimore, The Johns
Hopkins University Press.
Panchenko, A. & Buchholz, P. (2007) A hybrid algorithm for parameter fitting of
Markovian Arrival Process, in: Al-Begain K, Heindl A, Telek M. (Eds.). 14th Int.
Conf. "Analytical and stochastic modelling technique and applications". Prague, 2007, pp.
7-12.
Pattavina, A. & Parini, A., (2005) Modelling voice call inter-arrival and holding time
distributions in mobile networks, in: Performance Challenges for Efficient Next
Generation Networks - Proc.of 19th International Teletraffic Congress, Aug.-Sept. 2005,
pp. 729-738.
Riska, A.; Diev, V. & Smirni, E. (2002) Efficient fitting of long-tailed data sets into
hyperexponential distributions, Global Telecommunications Conference
(GLOBALCOM’02, IEEE), 7-21 Nov. 2002, pp. 2513-2517.
Telecommunication
Power System: energy saving, renewable sources and environmental monitoring 145
X8
Telecommunication Power System: energy

saving, renewable sources and
environmental monitoring
Carmine Lubritto
Department of Environmental Science,
II University of Naples
ITALY
1. Introduction
The considerable problems deriving from the growth of energetic consumptions and from
the relevant environmental “emergency” due to the emissions of greenhouse gases, push
people to find out new solutions and new technologies for the production of primary
energy fit for fulfilling the urging and growing energetic demands.
The global climate change, which is due to increased CO2 and other green house gases
concentration levels in atmosphere, is considered one of the most important global
emergency that requires immediate and effective policies (IPCC,2007). The CO2 emissions
are mostly due to the use of fossil fuels as energy source. Thus in the future the use of fossil
fuels has to be decreased. This can be obtained by improving energy efficiency and by using
large scale renewable energy sources.
This is also true in the telecommunication applications, which has seen, in the last years, a
remarkable increase in the number of installations present on the whole territory -
sometimes located in hardly reachable areas – and the relevant growth of energetic
consumptions, because of growing interest about new and reliable services in mobility calls
with an increase of the BTS operation hours and traffic management, in order to guarantee
the quality of the service anywhere and anytime.
As an example, in the last years the development of the telecommunications sector has
resulted in a significant increase of the number of Base Transceiver Station (BTS) on the
Italian territory: according to the official database of the non ionizing radiation observatory
of the National Agency for Environment (ISPRA,2007), the BTS present in Italy are about
60.000. On the other hand it has been shown (Roy, 2008) that the energetic consumptions of
a typical operator network varies between 1.5 TWh to 9 TWh for a year. What is alarming is
what we are faced in the future. On the wired line side it is expected a relevant growth of
the number of broadband subscribers with a power per subscriber rate that is 4 to 8 times
the traditional consumption, while on the wireless side the number of connected device with
high speed data services is growing. By extrapolation, it has been estimated (Roy, 2008) that
the telecom industry consumed last year about 1% of the global energy consumption of the
planet, that is the equivalent energy consumption of 15 million US homes and also the
equivalent CO2 emissions of 29 million cars.
Therefore, the reduction of the energetic consumptions of a Telecommunications Power

Systems represents one of the critical factors of the telecommunication’s technologies, both
to allow a sizeable saving of economic resources to the mobile communications system
management and to realize "sustainable" development actions. In other words improving
the energy efficiency of telecom networks is not just a necessary contribution towards the
fight against global warming, but with the rapidly rising prices of energy, it is becoming
also a financial opportunity.
Therefore clear and defined approaches must be taken to optimize actions of energy savings.
A telecom network is just like an eco-system: one cannot just apply any energy savings
actions without looking at the impacts on the other system components (Roy, 2008). It has
been proposed an “Energy Logic Method” which might be applied to both a wireless and a
wired line network. This approach is based on a holistic method to energy savings and
provides a complete roadmap of recommendations and quantifies their savings, reviewing
also the different impacts.
Starting from these considerations the research project “Telecommunication power systems:
energy saving, renewable sources and environmental monitoring” was launched by the
Department of Environmental Sciences of the Second University of Naples (DSA-SUN) and
the Institute for the Environmental Research (ISPRA), with the participation of the Italian
suppliers of mobile telecommunications (H3G, Vodafone, Telecom and Wind) and their
technological partners (Ericsson).
The general goal of the research project is to study a set of solutions which may allow:
a) to obtain a rationalization of the consumptions of a BTS through the intervention
on energy saving;
b) to produce, in the sites where the BTS are located, energy coming from renewable
sources - aiming to reduce the emissions of polluting agents in the atmosphere;
c) to implement intelligent monitoring systems for the energy consumptions and the
relevant impacts on the environment.
It has been evaluated, from a technical and economical point of view, the feasibility of some
solutions , including:
 Energetic auditing for a radio-telecommunication station in different operational

contexts (urban and rural areas, different periods in the year, different working
load, etc.);
 Interventions of efficiency and energy saving such as reduction of transmission
apparatus consumptions, optimization of air conditioning consumptions, efficiency
in the temperature control system;
 Evaluation and development of interventions and technical solutions based on the
production of a part of the energy used by radio-telecommunication apparatus,
through the use of photovoltaic cells on the infrastructures themselves;
Telecommunication
 Analyses of possible uses of other renewable sources (e.g. wind micro turbines)
generating energy usable for telecommunication power systems located in areas
not reached by the electricity network;
 Analyses of the social and environmental advantages in the introduction of
technologies based on renewable sources for covering a part of the energy
requirements of radio-telecommunication installations;
 Simulation studies useful to estimate the amount of energy that can be saved using
a software system that helps to use the BTS-GSM transmission power in a more
efficient way according to the telecommunication traffic features.
 Environmental monitoring of the sites where prototypal solutions has been
installed, aimed to compare the conditions before and after the intervention.
2. Wireless network energy consumption
The typical wireless network can be viewed as composed by three different sections:
 the Mobile Switching Center (MSC), that take care of switching and interface to
fixed network;
 Radio Base Station (RBS), which take care of the frequency interface between
network and mobile terminals;
 Mobile terminals, which is the subscriber’s part, normally limited to the handheld
device.
It is estimated that over 90% of the wireless network energy consumption is part of the
operator’s operating expenses (Scheck, 2008).
The key elements are the radio base stations because of the number of base stations is
relative high with relative high energy consumption. On the other hand as the number of
core network elements is low, the total energy consumption due to core network is relative
low. Finally the energy consumption of mobile terminals is very low due to the mobile
nature.
With these premises the ways to decrease energy consumptions of cellular network and thus
to reduce cost and CO2 emissions are:
 Minimizing BTS energy consumption;
 Use of renewable energy sources.
Moreover could also be considered a minimization of number of BTS sites in order to reduce
energy consumption: in this case the network design play an important role to implement a
telecom network with correct capacity and minimum number of sites at optimum locations.
The model used in this paper for a a typical Radio Base Station is shown in figure 1. It is the
same model analyzed and presented in the literature (Roy,2008; White Paper Ericsson 2007;
Lubritto, 2008). Analysing the proposed scheme it result that the system takes 10.3 kW of
electricity to produce only 120 watts of transmitted radio signals and to process the
incoming signals from the subscriber cell phones, with a total efficiency (output
power/input power) of about 1.2%.
Antenna
Total Power In Total Power
10,3 kW Out
Feeder
RF
DC Conversion &
AC Power Power
System
Signal
Cooling Processing &
Control
Fig. 1. RBS Block Diagram
In figure 2 is shown the energy allocation per function within the RBS (Roy, 2008). More
than 60% of the power is consumed by the radio equipment and amplifiers, 11% is
consumed by the DC power system and 25% by the cooling equipment, an air conditioning
unit, typical of many such sites. The Radio Equipment and the Cooling are the two major
sections where the highest energy savings potential resides.
11%
DC Power
Cooling
25%
Load
Feeder
62%
Radio Equipment
1%
1%
Fig. 2. Percent BTS Energy per function
In the framework of the energy saving, it is very important to consider a cascade effect that
represents in aggregate a benefit of 28 times: for example, saving 1W in the feeder cables
saves 17.3 watts of modulation and amplification losses, 3.3W of rectification losses and
7.1W of associated cooling energy (Roy, 2008).
Telecommunication
2.1 BTS Energy Savings Strategies

In the last years many Energy Savings Strategies for Wireless Network, applied both to the
radio equipment and to the cooling and power equipment. have been proposed (Roy, 2008,
White Paper Ericsson, 2007; Louhi, 2007; Cuccietti, 2006; Lubritto, 2008)
1. Remote Radio Units: consists in moving the RF converters and power amplifiers from the
base of the station to the top of the tower close to the antenna and connecting them via fiber
cables. This strategy offers the higher potential energy savings: most radio manufacturers
now offer this topology.
2 Radio Standby Mode The second strategy is very easy to implement and typically consists of
a software and basic hardware upgrade. Often termed ECO Mode or Power Saving Mode,
this strategy consists of turning radio transmitters and receivers off when the call traffic goes
down, typically during the night. When the ECO Mode is implemented, the power
consumption can be reduced by up to 40% under low traffic. Overall, this strategy will
reduce the consumption of the radio equipment between 10 to 20%, plus its associated
power conversion and cooling energy requirements. Overall this translates into cascaded
savings in the order of 6 to 7%.
3 Passive Cooling The third area of focus is the cooling. Such cooling requires, usually 1/3 of
the heat produced inside the RBS. It is also a noisy and maintenance intensive solution.
Depending on the geographic location, other cooling techniques such as free ventilation,
forced fan cooling with hydrophobic filtering or heat exchangers will change significantly
the energy consumption, and often yield a lower total cost of ownership. It is estimated that
passive cooling can provide energy savings above 10%.
4 Advanced Climate Control for Air Conditioners If an air conditioner remains necessary, one
can minimize its consumption operating at a higher temperature at opportune moments. By
doing so, the energy consumption is reduced for two reasons, one that the higher set point
means that the unit will be turned on less frequently, and two that it will run more
efficiently due to the higher temperature at the air exchange. A total savings of 3-4% can
safely be obtained without major availability impacts.
5 DC Power System ECO Mode The last two strategies relate to the DC power plant.
Evidently, at this stage, we have already reduced the load through the previous measures
and introduced the radio ECO mode to further reduce the load during low traffic periods.
An advanced system controller scheme can ensure that rectifiers will operate at their peak
efficiency over virtually all conditions
6 Higher Efficiency Rectifiers Last strategy is the use of higher efficiency rectifiers.
When all strategies are considered, total savings above more than 58% are possible
(Roy,2008):
 On the radio side, going to a Remote Radio concept and applying the Radio ECO
functionality will reduce the energy consumption by 40%.
 On the infrastructure side, the cooling costs can be improved by optimizing the use
of the air conditioner or preferably, by migrating to a more passive approach.
These will take to down by an additional 3% and 11% respectively, cumulatively
down by 54%.
 Finally, the last 4% of reduction will come from the DC plant by implementing
energy management to keep the rectifiers at their peak efficiency level and by
opting for higher efficiency rectifiers.
In the paragraph 3.2 results of specific studies concerning Radio Standby Mode Strategy will
be presented.
2.2 Renewable energy sources

As mentioned above a second way to reduce cost and CO2 emissions is the evaluation and
development of interventions and technical solutions based on the production of a part of
the power energy used by radio-telecommunication apparatus, through the use renewable
sources (e.g. photovoltaic cells, wind micro turbines or new alternative power based on fuel
cells) installed on the infrastructures themselves or usable from off-grid
telecommunications power systems. The use of alternative energy sources has been studied
in particular for sites that are beyond the reach of an electricity grid, or where the electricity
supply is unreliable or sites remote enough to make the regular maintenance and refueling
of diesel generators prohibitive (Morea, 2007; Boccaletti, 2007). The choice of alternative
energy source will depend on local conditions, BTS typology and energy consumption.
Solar and wind power can provide virtually free energy. Solar power is a mature technology
and can be used for low- and medium capacity sites. Apart from having very low
environmental impact, solar-powered sites also have the advantage of being very low-
maintenance, with a technical lifetime of 20 years or so, and much more reliable than diesel
generator-powered systems. Also, solar power scales with the load, so the size of the solar
installation can be matched to actual needs without unnecessary capacity.
On the other side a micro-wind turbine can support a traditional RBS site without too large
impact on cost, where on average a wind velocity is about 5 m/s. In most case a hybrid
solution combining of solar and wind is the actually the most feasible solution for
autonomous BTS site. Anyway the size of solar cell and wind turbine have to be defined
based on BTS load and on-site availability solar and wind. In the paragraph 3.4 a typical
hybrid solution for off-grid BTS is presented.
Finally fuel cells are increasingly being considered as a viable alternative site energy
solution for telecoms. They can be deployed in place of diesel generators, and partly replace
batteries, at remote sites with long back-up requirements. In addition to improving energy
efficiency, they can also improve network up-time and reliability. Moreover environmental
advantage in terms of special waste disposal will be obtained by using fuel cell in
substitution of backup battery.
Telecommunication
3. Telecommunications power systems and energy saving

Energetic auditing of a BTS is the most important step in the understanding of energy
management of wireless telecommunication power system. With this aim it has been
realized a campaign of measurements for a radio-telecommunication apparatus starting
from on-site measurements, performed in collaboration with Italian companies of mobile
communications systems (Vodafone, H3G, Telecom and Wind), which take into account
different technologies (GSM, UMTS, DCS+GSM+UMTS), different typologies of apparatuses
(outdoor, room, shelter), different locations (North, Centre, South of Italy) and different
working loads.
Thanks to the collaboration of the Italian mobile telecommunication providers, it has been
possible to retrieve data coming from a statistic sample of around 100 radio base stations
located on the whole national territory, that corresponds to more than 1000 monitoring
days. All the field measurements are performed by using specific monitoring systems
(Pizzuti, 2008).
For carrying out the statistic analyses and the correlations we considered separately the
following characteristics of the systems:
- Systems typologies (Shelter, Room, Outdoor)
- Systems technologies (UMTS, GSM, DCS)
- Localization (North, Centre and South)
and the following functioning parameters:
- Energy consumption (Wh)
- Instantaneous Power (W)
- Internal temperature (°C)
- External temperature (°C)
- Phone traffic for cells (erlang)
A specific database has been built containing all data related to: energetic consumptions,
BTS localization, typology and technologies, environmental parameters.
3.1 Energetic consumptions associated to a radio base station

Aim of our studies is to find statistic correlations between the energetic consumptions and
the operational parameters of the BTS. Moreover we are interested to study both the
energetic consumption correlated to the transmission function of the apparatuses and the
energetic consumptions related to the cooling of the equipments and infrastructures. To
achieve this goal, we made statistical analysis by using the software "R" of the "R-
Foundation for Statistical Computing" R-foundation for statistical computing” (www.r-
project.org/foundation/).
From these analysis one can establish the following:
 The average yearly consumption of a BTS is ca. 35500 kWh compatible with
the average consumption of 10 Italian families. If we consider that in Italy
there are 60.000 BTS (data 2007 coming from the NIR Observatory – ISPRA),
the total average yearly consumption of all the BTS systems present in Italy
is ca. 2,1 TWh/year; which is the 0,6 % of the whole national electrical
consumption (data source: TERNA 2007). In terms of economical and
environmental impact, the data correspond to ca. 300M€ yearly energy costs
and ca. 1,2 Mton of CO2eq emitted in the atmosphere every year.
 If we carry out analyses on the average energy consumptions associated to

the different technologies, we will note that the GSM energy consumptions
are considerably higher than the UMTS technology - as it is expected
because of the different mode of operation of the two technologies (Table 1)
Energy Consumption/technology
Technology kWh/day kWh/year

UMTS 72,97 26268
GSM 111,35 40085
Table 1. Energy consumption for GSM and UMTS technology
 To evaluate the daily energy consumptions and the contributions due to the
transmission and air-conditioning, in the following graph is represented the
daily trend of the energy consumptions of a BTS. We can clearly distinguish
two different trends: a constant energy consumption value of about 800Wh
for the nighttime and the morning, and an oscillating trend, with an average
value of about 1100Wh for the hottest hours and the late evening. This trend
is comprehensible if we consider that in the first case the consumption is due
only to the transmission functions (constant trend), whereas in the second
case at the transmission functions are added the conditioning energy
consumption, with a “saw tooth” trend generated by the switching on/off
of the air-conditioning systems. We can therefore divide the energy
consumptions in two contributions; i.e.: approx. 2/3 of the consumed daily
energy are due to the transmission, whereas 1/3 of the energy is used for
feeding the conditioning systems.
1400
1200
1000
800
Wh
600
Trasmission Trasmission +Conditioning
400
200
0
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
.0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0
00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00
0. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 2 0 21 22 23
Fig. 3. Daily Energetic Consumption of a BTS

Telecommunication
In order to verify our latter statement in figure 4 we show the power distribution
considering four selected days:
 15th December 2007: it represents the coldest day during the monitoring period in
which the internal temperature ranges from a minimum of 2,8°C to a maximum of
9,5 °C (the blue line);
 19th December 2007, it represents a typical day of the coldest period in which the
temperature ranges from a minimum of 7,0°C to a maximum of 17,6 °C (the violet
line);
 24th May 2008, it represents a typical day of the hottest period in which the
temperature varies from a minimum of 15,5°C to a maximum of 37,3 °C (the green
line);
 27th May 2008, it represents the hottest day of the monitoring period in which the
temperature ranges from a minimum of 21,8°C to a maximum of 42,5°C (the red
line).
There is a clear difference among the distribution functions on the coldest days (blue and
violet line), in which the instant power is presumably due only to the transmission function
and show a peak in correspondence of defined power values, and the distribution functions
on the warmest days (red and green lines), in which there are both the transmission and air
conditioning contributions .
In this way we can esteem an energetic consumption connected to the transmission
functions and another connected to the air-conditioning functions which are 2/3 and 1/3 of
the total energetic consumption, respectively.
Fig. 4. Density of states plotted as a function of power distribution

In figure 5 we plot the variation of the energetic consumptions versus the external
temperature, for different BTS typology (shelter, room, outdoor) and BTS technology (UMTS
and GSM).
Fig. 5. Energetic Consumption versus external temperature
A “universal trend” of the energy consumption versus the temperature is discovered

independently both from the technology used and from the typology of BTS. Furthermore
the higher dispersion of the consumption data in the room and shelter typologies, in
comparison with the outdoor one, can be comprehensible if we consider that in the first two
cases there is a higher thermal dispersion – with a consequent increase of the consumptions
– which is needed to air-condition the equipments as well as the environment where they
are implemented.
In figure 6 we plot the energy consumptions as a function of a different time of the year, for
different BTS technology (GSM and UMTS) and BTS typology of site (outdoor, room,
shelter). First of all we find a similar behaviour during the months with a greater energetic
consumption in summer respect to the winter, as expected.
Telecommunication
Fig. 6. Energetic consumptions versus. time of the year
Moreover comparing the graphs for the room and shelter typology, it is clear that the UMTS
technology has on average lower energetic consumption of GSM technology, because of the
different characteristic of the mobile communication standards. No difference was found for
energetic consumptions versus the time of year for the three BTS typologies.
For understanding an eventual correlation between the energy consumptions data and the
numbers of phone calls of a BTS, we show in figure 7, the energy consumptions as a
function of call traffic (in erlang). There seems to be no correlation between these two
parameters; that means that at the moment any intervention which may regulate BTS
turning on/off when the call traffic goes down, is realized.
Fig. 7. Energy consumption vs traffic for three different sites

3.2 Energy saving

Starting from the statistical analyses on the BTS energy consumptions, it is useful to study
the possible interventions for optimizing and saving energy consumptions. Our aim is to
individuate useful interventions both for saving action on conditioning systems and on
transmission consumption.
Regarding the saving of energy consumptions relevant to the air-conditioning we studied
two possible intervention strategies; the first one was based on “intelligent” algorithms for
the optimization and dynamical regulation of the air-conditioning functions, the second one
was founded on the local cooling of the single electronic equipment, avoiding to air-
condition the environment where they were located. Both hypotheses were based on the fact
that inside the shelter the thermo-dynamic parameters (temperatures, humidity, etc.) can
assume values ranges larger than in areas frequented by people. Therefore both the use of
air-conditioning intelligent systems and the local cooling technologies are useful strategies
for saving energy, basing on the possibility to eliminate useless conditioning actions of the
environment and mechanical parts. One can estimate that such interventions can achieve an
energy saving from 5% to 10% of the air-conditioning consumptions.
For optimizing the consumptions coming from the transmission functions we studied and
tested a software feature launched by Ericsson that helps to use the BTS-GSM transmission
power in a more efficient way (Hjorth, 2008). These algorithms can correlate the phone
traffic of a BTS and the energy consumptions. During periods of low network traffic, the
feature effectively puts transceivers that are not being used in standby mode - overcoming
the traditional practice of having radio equipment continually turned on, which results in
energy being wasted.
Depending on the network traffic pattern, the feature algorithm parameter settings and on
the type of apparatus, this innovation can save between 5 and 20 percent of the energy per
BTS when a base station is in use, while still providing the same services and quality to end
users.
The study has been carried out with a two fold approach:
 The direct investigation of test BTS in order to have direct hints about the
feasibility of the project, the practical problems, and the real savings. To reach this
goal on field measurements have been carried out in an operating BTS during
periods in which the feature was either activated or not activated.
 However, it is not feasible and practically unreliable to explore all the BTS power
consumption and to measure the power saved with different configurations of the
feature algorithm. Thus a simulation study has been carried out. The best
parameters ensuring good communications and best savings have been pointed
out.
A comparison between on field measurements and simulations have been realized, in order
to optimize the parameters used in the "BTS power saving algorithms".
Parameters useful in this kind of analysis are telephone traffic, BTS typology and location,
number of transmitters. Measurements of energetic consumptions and other environmental
parameters have been realized in a suburban BTS (provider Telecom Italia Mobile) located
in Agliana (Toscana), composed by 3 GSM transmitter and 3 DCS transmitter. On this BTS
the "power saving" function has been activated as experimental on field test of “power
saving algorithms”
Telecommunication
Fig. 8. BTS energetic consumption at the experimental site - Agliana
The results obtained by our monitoring are presented in the following figure 8, where we
reported the energy consumption of the basic radio station for three weeks; in two of those
was activated the “Power Saving” function. Rack 0-Rack 2 represents the energy
consumption of each DCS transmitter, Rack3 shows the energy consumption of all GSM
transmitters. It is clear that average value of energetic consumptions are lower in the days in
which the power saving function is active: a decrease of more than 10% of energetic
consumption is evident starting from the second period.
In the following graph (figure 9) it is presented the measured energy in the same day of the
week (Thursday), in the condition of power saving function ON (red line) and OFF (blue
line).
The considerable energy saving achieved during the nighttime - when the call traffic goes
down and the “power saving” algorithm can switch off many transmission supporters - is
here very remarkable. On the contrary, during the daytime, the curves trend coincides, since
the high traffic does not allow the switch off of any transmitters.
Fig. 9. Comparison of the measured daily energetic consumption in a day with Power
Saving ON (red line) and OFF (blue line).
The data of the experimental measures have been compared to the simulation analyses
realized through a specific software which has been implemented (Lubritto,2009).
The simulation software takes as input the available measured data (average number of
telephone calls started every ten seconds, the average length of calls, the cell parameters, the
parameters of the Power Savings algorithm) then it uses a Montecarlo algorithm for
simulation of telephone traffic.
The number of calls started every 10 seconds is supposed to follow a Poisson distribution
with a varying average value during the day. Length of calls follow an exponential
distribution with a given average. Both average values, measured every hour, are taken
from the collected data. The convolution of the two distributions gives the number of active
calls for any given time. However, the maximum number of calls in the cell is fixed by the
number of channel available.
Telecommunication
Fig. 10. Comparison of the simulated energy consumption in a day – with (red line) and
without (blue line) Power Saving features on.
In the above figure 10 we reported the values obtained by the simulation in a day; the blue
lines indicate the consumptions without Power Saving and the red lines indicate the
activation of the Power Saving. As it is evident, the simulation software faithfully
reproduces the data achieved by the field measurements.
The similarity with on field measurements is acceptable: here too the savings are evident in
periods of low traffic when there is the possibility to switch off the unused channels, as for
the measured data.
After established the goodness of the simulation algorithm, we can analyze the Power
Savings optimization parameters. In such a way we address the set of parameters giving the
maximum power saving, and thus lower environmental impact, while still guaranteeing the
communication quality.
The study has to be carried out with the following criteria:
[1] the intervals were chosen to cover the entire space of the suggested parameters;
[2] for each set of parameters were we took an average of many runs;
The resulting optimal set of parameters depends on the average communication load and
available channels, so should be site specific or at least cover “classes” of sites, with similar
load and operating behaviour. Optimal parameter sets for the sites under survey have
already been created, and an improvement in energy saving of about of 3% is expected.
The saving obtained through the introduction of the “Power Saving” algorithms might
change from 10% up to 15% of the transmission consumptions, according to the typology
and the location of the basic radio station. With the same algorithm we succeeded in
obtaining the Power Saving parameters set that maximizes the energy saving, increasing it
of a further 3%.
In conclusion, if we sum up the contributions of the energy saving to the air-conditioning

functions and the transmission functions, – considering the feedback processes of the two
components – we obtain a total yearly energy saving of 20% of the total consumptions (i.e.
7000 KWh per year) with an economical advantage of approximately 1000 €/year saved for
each BTS, that means a further environmental advantage in avoiding to emit, for each year
and BTS, 4 Ton of CO2eq in the atmosphere.
4. Telecommunications power systems and renewable energy

In order to introduce clean technology in the telecommunications power system
management, one has to consider the use of renewable sources technologies (photovoltaic,
wind, hybrid systems) installed on telecommunication systems infrastructures. We analyzed
the most advanced technological solutions in the photovoltaic sector (single crystal, multi-
crystal panels, amorphous silicon, thin films) and in the further renewable sources, useful
for producing the energy in situ in consideration of the functioning conditions and the
structural features of a radio-telecommunication station.
We chose to analyze the interventions in locations with different testing conditions for each
provider, in order to plan and realize photovoltaic systems for basic radio stations in urban
and rural areas (raw-land), to forecast the use of combustible cells for substituting the
batteries of a microcell site and to analyze the use of different renewable sources for grid-off
BTS (not connected to the electrical net), fed by generators
Therefore it has been realized an experiment in which use of photovoltaic systems and
other renewable sources for different typologies of contexts (urban and rural areas) and for
apparatuses not reached by the electricity network (stand-alone apparatuses), has been
tested.
Two rural sites in which a photovoltaic plant has been built in with provider Vodafone, in
order to understand how to use infrastructure of the BTS to obtain a total or partial
architectural integration of the photovoltaic plants on the shelter o support pole. It has been
shown that their energetic productivity depend on the geographical location, on the surface
available to implement the photovoltaic plants and on the effects of shadow.
In two of these pilot sites, photovoltaic plants have been realized both on shelter and on the
infrastructures; the area of PV modules varies from 16 to 20 m2, limited by the available site
space, to guarantee a production of 2.0 and 2,5 kWp. Figures 11 and 12 show the installation
on the two sites.
Telecommunication
Fig. 11. Photovoltaic system installed on the BTS infrastructure (Vodafone)
Fig. 12. Photovoltaic system installed on the BTS infrastructure (Vodafone)
These sites came into operation on 01/01/2008 and, according to data provided by
VODAFONE the two photovoltaic systems have produced, up to 30/05/2008, respectively
1100 and 1200 kWh; this implies an annual estimated production of 2640 and 2880 kWh. A
further important element of the application - made by the involved provider - was to
activate and complete the whole proceeding in order to obtain incentivizing by Italian
electricity Authority (i.e. “Conto Energia” fund). It is to be noted that such an application
gives an environment advantage of approximately of 3 Ton of not emitted CO2eq/year for
each BTS, apart all reduction of the pollution coming from furter physical agents.
In order to carry out specific controls on these pilot sites and to be able to compare the
conditions of energetic consumption pre and post installation of the photovoltaic panels, a
monitoring station has been used that allows to check operational parameters such as
electric consumption, external and inside temperature and other environmental parameters.
On these same sites we studied further technological solutions in the photovoltaic field. In
particular very important is to evaluate the possibility to implement further photovoltaic
technologies (amorphous, thin film) and to evaluate the technical feasibility and the
investment advantage for each solution.
A very convenient and realistic solution – from the technical and commercial point of view –
apart from the use of single and multi crystal panels - is the usage of panels based on the
amorphous and thin film.
In the amorphous solution one can consider an efficiency of about half of the crystalline,
well compensated by an implementing cost of 70% less than the crystalline solution. Using
the aforementioned option we get a 6 years payback time.
In the case of the thin film solution, we have to consider a reduction of efficiency of approx.
10% less than the amorphous solution. Furthermore the considered (sub-vertical) exposure
affects the energy production reducing it of the 30%. Beside this negativity, we noticed a net
gain in the installation costs, which is ca. 55% higher than the crystalline solution. Finally,
we could calculate the GSE government incentive basing on a system architectonically
integrated. With these hypotheses there would be 6 years payback time.
In general, in order to implement photovoltaic energy systems for feeding radio base stations
in a total or in a partial way in urban areas one have to take into account the presence of
obstacles which could cause shadowing effects and the real presence of useful areas.
A very interesting application of renewable energy source with telecommunication systems
sector is the situation when BTS is located far from the electrical network (grid-off systems).
In order to analyze innovative technologic solutions for producing energy close to the BTS,
we used as experimental site the BTS located in Sardinia on Gardaininu Mount (Nuoro,
provider TIM), for which to completely fulfill the BTS energetic requirements (approx. 35500
kWh a year), we considered to implement a micro-wind system (overhead generators with
an energy power up to 20 kW) coupled to a 4.5 kW photovoltaic system. Therefore it is
convenient to implement an hybrid solution composed by :
 a photovoltaic system with panels mounted on a support between the transmission
tower and the south side of the external border, since we noticed an available
surface of ca. 50 m2 was present;
 a micro-Wind system (power of 20kW), for which one can estimate energy production
starting from wind average speed and hours of productivity in the specific site;
In the aforementioned site in Sardinia, the annual average speed of the wind is 5 m/s (CESI
Wind Atlantis,2007) and the annual productivity is 2000 hours. Consequently electric energy
power production is 31567 kWh, and, the payback time is little more than 5 years.
5. Telecommunications power systems and environmental monitoring

In the framework considered in the present chapter, it becomes very important to study the
relation between energetic aspect, environmental impacts and radio-telecommunication
power systems.
One can consider at least three different relevant contexts:
 Impact of BTS on the landscape
 Electromagnetic pollution generated by the BTS
Telecommunication
 Environmental impact coming from further polluting agents (emission of

greenhouse gases, noise, etc.)
Regarding the impact on the environment, it is fundamental to build BTS integrated in their
territorial context. The possibility to use the structures of the plants (support poles, shelter,
etc.) in order to realize energy production systems deriving from renewable sources could
be an useful way to satisfy such request. It is evident that up to now such problems have not
been faced, but we believe it essential to define the right prescriptions in order to minimize
the impact of the BTS on the environment and, at the same time, to implement innovative
power energy solutions.
Besides, the possibility to decrease the emission quantity of the greenhouse gases deriving
from the BTS functioning is strictly joined to the introduction of the distributed production
system of renewable energy. In fact, the use of solutions based on renewable energy,
especially in grid-off conditions, can considerably reduce the impact on the environment
caused by the station and the feeding generators management. The usage of a micro-wind or
photovoltaic systems may avoid the implementing of feeding systems – which should be fed
with fuel every day with the relevant high use of vehicles. That would mean a remarkable
reduction of the emissions in the atmosphere.
Even more important is the impact on the environment and on the population of the
electromagnetic fields generated by the transmission systems. Especially during last years,
such a problem has attracted the attention of single citizens and of the entire civil community,
because of the considerable diffusion on the territory of low frequency electromagnetic fields
sources (power lines) and of high frequency sources (radio-television stations and mobile
phone stations). From a technical point of view, we should note that the field level in an area
can considerably change according to the status of the existing transmitters, to the emitted
power (at its turn variable according to the users’ need) and to the further territorial
characteristics, such as the presence of buildings and/or other obstacles which determine
reflections/refractions changeable in time. Basing on these premises and considering the
hypothesized energy power saving strategies, we evaluated the feedback on the minimization
of the electromagnetic fields emitted by the BTS. It is easy to understand that the use of a
“power saving” system can give a valid contribution to the reduction of the BTS emissions. In
fact, switching off the transmitters when the traffic goes down, means to get a null emission of
the electromagnetic fields and a reduction of the daily average value emitted by the radio base
stations. If we consider that, following to the activation of the “power saving” algorithm there
is a switch off of 70% of transmitters during nighttime (from 24 to 8), we can estimate a daily
average reduction of the electromagnetic emissions of 15-20%.
Acknowledgments
I would like to thank my colleagues and collaborator (in no particular order)– Salvatore
Curcuruto, Antonio D’Onofrio, Antonio Petraglia, Carmela Vetromile, Maria Logorelli,
Giuseppe Marsico, Floriana Caterina, Laura Miglio – for fruitful discussions and for the
many hours of working together on research project. This research project has been funded
by ISPRA.
6. References
Boccaletti C.; Fabbri G. & Santini E. (2007). Innovative Solutions for Stand Alone System
Powering, Proceedings of INTELEC, pp. 294-301, 978-1-4244-1628-8, Rome (Italy),
September 2007, IEEE, Rome
Efraimsson L. (2008). Halve the Power Related Costs for a Cluster of Diesel – Fed Telecom Sites,
Proceedings of INTELEC, pp. 397-400, 978-1-4244-2056-8, San Diego (CA), September
2008, IEEE, San Diego
Hjorth P. ; Lovehagen N. , Malmodin J. & Westergren K. (2008). Reducing CO2 Emissions from
Mobile Communications – BTS Power Savings and Tower Tube, Ericsson Review n. 1,
Ikebe H. ; Yamashita N. & Nishii R. (2007). Green Energy for Telecommunications, Proceedings
of INTELEC, pp. 750-755, 978-1-4244-1628-8, Rome (Italy), September 2007, IEEE,
Rome
IPCC, Fourth Assessment Report (AR4) Climate Change 2007 – International Panel of climate
Change, 2007
ISPRA, Non Ionizing Database - Italian Agency for Environmental Protection, 2007
Louhi J. T. (2007). Energy Efficiency of Modern Cellular Base Stations, Proceedings of INTELEC,
pp. 475-476, 978-1-4244-1628-8, Rome (Italy), September 2007, IEEE, Rome
C. Lubritto, A. Petraglia, C. Vetromile, F. Caterina, A. D’Onofrio, M. Logorelli, G. Marsico, S.
Curcuruto “New Energy” for Telecommunications Power System in Proceedings of
INTELEC, pp. 443-445, 978-1-4244-1628-8, Rome (Italy), September 2007, IEEE,
Rome
Curcuruto “Telecommunications power systems: energy saving, renewable sources and
environmental monitoring”- in Telecommunications Networks, Proceedings of INTELEC,
pp. 120-126, 978-1-4244-2056-8, San Diego (CA), September 2008, IEEE, San Diego.
Curcuruto, L. Miglio, F. Cenci “Simulation analysis and test study of BTS power saving
techniques” – Extended Abstract INTELEC 2009 – 18-22 /October/2009 Incheon
South Corea.
Masone M. (2007). Environmental Certification for Service Networks and Data Centers,
Proceedings of INTELEC, pp. 756-759, 978-1-4244-1628-8, Rome (Italy), September
2007, IEEE, Rome
Morea F. ; Viciguerra G. , Cucchi D. &Valencia C. (2007). Life Cicle Cost Evaluation of Off-Grid
PV-Wind Hybrid Power Systems, Proceedings of INTELEC, pp. 439-441, 978-1-4244-
1628-8, Rome (Italy), September 2007, IEEE, Rome
Pizzuti F. ; Rega G. & Grossoni M. (2008). Site Power Saving, Proceedings of INTELEC, pp. 84-
89, 978-1-4244-2056-8, San Diego (CA), September 2008, IEEE, San Diego.
Roy S. N. (2008). Energy Logic: A Road Map to Reducing Energy Consumption in
Telecommunications Networks, Proceedings of INTELEC, pp. 90-98, 978-1-4244-2056-8,
San Diego (CA), September 2008, IEEE, San Diego
White Paper Ericsson (2007). Sustainable Energy Use in Mobile Communications White Paper
EAB-07:021801 Ericsson AB 2007.
Propagation Models and their
Applications in Digital Television Broadcast Network Design and Implementation 165
X9
Propagation Models and their Applications in

Digital Television Broadcast Network Design
and Implementation
Armoogum V., Member IEEE, University of Technology, Mauritius
Soyjaudah K.M.S., University of Mauritius, Mauritius
Mohamudally N., University of Technology, Mauritius
Fogarty T., London South Bank University
1. Introduction
In this work, we will discuss the importance of propagation models when designing new
broadcast networks. We will hence consider the Island of Mauritius as a case study.
Mauritius is a tropical island, located in the Southern Hemisphere with geographic
coordinates 20o 17’ S, 57o 33’ E and has two climates. A humid and subtropical climate
dominates and becomes more effective below the 400-meter level on most of the south-
eastern side of the island as well as below 450 meters on the leeward side. Above these
altitudes, the climate is more temperate (Metz 1994). The island has two seasons. Summer is
hot and wet and lasts from November to April. The warmest months are January and
February with average temperatures of 35 ºC in the lowlands and 32 ºC on the central
plateau. The problem with warm weather is that cyclones are frequent. Cyclones, with
strong winds and heavy rain, can occur between November and April. Mauritius will
normally experience about three or four cyclones a year during this period, each usually
lasting two to three days. Winter season, which is influenced by the south-east trade wind,
is from May to October. During this period the weather is cool and dry. The average
temperature is 22 ºC in the lower parts of Mauritius and 16 ºC on the plateau and Plaines
Wilhems. Rainfall ranges from 90 cm per year in the western lowlands to 500 cm in the
central plateau with an average of 200 cm per year overall (Metz 1994). Humidity is
frequently high in Mauritius and is above 80% in the south and the central plateau. Hence,
the effects of climate can be investigated to know the degree of signal degradation.
The original motivation for this work came from a report by Pather (2000), in which he
described the penetration of digital television in Mauritius and how this would affect the life
of people. He predicted at that time that analogue television broadcasting would become
part of history and hence there would be no alternative than to venture within the digital
arena. He added that interactive digital television and T-Services would form part of
everyday life, as was the case for NTSC, SECAM and PAL fourty years ago. This new
technology would give better quality of service (QoS) and would be more economical.
Since the first proposal in 1990 for digital systems by General Instrument Corporation (GIS:
Company History online n.d.), there has been significant development in this area. The
United States, Europe and Japan developed their own standards between 1993 and 1994. In
this work we constrain our attention on the DVB-T standard used in European countries
and in Mauritius. The DVB-T standard was produced in 1997 following several
measurements and testing by various projects. Digital television is now an integral part of
the information superhighway that is being built to deliver large amounts of information at
very low cost compared to analogue technology and can be fully integrated into completely
digital transmission networks. Digital television can deliver more programs than traditional
analog television over one transmission channel and can be manipulated and treated in
various ways which were never possible with analog television. We can therefore store
digital images on computers and discs and play them continuously over digital networks
without signal degradation until a certain threshold value is reached. Pictures can be
modified, compressed, stored and transmitted. One advantage of the digital format is that it
can be integrated with telephone conversations as well as computer data and then
transmitted from one network to other broadcast networks. Furthermore, any program can
be stored on multimedia servers (soap and movie, fillers and jingles, advert, songs) and
retrieved instantly for broadcast to a single or multiple viewers on demand.
However, having good Quality of Service (QoS) and cheaper services do not mean that the
problems of transmission and reception of digital signal broadcasting are completely solved.
A correctly-formatted DTV signal is exposed to various factors which can detriment the
sound or picture quality before it reaches the intended customers. For that reasons, many
organizations and researchers are working on various areas such as transmitter and antenna
models, propagation and coverage failures, compression techniques and standards in order
to overcome implementation problems in view of setting frameworks and standards for
digital television implementation. In this study, we constrain our attention to propagation
models, coverage areas and failures.
2. Problem Statement
Following the liberalisation of the airwaves by the Government of Mauritius in 2002 and the
official launching of digital terrestrial television in October 2005, there are now more private
radio companies and will, in the near future, be more private television companies
operating in the country. The digital TV (DTV) signal, just like the analogue signal, suffers
from noise giving rise to problems such as ‘ghosts’ or tidal effects. A decoder in general is
able to tolerate loss in field strength. Independently of the threshold noise level that the
decoder can tolerate, our aim is to design a propagation model such that the signal despite
suffering from maximum attenuation does not go below the threshold level for the decoder,
as once the bit error rate crosses 2X10-4, a total loss of signal is obtained. The digital
coverage, therefore, has to be studied to obtain good QoS.
In the coming decades, the number of channels which will beam into homes and the number
of potential signal paths to the consumer will multiply. Due to a more sophisticated multi-
channel environment, there will be an even larger number of separate processes and
switching stages through which a radio or television signal will pass and, at any moment,
one signal may fail or detrimentally affect the picture or sound quality. Hence, the
opportunity for faults and failures occurring between the broadcaster and the consumer is
increasing. It will become more and more difficult for a broadcast engineer to know whether
a signal, after passing through all these separate processes and transmission paths, will
reach the consumer in a correct audio and picture format. An alternative solution to the
problems needs to be found.
3. Propagation Models
3.1 Overview
The common approaches to propagation modelling include:
(i) Physical models
Physical models of path loss make use of physical radio waves principles such as
free space transmission, reflection or diffraction.
(ii) Empirical models
Empirical models use measurement data to model a path loss equation. Examples
of empirical propagation models include the ITU-R and the Hata models. Empirical
models use what are known as predictors or specifiers in general statistical
modelling theory (Saunders 2005). To conceive these models, a correlation was
found between the received signal strength and other parameters such as antenna
heights, terrain profiles etc through the use of extensive measurement and
statistical analysis.
Prediction of path loss is an important element of system design in any communication
system. In the radio and TV broadcast systems, the prediction of path loss is very important
as the environment is constantly changing with time. The question that is always asked is
how to calculate the path loss with maximum accuracy. One solution is to use a propagation
model. A reliable propagation model is one which calculates the path loss with small
standard deviation. This will, hence, help network engineers and planners to optimise the
cell coverage size and to use the correct transmitted powers. Suitable models must be chosen
for prediction. An accurate and reliable prediction method helps to optimize the coverage
area, transmitter power and eliminates interference problems of other radio transmitters. All
the prediction methods are divided into empirical and deterministic/physical models.
The choice of the coverage prediction model depends on the propagation environment and
the coverage area. In communications, propagation takes place through multiple diffraction,
reflection and scattering among others from an extremely large number of objects. Since it is
very difficult to locate scatterers deterministically, characterisation of the signal within the
coverage zone is done statistically. For this reason, prediction models have been developed
using empirical or statistical methods. The accuracy of a particular model in a given
environment depends on the fit between the parameters required by the model and those
available for the area concerned (Rama Rao et al. 2000). Examples of these models are
Ikegami (Tapan et al. 2003), Ibrahim and Parsons (Tapan et al. 2003), Free-Space (Friis 1946
cited in Saunders 205 and Tapan et al. 2003), Extended COST-231 (COST 231 Final Report
1999 cited in Tapan et al. 2003), Perez-Vega and Zamanillo’s model (Perez-Vega and
Zamanillo 2002), Plane Earth Loss (Perez-Vega and Zamanillo 2002), Hata model (Hata
1980), Lee model (Lee 1985), COST231 Walfisch–Ikegami model (Ikegami et al. 1984),
Walfisch–Bertoni model (Walfisch and Bertoni 1988), and ITU-R (ITU Report 1998, p.370).
3.2 Applications of Propagation models

The prediction techniques or models described in this study are most often implemented for
practical planning within computer software. The development of such software has been
motivated and enabled by a number of factors (Saunders 2005):
(i) The enormous increase in the need to plan digital broadcast systems for TV
services and cellular systems accurately and quickly
(ii) The development of fast and affordable resources
(iii) The development of graphical information systems, which index data of terrain,
clutter and land usage in an easily accessible and manipulated form giving better
frequency management etc.
Such techniques have been implemented in a wide range of commercially available and
company-specific planning tools. Some of the prediction tools are listed in Table 1. Although
most are based on combined empirical and simple physical models, it is anticipated there
will be a progressive evolution in the future towards more physical or physical-statistical
models as computing resources becomes cheaper and cheaper, as clutter data improves in
resolution and as researchers develop more efficient path loss prediction algorithms.
Tools Description
PACE The original propagation systems is now integrated into Vodafone Geographical
Information System
ASTRIX Advanced SysTem for RadioInterface eXploration is intended for macrocells. It
incorporates path loss models Okumura-Hata, Blomquist-Ladell, Walfisch-Bertoni and
diffraction loss models Deygout and Epstein-Peterson. It treats also 3D terrain
scattering.
PathPro It treats propagation models COST 231 microcell and macrocell, Hata and Longley-
Rice.
CelPlanner It incorporates propagation models Lee-Picquenard, Okumura-Hata, COST 231 and
Korrowajczuk.
CRUMPET It incorporates the propagation model UK Army EMC Agency PR03.
Planet It incorporates propagation models Okumura-Hata and Walfisch-Ikegami.
NetPlan It incorporates propagation models Walfisch-Ikegami COST 231-Hata and Walfisch-
Xia
Table 1: Prediction Tools (Source from Saunders 2005)
3.3 Survey of various propagation models and their technical background

The two basic propagation models (Free-Space and Plane Earth Loss) have all the
mechanisms which are encountered in macrocell prediction. Many researchers use these
models and predict the total signal loss. Other models require detailed knowledge of the
location, dimension and parameters for every tree or building and terrain feature in the area
to be covered. The models are complex and yield an unnecessary amount of details as the
network designer is not interested in the particular locations covered, but the overall extent
of the coverage area. One appropriate way of removing these complexities is to adopt an
empirical model. These models use, as parameters, the received signal strength, frequency,
antenna heights and terrain profiles, derived from a particular environment through the use
of extensive measurement and statistical analysis. The models can then be used to design
systems operated in similar environments to the original measurements.
3.3.1 The Okumura-Hata model

The simple modeling of path loss is still dominated by the Hata empirical model (Hata
1980), where the propagation results are fitted to a simple analytical expression, which
depends on antenna height, environment, frequency and other parameters. Hata’s method is
basically an extension of Okumura’s method (which is somewhat cumbersome due to
numerous correction factors) and employs propagation curves instead of parametric
equations. It is a model based upon an extensive series of measurements made in and
around Tokyo city between 200 MHz and 2 GHz. Predictions are made via a series of
graphs. The thoroughness of work has made the model the most widely used macrocell
prediction model and is often regarded as a standard against which researchers can
benchmark new approaches. The model for urban areas has been standardised in 1997 for
international use as Rec ITU-R P.529 model (ITU Report 1997). The Hata model does not
have any of the path-specific corrections which are available in Okumara’s model. Okumura
takes urban areas as a reference and applies correction factors for conversion to the
classification of terrain. Hence the model will involve dividing the prediction area into a
series of clutter and terrain categories as follows:
(i) Open area: Open space, no tall trees or buildings in path, plot of land cleared for
300-400 m ahead, e.g. farm land, rice fields, open fields;
(ii) Suburban area: Village or highway scattered by trees and houses, some obstacles
near the receiving antenna but not very congested;
(iii) Urban area: Build up city or large town with buildings and houses with two or
more storages, or larger villages with close houses and tall and thickly grown trees.
The negative side of the Okumura-Hata model is that it is valid only for frequency between
150 MHz and 1500 MHz, with base antenna height between 30 m to 200 m and receiving
antenna between 1 m and 10 m. However, this model will not be a problem to use in this
research as measurements are taken within the ranges mentioned above. Another problem
encountered by this model is that in some countries measurements have been in
disagreement with the predictions. The reason cited is the difference in characteristics of
Tokyo city. Kozono and Watanabe (1977) have tried to modify the model by including a
measure of building density, but such approach has not found common acceptance. The
third problem is that the model has been developed for only three categories of land usage
(rural, sub-urban and urban) as in practice the classification of land usage of a country (e.g.
Mauritius, England, India) can exceed 10 categories.
3.3.2 Other Standard Models

In this study, the Lee model (Lee 1985), and the approximate model Extended COST 231-
Hata (COST 231 Final Report 1999 cited in Rama Rao 2000) are considered apart from the
Hata and Free Space model as explained in the previous section. Approximate models COST
231/Walfisch-Ikegami (1984) and Walfisch–Bertoni (1988) have some restrictions because
they do not include information on the environment. Hence, the risk of incorrect prediction
is high since there are no correction factors for conversion according to the terrain
classification. Moreover, COST 231/Walfisch– Ikegami model is valid for frequency
between 800 and 2000 MHz, that is, it is not applicable to macrocells but to microcells. In a
later stage in 1999, COST 231 was improved and a new model was created which is adopted
for various terrains. The model which was derived from Hata is known as Extended COST
231-Hata model (COST 231 Final Report 1999). The Walfisch–Bertoni model considers the
impact of rooftops and building height by using diffraction to predict average signal
strength at street level. These methods describe urban propagation loss as a sum of three
terms: free space losses, rooftop to street losses and multiple diffraction losses. The
approaches of Walfisch–Ikegami and Walfisch–Bertoni are restricted by definition to radio
paths that are obstructed by buildings. The models account for local terrain slope in the
vicinity of the receiving antenna and do not incorporate terrain roughness factors and do
not treat obstructing terrain features such mountains and gorges. Finally, the empirical
model developed by Blomquist-Ladell (1974) has some limitation since it includes only the
sum of free space loss, the sum of smooth spherical earth loss, obstacle diffraction loss, urban
loss and vegetation loss. It does not include loss due to reflection, climates and seasons.
Research on propagation models and path loss have been carried out in the past by several
researchers like Grosskopf (1987) in Germany, Rama Rao et al. (2000) and Prasad (2006) in
India, The Perez-Vega-Zamanillo (2002) in Spain and Hosseinzadey (2003) in Iran among
may others. The Perez-Vega-Zamanillo model is a simple propagation model for the VHF
and UHF bands. The model is a computational form of the data provided by the FCC
F(50,50) propagation curves. The model is not frequency dependent in the band of interest
and can be used to predict the path-loss for television broadcasting. One disadvantage is
that it does not provide information on issues such as fade margins, angles of arrival, or
delay spread, which must be estimated by another way.
The model developed by Grosskopf (1987) has no classification of urban, suburban and
open areas or correction factors which are very important for a model. His technique is to
predict path loss for only hilly and mountainous land in Germany. Prasad (2006) has done
intensive work where field strength measurements are taken over Indian subcontinent and
the calculated path loss is compared with other models including the Perez-Vega-Zamanillo
(2002) one. However, no model was proposed from the observations and results.
3.4 Factors to improve accuracy of models

Below are factors that can be considered to improve accuracy of propagation models:
(i) Reflection is the result of digital TV signal hitting on obstructions with properties
(thickness, length) much larger than the wavelength of the radio wave (e.g. smooth
surface of walls and hills/mountains).
(ii) Diffraction occurs when radio waves strikes the edges or corners of obstacles.
These act as secondary sources re-radiating into the shadow region. It is due to the
diffraction effect that radio frequency energy travels in dense urban environments
where there is no clear Line-of-Sight between two antennas (e.g. from edges such
as building rooftops and mountaintops).
(iii) Scattering occurs when the properties of the object interacting with the radio wave
is on the order of the colliding wavelength (e.g. from rough surfaces such as sea,
rough ground and the leaves and branches of trees).
(iv) Absorption (e.g. by walls, foliage and by atmosphere)
(v) Refraction (e.g. due to atmospheric layers or layered and graded materials)
(vi) The directional characteristics of both the transmitter and the receiver antennas.
(Saunders 2005).
All these factors are called multiplicative noise. It is more conventional to subdivide these
factors as path loss, shadowing or slow fading and fast fading or multipath fading.
Shadowing
Shadowing is the loss of field strength typically contributed to a diffracted wave emanating
from an obstacle between transmitter antenna and receiver antenna (Saunders 2005). As
passing through a shadow area requires considerable time, the name ‘slow fading’ is
commonly used. The shadow effect is modeled with a log-normal distribution of the mean
signal.
Fast Fading (Multipath Propagation)

As radio waves are reflected or diffracted or scattered by trees, hills and mountains,
buildings and other obstacles, they establish various transmission paths from the transmitter
to the receiver antennas. Many reflections are produced in an urban environment and few
reflections in rural areas. The multipath creates the most difficult problem in the digital
broadcast environment.
3.5 The Technical Background of Propagation Models

This section provides methods for predicting path loss used in macrocells (above 1 km). The
models presented here treat the path loss associated with a given macrocell as dependent on
distance between a transmitter and a receiver, provided that the environment is fairly
uniform. The free space propagation model is discussed briefly whereas the plane earth loss
model is not treated in this work as the latter will require detailed knowledge of the
location, dimension and constructive parameters of every tree, building and terrain feature
in the area to be covered. It will be too complex and will yield an unnecessary amount of
details since the broadcasting designer, network engineer and planner will not be interested
in the particular locations being covered, but rather in the overall extent of the coverage area
(Saunders 2005).
In the design of any broadcasting system, the fundamental task is to predict the coverage of
the proposed system. Digital television service coverages are characterised by a very rapid
transition from near perfect reception to no reception at all (Smith 2003). Hence, it becomes
critical to be able to define which areas are going to be covered and which are not. As it is
the case for Mauritius, it becomes necessary to increase the transmitter powers or to provide
a large number of transmitters in order to guarantee coverage to the last few percent of the
worst served small areas.
A wide variety of techniques have been developed over the years to predict coverage using
what are known as propagation models (Saunders 2005). Propagation, in this context, means
the transmission of signals from the transmitter to the receiver. En route from the
transmitter to the receiver, the signal gets weaker and may experiences shadow or multipath
effects (Ong et al 2004).
As said by Saunders (2005), based on the path loss information, to improve reception in a
particular situation the following factors can be considered:
(i) Use a more directional receiving antenna with a higher gain
(ii) Find a better position for the receiver-antenna
(iii) Use of a low-noise antenna amplifier (as in the case of fixed antenna reception).
In general terms, path loss occurs when the transmitted signal suffers a loss proportional
1 R 2 , where R is the distance between transmit and receive antennas.
3.5.1 Free Space Propagation Model

The free space propagation model is used to predict received signal strength when the
transmitter and receiver have a clear, unobstructed line-of-sight path between them (Friis
1946). As with most large-scale radio wave propagation models, the free space model
predicts that received power decays as a function of the Transmitter-Receiver separation
distance raised to some power (i.e. a power law function) (Saunders 2005).
The free space power received by a receiver antenna which is separated from a radiating
transmitter antenna by a distance d, is given by the Friis free space equation (Friis 1946),
P G G 2
Pr (d )  t t r (1)
(4 ) 2 d 2
where Pt is the transmitted power, Pr(d) is the received power, Gt is the transmitter antenna
gain, Gr is the receiver antenna gain, d is the T-R separation distance in meters and λ is the
wavelength in meters.
The Friis free space equation shows that the received power falls off as the square of the
Transmitter-Receiver (T-R) separation distance. This implies that the received power decays
at a rate of 20 dB/decade with distance.
The path loss, which represents signal attenuation as a positive quantity measured in dB, is
defined as the difference (in dB) between the effective transmitted power and the received
power, and may or may not include the effect of antenna gains(ITU Report 1998).
The path loss for the free space model when antenna gains are included is given by:
Pt
PL ( dB )  10 log
Pr
(2)
 G G 2 
 -10 log  t r 
 ( 4 ) 2 d 2 
 
Equation (3.2) can be expanded to give an equation in terms of distance, d (km) and
frequency of operation, f (MHz):
 ( c  10  3 ) 
PL ( dB )   10 log 10 ( G t )  10 log 10 ( G r )  20 log 10    20 log 10 (1 / d )
 4   f  10 6 
 
(2.1)
 - G t ( dB )  G r ( dB )  32 . 44  20 log 10 ( d / km )  20 log 10 ( f / MHz )
where c is the speed of light ( 3  10 8 ms 1 )
3.5.2 Okumura-Hata path loss model

The Okumura-Hata model (1980) is an empirical formulation of the graphical path loss data
provided by Yoshihisa Okumura, and is valid from 150 MHz to 1500 MHz. The Hata model
is, basically, a set of equations based on measurements and extrapolations from the curves
derived by Okumura. Hata presented the urban area propagation loss as a standard
formula, along with additional correction factors for application in other situations such as
suburban and rural.
Only four parameters are required in the Hata model. Hence, the computation time is very
short. This is an advantage of the model. However, the model neglects the terrain profile
between the transmitter and receiver, that is, hills or other obstacles between the transmitter
and receiver are not considered. This is because both Hata and Okumura made the
assumption that the transmitters would normally be located on hills. Figure 1 shows a
typical scenario for Hata-Okumura model.
Fig. 1. Scenario for Hata model
The above model assumes a direct line-of-sight path from transmitter (tx) to receiver (rx) but
the actual path is obstructed by two hills. Hence, the prediction would be too optimistic.
The standard Hata formula for median path loss in urban areas is given by:
L(urban)(dB)  69.55  26.16 log f c  13.82 log htx

(3)
 a(hrx )  (44.9  6.55 log htx ) log d
where:
fc is the frequency (in MHz) from 150 MHz to 1500 MHz,
htx is the effective transmitter antenna height (in m) ranging from 30m to 200m,
hrx is the effective receiver antenna height (in m) ranging from 1 m to 10 m,
d is the T-R separation distance (in km),
a(hrx) is the correction factor for effective antenna height which is a function of the size of the
coverage area.
To obtain the path loss in a suburban area, the standard Hata model formula in equation (3)
is modified to:
L ( dB )  L ( urban )  2 log( fc 28 ) 2  5 . 4 (3.1)
For a terrain category such as the north of Mauritius, the antenna correction factor is given
by:
a ( h rx )  (1.1 log f c  0.7) h rx  (1.56 log f c  0.8) dB (4)

3.5.3 Extended COST-231 Hata model

This model (COST 231 Final Report 1999 cited in Tapan et al. 2003 and Zreikat and Al-
Begain n.d.) is derived from the Hata model and depends upon four parameters for the
prediction of propagation loss: frequency, height of a received antenna, height of a base
station and distance between the base station and the received antenna.
From equation (3), the urban model is given by:
L(urban)(dB)  46.33  33.9 log fc  13.82 log htx

(5)
 a(hrx )  (44.9  6.55log htx ) log d
The path loss in a suburban area is given by:
L ( dB )  L ( urban )  2 log( f c 28 ) 2  5 . 4 (5.1)
where a(hrx) is obtained from equation (4).
3.5.4 Lee Model

The Lee model (1985) is a power law model with parameters taken from measurements in a
number of locations. The model is expressed as follows:
L ( suburban )( dB )  10 n log d  20 log htx  Po  10 log hrx  29 (6)
where n = 3.84 and Po=-61.7. Here it has been assumed that htx is the effective base station
height.
3.5.5 Rec ITU-R P.370 Propagation prediction method

Prediction of the coverage provided by a given transmitting station is normally done on the
basis of the field strength for the wanted signal predicted.
Rec ITU-R P.370 (ITU Report 1998) is a commonly agreed field strength method for
broadcasting services. The propagation curves given in this recommendation represent field
strength values in the VHF and UHF bands as a function of various parameters.
The power received at a distance d, Pr is given by:
2
E
Pr  Ae
120 
Pr ( dB )  20 log 10 E  10 log 10 (120  )  10 log 10 Ae
(7)
Pr ( dB )  2 E min  10 log 10 (120  ) /   Ae ( dBm 2 )
where:
Emin is the equivalent minimum field strength at receiving place
Ae is the effective antenna aperture (dBm2)
120  is the value of intrinsic impedance of free space (ohms).
However, the above equation relates electric field (with units of V/m) to received power
(with units of watts). Often, this equation is used to relate the received power level to a
receiver input voltage, as well as to an induced electric field at the receiver antenna.
In situations where practical values of field strengths are available in dBV / m from
measurements, the corresponding path loss in dB can be calculated as follows if the values
for transmitted power and effective receiver antenna aperture are known:
PL ( dB )  P t ( dB )  P r ( dB )
PL(dB)  Pt (dB)  Emin  Ae (dB)  10 log10 (120 )
PL(dB)  Pt (dB)  2Emin (dBV / m)  240  Ae (dB)  10 log10 (120 ) (7.1)
where E min  E min ( dB  V / m )  120 .
4. Comparative Field Strength and Path Loss Analysis – A Case Study

4.1 Data Collection, Experimental Details and Methodology
Measurement locations are divided into three groups for each region, that is, three
concentric circles CC1N, CC2N and CC3N for the north and CC1S, CC2S and CC3S for the
south. The radius determines the horizontal distance between the measured point (receiver
antenna) and the transmitter. The radii of CC1N/CC1S, CC2N/CC2S and CC3N/CC3S are
5 km, 10 and 15 km respectively. Points are selected on the circles as shown in Figure 2 and
determined the measuring sites. It should be noted that at each measuring site, three to four
times, measurements were taken every five minutes for a period of 20 minutes. The exercise
are conducted at 78 locations sites in the North (26 for CC1N, 26 for CC2N and 26 for CC3N)
and at 99 locations sites in the South (33 for CC1S, 33 for CC2S and 33 for CC3S)
Fig. 2. View of one part of Mauritius

A measuring vehicle is used to carry the equipment and the measurements are done
manually. As shown in Figure 3, the vehicle is stopped at each site and the log periodic
antenna from Fracarro (Antennas online 2007, Fracarro Antenna online 2007) is raised up to
a height and is properly oriented towards the transmitter to achieve maximum signal
strength. The measurements are conducted around the relay station and are repeated for
two receiving antenna heights of 4 m and 6 m. Once the received signal has been captured
the field strength, the BER value, the minimum Carrier to Noise (C/N) and CSI are recorded
for digital TV signal. Measurements were taken at antenna heights 4 m and 6 m.
Fig. 3. The Measuring Vehicle
4.2 Signal Strength and Interference

The aim of this work is to study the variation of the received field strength, carrier-to noise
and Bit Error Rate for different locations in the northern region of Mauritius (Fogarty,
Soyjaudah and Armoogum 2006). Similar works have been carried out in various countries
(India, Spain, Canada, Korea) as explained by Rama Rao et al. (2000), by Prasad (2006), by
Arinda et al. (1999a, 1999b), by Assia Semmar et al. (2006), by Sung Ik Park et al. (2007) and
more recently by Martinez et al. (2009). All these studies have analysed the quality of digital
TV signal reception. In Mauritius, this exercise is important so as to give a realistic picture of
the situation. Few measurements at UHF frequencies for digital television have been made.
The experiment (Figure 4) shows that at a distance of 5 km the field strength is high enough
(minimum threshold value is 53 dBµV/m) for the COFDM component of the AFSM to
decode picture to achieve a BER lower than 2x10-4. For both regions, almost all locations will
have high performance of digital transmission. For an ideal case, that is, a topography of flat
earth with Line-Of-Sight propagation, the received field strength at all locations over
concentric circle CC1 must be constant. At a distance of 10 km and more, the cell covered by
the base station does not include all the measuring points since the received signal strength
is less than 53 dBµV/m (the signal level and quality at some points are not high enough to
decode the information stream). However, for both regions at a certain constant distance
round the station, the graph obtained shows that the field strength is varying with location.
This may be explained by the fact that due to Non Line-Of Sight (NLOS), there may be
different diffraction losses due to different buildings in the north or due to dense forests and
mountains (knife-edge) in the south in the path linking the base station and the location of
interest. Multipath effects may also be the cause, giving rise to a graph of varying field
strength with location.
The field strength in the south decreases more than that of the north. Besides, when
comparing the standard deviations, the C/N deviates too much from the mean values in the
south, indicating that there are more obstructions in the south. Though the south is
classified as a rural area compared to the northern area (sub urban), there are more factors
affecting the signal in that region.
Fig. 4. Variation of Field strength for various antenna heights in the North and South at a
distance of 5 km from transmitter
4.3 Path Loss Analysis

The main aim is to study the variation of the path loss at various locations in the south and
compare them with those from the north (Armoogum et al. 2007a, Armoogum et al. 2007b).
In theory, the path loss at a constant distance from a transmitter is the same for any point
around it. For both areas, as depicted in Figure 5, the path loss is not constant at various
locations for a constant distance around the transmitter which therefore indicates about the
irregularity of both terrains of the island. The path loss for an antenna height of 6 m is lower
than that of 4 m as a result of a reduction of multipath effects with a higher antenna.
Fig. 5. Variation of Path Loss for both regions at 5 km from transmitter

4.4 Comparative Study of Pass Loss with Various Existing Propagation Models and
Selecting the Best One(s)
The measured pass loss is compared with models such as Free-Space, Okumura-Hata,
Extended COST-231 and Lee. The aim is find out which of the model(s) gives/give better
agreement with the measured pass loss. From figures Figure 6 and Figure 7, it is clear that
the Lee model and the Free-Space model deviate too much from the measured values, which
implies that these models are not suitable for modelling the south of Mauritius. For both
regions, the path loss using Okumura-Hata model and Extended COST-231 are much closer
to each other and give better agreement with the measured values. The slight difference
between these two models can be considered negligible.
Fig. 6. Variation of Path Loss in the South using various propagation models at 5 km from
transmitter
Fig. 7. Variation of Path Loss in the North using various propagation models at 5 km from
transmitter
There is no need to transform the two models into a discrete-time mode as the signal
strength depends only upon the power transmitted and received. The results show that the
analogue models can be used in digital broadcast systems. These models can be used to
develop a novel digital model for Mauritius. Since the north and the south of Mauritius are
inhomogeneous ones and consist of rural, sub-urban and urban areas unevenly located, we
can take sub-urban areas as a reference and applies correction factors for conversion to the
classification of terrain.
Using equations (3) and (3.1) from Section 3.5, the digital Hata-Okumura model is given by:
L ( dB )  69 . 55  26 . 16 log f c  13 . 82 log h tx
(4.1)
 a ( h rx )  ( 44 . 9  6 . 55 log h tx ) log d  2 log( f c 28 ) 2  5 . 4
Using equations (5) and (3.5.1) from Section 3.5, the digital Extended COST-231 Hata model
is given by:
L ( dB )  46 . 33  33 . 9 log f c  13 . 82 log h tx
 a ( h rx )  ( 44 . 9  6 . 55 log h tx ) log d  2 log( f c 28 ) 2  5 . 4
5. Proposed way for Modelling Broadcast Networks

The best way to design new broadcast networks or optimize existing networks is to develop
new models or improving existing ones. This section provides a new definition for the types
of terrain of Mauritius (a case study) and secondly proposed the techniques that could be
used in developing new models.
5.1 Redefining the Classification of Land Usage for Mauritius

The first problem with the empirical models is the classification of environment in which the
system is operating. Once the real classification through the analysis parts is known, better
results are obtained if the appropriate category of terrain is reviewed. The categories must
be numerous so that the properties of different locations classed within the same category
are not too variable. A proposed definition of categories is depicted in Table 2.
5.2 The Approaches

The two empirical models stated in Section 4 are fundamental for the prediction of path loss.
The new models may thus be developed as means of improving accuracy using two
approaches: deterministic empirical-physical approach or empirical-statistical approach
using some statistical parameters.
What are Physical and Statistical approaches?

The physical approach:
A Digital TV wave signal can be reflected or refracted when hitting a surface. In these cases
the Snell’s laws of reflection and refraction will be applied. Rays can hit different types of
surface. In Mauritius, there are dry land, average ground, wet ground, sea water and fresh
water and the reflection will differ according to the conductivity and relative dielectric
constant. These values are available in ITU Report-527 (ITU Doc. 1992). These reflection
processes are so far applicable to smooth surfaces and are termed specular reflection. When
the surface is rougher, the reflected signals become scattered from a larger number of
positions on the surface, hence, reducing the field strength and increasing the attenuation.
Refraction also occurs due to atmospheric layers. A higher permittivity (the denser medium)
causes the transmitted signal to bend more toward the surface normal. This change of
direction changes the velocity of the signal with respect to refractive index and might cause
delay spread. In many situations in Mauritius, diffraction over obstructions such as hills,
mountains and buildings may be treated as if they are absorbing knife-edges (single knife-
edge and multiple knife edge diffraction). Using physical approach means modelling all the
above factors.
The statistical approach:

The region is divided into small homogeneous areas and categories. The smaller the sector,
the more homogeneous will be the area. A forecasting technique is used to predict path loss
from 0 to 15 km (interpolation) and above (extrapolation). Mathematical equations are
derived for each homogeneous area. Finally, a general Mathematical equation is derived for
similar homogeneous areas, that is, for a specific category of land.
For the development of novel models, it is believed that the empirical-statistical approach is
better for the following reasons
(i) The physical approach will lead to entirely incorrect predictions when considering
fields in the shadow region behind an obstruction.
(ii) Using the physical approach, it would be very difficult to get the exact parameters
of obstructions (size of buildings, heights, distance between buildings, density of
forests, types of soil etc).
(iii) The empirical-statistical model is more economical as the physical models involve
a large number of expensive input data requirements.
Category Details
0 Reservoirs, Lakes and Sea1.
1 Open Rural Area with Plantation (Sugar cane and Tea).
2 Open Rural Area with Forest in between.
3 Dense Forested Area.
4 Mountainous region
5 Hilly and Mountainous Forested Area.
6 Sub Urban Area – Small villages2 of low-density houses of up to two storeys, with
some open areas and trees.
7 Sub Urban Area – Big villages3 with houses of up to two storeys and industries
(industrial zones) and with some open space and trees in between.
8 Urban Area – Big villages or towns with buildings of up to four storeys
9 Higher Urban Area – Town4 with buildings of up to four storeys and closed to each
other.
10 Dense Urban Area – Town with buildings very closed to each other in which some of
them are up to eight storeys.
11 Very Dense Urban Area – Big town or cities5 with buildings very closed to each other
in which most of them are eight storeys and above.
Table 2. Classifications of Land Usage for Mauritius
6. Conclusion
The increasing demand of various fixed and mobile services has placed considerable
pressure on the limited frequency spectrum. For the efficient utilisation of this resource, as
well as for performance assessment of the existing systems, modeling and coverage
predictions are essential. A reliable model of predicting path loss helps in reducing load on
base stations and helps in designing digital broadcasting networks including TV services.
We have considered Mauritius Island as a case study. From the observations and results
obtained, it is concluded that the existing empirical models are not accurate and therefore
cannot be used in Mauritius since the focus here is on small area propagation and for
tropical, mountainous regions. The first limitation of existing models is that they are
developed for limited categories of land (open area, sub-urban and urban areas). The table
of classification of land usage was defined for Mauritius. For qualitative classification, the
categories are numerous so that the properties of different locations classed with the same
category are not too variable. Analysis and comparisons of field strength and height gain
analysis in North and South were conducted to find out the types of terrain of Mauritius
and the reasons why TV signals suffer a loss. The path loss analysis was conducted and
tested using several models. Extended COST-231 and Okumura-Hata give better agreement
in regions of Mauritius. Before developing novel models, the regions were divided into
small homogenous areas and categorised. Two techniques are proposed for the
development of new propagation models.
7. Future Work
In section 5, the technicalities toward the development of new models are not discussed.
Then, performance analysis and testing need to be done using new measured data. The
usual goal of performance analysis and testing is to determine the places of Mauritius where
the models work properly and the places there are high deviations. An extensive work
needs to be conducted for this purpose.
Models are widely used in prediction tools such as ASTRIX or CRUMPET. Computer
applications have to be developed for the prediction of path loss, the designing of broadcast
networks or optimizing existing network.
Some of the functionalities of these software tools with the integrated models are as follows:
(i) Design new broadcast networks,
(ii) Perform coverage prediction for various types of communication systems in order
to design optimize networks of transmitters,
(iii) Analyze interference problems,
(iv) Explore new coverage scenarios,
(v) Diagnose difficult coverage situations,
(vi) Evaluate new transmitter and receiver concepts.
(vii) Visualize and analyze predicted performance, as well as compare simulation with
experimental data.
Finally, to obtain a high quality image, these applications can be used with a geographical
information system (GIS) storing the geographical locations of points for every 10 m2 and
their corresponding categories as the main fields.
8. References
Anderson, J. 2005. Digital Transmission Engineering. 2nd Edition. Wiley Publication.
Antenna [Online] 2001. Available at www.fracarro.com/NR/rdonlyres/7456FB27-45EF-
489D-8373-F8E23BB3CCAD/0/Antennas.pdf. [Accessed 2007].
Armoogum, V. Soyjaudah, K.M.S. Mohamudally, N. and Fogarty, T. 2007. Height Gain
Study for Digital Television Broadcasting at UHF Bands in two Regions of
Mauritius. Proceedings of the 2007 Computer Science and IT Education Conference, pp.
017-025, ISBN 9789990387476.
Armoogum, V. Soyjaudah, K.M.S. Mohamudally, N. and Fogarty, T. 2007. Comparative
Study of Path Loss with some Existing Models for Digital Television Broadcasting
for Summer Season in the North of Mauritius at UHF Band. IEEE The Third
Advanced International Conference on Telecommunications (AICT-07), ISBN 0-7695-
2443-0.
Armoogum, V. Soyjaudah, K.M.S. Mohamudally, N. and Fogarty, T. 2007. Path Loss
Analysis between the north and the south of Mauritius with some Existing Models
for Digital Television Broadcasting for Summer Season at UHF Bands. Proceedings
of the 8th IEEE AFRICON 2007, ISBN 0-7803-8606-X.
Arrinda, A. Ma Velez, M. Angueira, P. de la Vega, D. and Ordiales, J. L. 1999. Digital
Terrestrial Television (COFDM 8k System) Field Trials And Coverage
Measurements In Spain. IEEE Transactions On Broadcasting, 45(2), No. 2, pp. 171-
176.
Arrinda, A. Ma Velez, M. Angueira, P. de la Vega, D. and Ordiales, J. L. 1999. Local-Area
Field Strength Variation Measurements Of The Digital Terrestrial Television Signal
(COFDM 8k) In Suburban Environments. IEEE Transactions On Broadcasting, 45(4),
pp. 386-391.
Assia Semmar, Jean-Yves Chouinard, Viet Ha Pham, Xianbin Wang, Yiyan Wu, and
Sébastien Laflèche 2006. Digital Broadcasting Television Channel Measurements
and Characterization for SIMO Mobile Reception. IEEE Transactions On
Broadcasting, 52(4), pp. 450-463.
BBC 1998. AC106 Verification and Launch of Advanced Television in Europe (VALIDATE),
Project Report.
Blomquist, A. and Ladell, L. 1974. Prediction and calculation of transmission loss in different
types of terrain, FOA report.
CEPT 1997. The Chester 1997 Multilateral Coordination Agreement relating to Technical
Criteria, Coordination Principles and Procedures for the introduction of Terrestrial
Digital Video Broadcasting (DVB-T).
COST 231 Final Report 1999. Digital Mobile Radio: COST 231 View on the Evolution
Towards 3rd Generation Systems, Commisiion of the European Communities and
COST Telecommunications, Brussels.
ETSI (European Te1ecommunications Standards Institute) 1997. Digital Video Broadcasting
(DVB), Framing structure, Channel coding and Modulation for Digital Terrestrial
Television (DVB-T), ETS 300 744.
ETSI (European Te1ecommunications Standards Institute) 1997. Digital Video Broadcasting
(DVB), Implementation guidelines for DVB services. Transmission aspects, ETS 101
190.
Fogarty, T. Soyjaudah, K.M.S. and Armoogum, V. 2006. Signal Strength Variation

Measurements of Digital Television Broadcasting for Summer Season in the North
of Mauritius at UHF Bands, Conference Proceeding of the 3rd International Conference on
Computers and Device for Communication, pp. 89-92.
Fracarro Antenna [Online] 2001. Available at www.tvantenna.com.au
/FARCARRO%20ANTENNAS [Accessed 2007].
Friis, H.T. 1946.The Free Space Transmission equation. Proc. IRE, vol. 34, p.254.
Gaudrel, R. Betend, C. 1997. DIGITAL TV BROADCAST Field Trials on the Experimental
Network of Rennes, Internal VALIDATE document from CCETT,
FT.CNET/DMR/DDH.
GIS Company History [Online] n.d. Available at http://www.fundinguniverse.com
/company-histories/General-Instrument-Corporation-Company-History.html,
[Accessed on June 2006].
Grosskopf, R. 1987. Comparison of different methods for the prediction of the field strength
in the VHF range. IEEE Trans. Ant. Propagation, 35(7), pp. 852–859.
Hata, M. 1980. Empirical formula for propagation loss in land mobile radio services. IEEE
Trans. Veh. Technol., 29(no issue number), pp. 317–325.
Homayoon, O. and Shahram Hosseinzadeh 2003. Determination of the effect of Vegetation
on Radio wave Propagation by the Parabolic Equation Method, Proceeding of 2nd
International Symposium of Telecommunications, IST2003, pp.340-344 Isfahan Iran.
Ikegami, F. Yoshida, S. Takeuchi, T. and Umehira, M. 1984. Propagation factors controlling
mean field strength on urban streets. IEEE Trans. Ant. Prop., vol. 32 (no issue
number), pp. 822–829.
International Telecommunication Union, ITU-R recommendation P.529-2 1997. Prediction
methods for the terrestrial land mobile service in VHF and UHF bands.
International Telecommunication Union, ITU-R recommendation P.370 1998. Prediction
methods for the terrestrial land mobile service in VHF and UHF bands, 1998.
International Telecommunication Union 1992. ITU-R Recommendation 527-3: Electrical
Characteristics of the surface of the Earth, Geneva 1992.
ITU Geneva 1989. Final Acts of the Regional Administrative Conference for the Planning of
VHF/UHF Television Broadcasting in the African Broadcasting Area and
Neighbouring Countries.
Kozono, S. Watanabe, K. 1977. Influence of Environmental buidings on UHF Land mobile
radio propagation, IEEE Trans. Commun., 25(10), pp.1133-1143.
Lee, W.C.Y. 1985. Estimation of local average power of a mobile radio signal. IEEE Trans.
Veh. Technol., 34(1), pp. 22–27.
Martínez, A. Zabala, D. Peña, I. Angueira, P. Vélez M.M. Arrinda, A. de la Vega, D. and
Ordiales, J.L. 2009. Analysis of the DVB-T Signal Variation for Indoor Portable
Reception. IEEE Transactions On Broadcasting, 55(1), pp. 11-19.
Metz H. C., [Online] 1994. Ed. Mauritius: A Country Study. Washington: GPO for the
Library of Congress, 1994. Available at http://countrystudies.us/mauritius/7.htm
[Accessed on May 2006].
Ong, J. T. Rao, S. V. B. Shanmugam, G. Hong Y. 2003, Results of the Mobile DTV
Measurements in Singapore, Nanyang Technological University, Singapore,
Conference Publication.
Ong, J.T. Rao, S.V.B. Hong Y. Shanmugam, G. 2004. Height Gain Measurements for DTV
Reception in Singapore. IEEE Transactions On Broadcasting, 50(4), pp. 396-398.
Pather, A. August 2000. The move to Digital Broadcasting – Challenges Ahead, MultiCarrier
(Mauritius) Limited.
Parsons, J.D. and Ibrahim, M.F. 1983. Siganl strength prediction in built-up areas—Part 2:
Siganl variability. IEE Proc., vol. 130, no. 5, pp. 385–391.
Perez-Vega, C. and Zamanillo, J.M. 2002. Path Loss Model for Broadcasting Applications
and Outdoor Communication Systems in the VHF and UHF Bands. IEEE
Transactions On Broadcasting, 48(2), pp. 91-96.
Prasad, M.V.S.N. 2006. Path Loss Deduced From VHF and UHF Measurements Over Indian
Subcontinent and Model Comparison. IEEE Transactions On Broadcasting, 52(3), pp.
290-298.
Prasad, M.V.S.N. and Iqbal, A. 1997. Comparison of some Path Loss Prediction Methods
with VHF&UHF Measurements. IEEE Transactions On Broadcasting, 43(4), pp. 459-
486.
Rama Rao, T. Vijaya Bhaskara Rao, S. Prasad, M.V.S.N. Mangal Sain, Iqbal, A. and Lakshmi,
D. R. 2000. Mobile Radio Propagation Path Loss Studies at VHF/UHF Bands in
Southern India. IEEE Transactions On Broadcasting, 46(2), pp. 158-164.
Regional Radiocommunication Conference for planning of the digital terrestrial
broadcasting service (RCC-04). ITU-R Report, Geneva 2004.
Saunders, S. R. 2005. Antenna and Propagation for Wireless Communication Systems. Wiley.
Smith, D. 2003. Digital Transmission Systems. 3rd Edition. Kluwer Academic Publishers.
Sung Ik Park, Yong-Tae Lee, Jae Young Lee, Seung Won Kim, and Soo In Lee 2007. Field
Test Results of the E-VSB System in Korea, IEEE Transactions On Broadcasting, 53(1)
pp. 14-22.
Tapan, K. S. Zhong, J. Kyungjung, K. Abdellatif, M. and Magdalena Salazar-Palma 2003. A
Survey of Various Propagation Models for Mobile Communication. IEEE Antennas
and Propagation Magazine, 45(3), No. 3.
Walfisch, J. Bertoni, H. L. 1988. A Theoretical Model of UHF Propagation in Urban
Environments. IEEE Transactions On Antennas And Propagation, 36(12), pp. 1788-
1796.
Zreikat, A. and Al-Begain, K. n.d. Simulation of 3G Networks in Realistic Propagation
Environments, I.J. of SIMULATION, 4(3&4), ISSN 1473-804x online, 1473-8031.
Interference Modeling for Wireless Ad Hoc Networks 185
10
X
Interference Modeling for

Wireless Ad Hoc Networks
Altenis V. Lima-e-Lima*, Carlos E. B. Cruz Pimentel*
and Renato M. de Moraes**
* University of Pernambuco (UPE), Recife, Brazil
** University of Brasília (UnB), Brasília, Brazil
1. Introduction
Interference effects constrain scalability performance of ad hoc networks as Gupta and
Kumar (Gupta & Kumar, 2000) showed that the throughput capacity of a fixed wireless
network decreases when the number of total nodes � increases. More specifically, they
showed that the node throughput decreases approximately like 1⁄√�. Grossglauser and Tse
(Grossglauser & Tse, 2001) presented a two-phase packet forwarding technique for mobile
ad hoc networks (MANETs), utilizing multiuser diversity (Knopp & Humblet, 1995), in which
a source node transmits a packet to the nearest neighbor, and that relay delivers the packet
to the destination when this destination becomes the closest neighbor of the relay. The
scheme was shown (Grossglauser & Tse, 2001) to increase the throughput capacity of
MANETs, such that it remains constant as the number of users in the network increases,
taking advantage that communication among nearest nodes copes the interference due to
farther nodes.
On the other hand, detailed and straightforward models for interference computation in
dense ad hoc networks have not been extensively studied. Grid models have been proposed
to compute interference (Gobriel et al., 2004), (Liu & Haenggi, 2005), which take advantage
of the regular placement of the nodes. This orderly topology is a good starting point for
static networks; however, it does not apply for MANETs. Also, some previous works have
assumed a transmission or a reception range for communication among nodes without
considering the effect from the entire network (Tobagi & Kleinrock, 1975), (Deng et al.,
2004). This approximation can be good for low density networks, but it may imply in
inaccurate results for dense networks. One problem with such approximation is the
difficulty in finding an analytical description for the random topology inherent to ad hoc
networks. In other cases, analytical models use graph theory (Rickenbach et al., 2005), (Qin-
yun et al. 2005). While they are good for higher layer analysis, like routing, such models
may not be appropriate for a more detailed communication channel study because they do
not consider physical parameters like Euclidean distance, fading and path loss, for example.
This chapter analyzes an improved channel communication model, from the model
proposed by Moraes et al. (Moraes et al., 2008) that permits to obtain the measured signal to
noise and interference ratio (SNIR) by a receiver node, and consequently its spectral
Shannon capacity (or spectral efficiency) (Cover & Thomas, 1991) at any point in the
network when it communicates with a close neighbor. This model considers Euclidean
distance, path loss and Rayleigh fading. The nodes are assumed to move according to a
random mobility pattern and the parameter θ represents the fraction of sender nodes in the
network. Monte-Carlo simulations (Robert & Casella, 2004) are used to validate the model.
Furthermore, previous works had assumed the receiver node located at the center of the
network (Lau & Leung, 1992), (Shepard, 1996), (Hajek et al., 1997). The results presented
here are more general which shows that the received SNIR and spectral efficiency tend to a
constant as � increases if a node communicates with its close neighbors when the path loss
parameter � is greater than two, regardless of the position of the node in the network, i.e.,
wherever the receiver node is at the center, or at the middle, or at the boundary of the
network area. For the case where � equals two, the limit SNIR and spectral efficiency go to
zero; however, they decay very slowly making local communication still possible for a finite
�. Another study performed here presents an autonomous technique for node state
determination (sender or receiver) for each node in the network as function of the θ
parameter.
The remaining of this chapter is organized as follows. Section 2 introduces the network
model. Section 3 presents the average number of feasible receiving neighbor nodes as a
function of the network parameters. Section 4 explains the interference and spectral
efficiency computation. Section 5 shows the results. Section 6 explains the autonomous
technique for node state determination. Finally, Section 7 concludes the chapter
summarizing the main results obtained.
2. Model
The modeling problem addressed here is that of a wireless ad hoc network with nodes
assumed mobile. The model consists of a normalized unit circular area (or disk) containing �
nodes, and resembles the Grossglauser and Tse’s model (Grossglauser & Tse, 2001).
Therefore, information flow in the network follows the two-phase packet relaying technique
as described in (Grossglauser & Tse, 2001). The position of node � at time � is indicated by
�� . Nodes are assumed to move according to the uniform mobility model (Bansal & Liu,
2003). This model satisfies the following properties (Bansal & Liu, 2003): (a) the position of
the nodes are independent of each other at any time �; (b) the steady-state distribution of the
mobile nodes is uniform; (c) the direction of the node movement is uniformly distributed in
�0,��, conditional on the position of the node.
Any node can operate either as a sender or as a receiver. At a given time �, a fraction of the
total number of nodes � in the network, �� , is randomly chosen by the scheduler as senders,
while the remaining nodes, �� , operate as possible receiving nodes (Grossglauser & Tse,
2001). A sender density parameter θ is defined as �� =��, where � � �0, 1�, and �� =�1 � ��.
Section 6 describes a technique that allows the nodes to adjust their communication status
(sender or receiver) in order to the network attain �� =��.
A node � at time � is capable of receiving data at a given transmission rate of W bits/sec from
sender node � if (Grossglauser & Tse, 2001), (Gupta & Kumar, 2000)
��

�� , (1)
1 1
�� ∑��
where the summation is over all sender nodes � � �, �� is the transmitting power of
sender node �, �� is the channel path gain from node � to node �, � is the SNIR level
necessary for reliable communication, �� is the noise power spectral density, L is the
processing gain of the system, and I is the total interference at node �. In order to facilitate
the analysis, let us assume that no processing gain is used, i.e., � � 1, and that �� . The
channel path gain is considered to be a function of the distance, fading and path loss, so that
� �
��
�� , (2)
��
�
where �� is the Rayleigh fading from node � to node �, �� is the Euclidean distance between
nodes � and �, and � is the path loss parameter.
The goal is to find an equation relating the total interference measured by a receiver node
that is communicating with a neighbor node as a function of the number of total users � in
the network. More precisely, we aim to obtain an expression for Eq. (1) as a function of �
and calculate the limit of the SNIR and consequently the limiting spectral efficiency, as �
goes to infinity.
3. Feasible Receivers Near a Sender

In order to obtain the interference generated by nodes outside the neighborhood of a
receiver node, we first need to find the average radius size containing a sender node and
how many feasible receivers are within this range.
If the density of nodes in the disk is
� �
�� , (3)
�� 1
then the average radius for one sender node (�� ), for a uniform node distribution, is given by
1
1 � �� . (4)
√��
Hence, the average number of receiving nodes, called �, within �� , assuming a uniform
node distribution (Shepard, 1996), is
1 � 1
� � �� 1 � ��
� � � � 1� (5)
√��
which is constant for a given θ. Eq. (5) is a benchmark for obtaining the average number of
receiving nodes as a function of the network parameter θ.
Thus, the radius �� defines a cell (radius range) around a sender where �� receiver nodes are
�
nearby on average. The feasibility that all of those � nodes successfully receive the same
data being transmitted by the sender is the subject of the next section.
4. Interference and Capacity Computation 1
In the previous section, the average radius �� containing one sender with � � receiver nodes
around on average was obtained. Suppose that one of the � � receiving nodes is at the
neighborhood¹ distance �� . We want to show how the SNIR measured by this receiver
behaves as the number of total nodes in the network (and therefore the number of total
interferers) goes to infinity. We are interested in determing whether feasible communication
between the sender and the farthest neighbor (with distance �� ) is still possible, even if the
number of interferers grows.
Fig. 1. Snapshot of the unit area disk at a given time � . At this time, the receiver node being
analyzed is located at distance � � from the center while the sender is at distance �� from the
receiver node.
1 This represents the worst case scenario, because the other K  1 neighbors are located
either closer or at the same distance r0 to the sender, so they measure either a stronger or the
same SNIR value.
For a packet to be successfully received, Eq. (1) must be satisfied. Hence, consider a receiver
at any location in the network for a given time �. Its distance from the center �� is shown in
�
Figure 1, where 0 � � � � � �� .
√�
2
Let us assume that the sender is at distance �� from this receiver and transmitting at constant
power �, so that the power �� measured by this receiver is given by
��
�� , (6)
��
where �� is the Rayleigh fading from sender to receiver.
In order to obtain the overall expected interference at the receiver caused by all transmitting
nodes in the disk, let us consider a differential element area �� that is distant � units
from the receiver and �� units from the center of the network (see Figure 1). As consequence
of the uniform mobility model, the steady-state distribution of the nodes is uniform (Bansal
& Liu, 2003). Thus, the probability density function of the distance �� to the center of the
network is given by (Lau & Leung, 1992)
1
�� if 0 � �� √� (7)
0 otherwise.
Because the nodes are uniformly distributed in the disk, the transmitting nodes inside the
differential element of area generate, at the receiver, the following amount of interference²
��
��
�� . (8)
� �
The total interference is obtained by integrating Eq. (8) over the disk area and the result
depends on the value of �. Accordingly, the following two cases are considered.
A. The case � � �
For some propagation environments (Rappaport, 2002) the path loss parameter is modeled
to be always greater than two, i.e., � � �. In this case, the SNIR at the receiver located at
distance � � from the center for a total of � nodes in the network is given in the following
lemma.
Lemma 1. At a given time �, for a receiver node located at distance � � from the center in a unit area
disk network containing � mobile nodes uniformly distributed, where � � �, and assuming the
sender located at distance �� from this receiver, then the receiver SNIR is given by
2 Because the nodes are uniformly distributed in the disk and n grows to infinity, we
approximate the sum in Eq. (1) by an integral.
Pe
 s 
SNIRr ' (n)  2
2 Pe s 

N0 (9)
 q (n)

  2 r ', ,
 n  2
��
where � � �� , σ� is the standard deviation of the attenuation Gaussian random variable in decibels
due to shadowing (Akl et al., 2001), and
��
� 1
��
�
��
� �
�� ,�,� �� 1 � (10)
� �� .
� � � ��
� �
Proof of Lemma 1. By integrating Eq. (8) over the area of the disk, for � � �, we obtain the
interference at the receiver located at a distance � � from the center for a total of � nodes in
the network. Hence,
�� , ,��
��
��
��
��
�� , ,��
� �� (11)
� ��
�
�� 1 1
� � � �� .
�� , ��
�� is the maximum radius that � can have and is a function of the location � � and the angle �
(see Figure 1). To find this function, we can use the boundary disk curve (or circumference)
equation expressed as a function of the x-axis and y-axis shown in Figure 1, i.e.,
1 �
�� . (12)
√�
Define � � � � � � � , � � � �� , and � � �� . Then, Eq. (12) becomes
1 � 1
�� , �� . (13)
√ � �
By substituting this result in Eq. (11), we arrive at
�
��
�� ′��, (14)
��
where
�
��
�� ′� � � ��2
0 1 (15)
�� in ��2 � ��o� ��
�
is a constant for a given position � � . For the case in which � � 4, Eq. (15) reduces to
�2
�� ′� � . (16)
2 4
1 � 2�� 2 ��
The SNIR can be obtained by using Eqs. (1), (4), (6), and (14) to arrive at
�
��
��
�� 2�� 1
� �
�� 2 · 1
1� � �� · �� ′�
� � �� (17)
�
��
� � ,
�� 2��
� � · ��′,�,� ��
�� 2
Where
��
1
��,�,� �� 1 � � �� ′�� , (18)
� ��
� �
which finishes the proof. ■
From Eq. (9), taking the limit as � � �, we obtain
�
��
�� i� �
�� 2��
� � ��
� � 2 · �� ,�,�
�
��
��2 1
� if 0 � �′ � � �� (19)
� 2 √�
� ��2 1
� �� ,�,� �� if � ′ � � �� , i. e. ,
� 2 √ �
� the network boundary.
From Eq. (10), ��′,�,� �� ∞� � ��′,� �� ∞� because � is a scale factor on � and it does not
change the limit. Thus,
�
��
√�
� �.��
�
�
� ��
�
� √�
�
�� .�� (20)
� √�
�
� �.��
� √�
� �
��.�� .
√�
Therefore, from Eqs. (19) and (20), for � � �, the SNIR tends to a constant as � � ∞.
From Lemma 1, the spectral efficiency (C ) is straightly obtained and is given (in units of
bits/s/Hz) by (Cover & Thomas, 1991)
� � �
� ��
� � �� . (21)
� � ��
� ��
�
� � · � �
� ��
Accordingly, from Eqs. (9), (20) and (21), we conclude that the limiting spectral efficiency
goes to a constant as � � ∞ for � � �.
B. The case � � �
For the free space propagation environment (Rappaport, 2002), the path loss parameter is
modeled to be equal to two, i.e., � � �. Thus, the total expected interference at the receiver
located at distance �� from the center for a total of � nodes in the network is obtained by the
following lemma, which proof is analogous to Lemma 1.
Lemma 2. At a given time t, for a receiver node located at distance r from the center in a unit area
disk network containing � mobile nodes uniformly distributed, where � � �, and assuming the
sender located at distance �� from this receiver, then the receiver SNIR is given by
�
��
�� .
��
�
� (22)
· �� √��
��
Consequently, the spectral efficiency is obtained (in units of bits/s/Hz) by (Cover &
Thomas, 1991)
� � �
� ��
� � �� . (23)
� ��
� �
� · � �� √��
� ��
� ��
�
From Eq. (22), it is straightforward that �� as � � �. Therefore, the limiting
spectral efficiency goes to zero as � � � for � � �.
5. Results
In this section, the analytical results elaborated in Section 4 are compared with Monte-Carlo
simulations (Robert & Casella, 2004).
Figure 2 shows the spectral efficiency as function of � for � � �, � � �� for distinct values
of � � . Also, Figure 2 shows that the spectral efficiency remains constant when � goes to
infinity and its does not depend on � � if � � � � � √� �
� �� , and has the same value for any
position of the receiver node, whether the position is at the center, close to the boundary, or
at the middle region of the radius disk. Nevertheless, if the receiver node is at the boundary
(� � � √��
� �� ), then the limiting spectral efficiency is still a constant when � scales to infinity
but it has a greater value.
Fig. 2. Spectral efficiency curves as a function of � for � � ��

6 and �� for the receiver node located at different positions in the network. In the
legend, model is used for Eq. (21), while simulation is used for Monte-Carlo simulation.
Figure 3 illustrates the spectral efficiency behavior for different values of �� when � � �� .
As expected, the capacity diminishes when noise increases; however, the limiting capacity
is the same regardless of noise because the interference effect dominates the denominator of
Eq. (1) as � � �.
Fig. 3. Spectral efficiency curves as a function of ݊, for ߙ ൌ ͵ǡ ߠ ൌ ͳȀ͵ǡ ܲ ൌ ͳܹandߪ௦ ൌ ͸,

for different values of ܰ଴ . In the legend, model is used for Eq. (21), while simulation is used
for Monte-Carlo simulation.
Figure 4 confirms that the limiting spectral efficiency does not depend on ߠ as observed in
Section 4-A.
Fig. 4. Spectral efficiency curves as a function of ݊, for ߙ ൌ ͵ǡ ܰ଴ ൌ ͷǡ ܲ ൌ ͳܹandߪ௦ ൌ ͸, for

different values of ߠ. In the legend, model is used for Eq. (21), while simulation is used for
Monte-Carlo simulation.
Figure 5 shows the spectral efficiency as function of ݊ for different values of ߪ௦ . It also
illustrates that the capacity tends to a constant value as ݊ scales to infinity regardless of ߪ௦ .
Fig. 5. Spectral efficiency curves as a function of �, for � � �� 1� and � � 1��,

for different values of �� . In the legend, model is used for Eq. (21), while simulation is used for
Monte-Carlo simulation.
Fig. 6. Spectral efficiency curves as a function of � for � � �� 1�� 1��

6 and �� for the receiver node located at different positions in the network. In the
legend, model is used for Eq. (23), while simulation is used for Monte-Carlo simulation.
Figure 6 shows curves for spectral efficiency as function of n when � � �. Although the
limiting capacity goes to zero as already observed in Section 4-B, the decay is not fast. We
see that the spectral efficiency for a receiver node reaches 0.04 bits/s/Hz as the number of
interferers approaches 10�� , i.e., the capacity falls very slowly, and it can allow feasible
communication between neighbor nodes for a finite number of users �. In addition, the
capacity is about 0.066 bits/s/Hz for a receiver at the boundary of the network for this same
number of interferers. Note that we have plotted points up to � � 1�� for the model, while
the simulations were plotted up to � � 1�� due to computer limitation.
6. A Technique to Attain a Desired Value for The θ Parameter

Another challenge associated with this study is how a node can efficiently set its state
(sender or receiver) in order to the network attain a given �. The work presented by
Grossglauser and Tse (Grossglauser & Tse, 2001), and supplemented by Moraes (Moraes et
al., 2007) shows analytically and by simulation that the maximum throughput for a wireless
ad hoc network is achieved when the fraction of sender nodes (�) is approximately 1�3 of
the total nodes in the network. However, to the best of our knowledge, there are no studies
in the literature on MAC protocols that seek this distribution autonomously and in a
distributed way.
The technique suggested here consists of a simplified part of a MAC layer protocol. Similar
to the Traffic Adative Medium Access (TRAMA) protocol (Rajendran et al., 2003), our
scheme has the requirement to be synchronized with cyclical periods of contention followed
by transmission in which some nodes are capable of taking control of close neighbors, as
found in IEEE 802.15.4 - ZigBee (ZigBee Alliance, 2009). This technique, restricted only to the
contention period, is intended to be autonomous and able to distribute the states of the
nodes according to the � parameter.
Considering the node distribution as described in Section 2 and that each node has its
unique identification (ID), the node with the lowest ID controls the network and it is called
the coordinator node of the network.
The communication among nodes follows cycles which are divided in contention and
transmission phases. The contention period is divided in following three phases,
respectively. The announcement is the period in which each node sends its packet
identification number. The dissemination is the phase when the coordinator node sends its
identification to all nodes of the domain. Finally, there is the distribution phase where the
node coordinator sends a random sequence indicating the status (sender or receiver) that
each node in the network must assume during the following transmission period according
to the � parameter previously scheduled.
Figure 7 presents the results of a simulation implemented in JAVA (Java, 2009), using the
shuffle method of Class Collections (JavaClassCollections, 2009) for the random distribution
of the states (sender and receiver), which are displayed as fraction of times that three nodes
randomly chosen, over 100 cycles, were senders. It is observed that the three randomly
chosen nodes tend to converge their sender fraction of times to � � 1�3 as expected.
Fig. 7. Evolution of the fraction of times that three randomly chosen nodes were sender over
the simulated cycles for � � 1�3 and � � 1��.
The technique suggested here does not consider other medium access issues like channel
admission control, collision resolution, node failure, etc., which is subject of future work.
7. Conclusions
We have analyzed interference effects and spectral Shannon capacity (or spectral efficiency)
for mobile ad hoc networks using a communication channel model, which considers
Euclidean distance, path loss, fading and a random mobility model. We found that, for a
receiver node communicating with a close neighbor where the path loss parameter α is
greater than two, the resultant signal to noise and interference ratio (SNIR) and
consequently the spectral efficiency tend to a constant as the number of nodes n goes to
infinity, regardless of the position of the receiver node in the network. Therefore, for the
studied model, communication is feasible for near neighbors when the number of interferers
scales. Furthermore, for the receiver nodes located at the boundary of the circular network,
we show that they suffer less interference than those located inside attaining higher
capacity. Also, for the case where α is equal to two, the capacity was shown to go to zero as
n increases; however, the decay is very slowly making local communication still possible for
a finite n. Model and Monte-Carlo simulation results present good agreement and validate
the interference and Shannon capacity investigation performed.
It was also proposed a technique for autonomous and distributed allocation of states (sender
or receiver) of nodes based on the parameter �. Future work can consider MAC layer issues
like admission control and collisions, as well as power control and other types of mobility
which results in other distributions of nodes in the network, and how they affect the
capacity.
Acknowledgements
This work was supported in part by PIBIC/POLI, by Fundação de Amparo à Ciência e
Tecnologia do Estado de Pernambuco (FACEPE) and by Conselho Nacional de
Desenvolvimento Científico e Tecnológico (CNPq), Brazil.
8. References
Akl, R., Hedge, M., Naraghi-Pour, M., & Min, P. (2001). Multicell CDMA Network Design,
IEEE Transactions on Vehicular Technoogy, vol. 50, no. 3, pp. 711-722.
Bansal, N. & Liu, Z. (2003). Capacity, delay and mobility in wireless ad-hoc networks, Proc.
of IEEE Infocom, San Francisco, CA.
Cover, T. M. & Thomas, J. A. (1991), In: Elements of Information Theory. John Wiley & Sons.
Deng, J., Han, Y. S., Chen, P. N. & Varshney, P. K. (2004). Optimum transmission range for
wireless ad hoc networks, Proc. of IEEE WCNC, Atlanta, GA.
Gobriel, S., Melhem, R. & Mossé, D. (2004). A unified interference/collisionanalysis for
power-aware ad hoc networks, Proc. of IEEE Infocom, Hong Kong.
Grossglauser, M. & Tse, D. (2001). Mobility increases the capacity of wireless ad-hoc
networks, Proc. of IEEE Infocom, Anchorage, AK.
Gupta, P. & Kumar, P. R. (2000). The capacity of wireless networks, IEEE Transactions on
Information Theory, vol. 46, no. 2, pp. 388–404.
Hajek, B.; Krishna, A. & LaMaire, R. O. (1997). On the capture probability for a large number
of stations, IEEE Transactions on Communications, vol. 45, no. 2, pp. 254–260.
Java Class Collections (2009). On
http://java.sun.com/j2se/1.3/docs/api/java/util/Collections.html
Java Language (2009). On http://java.sun.com
Knopp, R. & Humblet , P. A. (1995). Information capacity and power control in single-cell
multiuser communications, Proc. of IEEE ICC, Seattle, WA.
Lau, C. T. & Leung, C. (1992). Capture models for mobile packet radio networks, IEEE
Transactions on Communications, vol. 40, no. 5, pp. 917–925.
Liu, X. & Haenggi, M. (2005). Performance analysis of rayleigh fading ad hoc networks with
regular topology, In Proc. of IEEE Globecom, St. Louis, MO.
Moraes, R. M. de, Buarque, C. E. & Araujo, F. P. (2008). Shannon Capacity for a Simple
Communication Channel Model in Dense MANETs, Proc. Of IEEE IWCMC, Crete,
Greece.
Moraes, R. M. de, Sadjadpour, H. & Garcia-Luna-Aceves, J. J. (2007). Taking full advantage
of multiuser diversity in mobile ad hoc networks. IEEE Transactions on
Communications, vol. 55, pp. 1202-1211.
Qin-yun, D.; Xiu-lin, H.; Hong-yi, Y. & Jun, Z. (2005). On the capacity of wireless ad-hoc
network basing on graph theory, In: Lecture Notes in Computer Science, vol.
3421/2005, pp. 353–360.
Rajendran, V. ; Obraczka, K. & Garcia-Luna-Aceves, J. J. (2003). Energy-Efficient, Collision-
Free Medium Access Control for Wireless Sensor Networks, Proceedings of ACM
SenSys 03, Los Angeles, CA.
Rappaport, T. S. (2002). In: Wireless Communications: Principles and Practice. Prentice Hall.
Rickenbach, P. von ; Schmid, S. ; Wattenrofer, R. & Zollinger, A. (2005). A robust

interference model for wireless ad-hoc networks, Proc. of IEEE WMAN, Denver, CO.
Robert, C. P. & Casella, G. (2004). In: Monte Carlo Statistical Methods. Springer-Verlag.
Shepard, T. J. (1996). A channel access scheme for large dense packet radio networks, Proc. of
ACM SigComm, San Francisco, CA.
Tobagi, F. & Kleinrock, L. (1975). Packet switching in radio channels: Part ii - the hidden
terminal problem in carrier sense multiple-access and the busy-tone solution, IEEE
Transactions on Communications, vol. COM-23, no. 12, pp. 1417–1433.
ZigBee Alliance (2009). IEEE 802.15.4, ZigBee standard. On http://www.zigbee.org
Energy Saving Drives New Approaches to Telecommunications Power System 201
11
X
Energy Saving Drives New Approaches

to Telecommunications Power System
Rais Miftakhutdinov
Texas Instrument Inc.
United States of America
1. Introduction
Steady growth in the telecommunication industry providing data, voice and video is very likely
to continue in the foreseeable future. This growth is supported by expansion into the new
markets, especially in Asia, accelerated widespread of wireless and broadband technology, and
strong demand for more efficient, power saving solutions. As the result of the growth, the
telecommunication infrastructure becomes significant energy consumer and contributor to
greenhouse emissions. Based on International Telecommunication Union estimation, the
information and communication technology contributes 2-2.5 per cent into the worldwide
greenhouse gas emissions (http://www.itu.int/themes/climate/index.html). To reduce the
impact on global warming, more efficient distribution, conversion and use of electrical energy
by telecommunication industry is required. Worldwide movements for energy saving and
“Green power” generation and distribution, have resulted in number of voluntary initiatives
and mandatory regulations by international and government organizations for increased
efficiency of electronic equipment including data and telecommunication power systems.
Examples of such organizations and initiatives are United States ENERGY STAR® program,
German Blue Angel, Japan Environment Association, European Code of Conduct and others
(http://www.energystar.gov/index.cfm?c=ent_servers.enterprise_servers; Mammano, 2006).
The focus of this chapter is efficient and low power consumption DC power systems for a
central office and base station of telecommunication infrastructure. According to (Fasullo et
al., 2008) telecommunication industry consumes 160 Billion kWh each year, and majority of
this electrical energy passes through DC power distribution system.
Telecommunication DC power systems have come long way from simple rectifier/battery
system to complex switching power supplies, from centralized power to distributed
architecture (Thorsell, 1990). At the same time, required tasks and functional complexity of
power systems continue to grow. To effectively reduce the overall system power
consumption per required functionality, all design levels from system architecture level
down to each specific function and component must be optimized. This chapter limits its
scope to energy saving considerations of power system at facility level, then down to power
distribution in a rack, or cabinet, and finally focuses on the specific power conversion
topologies and control algorithms implemented in power supplies.
At the facility level, intensive research and evaluation of 380-V DC distribution bus is
reported to replace traditional 208 V (230 V) AC mains (Pratt et al., 2007; Akerlund et al.,
2007). At the cabinet level, intermediate bus architecture (IBA) has become widespread to
address increased requirements for supply voltage quality, accurate power sequencing,
flexibility and availability of power system (Morrisson, 2002; White, 2003; Miftakhutdinov,
2008a). Currently, demand for high efficiency over wide output power range and low power
consumption reshapes the telecom power distribution system once again. Typical cabinet
level power system includes AC/DC front end power supply providing system bus voltage
that can be –48 V, 24 V, 12 V or 130 V depending on specific system and application. The
same power supply in most cases is used as a charger for the backup battery. Driven by
government regulations and market demand, the telecom and server power supply is now
required to be efficient over output power range from 10% (sometimes even 5%) up to 100%
(http://www.energystar.gov/index.cfm?c=ent_servers.enterprise_servers).
Efficiency was always important for data and telecommunication power supply to achieve
high power density and improve thermal performance. So far, only high efficiency at
maximum load was required because it determines reliability, size and cost of equipment
and cooling. Currently, the focus is shifted to energy saving and high efficiency over the
entire output power range.
Overall, the design procedure includes power system architecture selection and identifying
power conversion topologies and related control strategy. Use of the best in class
components is also critical to meet the design goals. In the chapter, all these critical stages of
telecom power system design are discussed in details including comparison of alternative
solutions.
Optimal control algorithm is critical not only to meet static and dynamic requirements of
telecom power system. It also opens new opportunities to increase the efficiency by
transitioning into different optimal power saving modes depending on system conditions.
Here, the flexibility, programmability and auto tuning capability of digital controllers must
be weighted against the lower cost, simple, and usually faster analog control ICs. Promising
control strategies along with the examples of advanced analog and digital controllers
addressing new requirements for high efficiency will be provided in the chapter.
The interface between IC controller and power stage, that includes power switch drivers,
current, voltage, and temperature sensing, auxiliary bias supply, has critical role and
deserve careful consideration as well.
The chapter discusses requirements for telecom rectifiers and front-end server power
supplies: the key functional parts of any data- and telecommunication power system.
Special attention is provided to intermediate bus converters (IBC) that are the enabling part
of any IBA. The IBC requirements and parameters, popular topologies, design challengers
are discussed in details. The design examples and test results of 600-W unregulated IBC
converter with 48-V input and 5:1 transfer ratio are provided to illustrate and verify the
recommended design approaches and solutions.
2. Strive for Efficiency and Power Saving

2.1 Energy Saving Trends and Regulations
High efficiency was always critical requirement for data and telecom power system as
precondition to achieve high power density and improve thermal parameters. So far, only
the efficiency at maximum load was usually being taken into consideration. This is because the
size, cost, temperature profile of components and their cooling selection is determined at the
maximum output power, where power losses are the highest. However, currently the
paradigm is shifted and the new requirements focus primarily on energy saving. Therefore, it
is critical to have high efficiency even at mid and light loads, where, as it turned out, power
system operates significant amount of time. Driven by government regulations and market
demand, the data and telecommunication power supply efficiency is now specified from 10%
(sometimes 5%) up to 100% of its output power range. At the same time, the power supply
and entire system must not exceed the power consumption limits specified for idle operation
modes. One example is ENERGY STAR®, which is a joint program of the U.S. Environmental
Protection Agency and the U.S. Department of Energy. The program sets efficiency and power
consumption recommendations and regulations for different types of electronic equipment.
The version 1 of ENERGY STAR® Program Requirements for Computer Servers was effective
starting May 15, 2009 (http://www.energystar.gov/ia/partners/product_specs/
program_reqs/computer_server_prog_req.pdf ). Table 1 below shows related efficiency
requirements at 10%, 20%, 50% and 100% output power of single-output AC/DC and
DC/DC converters.
Rated Output Power 10% Load 20% Load 50% Load 100% Load
≤ 500 W 70% 82% 89% 85%
501 – 1000 W 75% 85% 89% 85%
> 1000 W 80% 88% 92% 88%
Table 1. Efficiency requirements for single output AC/DC or DC/DC server power supply
For AC/DC Server Power Supply the ENERGY STAR® Program also defines the minimum
Power Factor Coefficient as 0.9 for loads from 50% to 100%. This practically means
mandatory use of active power factor corrector block in power supply. The Program also
limits maximum power dissipated at Idle State to 55 W for single processor based standard
server. By definition, during the Idle Operational State, the operating system and other
software have completed loading and the server is capable of completing workload
transactions, but not processing of any useful work. Adding redundant power supplies to
the system allows extra 20 W of power per each additional power supply. Another words
only 20 W power can be consumed by the power supply at no load condition.
Similar power saving programs are currently implemented or under development
worldwide by government organizations like German Blue Angel, Japan Environment
Association, European Code of Conduct and others (Mammano, 2006). It becomes
widespread practice that large data and telecommunication providers sometimes set even
stronger efficiency and power saving requirements to power system manufactures in
attempt to reduce the cost of service and stay competitive.
2.2 Power System Architecture at Facility Level

To meet new efficiency and power saving requirements all system and design levels must be
reviewed and optimized. These levels include general power system architecture, power
stage topologies for each power conversion, optimal power stage component selection and
control algorithms providing optimal and efficient operation of the entire system.
Typical power system of data center at the facility level is shown in Figure 1. Such system
generates uninterruptable 208 V AC line. There is double power conversion from DC to AC
in UPS and from AC back to DC in the front-end power supply.
Facility
Cabinet
PDU Front- POL
UPS (X-ER) End
City Facility DC/DC
480V AC 208V AC 12V DC
AC/DC DC/AC AC/DC
DC/DC
(PFC) (INV) (PFC)
DC/DC
Battery
Fig. 1. Typical power system of data center
If to replace AC uninterruptable distribution power line at facility level by the DC line, as it

is shown in Figure 2, more than 7% overall efficiency improvement (Pratt et al., 2007) and
10% to 30% saving in cost of operation (Akerlund et al., 2007) can be achieved.
Advantages of the power system with DC distribution bus at facility level are obvious from
the power saving view however, some safety and technical questions must be resolved
including certified DC power distribution units and availability of UPS with high voltage
DC output. The European Standard EN 300 132-3 issued by ETSI includes DC bus up to 400
V as an option for powering telecommunication equipment, thus setting guidelines for
development and use of such power architecture (ETSI, 2003).
Facility
X-ER Cabinet
for POL
USA PDU
City UPS DC/DC
Facility
480V AC 230V AC
AC/DC 380V DC 12V DC
(PFC) DC/DC
Battery DC/DC
Fig. 2. Power system with 380 V DC distribution bus at facility
3. Evolution of Telecommunication Power System at Cabinet Level

Power distribution systems for tele- and data-communication equipment at cabinet level
have undergone dramatic changes within last two decades because of fast progress of
modern digital-processing technology, requiring high quality supply voltages with specific
power sequencing. significant increase in economic losses in case of service interruption was
another key factor demanding highly reliable, flexible and available power system. And the
most recent changes are driven by push for the efficient, “green” power with the reduced
cost of ownership. The evolution of cabinet power system from centralized power to
distributed power architecture (DPA) and then to the intermediate bus architecture (IBA) as
subset of DPA is the focus of this section.
3.1 Centralized Power System

Originally, the only voltage needed for telecommunication electromechanical switching
systems was –48 V provided by AC/DC rectifiers and back up batteries. Since 1960s, the
transition from electromechanical relays to electronic semiconductor switchers added to
power system the DC/DC converters generating +5 V and ±12 V from –48V supply. These
centralized power supplies, typically located in the bottom of a rack or cabinet, included
AC/DC front-end rectifier/charger, a power backup battery and DC/DC converter. Large
and costly supply bus bars routed the required voltages to each shelf inside the cabinet,
which contained replaceable line cards with switching, diagnostic and monitoring
equipment. Figure 3 shows typical configuration of centralized power systems that were
dominant till mid 1980s (Thorsell, 1990, Ericsson Inc., 1996)
(a) Centralized power system (b) Multilocation centralized power system

Fig. 3. Different types of centralized power system with battery backup
In multilocation centralized power systems, the DC/DC converters were physically located
in different places, thus requiring safety shielding because of the presence of the high-
voltage bus. The centralized power system is still used in “silver” box power supplies for
low end desktop and server computers, but it has become obsolete in relatively large
telecommunication power distribution systems because of the following reasons:
 Centralized, custom power supplies require longer time to market and lack flexibility for
quick modification.
 Failure of any part of the power system means failure for the electronic equipment in the
whole cabinet.
 Custom, bulky power-delivery bus bars are expensive.
 Static and dynamic regulation of the supply voltage is poor and varies from shelf to shelf
3.2 Distributed Power Architecture

A dramatic step happened in early 1990s when the market largely adopted distributed
power architecture (Tabisz et al., 1992; Lindman & Thorsell, 1996). The bulky centralized
power supplies were replaced by AC/DC rectifier/charges providing –48-V backplane
voltage to each shelf and line card. The line cards allow hot-swap replacement to reduce
failure downtime. Each line card includes a number of –48-V input isolated DC/DC
modules, that provide all required voltages to the electronic functional blocks (Figure 4).
Fig. 4. Example of distributed power architecture
The introduction of distributed power architecture (DPA) was driven by the following:
 A trend towards digital processing blocks with increased power consumption, lower
voltages, and specific power sequencing
 A broad market introduction of modular, high density, and reliable isolated DC/DC
converters at a reasonable cost
 A demand for a more flexible, shorter design cycle power distribution systems allowing
quick changes and updates
 A need for systems with high reliability and availability that supported hot swapping
and had lower maintenance costs
3.3 Hybrid Power System

DPA-based systems addressed new power requirements, but the system cost remained
relatively high. When the required number of supply voltages per line card exceeded the
initial four to five, the excessive number of isolated DC/DC converters was questioned
(Narveson, 1996). In this paper there was suggestion to use only one isolated DC/DC
converter. This converter provides most power demanding supply voltage in the system
and also supplies non-isolated point-of-load (POL) regulators, which provide the remaining
supply voltages to electronic blocks. This architecture, commonly called hybrid power
system (Figure 5), was the first step towards the IBA. The hybrid power system reduces
power distribution costs and allows placing POLs right next to the related load, thus
reducing the impact of supply plane parasitics and improving high di/dt transient response.
If power sequencing is needed, an additional switch can be added between the isolated
converter output and the electronic load (Figure 5).
Fig. 5. Hybrid power system
The hybrid power system is preferable solution when one of the output voltages requires
relatively high power. In this case, a single regulated isolated converter improves the
efficiency of the whole system when the converter’s output voltage is 3.3 V or higher. With
the 3.3-V bus voltage, the hybrid system’s overall output power might be limited to about
200 W. This limit is suggested because high currents circulating through the power and
ground planes can cause significant losses and EMI issues as the system power increases.
3.4 Intermediate Bus Architecture

Driven by digital- and analog-IC industry demands for the low-level supply voltages in the
0.5-V to 3.3-V range and for the low-cost POLs, since early 2000s the market adopted the IBA
(Morrison, 2002; White, 2003, Mills, 2004). In many applications, the IBA-based power
system includes a front-end AC/DC power supply with a typical output of –48 V, 24 V, 12 V
or 130 V. In some data-communication and medical equipment the input DC voltage can be
380 V taken directly from a power factor corrector output (Zhu & Dou, 2006) This voltage is
supplied to an input of intermediate bus converter, that provides isolation and conversion to
the lower level intermediate bus voltage, typically within 5 to 14 V. This intermediate bus
voltage is supplied to non-isolated, POL regulators that provide high quality voltages for a
variety of digital and analog electronic functional blocks (Figure 6).
Fig. 6. Example of intermediate bus architecture
The following are advantages of IBA:

 System cost is reduced because only one isolated converter is needed and low cost,
standardized, non-isolated POL regulators are available in the market.
 IBC circuit can be made simple because typically intermediate bus voltage variation is
relaxed.
 Quality of supply voltages is increased because non-isolated POLs are located next to the
electronic functional blocks.
 System is flexible for modifications and updates.
 Overall system reliability is higher.
 Housekeeping, power sequencing, diagnostics, optimized power saving modes are
easier to implement because all major control signals are on the secondary side.
The following are challengers that IBA needs to address:

 The IBC must have highest efficiency and power density to provide a competitive edge
for IBA versus DPA.
 The overall line card power can be limited because of high currents circulating through
ground and bus-voltage planes.
 Parallel operation of highly efficient unregulated bus converters can be difficult.
 Specialized IBC controller ICs are needed to address specific IBC requirements.
3.5 Comparison and Trade-Offs of IBA versus DPA

IBA is a continuation of DPA at the line card level. An optimal choice between IBA and
standard DPA for each specific case depends on many factors, including the number of
supply voltages, the required voltage and power levels, the system-bus input voltage range,
and the specified static and dynamic regulation for supply voltages. It is obvious that cost
and efficiency are the most significant trade-off. Table 2 shows the pros and cons between
IBA- and DPA-based systems in very general terms. A detailed analytical comparison is
needed to make the right design decision. Examples of such analysis can be found in
literature (Sayani & Wanes, 2003).
System Requirement IBA DPA

Wide ─ Best
Input Voltage Range
Narrow Best ─
<4 ─ Best
Number of Outputs
≥4 Best ─
─ Good
One Regulated Output Demands Most of the Power
Hybrid system could be the best in such case
Cost Best ─
Efficiency Better Best
Load Supply Voltage Quality Best Good
Power Density Best Good
Table 2. Comparison of IBA versus DPA for different system requirements
3.6 Selection of Optimal Bus Voltage

Optimal selection of intermediate bus voltage is critical for the overall performance and
lowest cost of IBA based power distribution system. For higher bus voltages, IBC is more
efficient; however, POL regulators perform more efficiently at lower bus voltages. A lower
bus voltage means higher currents circulating through the power and ground planes, thus
adding additional losses. Obviously, there are some trade-offs to consider when defining a
bus voltage optimized for the lowest overall power losses.
In general, the power losses, Ptot, associated with any switching power conversion can be
expressed as
Ptot  Pconst  Kv  V 2  Re q  I 2 (1)
where Pconst is nearly-constant power losses consumed by the control and housekeeping
circuits; Kv  V 2 is the power losses associated with the switching process (a function of
switching voltage, frequency and in some cases, the load current); Kv is a coefficient
measured in W/V2 that reflects module losses dependence on the switching voltage;
Re q  I 2 is the conduction power losses that are dependent on load current, I, and
equivalent resistances, Req, of the components and traces. It is assumed that the switching
frequency is constant.
The optimal bus voltage has to be analyzed for each design case because the selected IBC
converter and POL regulators differ in terms of their power losses dependence from the bus
voltage and current. The following example of bus-voltage optimization is for a DPA
consisting of an unregulated IBC converter and POLs providing five different output
voltages. It is assumed that for the bus voltage ranges from 5 V up to 15 V, the MOSFET
switches for the selected IBC converter and POL modules remain the same. The key
optimization parameters are shown in Table 3. These data is taken from the IBC converter
and the POL modules available in the market. The parameters Req and Kv are specific for
the selected modules and might be different for other practical examples.
Kv,
Module Vout, V Iout, A Pout, W Pconst, W Req, mΩ Ploss(I), W
W/V2
POL #1 0.7 60 42 0.46 2.5 9 0.038

POL #2 1.0 120 120 0.92 1.25 18 0.076
POL #3 1.5 60 90 0.46 2.5 9 0.038
POL #4 2.5 60 150 0.46 2.5 9 0.038
POL #5 3.3 30 99 0.23 5 4.5 0.019
Total - - 501 2.53 - 49.5 0.209
Bus
- - - - 2 Pplane(Vbus) -
Plane
IBC Vbus Ibus Pbus(Vbus) 0.5 4 Req x Ibus2 0.056
Table 3. IBA power system parameters for optimal bus voltage analysis
The sum of the constant losses of each POL module (2.53W) and the sum of the output-
current related losses (49.5W), can be used to define the total losses in the POLs as function
of Vbus:
W
Ppol (Vbus)  2.53W  0.209  Vbus 2  49.5W (2)
V2
The bus-voltage power and ground planes have a resistance, Rbus, equal to 2mΩ, and the
overall output power, Pout total, is equal to 501 W. Thus, the plane losses are defined as
function of Vbus:
2
 Ppol (Vbus)  Pouttotal 
Pbus(Vbus)  Rbus    (3)
 Vbus 
IBC converter Vbus-dependent losses, Pibc(Vbus), are shown in Equation (4) after
substituting the related parameters from Table 3 and the Equations (2) and (3):
2
W  Pbus(Vbus)  Ppol (Vbus)  Pouttotal 
Pibc(Vbus)  0.5W  0.056 2
 Vbus 2  4m    (4)
V  Vbus 
Therefore, total IBA-based power system losses can be defined as:
Ptotal (Vbus )  Ppol (Vbus)  Pbus (Vbus )  Pibc(Vbus) (5)
Figure 7 shows power losses plots as a function of bus voltage. The optimal bus voltage for
minimal overall power losses can be chosen from the plot. In this particular case, the curve
showing total power losses is relatively flat in the region of minimum losses for bus-
voltages between 8 and 10.5 V. With this wide optimal bus-voltage range, the unregulated
IBC converter can be good fit depending on its input voltage range.
The optimal bus voltage is usually lower for the higher switching frequencies of POLs and
the lower total system power. This trend supports a balance between the voltage-dependent
losses like switching losses and the current-dependent losses like conduction losses.
150
140
130
120
110
Power Losses, W 100
90
80
70
60
50
40
30
20
10
0
5 6 7 8 9 10 11 12 13 14 15
Bus Voltage, V
Ppol
Pbus
Pibc
Ptot
Fig. 7. Power losses over bus voltage
4. Telecom Rectifier and Front-End Server Power Supply

Practically every telecom rectifier, or server power supply have the following key functional
blocks, which are usually associated with any over 500 W AC/DC power supply:
 EMI filter
 Power factor corrector (PFC) with hold up capacitor
 Isolated post-PFC DC/DC converter
 Auxiliary bias or standby power supply
 Fan and its regulator
Regulations and specifications define the overall efficiency of AC/DC power supply. It is
the responsibility of designer, based on previous designs and future forecast, identify the
efficiency and power losses of each functional block to meet the total efficiency goals.
4.1 Power and Efficiency Distribution

For a power and efficiency distribution analysis between the key functional blocks of
AC/DC power supply the following approach can be used. Usually the EMI filter and PFC
are considered together because it is convenient from the test procedure as well. The output
of the PFC (typically 400 V) supplies the main isolated DC/DC converter and the standby
power supply. The typical standby power-supply output-power range Can be from 5 W up
to 30 W depending on application. It is much lower than the output power of main DC/DC
converter. But, the efficiency and power consumption of standby power supply can not be
neglected, because the regulations specify the efficiency down to 10% or even 5% of
maximum output power. The fan regulator is usually supplied from the output of DC/DC
converter and thus, it is included into the efficiency of converter. Table 4 below is an
example of power and efficiency distribution analysis between the PFC, main DC/DC
converter and standby power supply. It is fulfilled for 12-V, 660-W output server power
supply.
Rated Output Power 10% Load 20% Load 50% Load 100% Load
Efficiency from Table 1 75% 85% 89% 85%
Overall Power
89 W 158 W 376 W 788 W
Consumption
PFC Efficiency 95.3% 96.4% 97.6% 97.7%
PFC Output Power 85 W 152 W 367 W 770 W
Standby Power 6W 7W 10 W 10 W
Standby Power Efficiency 80% 82% 85% 85%
Standby Power
7.5 W 8.5 W 12 W 12 W
Consumption
DC/DC Input Power 77.5 W 143.5 W 355 W 758 W
DC/DC Output Power 66 W 132 W 330 W 660 W
DC/DC Efficiency Goal 85.2% 92.7% 93% 87.1%
Table 4. Power and efficiency analysis of 660-W server power supply
4.2 Power Factor Corrector

Efficiency of power factor corrector (PFC) depends significantly on input AC line range
(Cohen & Lu, 2008). Typically, for the more than 500-W PFC, the boost converter based
power stage remains the most popular option. The boost converter achieves its highest
efficiency at high input line, and the efficiency gradually degrades at lower input voltages.
The efficiency and power factor specified by ENERRGY STAR® test procedure for the single
output server power supplies has to be confirmed by measurements at 230 Vrms AC line
(http://www.energystar.gov/ia/partners/product_specs/program_reqs/computer_server
_prog_req.pdf ). However, if the design targets the 85 to 265 Vrms universal range, all
critical thermal and electrical parameters of PFC have to be verified in the whole operating
range. Usually, the output power capability rated at 230-Vrms input voltage, for the same
front-end AC/DC power supply is de-rated for the 115-Vrms AC line.
Currently the interleaved PFC and bridgeless PFC are two major directions where most of
the research and development is focused. The interleaved PFC is already established
solution in mass production supported by available in the market controllers from different
vendors (see in http://focus.ti.com/docs/prod/folders/print/ucc28070.html). Typical application
diagram of the two-phase interleaved, continuous current mode PFC using UCC28070 from
Texas Instruments is shown in Figure 8.
+
–
Fig. 8. Two-phase interleaved PFC converter using UCC28070 controller
Advantages of interleaved PFC include:

 Reduced input current ripple because of ripple cancellation effect caused by 180O phase
shifted operation;
 Reduced EMI filter because of lower input current ripple;
 Lower RMS current through the output capacitor because of ripple cancellation effect.
This means less number of capacitors is needed, or increased reliability when the
output capacitance can not be reduced because of required hold up time;
 Better, equalized temperature profile because the power dissipated components are
spread between phases. This also results in the higher overall efficiency.
The first bridgeless PFC circuit has been patented as far as in 1983 (Mitchell, 1983), but the
concept is still mostly at the research stage. The practical implementation has been limited
by the high voltage MOSFET and diode performance, EMI issues, difficulties of voltage and
current sensing. Latest achievements in components technology, especially availability of
CoolMOSTM transistors and Silicon Carbide diodes, renewed interest to the bridgeless PFC
(Hancock, 2008). The analytical and experimental comparison of few different bridgeless
PFC topologies is provided in (Huber et al., 2008). The analysis claims that the bridgeless
PFC with two boost circuits (Souza & Barbi, 1999) shown in Figure 9 has the efficiency
advantages and less EMI issues versus other bridgeless PFC topologies. In general,
publications claim up to 1% efficiency improvement when using the bridgeless PFC versus
standard approach.
Fig. 9. Bridgeless PFC with two boost circuits (Souza & Barbi, 1999)
4.3 Isolated DC/DC Converter for Front-End Power Supply

The isolated DC/DC converter topology selection is critical for total efficiency of front-end
power supply. Zero voltage switching (ZVS) enabling topologies are preferable in such
applications because of the relatively high input voltage usually, from 350 to 420-V range.
Attractive solutions include phase shifted full-bridge, asymmetrical half-bridge, LLC
resonant converter and variations of these topologies (Zhang et al., 2004; Miftakhutdinov et
al., 1999; Fu et al., 2007).
For the interleaved topology, the asymmetrical half-bridge converter suits best because of its
relative simplicity (Miftakhutdinov et al., 1999). One possible example of interleaving with
four phases is shown in Figure 10.
Fig. 10. Four phases interleaved asymmetrical half-bridge.
The LLC resonant topology is recently gaining popularity as post-PFC isolated DC/DC
converter (Figure 11).
Fig. 11. LLC resonant converter power stage
Its main advantage is ZVS for the primary side switches and zero current switching (ZCS)
for the secondary side synchronous rectifier MOSFETs (Fu et al., 2007). Variable switching
frequency, special attention to light load operation and difficulties with interleaving limit
this topology to sub-kW range.
One implementation of classical phase shifted bridge topology using specialized analog
controller is shown in Figure 12. The efficiency improvement of this circuit is achieved by
using synchronous rectification, adaptive control algorithm providing ZVS condition over
wide operating range, accurate adaptive timing of control signals for primary and secondary
power FETs and light load management block providing the highest efficiency and power
savings at low output power conditions.
+ CT
Vdd Vdd
QA QC
A C
Vin
Vdd Vdd
QB QD
B D
_
+
QE QF
E F Vout
_
Vsense
1 VREF GND 24
2 EA+ VDD 23 Vdd
3 EA- OUTA 22 A
4 COMP OUTB 21 B
Vsense C
Enable 5 SS/EN OUTC 20
6 DELAB OUTD 19 D
7 DELCD OUTE 18 E
8 DELEF OUTF 17 F
9 TMIN SYNC 16 SYNC
10 RT CS 15
11 RSUM ADEL 14
12 DCM ADELEF 13
Fig. 12. Phase shifted full-bridge converter with advanced analog controller
4.4 Control Algorithms for High Efficiency

Optimal control algorithm is critical not only to meet static and dynamic requirements of
telecom power system, but it also opens new opportunities to increase efficiency by
transitioning into optimal power saving modes depending on system conditions. When
selecting the controller, the flexibility, programmability and auto tuning capability of digital
controllers have to be weighted versus lower cost, simple and generally faster analog control
ICs. The list of most popular power saving control strategies is provided below.
 Interleaving of few phases for better current and temperature distribution at maximum
output power and gradual phase shedding when the load is reduced (Figure 10);
 Synchronous rectification using MOSFETs with the diode emulation technique at light
load to avoid current circulation. It could be beneficiary to switch off the drive circuit of
rectifier MOSFETs at very light load where the drive losses exceed the conduction
losses. Performance of synchronous rectifier significantly depends on accurate timing
between primary and secondary side switches (Figure 12);
 Proper use of zero voltage (ZVS) and zero current (ZCS) switching technique to reduce
switching losses in power MOFETs. This requires optimal adaptive or predictable set of
delays between switching events depending on operation conditions (Figs. 10 - 12);
 Optimal adjustment of intermediate bus voltage, drive voltage and other system
parameters to maintain highest efficiency at different operation conditions;
 Smooth transition between operation modes to maintain highest efficiency depending
on operating conditions, for example from continuous mode to discontinuous, from
fixed frequency to frequency foldback etc (Figure 12);
 Proper use of pulse skipping or burst mode at light load or no load to reduce the power
consumption (Figure 12)
This list shows benefits of wide use of digital controllers to address power saving technique
because of their programmability and flexibility. The digital controllers for power supplies
are available from few vendors at reduced cost that make these devices competitive with
analog controllers, even for relatively low power applications in sub-kW range (see in
http://focus.ti.com/docs/prod/folders/print/tms320f28023.html ). Specifically designed
for these applications high end analog controllers also have their niche. Analog controller
ICs remain popular in mature, high volume applications where the operating conditions are
well known and established, and thus, cost is more critical than programmability and
flexibility (Figure 12).
4.5 Design Considerations and Component Selection

Optimal selection of power stage components provides foundation for high efficiency power
system design. Magnetics and power switches are major contributors into the total power
losses budget. In this chapter the main focus is on power MOSFETs and high voltage diodes
where the significant progress has been achieved lately. The new super junction technology
for high voltage MOSFETs significantly reduces Rdson, drain-source and gate-source
capacitances providing lower conduction losses and switching losses (Bjoerk et al., 2007).
Still accurate ZVS condition analysis over operating conditions remains critical to ensure the
highest efficiency. Because of significant non-linear behavior of drain-source capacitance,
the super junction MOSFETs, like CoolMOSTM, require new analytical model to estimate
switching losses and determine ZVS conditions. The following Equation (6) is adequate for
energy calculation stored in the output capacitance of high-voltage regular MOSFETs
(Miftakhutdinov, 2008a)
3
2
Ecds   Coss  Vdsoss  Vds 2 (6)
3
Here, Ecds is the energy, Coss is the output capacitance at Vdsoss = 25V from datasheet, and
Vds is the voltage where the energy should be calculated. The new super junction MOSFETs
require different model because of significant non-linear behavior of drain-source
capacitance. The following approximated Equation (7) provides good practical results for
super junction FETs:
Cinit  Vds 
2
Coss Vds  5V (7)
Ecds   (Vdsoss) 2  ln( )
Kc V 2
where Kc = 2.2 and Cinit = 40 pF for SPA11N60FCD type MOSFET from Infineon.
The plots in Figure 13 compare calculated energy using Equation (7) with the plot provided
in the datasheet.
8
5
Eoss (microJ)
0
0 100 200 300 400 500 600
Vds (V)
a) analytically derived plot b) experimental plot from datasheet

Fig. 13. Energy Ecds over Vds for SPA11N60FCD type MOSFET
The pairing of super junction MOSFETs with silicon carbide diodes in PFC applications
results in significant power losses reduction (Miesner et al., 2001). The use of silicon carbide
diodes practically eliminated the need for complicated snubbers in PFC boost power stage.
This is because these Schottky type diodes have very fast recovery time versus the p-n
junction silicon diodes. Regardless of the extra cost of such diode, the industry widely
accepts silicon carbide diodes for PFC applications because the overall efficiency gain could
be 3% or higher.
5. Intermediate Bus Converter

This section discusses major requirements to IBC converters, compares key parameters of
the available in the market products, considers preferable topologies and focuses on design
challengers that must be taken into account. An example of practical implementation based
on the IBC controller UCC28230 is also provided and supported by test results. Additional
analysis and design information related to IBC as part of IBA can be found in publications
(Barry, 2004; Miftakhutdinov & Sheng, 2007; Miftakhutdinov et al., 2008; Miftakhutdinov,
2008a; Miftakhutdinov, 2008b)
5.1 Major Requirements and Parameters of Modern IBCs

IBA includes an additional DC/DC conversion stage provided by IBC to supply
intermediate bus voltage. It is important for the IBC to be highly efficient with high power
density at the lowest possible cost. The first bus converters in the market were slightly
modified versions of fully regulated DC/DC modules. However, the IBC’s strict
requirements in a short time have made it a stand-alone, specialized product in module
manufacturer’ portfolios. A list of major IBC parameters follows:
 Efficiency: 96% to 97% typical
 Power density: >250 W/inch3
 Cost: $0.1 to $0.2 per watt
 Input voltage range:

 43 to 53 V for servers and storage
 38 to 55 V for enterprise systems
 36 to 60 V for narrow telecom range
 36 to 75 V for wide telecom range
 380 to 420 V for data center high-voltage systems
 Power range: 150 to 600 W and higher
 Mechanical form factor:
 1/4 brick for > 240 W of output power
 1/8 or even 1/16 brick for < 240 W output power
 Most popular transfer ratios: 4:1, 5:1 and 6:1 for –48-V nominal input voltage
 Switching frequency: relatively low at 100 to 200 kHz
 Most popular power stage topologies: Full-bridge, half-bridge, and push-pull
 Secondary-side rectification: Almost entirely uses synchronous MOSFETs, self- or
control-driven
 Control approaches: Fully regulated, semi-regulated, or unregulated
Because of the growing popularity of IBA, the IBCs for different power levels and transfer
ratios are readily available from different vendors. Table 5 shows the major parameters of
currently available IBCs in the market. This data is based on review of products from the
popular vendors in the first half of 2008.
Form
Manu- Input, Pout, Trans. Output, Eff., Density,
Model Fact.
facturer V W Ratio V % W/inch3
Brick
6.5-11.5
Tyco EUK240S9R0 36-60 1/8 240 5:1 95.5 272
unreg.
11.4-12.6
Tyco QBK033AOB 36-60 1/4 396 4:1 94.5 285
reg.
7.1-11.0
Ericsson PKM 4402NG PI 38-55 1/4 587 5:1 96.4 403
unreg.
11-12.5
Ericsson PKM400B PI 36-75 1/4 286 4:1 95.9 191
reg.
6.8-11.5
Delta Q48SB9R650NRFA 36-57 1/4 500 5:1 96.4 312
unreg.
7-11
Delta ES8SB9R625NRFA 38-55 1/8 240 5:1 96.5 258
unreg.
8.9-13.75
Delta V48SB12013NFRA 38-55 1/16 150 4:1 95.2 347
unreg.
Table 5. Major parameters of modern IBC converters
5.2 Control Approaches

Depending on the input voltage range and the requirements for output voltage tolerances,
the IBC can be regulated with the feedback loop taken from its output; semi-regulated with
the input voltage feed-forward circuit; or unregulated (Barry, 2004; Ericsson Inc., 2005).
The IBC with a closed feedback loop requires an additional isolation barrier for feedback
signal transfer. It is more expensive than semi-regulated because of more complex control
circuit and less efficient than unregulated IBC because it operates in a wide duty cycle
range. However, full regulation is justified for the hybrid power system where the IBCs
output is the supply voltage for the most power consuming load. If the power sequencing is
needed, an additional switch can be added between the IBC output and the load as it is
shown in Figure 5 (Ericsson Inc., 2005).
The semi-regulated IBC with input feed-forward control is usually a lower cost solution
than the fully regulated converter, but it also has lower density and efficiency than the
unregulated converter. This is because the semi-regulated IBC is designed to operate over a
wide duty cycle range, even at steady state. The semi-regulated IBC is usually used in a
system with a relatively wide input voltage range.
The unregulated IBC provides the solution with the highest efficiency and power density
and the lowest cost because it operates at almost 100% duty cycle at steady state. There is no
additional communication through the isolation barrier except for the energy transfer
through the power transformer. The size of the transformer and output and input filters is
small because converter operates at maximum duty cycle. However, overstresses during
transient conditions like start up, current limiting, and shut down need to be addressed
during the design.
5.3 Major IBC Topologies

IBCs usually employ forward type full-bridge, half-bridge and push-pull topologies with
the synchronous MOSFET rectification technique to achieve highest efficiency. Figure 14
shows three such IBCs in their very simplified forms.
a) Full-bridge IBC b) Half-bridge IBC c) Push-pull IBC

Fig. 14. Popular power stage topologies for IBC
Using a self-driven synchronous MOSFET rectifier is a very popular choice, especially for
unregulated converters, but practical solutions might require additional control windings
and snubber circuits for improved efficiency and reliability. For the high power applications
and, especially for the fully regulated and semi-regulated converters, the control driven
MOSFET rectifiers can be preferable. The advantages of using control driven rectifiers are a
simplified power transformer and a gate drive voltage that is independent from input
voltage and load current variations. A detailed review, classification and comparison of
synchronous rectification techniques can be found in Reference (Miftakhutdinov, 2007).
The double-ended topologies shown in Figure 14 are preferred for bus converter
applications because they can operate at almost 100% duty cycle applied to the output filter,
thus significantly reducing the size of the output inductor. Currently available IBCs usually
operate at about 100 kHz switching frequency. IBCs with 48-V (nominal) input voltage can
operate in the hard switching mode, but the zero voltage switching technique is preferred
for the IBCs with 400-V (nominal) input voltage. The full-bridge topology is preferred for a
250-W or higher output power. The half-bridge topology provides a low cost solution for the
output power range below 250 W. The bridge based topologies have primary MOSFETs
with a drain-to-source voltage rating equal to the input voltage, with some reliability
margin. These topologies are better choice for input voltages higher than 24 V. For a 24-V or
lower input, the push-pull topology is attractive because of simple drive circuit of primary
MOSFETs. However, the center tapped primary winding is a drawback for the planar
transformer in push-pull topology.
Table 6 provides a general comparison of IBC topologies. However, to select the right
topology during practical design, detailed calculations and a review of power system
specifications are needed for each specific case.
Topology Full-Bridge Half-Bridge Push-Pull

Primary MOSFETs Vds = Vin Vds = Vin Vds > 2Vin
Transformer Good utilization Issue with 5:1 transfer ratio Poor utilization
because planar transformer has
to be 2.5:1
Rectifier MOSFETs Primary winding clamping No primary winding clamping No primary winding
to zero is possible ability clamping ability
Output inductor The Same
Cycle-by-cycle Only a problem if a DC Inherent issue Not a problem
current limit blocking capacitor is used
Table 6. Comparison of popular IBC topologies
5.4 Using a Resonant Converter as an Unregulated IBC

Recently, high frequency resonant topologies for IBC application have been suggested and
their high performance reported (Ren et al., 2005). In this research the resonant topology has
been successfully used for a 48-V input, 12-V, 500-W output IBC at switching frequency up
to 800 kHz and the 95.5% efficiency achieved. Nevertheless, the resonant IBC approach has
not yet become mainstream in the industry.
5.5 Unregulated IBC Design Challenges

The design of unregulated IBC with self-driven MOSFET rectification has its own challenges
and trade offs. The design goal is to achieve the highest efficiency and power density at the
lowest cost. The challenges include the following:
 High ripple current during transitional states
 Start up problems
 Optimal synchronous rectification
 Reverse energy flow and self-oscillation
 Parallel operation issues
 Flux balancing of power transformer
A. Operation at Transitional States

At steady state, an unregulated converter operates at almost 100% duty cycle with very low
output inductor current ripple. However, during soft start or cycle-by-cycle current limiting,
the duty cycle varies from 0% to 100%, which can cause significant ripple increase in the
middle of this range. This ripple can overstress the power stage and limit the start up
capabilities of the IBC, especially when there is a large output capacitance. The output
inductor’s peak-to-peak ripple current, ΔIL, is defined for the whole duty cycle range with
Equation (8):
Vin  D  (1  D)
IL  , (8)
2  Ntr  Lo  Fsw
where Fsw = 1/Tsw is the switching frequency, D = Ton/(0.5 x Tsw) is the duty cycle after
rectification, Ntr = Wpr/Wsec is the transformer’s turns ratio, Wpr is the primary winding
turns, Wsec is the secondary winding turns, Lo is the inductance of the output inductor, and
Vin is the voltage applied to the transformer’s primary winding. Note that duty cycle
calculations and related equations assume a D value between 0 and 1. The following
discussion and plots refer to duty cycle in percent, which is D x 100. The plots in Figure 15
show that the output inductor’s ripple current is very low in the vicinity of D = 0 and 100%,
but can reach 120 A at D = 50%.
Fig. 15. Output inductor’s ripple current versus duty cycle for Vin = 50 V, Ntr = 5, Lo = 0.1
μH, and 100, 200 and 500 kHz switching frequencies in accordance with Equation (8)
Thus, the size and cost of power stage components, especially of the output inductor,
become significantly higher when this increased transitional ripple and peak current has
been taken into account.
One way to avoid the issue of high ripple current is to use a special frequency control circuit
that limits the output inductor’s ripple current during duty cycle transitions between 0%
and 100%. The desired change in switching frequency over the duty cycle range is
Fsw  k  D  (1  D ) , (9)
where k is a constant based on circuit implementation. Substitution Equation (9) into

Equation (8) gives the inductor ripple current as
Vin . (10)
IL 
2  Ntr  Lo  k
The result is that the switching frequency changes as the duty cycle changes to maintain the
inductor’s ripple current at a constant value. This idea has been implemented in Texas
Instruments UCC28230/1 controller
(http://focus.ti.com/docs/prod/folders/print/ucc28230.html).
A measured plot of the switching frequency change versus the duty cycle is shown in Figure
16. In this case, the nominal switching frequency is set at about 100 kHz.
Fig. 16. Measured switching frequency versus duty cycle with frequency control circuit
The frequency is maintained constant at steady state operation when the duty cycle is above
90 % or less than 10%. During start up or cycle-by-cycle current limiting, the duty cycle
varies significantly such that the inductor’s ripple current reaches a maximum value at 50 %
duty cycle. The frequency control circuit maintains the maximum frequency at about 420
kHz when the duty cycle is between 30 % and 70 %. The higher frequency significantly
reduces ripple current and allows the output inductor to be approximately 25% of the value
needed without the frequency control circuit. When the frequency control circuit is used,
inductor selection is based on a maximum frequency of 420 kHz at 50% duty cycle instead of
on 100 kHz as it would be without frequency control.
B. Start Up Problems
The ripple increase described in previous section also impacts IBC start up. The inductor’s
ripple current increase during start up may activate the over-current protection circuit,
possibly causing the converter not to start at all. Increasing the over-current limit threshold
and adding more filtering are not recommended for correcting a start up problem. If a real
over-current or output short circuit occurs, these methods of correction will probably
overstress the converter. To meet reliability and current stress margin requirements for the
power stage components, a much larger output inductor must be selected or the switching
frequency must be increased to reduce the ripple (Figure 15).
To further illustrate the start up issue, the plots in Figure 17 of output voltage versus
average load current are presented based on the following analysis. During the cycle-by-
cycle current limiting, the inductor’s average output current, Iout, and converter’s output
voltage, Vout, can be described by Equations (11) and (12):
Ntr  Vin  D Vin  D  (1  D ) , (11)

Iout  Io lim 
4  Lm  Fsw 4  Ntr  Lo  Fsw
where Iolim is the output current limit and Lm is the primary magnetizing inductance of the
power transformer. For any Iout range the output voltage, Vout, can be determined as
follows:
(Vin  Iout  Rpr / Ntr )  D

Vout   Iout  R sec , (12)
Ntr
where Rpr is the equivalent series resistance of the power stage primary side, and Rsec is the
equivalent series resistance of the secondary side.
a) Without frequency control Fmax = 100 kHz b) With frequency control Fmax = 420 kHz
Fig. 17. IBC start up capability at 75 A current limit threshold and 100-kHz nominal
switching frequency
The plots in Figure 17a show the output voltage versus the average load current at steady
state and during start up operation with cycle-by-cycle current limiting. These plots were
determined after substituting Equation (11) into Equation (12) with the following conditions:
Vin = 48 V, Fsw = 100 kHz, Ntr = 5, Lo = 0.1 μH, Lm = 75 μH, Iolim = 75 A, Rpr=25 mΩ and
Rsec=4 mΩ.
Also included in Figure 17a are the load curves for the resistive load of 0.45 Ω, a constant
current load of 11.5 A, and a constant power load of 60 W. The constant-resistance and
constant-current load curves are touching the start up Vout versus Iout curve without
crossing it. With the constant-power mode replicating POL regulator behavior, it is assumed
that the POL regulator starts operating and draws current only after Vout exceeds the under
voltage lockout threshold (UVLO) set at 5 V. Until then, the POL regulator does not draw
any current. Thus, the load curves indicate the maximum start up load current of the
converter designed for 60-A nominal output with a current limit set at 75 A. Obviously, the
fold back type of behavior of Vout versus Iout limits the start up capability of this
unregulated IBC. The load curves cross the steady state Vout (upper) plot at 21 A for the
constant-resistance mode, at 11.5 A for the constant-current mode, and at 5 A for the
constant-power mode. The start up performance of the converter is reduced dramatically
because of the inductor’s large ripple current at 50% duty cycle. Without the frequency
control circuit suggested earlier, the only way to override this limitation is to either increase
the output inductance or increase the nominal switching frequency. Either way, power
losses and converter cost increase.
Figure 17b illustrates the advantage of a start-up frequency-control circuit. The conditions
are the same as for Figure 17a except that the converter operates at 420 kHz for most of the
start-up time and at 100 kHz when it reaches the steady-state condition. With the same 0.1-
μH output inductor, the start-up capability is significantly improved over that shown in
Figure 17a where Fsw(max) = 100 kHz.. The load curves cross the steady-state Vout curve at
75 A for constant-resistance mode, at 59.5 A for constant-current mode, and at 32 A for
constant-power mode.
This start-up analysis is based on the assumption that the IBC’s output capacitance is not
very large. Obviously, if the allowable start-up time is short and the output capacitor is
large, an additional current to charge the high capacitance must be taken into account. The
frequency-control circuit increases the average charge current available for start-up even
with a large output capacitor. The average charge current, Ich, for the output capacitor, Cout,
that satisfies the selected soft-start time, tss, can be determined by Equation (13):
Vin (13)
Ich  Cout 
tss  Ntr
Figure 18 shows the IBC’s average output current required for charging different output
capacitances for the selected soft-start time. These curves do not account for the extra
current drawn by the load. The effects of different output capacitances can be estimated
with and without a frequency-control circuit by comparing the plots in Figs. 17 and 18. With
the frequency-control circuit, a charge current of at least 59.5-A is available per Figure 17b.
A 10-A portion of this current can be used to charge the 10,000-μF output capacitor within
10 ms per Figure 18. The remaining 49.5-A current is available to the load. Without the
frequency-control circuit, the available current per Figure 17a is only 11.5 A. This current is
barely sufficient to charge the 10,000-μF output capacitor within 10 ms. If the load draws
more than 1.5 A in addition to the capacitor’s charge current, the converter will not start
because the over-current protection circuit will be activated due to the large ripple current.
3
Charge current over soft start time
1 10
Charge current for specified Cout, A

100
10
0.1
0.01
0.1 1 10
Soft start time, ms
Cout = 100 uF
Cout = 1000 uF
Cout = 10000 uF
Fig. 18. IBC’s required charge current for different output capacitances at selected soft start
time
C. Optimal Synchronous Rectification Technique

For IBCs with a 12-V or lower output voltage, the synchronous-rectification technique is
mandatory to achieve the required efficiency. Compared to Schottky diodes, low-RDS(on)
rectifier MOSFETs can increase IBC efficiency by more than 5%. There are many
publications and patented solutions for how to drive the rectifier MOSFETs. Most designs
can be divided into self-driven, control-driven, and diode-emulator categories.
Classification of synchronous rectification and additional details can be found in Reference
(Miftakhutdinov, 2007). For the unregulated IBC, a self-driven rectification approach that
uses a secondary-side transformer winding (Figure 14) or an additional control winding is
quite popular because of its simplicity. The proper timing in either self-driven or control-
driven synchronous rectifiers is critical to reduce power losses. To avoid overshoot, it is
important that the conducting rectifier MOSFET on the secondary side turn off before the
primary-side MOSFET is turned on. This is achieved by proper OFF-time switching control
of primary-side MOSFETs for half-bridge (Figure 14b) and push-pull (Figure 14c)
topologies. For the full-bridge topology (Figure 14a), the OFF time is specified to be the time
between primary current switching of MOSFETs on one diagonal to MOSFETs on the other
diagonal. The optimal OFF time, Toff(opt), depends on power-stage parameters and the
load current.
With light loads, the optimal OFF time is longer. This relationship is illustrated in the drain-
source and gate-source switching waveforms of the synchronous-rectifier MOSFETs shown
in Figure 19a for no load and in Figure 19b for nominal current conditions.
a) No load, 50 ns/div b) 44-A load current, 25 ns/div

Fig. 19. Secondary side MOSFET rectifier switching waveforms
Optimal switching of rectifier MOSFETS over a wide load-current range is possible when the
OFF time is allowed to increase to some degree at light loads but is kept as short as possible
with nominal loads. A special OFF-time control circuit can be designed to allow the desired
output-current threshold to be set such that the OFF time, Toff, starts increasing and reaches its
maximum at no-load condition (Figure 20). This increase can be implemented in a linear
manner as shown in Figure 20a, or as a step function with hysteresis as shown in Figure 20b.
The method can vary depending on the specific design and application. Texas Instruments’
specialized UCC28230/1 bus-converter controller implements a comparator based approach
as shown in Figure 20b. This controller has dedicated pins (OS and OST) to allow
programming of the nominal OFF time, Toff, and the output current threshold so the OFF time
steps up to the new Toff(max) value at the desired current level. The gray area designated
“Tclamp” in Figure 20 represents the time when both rectifier MOSFETs are turned off. The
purpose of Tclamp is to prevent reverse energy flow, which is described in detail in the
following Section D.
a) Linear approach of controlling off time b) Step change approach of controlling off time
Fig. 20. Setting Toff, Td and Tclamp versus load current with off time control circuit
Since the control circuit of unregulated IBC does not have direct access to the secondary side,
so the primary current sensing with a current sense transformer or resistor is usually used to
monitor the output current indirectly. Primary side current sensing includes not only the
reflected load current, but also magnetizing current. However, in most applications, the
magnetizing current is only a small percentage of total current and can be ignored.
The impact of increasing off time at light load to the output voltage Vout is shown by
Equation (14):
Vin Ts  Toff (14)

Vout    Iout  Rout
Ntr Ts
For the comparator based approach shown in Figure 20b, the output voltage, Vout, can jump
a few hundred millivolts (with hysteresis) as shown in Figure 21. This jump is not desirable
if IBC’s operate in parallel with droop current sharing. For such applications, the off time
control circuit can be disabled to allow constant-slope output voltage.
Vout Droop of Vout vs Iout in

unregulated IBC because of
Toff min equivalent output resistance
Toff max 1
Toff min
Toff max2
Toff max 1
Toff max2
Hysteresis to avoid
instability because of
LC filter ringing
Selected OFF Time

Control Threshold
Idc of Lout
Fig. 21. Impact of comparator based off time control circuit on output voltage (not to scale)
The impact on Vout of changing off time is different for linear based off time control circuit.
Depending on the gain of the control circuit shown in Figure 20a and the output impedance
of the IBC, the slope of Vout versus Iout below the off-time-set threshold can be positive,
negative or zero (Figure 22).
Droop of Vout vs Iout in

Vout
unregulated IBC because of
Toff min equivalent output resistance
Always Stable Zone
Toff max 1
Proper Bandwidth is
needed to stabilize Toff min
Toff max2
Toff max 1
Toff max2
Selected OFF Time

Control Threshold
Idc of Lout
Fig. 22. Impact of linear based off time control circuit on output voltage (not to scale)
D. Reverse Energy Flow and Self-Oscillation

All topologies shown in Figure 14 are capable of transferring energy in the reverse direction,
that is, from the output to input. This is because MOSFETs can conduct current in either
direction when turned on. This is not true for a converter using a diode rectifier. During
shutdown or a sudden input voltage drop, it is possible for the self-driven MOSFET rectifier
to start oscillating and pumping energy backwards, thus causing large current and voltage
spikes at the rectifier MOSFETs (Bottrill, 2007). The reverse current flow is also possible
during quick converter re-start because the output bus capacitor has not been completely
discharged from the previous operation. Another potential condition for reverse energy
flow is the parallel operation of several bus converters.
One possible way to address this issue is to forcibly turn off the secondary side rectifier
MOSFET during primary-side MOSFET off time. To understand this technique let us refer to
Figs. 23 and 24. In this implementation, the controller uses additional output signals
O1_DIN and O2_DIN, to turn off rectifier MOSFETs during Toff time as shown in Figure 23.
Figure 24 shows the controller’s push-pull outputs, O1_D and O2_D, driving the high side
MOSFETs in the full-bridge power stage, and complementary 1-D outputs, O1_DIN and
O2_DIN, driving the low side MOSFETs via external drivers. There is always dead time, Td,
between the D and 1-D pulses that is necessary to avoid shoot-through currents in each leg
on primary side. If the duty cycle is less than maximum, there is the overlapping time,
TCLAMP, when the primary winding is shorted by the lower MOSFETs because they are both
in the ON state (Figs. 23 and 24). This specific timing algorithm has been implemented in
UCC28230/1 controllers (http://focus.ti.com/docs/prod/folders/print/ucc28230.html).
Fig. 23. Timing of UCC28230 controller’s output signal
As mentioned earlier, this timing technique addresses the problem of reverse current flow
during output pre-bias start up, shut down, input voltage drop, or parallel operation. For
the half-bridge (Figure 14b) or push-pull (Figure 14c) IBC topologies, the primary winding
of the power transformer can not be shorted by the primary power MOSFETs. To turn off
the secondary side rectifier MOSFETs during the TCLAMP interval, an external pulse
transformer can be used as shown in Figure 25. In this case the synchronous rectifier scheme
uses the control-driven technique for the unregulated IBC.
VDD
VDD UVLO Thermal
VDD Comp. Shutdown
12 EN Reference
6.3V rise VDD Generator
5.7V fall
VREF
1
5V/3.3V O1_DIN
LDO
R1 OS 10
3 Off Time
R3 Control
Short Circuit
Circuit
R2 2 Shutdown O2_DIN
OST 8
CS
Is
6
CT
Logic O1_D
Cycle-by- 11
Cycle Current Block
Limit
O2_D
9
Oscillator & CLK
Vin RT Start Up Soft Start &
4
Frequency Hiccup Current GND
Control Limit Circuit 7
5
SS
Fig. 24. Simplified diagram of typical full-bridge unregulated IBC
VDD
VDD UVLO Thermal
VDD Comp. Shutdown
12 EN Reference
6.3V rise V DD Generator
5.7V fall
VREF
1
5V/3.3V O1_DIN
LDO
R1 OS 10
3 Off Time
R3 Control
Short Circuit
Circuit
R2 2 Shutdown O2_DIN
OST 8
CS
Is
6
CT
Logic O1_D
Cycle-by- 11
Cycle Current Block
Limit
O2_D
9
Oscillator & CLK
Vin RT Start Up Soft Start &
4
Frequency Hiccup Current GND
Control Limit Circuit 7
5
SS
Fig. 25. Typical half-bridge unregulated IBC with control driven synchronous rectifier
E. Parallel Operation Issues

Parallel operation of IBCs is desirable in cases when the physical height is limited or when
there may be a future need to easily upgrade to higher power levels. Paralleling can also be
used for N+1 redundancy, but in this case, diodes in series with the outputs are needed to
isolate a failed converter from the rest of the system. It is impossible to use any kind of
active current-sharing technique with unregulated converters in parallel. The only option is
to use a droop-current-sharing mechanism that depends on the output impedance of the
converters sharing the current. Obviously, an accurate droop-current-sharing approach
becomes more difficult as new IBC designs become more efficient. Additional problems
related to sharing steady-state current can occur if all parallel IBCs do not start
simultaneously. These problems include power circulation and tripping the over current
protection circuit. Maintaining the secondary-side rectifier MOSFETs in the off state during
the 1 – D cycle previously described is one way to prevent reverse current flow during
parallel operation of unregulated IBCs.
F. Flux Balancing of Power Transformer

To reduce switching losses, unregulated IBCs use a relatively low 100- to 200-kHz switching
frequency. The power transformers in the topologies shown in Figure 14 are expected to
operate with a symmetrical B-H loop with no flux unbalancing for smaller size and reduced
losses. One option to avoid flux unbalancing is using the gapped transformer. However, this
approach increases a magnetizing current. Another option is to use a DC blocking capacitor
in series with the primary winding of full-bridge converter shown in Figure 14a. For the
half-bridge topology, this capacitor is already present as a necessary part of the power stage
(Figure 14b). The potential issue with the DC blocking capacitor is that during cycle-by-cycle
current limiting, significant variations in pulse amplitudes applied to the primary winding
each half-cycle might occur. This pulse variation occurs because significant DC voltage can
build up across the blocking capacitor that maintains volt-second balance of the transformer
each half-switching cycle. Unequal amplitude pulses to the transformer windings cause over
voltage stresses at the secondary side rectifier MOSFET. In many cases, careful layout and
symmetrical matched output pulses from the controller and drivers can eliminate the need
for DC blocking capacitor in the full-bridge converter. The simplest way to avoid
unbalancing in a push-pull converter is to use cycle-by-cycle current limiting or a gapped
transformer, because the DC blocking capacitor can not be used with this topology.

The described advanced control improvements to an unregulated IBC were verified with a
DC/DC module that had a 600-W output, a 48-V input, a 5:1 turns ratio, and a quarter-brick
form factor. The controller used for these experiments was the UCC28230/1. More details
about this controller can be found in (Texas Instruments, 2008.
http://focus.ti.com/docs/prod/folders/print/ucc28230.html) The measured module
efficiency, power losses, and output voltage are shown in Figs. 26, 27, and 28, respectively.
In this example, the off time was set to a fixed time period. For this reason the output
voltage shown in Figure 28 has an almost constant slope. The input-voltage measurements
were Vin1 = 38 V, Vin2 = 44 V, Vin3 = 48 V, and Vin4 = 53 V. A comparison to the old
controller that had an identical power stage, revealed an efficiency improvement of at least
1% over the full load-current range.
Fig. 26. Efficiency at 38V, 44V, 48V and 53V inputs over 0A to 56A output current range
Fig. 27. Power losses at 38V, 44V, 48V and 53V inputs over 0A to 56A output current range
Fig. 28. Output voltage versus load current measurements

6. Conclusion
General market trends and new regulations to the telecommunication power system are
discussed. It was shown, that to meet the new efficiency and power saving requirements, all
system and design levels must be considered. Therefore, the focus was on review and
comparison of the efficient, power saving solutions from the facility-level power system, to
the cabinet level, followed by discussion of the specific requirements and solutions for the
key functional blocks.
At the facility level, the new high voltage DC bus distribution system and its pros and cons
have been described and compared. At the cabinet level, the brief history of power system
evolution was shown. Pros and cons of different distribution power architectures were
provided. Advantages and challengers of the evolving intermediate bus architecture were
discussed in details including the optimal bus voltage analysis and selection.
The chapter discussed the requirements for telecom rectifiers and front-end server power
supplies: the key functional parts of any data- and telecommunication power system.
Special attention was provided to the intermediate bus converters that are an enabling part
of any IBA. Their requirements, key parameters, popular topologies, and design challenges
were discussed in depth. The design example and test results of 600-W unregulated IBC
converter with 48-V input and 5:1 transfer ratio was provided to illustrate and verify the
recommended design approaches and solutions.
7. References
Akerlund, J.; Boije af Gennas, C.; Olsson, G. & Rosin, D. (2007), One year operation of a 9
kW HVDC UPS 350 V at Gnesta Municipality data center, Proceedings of 29th
International Telecommunications Energy Conference, INTELEC’2007, pp. 40-45, ISBN:
978-1-4244-1628-8, Rome, Italy, September 2007, IEEE
Barry, M. (2004). Design issues in regulated and unregulated intermediate bus
converters,Proceedings of 19th Applied Power Electronics Conference, APEC’2004, Vol. 3,
pp.1389-1394, ISBN: 0-7803-8269-2, Anaheim, CA, USA, February 2004, IEEE
Bjoerk, F.; Hancock, J. & Devoy, G. (2007). CoolMOSTM CP- How to make most beneficial
use of the latest generation of super junction technology devices, Infineon
Appication. Note, AN-CoolMOS-CP-01, February 2007
Bottrill, J. The effects of turning off a converter with self-driven synchronous rectifiers,
Bodo’s Power Systems, May 2007, pp.32-35
Cohen, I. & Lu, B. (2008). High power factor and high efficiency – you can have both, Texas
Instruments Power Supply Design Seminar, Topic 1, SEM-1800, 2008
ENERGY STAR® program requirements for computer servers, version 1.0,
http://www.energystar.gov/index.cfm?c=ent_servers.enterprise_servers
Ericsson Inc., (1996). A designers’ guide to distributed power architectures using DC/DC
power modules, Published by Ericsson Components AB, 1996.
Ericsson Inc., (2005). Selection of architecture for systems using bus converters and POL
converters, Design Note 023, Ericsson Inc., May 2005
ETSI (2003) Environmental Engineering (EE) EN 300 132-3 Standard, Power supply interface
at the input to telecommunication equipment; Part 3: Operated by rectified current
source, alternating current source or direct current source up to 400 V, 2003
Fasullo, G.; Kania, M. & Pitts, A. (2008). The Green revolution in DC power systems,
Proceedings of 30th International Telecommunications Energy Conference,
INTELEC’2008, pp. 1-7, ISBN: 978-1-4244-2056-8, San Diego, CA, USA, September
2008, IEEE
Fu, D.; Lu, B. & Lee, F. (2007). 1MHz high efficiency LLC resonant converters with
synchronous rectifier, Proceedings of 38th Power Electronics Specialists Conference,
PESC’2007, pp. 2404-2410, ISBN: 1-4244-0655-2/07, Orlando, FL, USA, June 2007
Hancock, J. (2008). Bridgeless PFC boosts low line efficiency, Power Electronics Technology,
February 2008, pp.14-20
Huber, L.; Jang, Y. & Jovanovic, M. (2008). Performance evaluation of bridgeless PFC boost
rectifiers, IEEE Transactions on Power Electronics, vol.23, no.3, pp. 1381-1390, ISSN:
0885-8993, May 2008.
ITU, (2009). International Telecommunication Union web site,
http://www.itu.int/themes/climate/index.html
Lindman, P. & Thorsell, L. (1996). Applying distributed power modules in telecom systems,
IEEE Transactions on Power Electronics., vol. 11, no. 2, pp. 365-373, ISSN: 0885-
8993(96)01921-7, March 1996
Mammano, R. (2006). Improving power supply efficiency - The global perspective, Texas
Instruments Power Supply Design Seminar, Topic 1, SEM-1700, 2006
Miesner, C.; Rupp, R.; Kapels, H.; Krach, M. & Zverev, I. (2001). ThinQ!™ Silicon Carbide
Schottky Diodes: An SMPS Circuit Designer’s Dream Comes True! Infineon white
paper, Ordering No. B112-H7804-X-X-7600, September 2001
Miftakhutdinov, R.; Nemchinov, A.; Meleshin, V. & Fraidlin, S. (1999). Modified
asymmetrical ZVS half-bridge DC-DC converter, Proceedings of 14th Applied Power
Electronics Conference, APEC’1999, Volume 1, pp. 567-574, ISBN: 0-7803-5160-6,
Dallas, TX, USA, March 1999, IEEE
Miftakhutdinov, R. (2007). Synchronous rectification technique: classification, review,
comparison, Proceedings of Power Electronics Technology Conference on CD-Rom,
Dallas, TX, USA, October 2007, http://home.powerelectronics.com
Miftakhutdinov, R. & Sheng, L. (2007). New generation intermediate bus converter,
Proceedings of Power Electronics Technology Conference on CD-Rom, Dallas, TX, USA,
October 2007, http://home.powerelectronics.com
R. Miftakhutdinov and L. Sheng, J. Liang, J. Wiggenhorn and H. Huang, (2008). Advanced
control circuit for intermediate bus converter, Proceedings of 23th Applied Power
Electronics Conference, APEC’2008, pp. 1515-1521, ISBN: 978-1-4244-1873-2, Austin,
TX, USA, February 2008, IEEE
Miftakhutdinov, R. (2008a). Power distribution architecture for tele- and data
communication system based on new generation intermediate bus converter,
Proceedings of 30th International Telecommunications Energy Conference,
INTELEC’2008, Session 34-1, pp. 1-8, ISBN: 978-1-4244-2056-8, San Diego, CA, USA,
September 2008, IEEE
Miftakhutdinov, R. (2008b). Improving system efficiency with a new intermediate-bus
architecture, Texas Instruments Power Supply Design Seminar, Topic 4, SEM-1800, 2008
Mills, F. M. (2004). An alternative power architecture for next generation systems,
Proceedings of 4th International Power Electronics and Motion Control Conference,
IPEMC’2004, pp.67-72, ISBN: 7 - 5605 - 1869 – 9, Xian, China, August 2004, IEEE
Mitchell, D. M. (1983). AC-DC converter having an improved power factor, U.S. Patent 4 412
277, October 25, 1983
Morrison, D. (2002). Distributed power moves to Intermediate Bus Voltage, Electronic
Design, pp. 55-62, 16 September, 2002
Narveson, B. (1996). How many isolated DC-DC’s do you really need?, Proceedings of 11th
Applied Power Electronics Conference, APEC’1996, Volume 2, pp.692-695, ISBN: 0-
7803-3044-7/96, San Jose, CA, USA, March 1996, IEEE
Pratt, A.; Kumar, R. & Aldridge, T.V. (2007). Evaluation of 400V DC distribution in telco and
data centers to improve energy efficiency, Proceedings of 29th International
Telecommunications Energy Conference, INTELEC’2007, pp. 32-39, ISBN: 978-1-4244-
1628-8, Rome, Italy, September 2007, IEEE
Ren, Y.; Xu, M.; Sun J. ; & Lee, F. (2005). A family of high power density unregulated bus
converters, IEEE Transactions on Power Electronics, Vol. 20, No. 5, pp. 1045-1054,
ISSN: 0885-8993, September 2005, IEEE
Sayani, M. & Wanes, J. (2003). Analyzing and determining optimum on-board power
architectures for 48V-input systems, Proceedings of 18th Applied Power Electronics
Conference, APEC’2003, Volume 2, pp.781-785, ISBN: 0-7803-7768-0, Miami Beach,
FL, USA, February 2003, IEEE
Souza, A. & Barbi, I. (1999). High power factor rectifier with reduced conduction and
commutation losses, Proceedings of 21th International Telecommunications Energy
Conference, INTELEC’1999, Session 8.1, pp.1-5, ISBN: 0-7803-5624-1, Copenhagen,
Denmark, June 1999, IEEE
Tabisz, W.; Jovanovic, M. & Lee, F. (1992). Present and future of distributed power systems,
Proceedings of 7th Applied Power Electronics Conference, APEC’1992, pp.11-18, ISBN: 0-
7803-0485-3/92, Boston, MA, USA, February 1992, IEEE
Texas Instruments, Datasheet: Two-phase, interleaved CCM PFC controller, UCC28070,
http://focus.ti.com/docs/prod/folders/print/ucc28070.html
Texas Instruments, Datasheet: Advanced PWM Controller for Bus Converters, UCC28230/1
TI Document, SLUS 814, April 2008
http://focus.ti.com/docs/prod/folders/print/ucc28230.html
Texas Instruments, Datasheet: TMS320F28023 PiccoloTM microcontroller,
http://focus.ti.com/docs/prod/folders/print/tms320f28023.html
Thorsell, L. (1990). Will distributed on-board DC/DC converters become economically
beneficial in telecom switching equipment?, Proceedings of 12th International
Telecommunications Energy Conference, INTELEC’1990, pp. 63-69, Cat. No.90CH2928-
0, Orlando, FL, USA, October 1990, IEEE
White, R. (2003). Emerging On-Board Power Architectures, Proceedings of 18th Applied Power
Electronics Conference, APEC’2003, Volume 2, pp.799-804, ISBN: 0-7803-7768-0,
Miami Beach, FL, USA, February 2003, IEEE
Zhang, J. ; Xie, X.; Wu, X. & Qian, Z. (2004). Comparison study of phase-shifted full bridge
ZVS converters, Proceedings of 35th Power Electronics Specialists Conference,
PESC’2004, pp. 533-539, ISBN: 0-7803-8399-0, Aachen, Germany, June 2004, IEEE
Zhu, J. & Dou, S. (2006). Intermediate bus voltage optimization for high voltage input VRM,
Proceedings of 7th International Conference on Electronic Packaging Technology,
ICEPT’2006, pp.1-4, ISBN: 1-4244-0620-X/06/, Shanghai, China, August 2006, IEEE
Directional Routing Protocol in Wireless Mobile Ad Hoc Network 235
12
X
Directional Routing Protocol in Wireless

Mobile Ad Hoc Network
L.A.Latiff1, N. Fisal2, S.A. Arifin3 and A. Ali Ahmed4,
1,2,3Telematics Research Group, Universiti Teknologi Malaysia, Malaysia
Yemen University of Science and Technology,Yemen
1. Introduction
Advancement in wireless communication technology and portable computing devices such
as wireless handhelds, Personal Digital Assistants (PDA) and other mobile information
terminals have led to a revolutionary change in our information society towards the era of
mobile computing. The ubiquitous access to a variety of digital devices and multimedia
tools makes it possible to create, analyze, synthesize and communicate knowledge using a
rich variety of media forms. Additionally, the mobile devices are getting smaller, cheaper,
more convenient, and more powerful and have contributed to the explosive growth of the
mobile computing equipment market. Vast interest and concerted work in developing and
enhancing wireless and mobile network protocols are being driven by the ever increasing
demand for an anytime and anywhere Internet access.
To date, the type of network that have been widely deployed is based on a centralized
approach which requires a network point of access, commonly called the Access Point (AP)
that act as a gateway for the mobile device to the Internet. Even though these infrastructure-
based networks provide the path for mobile devices connectivity, time and potentially high
cost are required to set up the necessary infrastructure (Xue & Ganz, 2003). Also due to the
limited radio range, the devices must be in the vicinity of an AP in order to be connected. It
is important to note that when a natural catastrophe, war, or geographic isolation occurs,
communication may break down; thus, unavailability of the network connection (Milanovic
et al.,2004). Hence, the provision of required connectivity and network services at this
instance becomes a real challenge. With this scenario, ad hoc technology emerges with the
aim to solve this problem.
1.1 MANET Features

Mobile Ad Hoc Network (MANET) is a large collection of mobile nodes which are equipped
with wireless communication devices. These devices communicate peer-to-peer in a
network with no fixed infrastructure even when moving. Nodes in MANET can serve as
hosts and routers where communication between nodes beyond their transmission radius
can be achieved via several hops as inherent in collaborative communication of
neighbouring nodes. Due to its portability, MANET nodes depend on batteries for their
source of energy (Sun, 2001).
The wireless media used will remain to have a significantly lesser capacity compared to the
wired media (Macker & Corson, 1997). Also, the communication media, which is shared
with neighbours that are within communication range, warrants implementation of a
multiple access protocol to support systematic and efficient sharing.
Each mobile device in MANET is an autonomous node (Macker & Corson, 1997) and this
means that nodes not only perform basic processing abilities to initiate and receive data as a
host, it also performs routing functions. The routing algorithm in MANET can be a single-
hop or multihop which has different link layer attributes and routing protocols. Single-hop
communication is simpler in terms of structure and implementations but has lesser
functions and applications compared to multi-hop communication. In multi-hop
communication, the destination is beyond the transmission coverage of the source and
hence the packets are forwarded via one or more intermediate nodes. Figure 1 shows a
MANET network consisting of nodes and their transmission ranges. As shown in Figure 1,
Node 2 and Node 3 are neighbours of Node 1 whilst Node 4 and Node 5 are not. Therefore,
data transmission to Node 4 and Node 5 will have to be relayed by Node 2.
Fig. 1. MANET nodes and their transmission ranges
Since nodes in MANET use batteries as their source of energy which is easily depleted, it is
required to extend the longevity of nodes by utilizing effective and energy-saving
algorithm.
In MANET, nodes are free to move randomly. Hence, the topology of the network is ever
changing and the communication link created between a source and destination pair may
vary with time (Agrawal & Zeng, 2003). Nodes must be able to establish and maintain the
routes as they move to allow applications services to operate without interruptions.
Therefore, the control and management of the network must be distributed among the
mobile nodes to support multi-hop communication beyond the transmission range of a
node.
These characteristics do pose challenges to the roll-off of MANET which have motivated a
concerted time and effort of researchers in proposing new, innovative and improved routing
protocols for MANET. The important issues in routing are reaching the destination in
minimum time at minimum cost.
1.2 Routing Protocols in MANET

Topology in MANET is very dynamic and ever-changing where nodes are free to join or
leave arbitrarily. The goal of mobile ad hoc networking is to extend mobility into the realm
of autonomous, mobile, wireless domains, where a set of nodes themselves form the
network routing infrastructure in an ad hoc fashion (Macker & Corson, 1997). Traditional
routing protocol used in wired network cannot be applied directly to wireless and mobile
network. Several considerations are needed before we embark on the development of a
routing protocol for a wireless network which is non-trivial due to nodes’ high mobility.
Generally, there are two different stages in routing; they are route discovery and data
forwarding. In route discovery, route to a destination will be discovered by broadcasting
the query. Then, once the route has been established, data forwarding will be initiated and
sent via the routes that have been determined. Through broadcasting, all nodes that receive
the query will broadcast to all neighbours and hence, large number of control messages is
transmitted. It will be further compounded if the nodes move and new route need to be
recomputed. Frequent route discovery and in some instances, additional periodic updates
will cause more bandwidth being utilized and thus more energy wastage. Hence, to
conserve the power consumption, route relaying load, battery life, reduction in the
frequency of sending control messages, optimization of size of control headers and efficient
route reconfiguration should be considered when developing a routing protocol (Chlamtac
et al., 2003).
Over the past several years, many routing protocols have been proposed and can be
categorized into topology-based (Royer & Toh, 1999) and position-based protocols
(Giordano et al., 2004). Topology-based routing protocols route packets based on
information about the network links while position-based routing protocols uses physical
information about the participating nodes to decide on how to route packets. Topology-
based protocols can be further divided into proactive, reactive and hybrid routing protocols.
The network links are determined long before routing process in proactive protocols, when
routing in reactive protocols and a combination of before and when routing in hybrid
protocols. In the position-based protocols, location information of the destination are
known and used. There are two sub-divisions in position-based routing protocols, namely
greedy forwarding and restricted flooding. In greedy forwarding, nodes that have the best
progress will be selected and data packet will be forwarded to these nodes. Ideally, this
process is repeated until the packet arrives at the destination. Note there is no route
discovery in greedy forwarding. Restricted flooding, on the other hand, will mitigate
broadcast storm problem where only nodes in the direction of the destination will
participate in the route discovery until the route to destination is found. The participation
of nodes in routing will optimize broadcasting in MANET. Restricted flooding will
broadcast messages to a selected number of nodes which is usually more than one that are
located closer to the destination. It will significantly reduce not only energy but also reduce
the probability of packet collisions of messages rebroadcast by neighbours using the same
transmission channel (Stojmenovic, 2002; Mauve et all, 2001; Giordano et al, 2004). Figure 2
shows the categorization of routing protocols in MANET.
Fig. 2. Catergorization of MANET Routing Protocols
2. Literature Review
With the advent of Global Positioning System (GPS) and MANET environment-based self-
positioning (Latiff et al, 2005) and remote-positioning system (Li et al, 2000; Ali et al, 2004),
location information can be easily disseminated to the requesting node as required in the
position-based routing protocol. Besides availability of location information, complexity of
mathematical computations and issues pertaining to the implementation of the said
protocols should also be considered. Complex computational iterations will result in
processing delay and hence, higher latency while many routing packets transversing in the
network will result in high energy consumption and high probability of packet collisions.
Hence, position-based routing that restricts the broadcast region will reduce routing
packets, packet collisions and lower end-to-end delay with tolerable percentage of packet
delivered.
2.1 Greedy Forwarding

Greedy forwarding requires an up-to-date local topology via periodic beaconing which
eliminates route discovery and hence, only data packet forwarding are employed until it
reaches the destination. There are several forwarding strategies proposed that differ in the
way the node selects the next hop among its neighbours (Stojmenovic, 2002; Macker &
Corson, 1999). Figure 3 illustrates the various strategies, where S and D are the source and
destination nodes. The circle with radius r is the transmission range of S. The strategy is to
select and forward the packet to the node that has the best progress towards (or closest to)
destination.
Fig. 3. Greedy forwarding strategies
The first strategy is Most Forward within r (MFR) (Hou & Li, 1986) which select nodes that
will minimize the number of hops that a packet will traverse in order to reach D. The
selected node is C. Nearest with Forward Progress (NFP) ( Kranakis et al, 1999) proposed
to minimize interference with other nodes and also reduce the overall power consumption
by forwarding to the node that is nearest to S which is node A. In compass routing ( Chang
& Tassiluas, 2000), the forwarding decision will select the neighbour that is closest to the
straight line between S and D. In this approach, the selected node is B which minimizes the
spatial distance a packet travels. However, there are drawbacks in greedy forwarding where
it can only guarantee loop freedom for a certain kind of network topology. Greedy
forwarding works well in dense network but degrades in sparse networks. Hence, a path
towards destination cannot be found even though the path exists (Stojmenovic, 2002). As
described above, proactive information of one-hop neighbours obtained via HELLO
messages periodically transmitted that has information of the sending node location
information must be implemented. Hence, knowledge of the local topology will have to be
maintained in a neighbours table at each node. This will require storage of neighbour
information which meant additional cost. Also, these approaches will also require complex
computation at the nodes and hence,will incur delay at the intermediate nodes.
2.2 Restricted Flooding

The main approach in restricted flooding is to limit the flooding region which can be based
on distance, angle and distance covered by the next intermediate node. Using distance, only
nodes that are nearer to the destination will participate in the route discovery. Nodes that
are further away from source will not participate. LAR (Ko & Vaidya,2000) calculates
distance from the destination based on location information of the destination that will be
extracted from the request packet while (Cartigny et al, 2003) uses the relative neighborhood
graph (RNG) with local information of distance to neighbours and distances between
neighbours to decide whether its participation has better coverage compared to other nodes.
This will minimize the total energy consumption while still maintaining the whole network
coverage through broadcasting. (Ali et al, 2005) calculates distances to all nodes in the
network and will compare the distance information of the source to the destination
extracted from the request packet to determine its participation.
On the other hand, ARP (Kumar Bankar & Xue, 2002) and DREAM [Basagni et al, 1998) uses
the angle made from the straight line drawn from source to destination as the restricted
region whereby all nodes in this region will participate in the route discovery. However,
DDB (Heissenbutte et al, 2004) uses the location information of the destination node and
also of the intermediate node which are inserted in the request packet. With this additional
information, an intermediate node can calculate the estimated additional covered area that it
would cover with its transmission which is based on Dynamic Forwarding Delay (DFD).
The concept of DFD is to determine when to forward the packet and node with more area
covered will be given a smaller delay to broadcast and hence, will broadcast it first.
All the proposed protocols require quite complex mathematical computation of the distance,
angle and coverage at all intermediate nodes to determine the nodes’ participation.
Information of the source and destination are required and must be inserted in the incoming
packet.
In MANET, route discovery is initiated by total flooding of route request (RREQ) messages
that consume a large portion of the already limited bandwidth in MANET. As illustrated in
Figure 4, RREQ is broadcasted to all neighbours whereby frequent broadcast causes
network congestion and degrades the performance of routing protocol. This could be
proved by several performance observations that the number of RREQ in the network
increases linearly with the node population (Perkins, 2004). The ratio of control packet over
data packet even reaches 5000 in one of the experiments.
As such, we suggest utilizing restricted flooding mechanism to optimise the route

establishment phase of AODVbis. Restricted flooding is broadcasting messages to a selected
number of nodes which is more than one that are located in an area in the vicinity of the
destination. Position information of the destination can be obtained from any location
service while position location of the destination can be obtained with the aid of GPS or any
other self-positioning system proposed for MANET. Then if these information are piggy-
backed in the query packet, nodes will calculate its location with reference to the source and
destination and will then decide to broadcast the query or not. Figure 5 illustrates that the
same network topology shown in Figure 4 but with limited flooding. RREQ packets will be
broadcast by nodes located in the request zone which is a quadrant drawn with respect to
source node coordinates. Nodes participation is denoted by shaded circles with arrows
indicating the direction of broadcast while lesser-toned circles indicate non-participating
nodes. With this unique approach of using quadrant as the broadcast region, we proposed
Quadrant-based Directional Routing Protocol or Q-DIR.
Fig. 4. RREQs broadcast based on Total Flooding.
Fig. 5. RREQ broadcast based on Restricted Flooding.

2.3 Quadrant-based Directional Routing Protocol (Q-DIR)

Q-DIR is a limited flooding routing protocol that concentrates on a specified zone using
location information provided by a location service. It restricts the broadcast region to all
nodes in the same quadrant as the source and destination and does not require maintenance
of a separate neighbours table at each node as in (Ko & Vaidya, 2000; Kumar Banka & Xue,
2002; Cartigny et al, 2003; Heissenbutte et al, 2004). Q-DIR determines the quadrant of the
current node based on the coordinates of source, destination and the current node that will
direct the packet towards the destination. Even though (Cartigny et al, 2003) uses all these
information to determine the distance or area covered, it requires trigonometric
computations which will further incur delay if computed in kernel space.
Decision to broadcast or discard will be done as the RREQ packet is received by the node.
Unlike LAR scheme 2 (Ko & Vaidya, 2000), geocast-enhanced AODVbis (Ooi, 2005; Latiff et
al, 2006), nodes in Q-DIR do not keep a distance table or a neighbours table which must be
updated frequently. Keeping a distance table is much like a table-driven proactive routing
concept. The size of the table will increase for a larger network because nodes need to have
distance information of itself to every other node in the network which will vary from node
to node. This will pose a constraint to the maximum number of nodes in the network as the
memory allocation to store distance information of every node in the network at each node
is limited and scarce.
In Q-DIR, the RREQ packet which contains the coordinates of the source and destination
will be the only information the current node needs to decide to participate in the routing or
not. The decision to participate at each node is made immediately as the node receives the
RREQ packet and a neighbours table is not required to make the decision.
Q-DIR will significantly reduce not only energy but also reduce the probability of packet
collisions of messages rebroadcast by neighbours using the same transmission channel. This
will result in reduced routing overhead especially in a dense network.
3. Development of Q-DIR in Ns-2

In Q-DIR operation, the location information of the source and destination nodes is piggy-
backed in the route request (RREQ) packet and then broadcasted. Upon receiving the RREQ,
intermediate nodes will compare using a simple mathematical comparison based on the
coordinates of source, destination and the current node that directs the packet towards the
destination and as illustrated in Figure 6. This mathematical processing will be done in the
kernel environment to eliminate the cross-over from user to kernel space and vice versa.
Hence, the decision to participate is made immediately.
Quadrant of me compared to source?

Quadrant of destination compared to source?
If same, FORWARD
If not, DROP.
Fig. 6. Q-DIR decision at each intermediate node.

Once the decision to broadcast has been made, the intermediate node will insert its location
by replacing the source node coordinates and append its address and sequence number at
the end of the RREQ packet. It will then broadcast the packet. The process will repeat at
each intermediate node until it reaches the destination. The replacement of the source node
location information with the intermediate node coordinates will make the packet more
directed towards the destination since the comparison now is based on the previous node.
Upon receiving the RREQ, destination node will send a route reply message (RREP) back to
source via the path taken to reach the destination that was appended in the RREQ as it
traverses across the network. There is no need for the route discovery to the source node.
Figure 7 shows the format of the RREQ packet in Q-DIR where the source and destination
nodes location information are inserted are highlighted.
Fig. 7. RREQ format in Q-DIR.
There are several open source network simulators such as Commnet, OMNeT++ but
Network Simulator-2 (Ns-2) has been found to be a widely used tool for simulating inter-
network topologies and to test and evaluate various networking protocols (CMU Monarch
Project, 2006). It is a discrete event simulator written in C++ and uses Massachusetts
Institute of Technology (MIT) Object Tool Command Language (OTcl) as a command and
configuration interface. The most important characteristic of a discrete-event approach is
that the components of an actual network are represented within the software and real
events are simulated by the operation of the software.
Ns-2 can be installed on both Windows and Linux platforms. For Q-DIR, the simulation
work is done on a Red Hat Linux platform (Chakeres et al, 2005) The compiler used in Q-
DIR is the ns-allinone-2.28 version (CMU Monarch Project, 2006). The underlying protocol is
AODVbis which has the path accumulation feature (Gwalani et al, 2003). The Dynamic
MANET On-demand – University of Murcia (DYMOUM) (Ros & Ruiz, 2004) is based on
DYMO which is an internet draft dated June 2005. DYMO enables reactive, multi-hop
routing between participating node that wish to communicate. The basic operation of
DYMO is similar to AODVbis which are route discovery and route management and the
differences are in the new packet format, generic packet handling, unsupported element
handling and optional path accumulation. DYMOUM has There are three types of elements
that have been defined in DYMO. They are RE (Routing Element), RERR and UERR
(Unsupported-element Error) and RE can be further divided into RREQ and Route Reply
(RREP). From the description given, DYMO can be used as the underlying protocol in this
work. DYMO is reactive and implements route discovery and path accumulation. Even
though it uses a generic element structure but basically has the needed RREQ, RREP and
RERR packets as in the AODVbis routing protocol. Any modification work should be done
in the C++ hierarchy.
3.1 RREQ packet format

As shown in Figure 8, to modify the RREQ packet, the source and destination coordinates
are declared as a double precision integer. In the DYMOUM source file, when a new RREQ
is generated by the source node, the NS_CLASS re_create_rreq () procedure will create the
RREQ packet. The RREQ packet requires location information of the source node; therefore
the following syntax will extract the source coordinates from the ns-2 environment which is
searched by using the node address.
Node*node=Node::get_node_by_address(re_node_addr.s_addr);
((MobileNode *)node)->getLoc(&x,&y,&z);
To extract the destination coordinates, a declaration of the following were made at the
beginning of the source file that permits calling for those information using Tcl hooks in the
ns-2 platform. Description of the declaration for Tcl hooks will be described in the following
section.
extern dst_x, dst_y;
Fig. 8. Declaration of additional fields to the RREQ packet.

3.2 Tcl Hooks

Ns-2 consists of two hierarchies: compiled C++ hierarchy and the OTcl that make use of
objects in C++ through OTcl linkages that have a one-to-one correspondence to each other.
The objects that have already been linked are “no_path_acc_”, “reissue_rreq”, and “s_bit”.
Therefore, to link the coordinates of the destination node that will be declared in the tcl
script in the OTcl environment, links for both objects have to be created and the declarations
are as shown in Figure 9. This ns-2.28/dymoum-0.1/ns/dymo_um.cc file has other links to the
C++ hierarchy that are relevant to the DYMO configuration but will not be shown here. The
variables for dst_x and dst_y in the header file dymo_um.h to enable referencing by
dymo_um.cc have been declared. To enable calling DYMOUM from the tcl script, the agent
DYMOUM (Agent/DYMOUM), dst_x and dst_y in the ns-default.tcl file in ns-2 library are
inserted.
Fig. 9. Binding of Tcl objects to the C++ hierarchy
3.3 Processing RREQ

As described in Section 3, processing of RREQ consists of two events. They are Generating
RREQ when the current node has data to send and initiates the route discovery for a certain
destination and Receiving RREQ that is implemented at the intermediate nodes that receives
the query broadcast. In the same dymo_re.c file previously mentioned, in the function
NS_CLASS re_process(),variables such as temporary fields to store coordinates of current
node and value of quadrant are declared. Then, the syntax to search for the current node
coordinates and store these information in mynode_x and mynode_y will be made as shown in
Figure 10.
Fig. 10. Declaration of variables in dymo_re.c.
When receiving RREQ, nodes will compare the quadrant of destination and current node
and the codes are be inserted right after the declaration of the variables in function
NS_CLASS re_process(). The coordinates of the source are denoted by src_x and src_y,
while the coordinates of destination are denoted by dst_x and dst_y. The current node
coordinates are denoted by mynode_x and mynode_y. If the quadrant of the source is equal
to the quadrant of destination, the current node will broadcast the request. The code for this
receiving RREQ is shown in Figure 11.
Fig. 11. Code for broadcast decision at each node.

4. Verification of Q-DIR
4.1 Network Model
A network model N-1 as shown in Figure 12 is used which consist of 6 nodes and the
coordinates are carefully chosen so that there will be at least 2-hops transmission to the
destination. The imaginary x- and y-axis are drawn to show in which quadrant the nodes
are located with reference to their immediate neighbours.
Fig. 12. Topology N-1
For topology N-1, the 1-hop neighbours of node 0 are nodes 1, 2, 3 and 4 while the
neighbours of node 5 are 1 and 2. The configuration parameters used for both Q-DIR and
AODVbis routing are shown in Table 1. The values can be modified depending on the
network and its environment. The maximum number of hops between nodes have been set
to 10 while the estimated average of one hop traversal time is set to 0.6 s. From I-D (Perkins
et al, 2003), for correct operation, the route delete period must be greater than both (Allowed
HELLO loss* HELLO interval) and the total traversal time.
The MAC layer protocol used is IEEE 802.11 DCF CSMA/CA. The data rate have been set to
2 Mbps and the network protocol is IP. The path loss model used is the log-normal path loss
model (Rappaport, 2002) and the value for n is 2.4 and the standard deviation σ is 4. To
simulate in ns-2, the receive threshold power has to be determined first in order to set the
transmission range to 30 meters for all 1-hop neighbours. The default transmitted power is
0.28318 W. The receive threshold power calculated is 1.20475e-08 watts and the packet rate
is set to 1 packet/sec while the packet size is set at 64 bytes and 512 bytes with a CBR
(Constant Bit Rate) traffic pattern.
Configuration Parameters Value

Maximum number of possible 10
hops between two nodes
Average one hop traversal time 60 milliseconds
Route discovery time 2400 milliseconds
Route delete period 4800 milliseconds
Number of RREQ tries 3
Total traversal time 1200 milliseconds
HELLO interval 1000 milliseconds
Allowed HELLO loss 2
Table 1. Simulation Configuration Parameters.
4.2 Results
The simulation for topology N-1 was run and results show that for N-1, nodes 1, 2, 3 and 4
will all receive the RREQ packet from source node 0 destined for destination node 5.
However, nodes 1, 3, and 4 will drop the packet since they are in different quadrant from
the source and destination. Figure 13(a) shows the snapshot of the message displayed on the
screen when running the simulation. The snapshot shows that node 1 and 3 drop the RREQ
received from source node 0 while Figure 13(b) shows that node 4 drops the packet from
source node 0. On the other hand, from Figure 13(c), node 2 forwards the packet to
destination node 5 since it is in the same quadrant as destination compared to source.
(a)
(b)
(c)
Fig. 13. RREQ packet broadcasting decision for Topology N-1 (a) Node 1 and 3 drop. (b)
Node 4 drops. (c) Node 2 broadcasts.
5. Performance of Q-DIR in Dense Network

The study to evaluate the performance of Q-DIR in a large and densely populated network
were conducted and it is hoped that results will show that Q-DIR with reduced collisions,
and less contention of bandwidth, less routing overhead and consequently, power
consumption is inherent and reflects that this new routing protocols is implementable and
economical.
5.1 Dense Network Model

Figure 14 shows a network model of 49 nodes that forms a 7 by 7 grid model where the
distance from adjacent nodes are 30m. Based on this grid model, the density is 1 node per
661m2. In the network model, the x- and y-axis of the Cartesian coordinate system have been
drawn to denote in which quadrant the nodes are located. The source and destination are
denoted by the letter S and D respectively and destination node is at the top right edge of
the grid. The simulation configuration parameters used in the simulation are as shown in
Table 1.
Fig. 14. Simulation Network Model of 49 nodes
5.2 Performance Metrics

Two protocols were simulated and they are AODVbis which is a total flooding protocol and
Q-DIR which is based on restricted flooding. The performance metric used are as follows:
 normalized routing overhead - The number of routing packets transmitted per data
packet received at the destination.
 Effective energy consumption per data packet received - The total energy consumption in
the network for every data packet successfully received by the destination. This is the
metric on the effectiveness of energy consumption when routing data packets.
5.3 Simulation Results

A. Effect of Varying Simulation Time
The simulation time was varied from 100s to 800s in steps of 100s. The number of routing
packets that are broadcast and the corresponding data packet received at the destination in
the network are counted for both AODVbis and Q-DIR routing protocol. Figure 15 shows
the normalized routing overhead graphs for both protocols. As the simulation time
increases to 800s, both protocols show reduced routing packets and leveled to a constant as
it approaches 800s. The average normalized routing overhead in AODVbis is 338 packets
while in Q-DIR, the average normalized routing overhead is 128 packets per data packet
received. It is observed that 160% more routing packets are transmitted in AODVbis
compared to Q-DIR due the higher number of node participations in the network in
AODVbis.
Figure 16 shows graph for effective energy consumed per data packet received for both
protocols. Both protocols shows a reduced energy consumption as the simulation time
increases The average effective energy is 2.43 J in AODVbis and 1.48 J in Q-DIR. Q-DIR
consumes 64% less energy to send packets since only a quarter of the number of nodes
participated in the routing process which is a limited flooding protocol based on quadrant.
Fig. 15. Normalized routing overhead with simulation time.

Fig. 16. Effective energy consumed per data packet received in Q-DIR.
B. Effect of Varying Packet Rate

Both AODVbis and Q-DIR routing protocols are simulated in the 49 nodes topology for a
simulation time of 400s because the performance of both protocols remains constant. The
transmission rate was varied in steps of 32 kbits/s with initial rate of 16 kbits/s to a
maximum of 144 kbits/s. Figure 17 shows the average normalized routing overhead for
both protocols which increases as the transmission rate increases. The graph for AODVbis
shows large fluctuations as the transmission rate increases. AODVbis sends out an average
of 255.664 normalized routing packets compared to Q-DIR which sends out only 108.08
packets. The large fluctuations in AODVbis are due to the total flooding algorithm of
AODVbis and hence the routes taken vary for different transmission rate. However, the
graphs in Q-DIR remain consistent throughout due to the directed flooding based on
quadrant.
Figure 18 shows the graphs for effective energy consumed per data packet received for both
AODVbis and Q-DIR protocols. The effective energy for AODVbis fluctuates as the
transmission rate increases but for Q-DIR, it remains constant. Again, the fluctuation in
AODVbis is due to different route taken at different transmission rate. .AODVbis consumes
an average of 1.574 J of energy while Q-DIR consumes only 1.084 J of energy which 45% less
energy consumed compared to AODVbis. Based on this trend in energy consumption, less
power is consumed if only a section or an area of a network participates in the routing.
Fig. 17. Normalized routing overhead for 49 nodes.
Fig. 18. Effective energy consumed per data packet received

6. Conclusion
This paper has presented the performance of Q-DIR which is a restricted flooding algorithm
which uses location information of the source, destination and the intermediate node to
determine the broadcasting decision. Nodes that are in the restricted broadcast region will
broadcast while other nodes which are out of this region will ignore the RREQ packet. The
simple mathematical comparison is implemental in the kernel environment which does not
incur processing delay due the crossing from user to kernel space and vice versa. The
simulation results shows that implementing Q-DIR reduces the power by 160% as the
simulation time is increased and by 45% as the transmission rate increases compared to
AODVbis. The restricted flooding and directional routing reduces the number of
participating nodes as the RREQ traverses in the network towards the destination node and
hence reduced routing overhead and power consumption are achieved in Q-DIR.
7. References
Agrawal, D.P. and Zeng, Qing-An. (2003). Introduction to Wireless and Mobile Systems.
USA:Brooke/Cole-Thomson Learning.
Ali, A., Latiff, L.A., Fisal,N. (2004). GPS-free Indoor Location Tracking in Mobile Ad Hoc
Network (MANET) using RSSI. RFM 2004, Malaysia.
Ali, A., Latiff, L.A., Ooi, Chia-Ching, and Fisal, N. (2005). Location-based Geocasting and
Forwarding (LGF) Strategy in Mobile Ad Hoc Network (MANET) ICT 2005, Cape
Town, South Africa, May 2005.
Basagni, S., Chlamtac, I. and Syrotiuk, V.R. (1998). A Distance Routing Effect Algorithm for
Mobility (DREAM). ACM/IEEE International Conference on Mobile Computing and
Networking. Oct. 1998. 76-84.
Cartigny, J., Simplot, D. and Stojemenovic, I. (2003). Localized minimum-energy
broadcasting in ad hoc networks. INFOCOM 2003. 30 March-3 April 2003. 3:2210 –
2217.
Chakeres, I. D., Royer, E.M. and Perkins, C. E. (2005). Dynamic MANET On-demand
(DYMO) Routing . Internet Draft of IETF MANET WG. draft-ietf-manet-dymo-02.txt.
June 2005.
Chang, J. and Tassiluas, L. (2000). Energy Conserving Routing in Wireless Ad Hoc
Networks. Proceedings of IEEE INFOCOM 2000. 22-31.
Chlamtac, I., Conti, M. and Liu, J. J. (2003). Mobile Ad-hoc Networking: Imperatives and
Challenges. Elsevier Ad Hoc Networks Journal. 2003. Vol. 1(1): 13-64.
Department of Computer Science, Rice University. (2006). CMU Monarch Project.
http://www.monarch.cs.cmu.edu/. Accessed June 2006.
Giordano, S., Stojmenovic, I. and Blazevic, L. (2004). Position-based Routing Algorithm for
Ad Hoc Networks: A Taxanomy. In: Ad Hoc Wireless Networking. X. Cheng, X
Huang and D.Z. Du (Eds.). 103-136. Kluwer Publishers.
Gwalani, S., Belding-Royer, E.M. and Perkins, C.E. (2003). AODV-PA:AODV with Path
Accumulation. International Conference on Communications. May 2003. 26(1):527-531.
Heissenbuttel, M., Braun, T., Walchli, M. and Bernoulli, T. Broadcasting in Wireless
Multihop Network with Dynamic Forwarding Delay Concept. Technical Report
IAM-04-010. Institute of Computer Science and Applied Mathematics University of
Berne, Switzerland, December 2004.
Hou, T.C. and Li, V.O.K. (1986). Transmission Range Control in Multihop Packet Radio
Networks. IEEE Transactions. In Communications. 34(1):38-44.
Ko, Y.B. and Vaidya, N.H. (2000). Location-Aided Routing (LAR) in Mobile Ad Hoc
Networks. ACM/Baltzer Wireless Networks Journal. 2000. 6:307-321.
Kranakis, E., Singh, H. and Urrutia, J. (1999). Compass Routing on Geometric Networks.
Proceeding of the 11th Canadian Conference on. Computational Geometry. Vancouver,
Canada.
Kumar Banka, R. and Xue, G. (2002). Angle routing protocol: location aided routing for
mobile ad-hoc networks using dynamic angle selection. MILCOM 2002 Proceedings.
7-10 Oct 2002. 1:501-506.
Latiff, L.A., Ali, A., Ooi, C.-C. and Fisal, N. (2005). Development of an Indoor GPS-free Self-
Positioning System for Mobile Ad Hoc Network (MANET). MICC-ICON 2005,
Malaysia.
Latiff, L.A., Ali, A., Ooi, C.-C. and Fisal, N. (2006). Implementation of a Quadrant-based
Directional Routing Protocol (Q-DIR) in Wireless Mobile Ad Hoc Network.
IASTED NCS 2006. 29-31 Mac 2006. Chiang Mai, Thailand. IASTED. 2006.
Li, J., Jannotti, J., De Couto, D.S.J. , Karger, D.R. , Morris, R. (2000). A Scalable Location
Service for Geographic Ad Hoc Routing. Mobicom 2000.
Macker, J. and Corson, S. (1997). Mobile Ad Hoc Networks (MANET). 1997.
http://www.ietf.org/html.charters/manet-charter.html
Mauve, M., Widmer, A. and Hartenestein, H. (2001). A Survey on Position-based Routing in
Mobile Ad Hoc Network. IEEE Network. 15(6):30-39.
Milanovic, N., Malek, M., Davidson, A. and Milutinovic, V. (2004). Routing and Security in
Mobile Ad hoc Network. IEEE Computer Journal. 37(2). 61-65.
Ooi, C.-C. (2005). Implementation of Geocast-Enhanced AODVbis Routing Protocol in MANET.
M.Sc. Thesis, UTM;2005.
Perkins, C.E., Royer, E. M and Chakeres, I. (2003). Ad hoc On Demand Distance Vector-bis
(AODVbis) routing. Internet Draft of IETF MANET WG. October 2003.
Perkins, C.E. (2004). Convergence in Ad hoc Networking Protocol. The IASTED International
Conference on WNET 2004. July 2004.
Rappaport, T.S. (2002). Wireless Communications: Principles and Practice. 2nd Edition. USA.
Prentice Hall. 2002.
Ros, F. J. and Ruiz, P. M. Implementing a New Manet Unicast Routing Protocol in NS2.
December 2004. http://www.masimum.dif.um.es/nsrt-howto/html/nsrt-howto .html.
Royer, E.M and Toh, C.K. (1999). A Review of Current Routing Protocol for Ad-Hoc Mobile
Wireless Network. IEEE Personal Communications. pp. 46-55, April 1999.
Stojmenovic, I. (2002). Position-based Routing in Ad Hoc Networks. IEEE Communications
Magazine. 40(7). 128-134.
Sun, J.Z. (2001). Mobile Ad Hoc Networking An Essential Technology for Pervasive
Computing, Intl. Conference on Info-tech & Info-net. Beijing, China.
Xue, Q. and Ganz, A. (2003). Adaptive Routing in Ubiquitous Mobile Access Networks.
Proceedings of IEEE VTC. 6-9 October, 2003. Orlando, Florida: IEEE. 3070-6074.
Free Space Optical Technologies 257
13
X
Free Space Optical Technologies

Davide M. Forin, G. Incerti
University of Rome Tor Vergata –Italy
G.M. Tosi Beleffi
ISCOM, Italian Ministry of Economic Development Comm. Department, Rome – Italy
A.L.J. Teixeira, L.N. Costa, P.S. De Brito Andrè
Instituto de Telecomunicações, Aveiro, Portugal
B. Geiger, E. Leitgeb and F. Nadeem
Graz University of Technology
1. Introduction (D. Forin – G.M. Tosi Beleffi – B. Geiger – E. Leitgeb)

Free Space Optics (FSO), also known as Optical Wireless or Lasercom (i.e. Laser
Communications), is a re-emerging technology using modulated optical beams to establish
short, medium or long reach wireless data transmission. Most of the attention on FSO
communication systems it was initially boost by military purposes and first development of
this technology was dedicated to the solution of issues related to defense applications.
Today’s market interest to FSO refers to both civil and military scenarios covering different
situations and different environments, from undersea to space. In particular, due to the high
carrier frequency of 300 THz and the consequently high bandwidth, the most prominent
advantage of Free Space Optical (FSO) communication links may be their potential for very
high data rates of several Gbps (up to 40 Gbps in the future (J. Wells, 2009)). Other
advantages like license-free operation, easy installation, commercial availability, and
insensitivity to electromagnetic interference, jamming, or wiretapping make FSO interesting
for applications like last mile access, airborne and satellite communication (L. Stotts et alt,
2009), temporary mobile links and permanent connections between buildings. Mainly, the
adoption of FSO is needed when a physical connection is not a practicable solution and
where is requested to handle an high bandwidth. As a matter of fact, FSO is the only
technology, in the wireless scenario, able to grant bandwidth of several Gigabits per second.
The interest in this technology is also due to the low initial CAPEX (Capital Expenditure)
requirement, to the intrinsic high-level data protection & security, to the good flexibility and
great scalability innate in this solution. For these reasons FSO possible applications cover
today, as mentioned, a wide range. Thus this technology generates interest in several
markets: the first/last mile in dense urban areas, network access for isolated premises, high-
speed LAN-to-LAN (Local Area Networks) and even chip-to-chip connections, transitional
and temporary network connection, undersea and space communication. Furthermore FSO
can be used as an alternative or upgrade add-on to existing wireless technologies when the
climatic conditions permit its full usage.
In spite of the growing interest in space and undersea applications, infact, the terrestrial FSO
still remains of primary significance and the performance of such links is highly dependent
on different weather conditions. Atmospheric effects affect the distance and the availability
of the optical wireless links so not all the geographical sites are suitable for this kind of
broadband solution. Links as long as 7 km are in operation, but prior to the deployment of
wireless optical links the average weather condition must be evaluated to estimate the
expected outage time on the link in that area. The outage will depend on the link length and
on the persistence of adverse weather conditions. So in general, it can be affirmed that short
enough links, hundreds of meters, will be operational also with the worst possible weather
conditions.
Besides commercially available opaque systems, where the optical signal is terminated to an
electronic receiver and subsequently sent through the atmosphere by means of a dedicated
laser, a new configuration, known as fully transparent, is under study. The bandwidth
achievable in these last systems is comparable to the optical fiber one. Being absent any kind
of optical to electrical bottleneck (OEO). For transparency, infact, is meant launching and
collecting power directly through single mode optical fibers. Such new extremely high
bandwidth wireless systems, although still in the research stage, are gaining more and more
interest especially in the last mile scenarios. Last but not least, compared to a microwave
link, an FSO link can support higher bit rates and its operational frequencies are license-free
in all jurisdictions apart for an year low cost fee that must be paid to the reference PA
(Public Administration). The Authors want to outline that the work carried out in this
chapter has been done in the framework of the European funded FP7 NoE BONE Project
(WP13) and the COST IC0802 Action.
2. History of Free Space Communications (G.M. Tosi Beleffi)

Telegraphy is a word coming from ancient greek and means in Italian “scrivere a distanza”
while in English sounds more or less like “writing to a distant place”. The human being has
from the very beginning tried to increase his capabilities to communicate with his far away
fellow men and so to transmit. Under this point of view, the mythology is full of interesting
examples with the most famous and known that is Ermes, the Gods messenger, able to move
faster than the wind and responsible to carry informations to the Gods.
First experiences in the ancient past can be found in the IVth century b.C. (before Christ),
where Diodoro Crono reports on a human chain used by the Persian king Dario I (522-486
b.C) to transmit informations from the Capital to the Empire’s districts.
In the IVth Century b.C., Enea il Tattico, reports on an hydraulic telegraph probably
invented by the Chartaginians. During the Roman and Greek age, was used to place in
geographical key points “fire towers” to be switched on in case of security breachs and/or
attacks on the borders. Eschilo (525-456 a.C.) reports in the Orestea that the news about the
falls of Troy arrived to Argo passing through the Cicladi islands covering, more or less, 900
km (Eschilo, 458 b.C.). This sort of tradition remained, for example, on the Italian territory
assuming and adopting different schemes, fire or mechanical systems, depending on the
time period, the geography and the geopolicy (Pottino, 1976).
In the Center-South of Italy, in particular, the use of fire based signals during night and of
smoke based signals during the daylight on the top of towers or hills, afterwards called
communications by the usage of fani, has been quite common in the XVI and XVII Centuries
a.C. (Agnello, 1963). During the day one smoke signal means the presence of one enemy
vessel, in the night was switched on a bundle of dry woods and moved up and down to
inform about the exact number of the enemy vessels. Several testimonies report on different
communications links and distances. The most interesting one has been established in 1657
between the city of Messina and the Malta island with mid span vessels used to cover the
Mediterranean sea (Castelli, 1700).
The use of mechanical systems to implement optical wireless systems is due to Claude
Chappe in 1792 (Huurdeman, 2003). Chappe introduces the “optical telegraph” in France.
The system was based on a regulator, 4.5m long and 0.35m wide, to which two indicators
were attached. This systems was placed on the top of stations in LOS (Line Of Site) at 9 km
each. Telescopes and human repeaters were, of course, needed to move the regulator and
the indicators via three cranks and wire ropes. The time usage was short because the system
was able to work only during the daylight and with good weather conditions. On the other
hand, it was long reach considering an average coverage in France equal to around 4830km,
with 29 cities connected using around 540 towers. Security was ensured by transmitting
secret codes with short preambles, this also to understand the accuracy of the transmission.
Chappe introduced, infact, a particular code in 1795, to increase the transmission speed. This
system helped to reduced the time to exchange informations from several days to minutes
and has been adopted in 1794. The subsequent studies on the electricity, the results from
Volta (1745-1827) and from Ampere (1775-1836) on the electrical pile and the introduction of
the electrical telegraph in 1838 (Morse), will carry to the dismission of the Chappe system
around the mid of 1800. The Chappe system was introduced also in other European
countries connecting the cities of Amsterdam, Strasbourg, Turin, Milan and Brussels.
At the end of the 19th century, Alexander Graham Bell experienced with excellent results
the so called Photophone (Michaelis, 1965) (Bova and Rudnicki, 2001). This system worked
using the sound waves of the voice to move a mirror, responsible to send pulses of reflected
sunlight to the receiving instrument. In particular Bell modulated with his voice, by the use
of an acousto-optic transducer, a lens-collimated solar beam. Bell used to consider this
invention to be his best work, even better than “his demonstration of the telephone”.
Although Bell’s Photophone never became a commercial reality, it demonstrated the basic
principle of optical communications.
Wireless Optical Communications, becomes from this point and year by year more
important boosting the research worldwide. We can in this case divide the wireless optical
experiments in three main areas depending on the time periods: in the 60s arrives the laser
concepts and rises up the idea of wireless communications, in the 90s becomes popular the
idea of ground to satellite and satellite to ground laser communications still using red and
green sources, after 2000 the explosion of the Free Space Optical technologies (FSO) faces
civil and military applications ranging from standard telecommunications up to
intersatellites & interplanets experiments and using different wavelengths from 1 up to 10
microns.
For these reasons, essentially all of the engineering of today’s FSO communications systems,
has been studied over the past 40 years, at the beginning for defense applications and
afterwards for civil ones. By addressing the principal engineering challenges FSO, this
aerospace/defense activity established a strong foundation upon which today’s commercial
FSO systems.
In particular, the realization of the first LASER, based on ruby, in 1960 by Maiman opened
wider possibilities for the communications involving beams propagating over long
distances in atmosphere. Low loss optical fibers (less than 20 dB/km), infact, will arrive only
in the 70s. In 1960s NASA started to perform preliminary experiments between the Goddard
Space Center and the Gemini 7 module. In 1968 the first experiment about FSO transmission
of 12 phone channels along 4km had been demonstrated in Rome (Italy) by researchers from
the Istituto P.T, CNR and Fondazione Ugo Bordoni under the management of Prof. Sette,
Phisic Insitute University La Sapienza. A red laser source (0.8 microns) was used to connect
two buildings between the Colombo and Trastevere Streets in Rome (Unknown, 1968). In
the same year Dr. E. Kube in Germany published on the viability of free space optical
communications considering both green (0.6 microns) and red (0.8 microns) laser sources
(Kube, 1968). The introduction of semiconductor light sources working at room
temperature, by Alferov in 1970, were decisive for a further development of integrated and
low cost FSO systems. On the point of view of the research, the first experiment using a
quantum cascade laser (Capasso 1994) can be considered fundamental today speaking about
new transmission wavelengths for FSO (up to 10 microns). Between 1994 and 1996 years the
first demonstration of a bidirectional space to ground laser link between the ETS-VI satellite
and the Communications Research Laboratory (CRL) in Koganey (Tokio) has been
accomplished. 1Mbps using 0.5 microns and 0.8 microns emitting lasers. With the ongoing
intensive and worldwide studies on FSO communications, especially re started after the
September 11 tragedy where the communications were supported by free space optics links,
the related scenarios changed extremely fast covering today different applications and
environments like the followings: atmosphere, undersea, inter satellites, deep space. We can
infact report on the SILEX experiment (Semiconductor Intersatellite Link Experiment) in
2001 demonstrating bidirectional GEO-LEO and GEO-ground communications. ARTEMIS
satellite (GEO) using a semiconductor laser at 0.8 microns directly driven at 2 Mbps with an
average output of 10mW towards a Si-APD on SPOT-4 satellite (LEO). In the same year, the
GeoLite (Geosyncronous Lightweight Technology Experiment) experiment successfully
demonstrated a bidirectional laser communication between GEO satellites, ground and
aircraft. We cannot forget afterwards the MLCD (Mars Laser Communication
Demonstration) program started in 2003 and ended in 2005 with the aim of covering the
distance between Earth and Mars planets using an optical parametric amplifier with an
average output of 5W and photon counting detectors working at 1.06 microns (Majumdar
and Ricklin, 2008).
3. Basic principles of the optical wireless communications

(E. Leitgeb – B. Geiger – F. Nadeem - A.L.J. Teixeira, P. Andre)
3.1 Introduction
Free Space Optical communication links transmit information by laser light through an
atmospheric channel. Relying on infrared light, these communication systems are immune
to electromagnetic interference (EMI), jamming, or wiretapping. Furthermore, they do not
cause EMI themselves and operate at frequency bands (around 300 THz) were the spectrum
is unlicensed. As a further advantage, FSO and fiber equipment can be combined without
intermediate conversion, since both the air and the material used for fiber cables have good
transmittance at the established wavelengths, namely 850 nm and 1550 nm. Currently, all-
optical fiber/FSO systems are a well populated field of research, developing solutions for
signal regeneration, transmission, and reception without an intermediate electronic signal.
FSO systems can be installed faster and cheaper than their wireless radio counterparts,
making them interesting for short-term installations for events, military purposes and
disaster recovery. Consequently, a multitude of FSO equipment is commercially available
for interconnection with standard fiber or Ethernet components. Acting as an alternative to
other wireless radio or high-bandwidth wired links (fiber optics, Gigabit Ethernet), it has to
fulfill general requirements such as low bit error rate (BER < 10-9) and high reliability.
As already mentioned, a prerequisite for these requirements is an unobstructed line-of-sight,
especially in long-distance outdoor environments. A major drawback therefore is the
susceptibility of FSO links to certain weather conditions, where especially fog causes severe
attenuation of the laser beam and subsequent total link loss. Even moderate continental fog
can result in an attenuation of 130 dB/km, whereas dense maritime fog can account for
attenuations up to 480 dB/km (E. Leitgeb et alt, 2006; M.S. Awan et alt, 2008).
Rain attenuation has very little effects on the availability of FSO systems, but these effects
strongly depend on the rain rate R. According to (T. Carbonneau and D. Wisley, 1998) and
the references therein an adequate relationship between rainfall and attenuation would be
τ rain  exp( ( 0.05556  0.00848  R

(1)
 3.66  105  R 2 )  l)
For light to moderate rain rates of R = 5 mm/h as they are occurring in the continental
climate of middle Europe the attenuation is only approx. 3 dB/km. Peak attenuations due to
tropical rain falls of R = 100 mm/h would result in higher attenuations (approx. 30 dB/km),
but such weather conditions occur rarely and only in burst in Europe and the United States
(J. Wells, 2009; H.Alma and W. Al-Khateeb, 2008). Similar considerations hold for heavy
snowfall (more than 5 cm over 3 hrs), where attenuations of more than 45 dB/km have been
measured (R. Nebuloni and C. Capsoni, 2008). Depending on seasonal and geographic
peculiarities, these values can vary to some extent. It may also happen that certain weather
events occur simultaneously, i.e. heavy rain in combination with fog, or fog in combination
with snowfall (V. Kvicera, 2008).
Another phenomenon occurring influencing FSO communication links is related to
scintillations and air turbulences. Air cells with different temperatures randomly distributed
along the link cause focusing and defocusing of the link due to changes in the refractive
index. Amplitude and frequency of these scintillation depend on the size of cells compared
to the diameter of the optical beam (S. S. Muhammad, 2005; A. Chaman Motlagh, 2008). FSO
systems usually cope with such variations in the optical received signal strength (ORSS) by
using multiple beams (so-called multi-beam systems) or by using saturated amplifiers (M.
Abtahi and L. Rusch, 2006). More detailed investigations can be found in (S. S. Muhammad,
2005) and the references there.
Other problems affecting visibility are mostly related to the narrow beam FSO systems use
(usually at the order of a few milli radians): sand, dust, birds, et cetera flying through the
beam cause momentary link losses, whereas misalignment due to tower sway or thermal
effects can be coped by auto-tracking systems (J. Wells, 2009). The sun itself acts as a noise
source, which may completely overdrive the receivers (W. Kogler, 2003) if they are directly
exposed to sunlight. Soiling and aging of the components, especially lenses and mirrors,
finalize the list of effects on FSO link availability. Summarizing, most of these effects can be
overcome by either granting a certain link margin (snow and rain attenuation, fluctuations
due to scintillation) or by adding complexity to the system (multi-beam and auto-tracking
systems). Fog, on the other hand, is the only remaining condition harmful for availability,
making carrier class availability (99.999%) for FSO systems highly questionable.
Depending on the geographic areas, fog mainly occurs during fall and winter months on a
persistent basis, whereas outages during summer and spring are caused by thunderstorms (E.
Leitgeb, 2004). Fig. 1 shows the average unavailability throughout the year in Graz, Austria.
Moreover, diurnal changes affect the probability of fog as Fig. 2 shows; it is low during noon
where the sun clears up the sky and high during dusk, dawn and the night (E. Leitgeb, 2004).
Fig. 1. Average unavailability throughout the year (based on measurements from Oct. 2000
to Sep. 2001 (with the permission of E. Leitgeb, 2004))
Fig. 2. Probability of failure during the daytime (based on measurements from Dec. 2000
(with the permission of E. Leitgeb, 2004))
Due to the complexity connected with phase or frequency modulation, current free-space
optical communication systems typically use intensity modulation with direct detection
(IM/DD). Like in fibre optics communications systems, the performance characteristics for a
free space optical communications system are dependent on the propagation medium.
However, in this case the propagation medium is randomly changed, and susceptible to
atmospheric conditions, resulting in alterations to the beam propagation constants.
In order to get a clearer picture of the behavior of FSO systems a numerical model for the
atmosphere is going to be presented [Andre, 2003]. This propagation model was
incorporated in a commercial available photonic simulator and used to predict the
behaviour of a point to point free space data link as function of the climacteric variables.
3.2 Atmosphere Model

As, referred, atmospheric effects can degrade free space data links by two mechanisms: (i)
reduction in the detected optical power level due to atmospheric attenuation and (ii)
random optical power fluctuations in the received beam which result in beam deformation,
scintillation effects and beam wander (Kim, 1998). All these factors can become impairing to
the communications if their influence is significant. For that it is going to be presented each
of the contributing parts model and therefore a complete evaluation with effects will be
made for better understanding of the real effects in the system.
A. Atmospheric attenuation
The atmospheric attenuation results from the interaction of the laser beam with air
molecules and aerosols along the propagation. Similar to other waves, the optical beam
power has an exponential decay with propagation distance. At a given distance from the
emitter, l, the optical transmittance is:
P(l)
τ  τa  τs   exp( α  l) (1)
P( 0 )
where  is the overall attenuation coefficient, determined by four individual processes:

molecular absorption, molecular scattering, aerosol absorption and aerosol scattering.
The molecular absorption includes the absorption by water, CO2 and ozone molecules. The
aerosol absorption results from the finely dispersed solid and liquid particles in the
atmosphere, such as ice and dust, with a maximum radius of 20 m. A simple approach to
calculate absorption, assumes that variations in the transmission are caused by changes in
the water content of the air. The precipitable water, w (in millimetres), encountered by the
light beam is (Wichel, 1990):
(2)
w  103  ρ  l
where  is the absolute humidity in g/m3. This value can be related with the relative
humidity percentage (RH) and with the temperature in degrees Celsius, T, by:
  RH  (  0.74  90.96  expT/ 13.67 ...

 85.4  expT/ 13.52  ) (3)
The absorptive transmittance can be then calculated for any transmission window, by the
following empirical expressions (Wichel, 1990):
τ a  exp   Ai  w1 / 2  , w  wi (4)
βi
w 
τ a  ki   i  , w  wi (4a)
w
The values typical values of the constants used are listed in table 1 (these are also used in
calculations following).
Window Ai ki i Wi
boundaries
(nm)
720 – 940 0.0305 0.800 0.112 54
940 – 1130 0.0363 0.765 0.134 54
1130 – 1380 0.1303 0.830 0.093 2
1380 – 1900 0.211 0.802 0.111 1.1
1900 – 2700 0.350 0.814 0.1035 0.35
2700 – 4300 0.373 0.827 0.095 0.26
4300 - 6000 0.598 0.784 0.122 0.165
Table 1. Constants used in expressions (4) and (4a).
Another attenuation process is the scattering, where there is no power loss, and there is only
a directional distribution. The two dominate scattering mechanisms are the Rayleigh
scattering, when the wavelength of the light is higher than the particle size, and the Mie
scattering when the particle size is comparable with the wavelength of the radiation. An
empirical relationship sometimes used to describe the scattering transmittance is [Wichel,
1990]:

τ s  exp l   C1  λ  δ  C2  λ 4   (5)
where, C1 and  are constants determined by the aerosol concentration and size distribution
and C2 = 0.00258 m3 accounts the Rayleigh scattering. These two constants can be related
with the visual range, V, in kilometres at 550 nm [1]:
3.91
C1   0.55δ (6)
V
For a very good visibility,  can take a value of 1.6, and for average visibility it have a value
of 1.3. If the visual range is inferior to 6 km, them the exponent  can be obtained by:
δ  0.585  V 1 / 3 (7)
The presence of precipitation (rain or fog) increases the scattering coefficient. The
transmittance can be related with the rainfall rate (R) in mm/hr, by:
τ rain  exp( ( 0.05556  0.00848  R

(8)
 3.66  105  R 2 )  l)
The propagation of a laser beam in dense fog or clouds much difficult and attenuations as
high as  50 – 150 dB/km can be found (Strickland, 1999).
The total attenuation to be considered in (1) is the sum of the several partial attenuation
factor (eqs (4), (5) and (8))
The geometrical beam expansion, resulting from the beam divergence, is also responsible for
a reduction of the optical power coupled to the receiver. It must be also take into account the
optical miss alignment between the emitter and the receiver, for systems without auto-
tracking (Kim, 1998).
B. Turbulence
The atmospheric turbulence arises when air parcels at different temperatures are mixed by
wind and convection. This effect produces fluctuations in the density and therefore in the air
refractive index. The parameter that describe the disturbances caused by turbulence is the
refractive index structure coefficient, Cn, which usually varies between 5  10-7 m-1/3 and 8
 10-9 m-1/3 for situation of strong and weak turbulence, respectively.
The value of Cn can be estimated by (Strohbehn, 1978):
 79  106  P 
 
 h -4 / 3   T  273.15  
2
2
Cn =  -4 / 3 

 3000  5.49  1013 (9)
  v  sin(θ )   -3
2

  2.2  10-53  300010      e + 10-16  e-2 
  27   
 
where h is the height in meters, P the air pressure is milibars, v is the wind speed in m/s and
 the angle between the beam and the wind.
The dominant turbulence scale size leads to different effects: (i) if the scale of the turbulence
cells is larger than the beam diameter then the dominant effect is the beam wander, that is
the rapid displacement of the beam spot, (ii) if the scale of the turbulence cells is smaller
than the beam diameter then the dominant effect is the beam intensity fluctuation or
scintillation.
The radial variance of beam wander can be described by (Zhu, 2002):
σ r 2  1.90  Cn 2   2  w 
1 / 3
 l3 (10)
where w is the spot size at the transmitter.

The scintillation is described by a log-intensity distribution (Clifford, 1981), with a variance
given by [Wichel, 1990]:
7/ 6
 2π 
σ i 2  1.23  Cn 2     l11 / 6 (11)
 λ 
The effect of scintillation can be smoothed by spatial averaging using a width detector area,
multiple apertures detector or a spatial diversity with several receivers or emitters (Kim,
1997).
The presence of atmospheric turbulence is also responsible for the beam spreading,
contributing to the beam divergence, which is given by [Wichel, 1990]:
at  2.01  λ 1 / 5  Cn 6 / 5  l 8 / 6 (12)
C. Thermal Blooming
The molecular absorption by the air of the beam energy, will lead to a temperature gradient
in the medium that induces density and index refraction changes. In the presence of air flow
(wind) results in a density wavefront destruction which leads in a beam bender directed to
the air flow.
The displacement of the beam at the receiver is (Strohbehn, 1978):
2
 I 
5  (γ  1 )  (n  1 )  l   0 
u=  w  (13)
6  γ  P  100  v  sin(θ )
where  is the ratio of specific heats (with a value of 1.4 in air), n is the refractive index of the
air and I0 is the beam optical power at the transmitter.
III. Simulation
In this subsection, and for sake of understanding of the modeling described above, a set of
simulations is presented based on the atmospheric model described in the previous section.
This model was implemented through Matlab in a commercial available photonic simulator,
VPI from Virtual Photonics .
The free space optical communications system used as reference for these simulations had
the following parameters. The propagation distance was 1000 m, oriented in a 158º heading.
The optical power of the beam at the emitter was 40 mW and at a 780 nm wavelength, with
a radius of 10 cm and a divergence of 1 mrad. The optical beam is modulated at 2.048 Mb/s
(E1) with an optical extinction ratio of 15 dB. The use of this low bitrate allows us to later
compare these results with some experimental ones (Almeida, 2001). For the receiver we
have considered a photodiode based O-E converter with a 0.85 responsivity and a 1.4
Mbit/s bandwidth pulse reformatting electrical filter, preceded by a 6 dB attenuator to
account the miss alignment losses.
We obtained, for several values of temperature and relative humidity, the attenuation of for
1 km path link. These results can be observed in figure 3.
35
0.4375
30 0.5469
0.6563
0.8750
Temperature (ºC) 25 1.094
1.313
1.531
20 1.750
1.969
2.188
15 2.406
2.625
2.844
10 3.063
3.500
5
40 50 60 70 80 90 100
Relative Humidity (%)
Fig. 3. Optical attenuation for one km path link as a function of the temperature and relative
humidity.
It is clear, from figure 3 that the effect of temperature and humidity in the losses, due to
absorption and scattering is important. The total attenuation vary from  0.5 dB/km to 3.5
dB/km.
The effect of rain fall can also be analyzed with this model. Keeping the values of
temperature and relative humidity constant, 25 ºC and 80 % respectively, the attenuation
was obtained as function of the rain fall rate, as displayed in figure 4.
1.9
1.8
1.7
1.6
Attenuation (db/km)
1.5
1.4
1.3
1.2
1.1
1.0
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Rain Fall Rate (mm/h)
Fig. 4. Optical attenuation for 1 km path link due to rain. The line is a visual guide.
The introduction of turbulence in the atmosphere model will result in the observation of
scintillation on the received power. In figure 5 is shown the BER (bit error rate) of the
received data for 1 km path direct optical link as function of the received optical power and
for several values of the refractive index structure coefficient.
0.01
1E-3
1E-4
1E-5
1E-6
1E-7
BER
1E-8
1E-9
-8 -1/3
1E-10 Cn = 5 x 10 m
-8 -1/3
1E-11 Cn = 1 x 10 m
-9 -1/3
Cn = 5 x 10 m
1E-12
Back to back
1E-13
1E-7 1E-6 1E-5
Optical Power Received (W)
Fig. 5. BER versus the received optical power for several values of Cn. The lines are visual
guides.
From the previous figure it is clear that the power penalty depends on the value of Cn, as it
can be observed in the eye diagrams of figure 6, obtained for a received power of –22 dBm
and for several values of Cn. The eye diagram of figure 6 a) corresponds to a high turbulent
medium with a Cn value of 1  10-7 m-1/3, 6 b) is a situation of medium turbulence with Cn
of 5  10-8 m-1/3, while 6 c) is obtained in a low turbulence medium with a value of 1  10-9
m-1/3 for Cn.
a) b) c)
Fig. 6. Eye diagrams obtained for several values of the refractive index structure coefficient:
a) 1  10-7 m-1/3, b) 5  10-8 m-1/3, c) 1  10-9 m-1/3.
A reasonable good approach to estimate the BER of a FSO system is to consider only the
attenuation (discarding the scattering and thermal blooming but considering the beam
wander), then the BER can be written as:
(14)
Where R is the detector responsivity, PR the optical power at the detector and  the receiver
thermal noise. The impact of the Cn factor in the BER can be observed in the figure 7, where
experimental Cn factor measured in Rio de Janeiro along the day in February 2003. For the
receiver it was considered a typical configuration with R=0.9 A/W and a receiver diameter
of 13 cm, the optical power at the emitter is 10 mW for a link length of 1 km.
Fig. 7. Cn influence in the BER for 1 km link.
4. Hybrid network infrastructures: FSO & RF

(B. Geiger – E. Leitgeb – F. Nadeem)
4.1 Introduction
Relying on an unobstructed line-of-sight, FSO links are strongly influenced by atmospheric
conditions reducing or influencing visibility, such as fog, precipitation, haze, and
scintillation. Fog, as one can expect, is the most critical effect affecting attenuation and,
subsequently, availability of the FSO link. Attenuation is caused by scattering, resulting
from the fact that the size of the fog particles is in the order of the wavelength of optical and
near-infrared waves (as they are used for FSO). Consequently, link distances in coastal or
metropolitan environments which are prone to fog are limited to a few hundred meters.
Radio Frequency (RF) links on the other hand show almost negligible fog attenuation if the
carrier frequency is chosen accordingly, while they usually suffer from other precipitation
types like rain and wet snow. Combining these two technologies to an FSO/RF hybrid
network may increase overall availability significantly, guaranteeing quality-of-service and
broadband connectivity regardless of atmospheric conditions.
Several types of hybrid system concepts have been introduced in the literature (L. Stotts et
alt, 2009; S. Bloom and W. S. Hartley; H. Izadpanah et alt, 2003; T. Elbatt et alt, 2001;
F. Nadeem et alt, 2009; E. Leitgeb et alt, 2004; J. Pacheco de Carvalho et alt, 2008; S. Vangala
et alt, 2007; A. Akbulut et alt, 2005; S. Gurumani et alt, 2008; J. Derenick et alt, 2005; T.
Kamalakis et alt, 2005; W. Kogler et alt, 2003; F. Nadeem et alt, 2009; S.D. Milner et alt, 2004;
O.I. Kim and E. Korevaar, 2001; H. Wu et alt, 2004; S. Vangala and H. Pishro-Nik, 2007),
focusing on increasing availability, bandwidth efficiency, or minimizing system complexity.
Resulting from these focuses, hybrid systems can be categorized in three different groups:
Redundant, load-balancing, and switch-over hybrids.
4.2 Description of the RF Communication Links

As already mentioned, FSO links are strongly influenced by fog attenuation and
consequently suffer from long periods of total link loss (E. Leitgeb, 2005). On the other hand,
FSO systems provide very high data rates without the requirement of licensing. As a
consequence, the RF link for a hybrid FSO/RF system has to be chosen according to the
following requirements:
 The RF link should be available whenever the FSO link is not, i.e. it should not be
influenced by fog or other weather effects reducing visibility.
 The RF link should provide a similar bandwidth as the FSO link, so that the hybrid
system does not suffer from performance degradation.
 The RF link should be operated in a frequency band which does not require
licensing, so that this advantage of FSO systems is not lost in a hybrid setup.
Unfortunately, some of these requirements are contradictory. High data rates, or

equivalently, bandwidths require high carrier frequencies, which on the other hand either
show strong attenuation due to fog or would result in a prohibitively high system
complexity. Moreover, these systems operate in license-free, but regulated bands and are
thus subject to stringent transmission power restrictions limiting the possible link margin
considerably. A higher geometrical loss further adds to the availability issues. Especially
during times when rain and fog occur simultaneously, as it often happens in continental
climate, both links are suffering from weather effects (E. Leitgeb et alt, 2004; E. Leitgeb et alt,
2005). Scintillations have little or no effect on RF links, since they are not susceptible to
changes in the refractive index rather than changes in humidity (S.S. Muhammad, 2005). A
more complete discussion of weather effects on RF links is available in (H. Wu et alt, 2004;
ITU-R, 2005). Commercially available bands with very high available bandwidths are
centered around 60 GHz and 70/80 GHz, respectively. While the former, license-free band
cannot be exploited for long link distances due to an oxygen absorption peak (15 dB/km),
the latter provides an interesting field of research for hybrid systems. Peak attenuation due
to moderate rain are usually well below 5 dB/km, whereas fog attenuation is as small as 0.4
dB/km for a fog density of 0.1 g/m3 – under these conditions a FSO system would suffer
from 225 dB/km (S. Bloom, 2005). Currently available equipment operating in the 70/80
GHz band can provide carrier class availability disregarding weather conditions over a
distance of 2-3 km achieving 1 Gbps. Unfortunately, this band can only be used after
obtaining a low-cost license. Spectra currently unregulated by the ITU lying at 275 GHz are
unreachable with current technologies (J. Wells, 2009).
Lower frequency bands (e.g. the license-free ISM bands at 2.4 and 5 GHz), on the other
hand, provide much less bandwidth to the user, leading to a greatly decreased bandwidth
performance of the overall hybrid system. However, these systems are not susceptible to fog
at all (ITU-R, 2005) and also show much smaller influence of rain and snow than systems
operating with carrier frequencies beyond 20 GHz (H. Wu et alt, 2004). Multi-user
interference, as it is common in license-free bands at lower frequencies, which are accessible
by low-cost technologies, can be mitigated by using directional links.
These bands, which also host users of IEEE 802.11a/b/g/n compliant equipment, are of
particular interest since they allow building a hybrid FSO/RF system with commercially
available equipment for the RF part as well. Especially the IEEE 802.11a standard is
interesting for FSO/RF hybrids, since it is operating in the less populated 5 GHz band and
allows a higher transmit power of up to 30 dBm EIRP. Although the standard claims
possible data rates of up to 54 Mbps, usual limits for long-range directional links are around
20 Mbps. Moreover, the use of this technology offers the possibility of an upgrade to IEEE
802.11n compliant equipment, claiming data rates of up to 600 Mbps with MIMO
functionality. Future studies will show if this technology can also be used for long-rage
directional links and thus build the RF component of the proposed hybrid systems.
4.3 Overview of Hybrid Systems introduced in the literature

Hybrid systems can be classified into three categories:
 Redundant systems: These systems duplicate data and transmit it simultaneously
over both the RF and the FSO link. As a consequence, the data rate of both links has
to be equal, resulting in either a requirement for very high frequencies on the RF
link or a relatively high FSO underutilization. Moreover, systems which duplicate
and recombine data are necessary. Redundant systems provide a high reliability,
but suffer from the fact that both links have to be active all the time, wasting a
significant amount of energy.
 Switch-Over systems: These systems transmit data only over one link, which is
chosen according to link availability. Usually, since the FSO link allows higher data
rates, it is chosen as a primary link whereas the RF link acts a backup.
Consequently, data rates of both links need not be identical, if one accepts a
reduced bandwidth during fog events. Switch-over systems require multiplexers
on both ends, algorithms choosing the active link, synchronization, and accurate,
timely measurement data of the optical signal strength. However, these algorithms
save energy by transmitting over one link only, and can be connected to standard
network equipment without protocol overhead (F. Nadeem et alt, 2009).
 Load-Balancing systems: These most sophisticated algorithms distribute traffic
among the links according to the quality of their connectivity, thus exploiting the full
available bandwidth each time. Besides a measurement of the link quality, these
systems require recombination systems on either sides of the hybrid link, often
resulting in either significant protocol overhead (F. Nadeem et alt, 2009) or high-
complexity codes which automatically distribute data among different links.
In the literature a wide field of hybrid systems can be found: AirFiber (S. Bloom, 2009), a US-
based company pioneered redundant transmission over FSO and RF links, the latter one
being a millimeter wave (MMW) link with a carrier frequency at 60 GHz. Data rates of
approx. 100 Mbps were achieved, but availability was well below the expectations. Wu et alt.
(H. Wu et alt, 2004) analyzed FSO and RF link separately and concluded that by using a
hybrid network link margin can be reduced significantly to achieve carrier class availability.
The FSO link was using a 1550 nm laser source, the RF link was operated in the 60 GHz
MMW band. Link distance was 1500 m.
Fig. 8. Redundant hybrid system and availability measurements (with the permission of
E. Leitgeb et alt, 2004)
Kim and Korevar (I.I. Kim and E. Korevaar, 2001) studied the distance limitation of FSO
systems for both carrier and enterprise applications and showed that carrier class
availability can be achieved for much longer link distances if the FSO link is combined with
an RF back-up transmitting data redundantly. Leitgeb et alt. performed an experiment in
Graz, Austria, over 15 months during 2002-2003, where data was sent simultaneously over
two FSO/RF links (E. Leitgeb et alt 2004). The RF link was designed with a carrier frequency
of 40 GHz, the FSO system transmitted using a 850 nm laser with a data rate of 155 Mbps.
The availability was measured over this time for each individual link and for the hybrid
combination. The redundant transmission achieved an availability of 99.93% (see Fig. 8.)
Hashmi et alt. (S. Hashmi and H. Mouftah, 2004) also proposed a redundant hybrid system,
calculating it based on rain data only for an FSO and a MMW link in the 60 GHz band. They
also mentioned that the hybrid system could be used in an asymmetric uplink/downlink
scenario where different traffic demands have to be served.
Akbulut et al (A. Akbulut et alt 2005) developed an experimental hybrid FSO/RF switch-
over system between the two of five campuses of Ankara University, Turkey, located at
different locations in the city. The optical link provided a 155 Mbps full duplex connection
by using a laser source at 1550 nm over a distance of 2.9 km. The RF link was compliant to
IEEE 802.11b WLAN, operated at 2.4 GHz linking the two terminals at 11 Mbps. The switch-
over algorithm was a power hysteresis. Pacheco de Carvalho et alt. (J. Pacheco de Carvalho,
2008) installed a similar system at the University of Aveiro, Portugal, where a 1 Gbps laser
link was backed up by a 75 Mbps (nominal) WiMAX (IEEE 802.16) link. The laser link was
operated at 1550 nm, switching was implemented on the network layer via switching
between static routes. Power hysteresis was employed, and the link distance was 1.14 km.
Milner and Davis (S.D.Milner and C.C. Davis, 2004) proposed a switch-over system for
tactical operations in a general manner, considering protocols for switching between links as
well as for traffic re-distribution after a change in the network topology. Their intention was
to use two 1550 nm FSO systems in combination with an RF link operated in the Ku-band
(12-18 GHz). Kamalakis et alt. (T. Kamalakis et alt, 2005) installed a hybrid switch-over
system at the University of Athens, Greece, using a 1 Gbps FSO link and a 100 Mbps MMW
link operated at 95 GHz over a distance of 800 m. Also, (L. Stotts et alt, 2009) report about
switch-over systems.
Dynamic load balancing is also proposed in literature: ElBatt and Izadpanah ( H. Izadpanah
et alt, 2003; T. ElBatt and H. Izadpanah, 2001) proposed a load balancing system distributing
traffic among one FSO and several RF links. In this work, however, it is assumed that the
amount of traffic on the FSO link affects it availability. Vangala and Pishro-Nik (S. Vangala
and H. Pishro-Nik, 2007; S. Vangala and H. Pishro-Nik, 2007) use special non-uniform low-
density parity check codes to distribute traffic among different links, showing increased link
utilization and availability, while BER could be reduced significantly. Finally, Nadeem et alt
(F. Nadeem et alt, 2009; F. Nadeem et alt, 2009) analyzed both switch-over and load-
balancing systems based on standard Ethernet equipment with minimum hardware
extension. A 155 Mbps FSO system with a 850 nm laser was used in combination with an
IEEE 802.11a link. It was shown that availability almost achieves carrier class values of
99.999%.
4.4 Hybrid FSO/RF switch-over system

Switch-over (SO) systems, as they were introduced in Section IV can be illustrated by Fig. 9:
Depending on the strength and availability of the links, only one of them is used for
transmission. While the FSO link is the primary link, the RF link acts as a back-up. In this
section the interested reader will find an overview of the problems in designing such an SO
system together with some possible solutions. In particular, synchronization between the
switches/multiplexers on both sides, SO algorithms and possible applications will be
analyzed.
Fig. 9. Application setup of FSO-WLAN switch-over system (with the permission of

(F. Nadeem et alt, 2009)
For the simulations, a commercial WLAN link with a carrier frequency in the 5 GHz ISM
band will be considered. The WLAN link was built with two embedded PCs using high-gain
grid antennas and miniPCI WLAN cards with fairly high receiver sensitivity (depending on
the antennas, distances of over 50 km can be covered). The FSO system is a GoC MultiLink
155/2 system. It supports data rates of 155 Mbps over distances up to 2 km and uses 4
transmitters at 850 nm. The properties of the FSO and the WLAN system are given in
Table 2. It is further assumed that the transceivers of both links are able to provide status
information.
Table 2. Properties of FSO and WLAN Systems (with the permission of F. Nadeem et alt,
2009)
Since FSO is the primary link switching is exclusively determined by the ORSS: As soon as
the ORSS indicates a total link loss for FSO, an SO operation to the WLAN link is performed.
Similarly, as soon as the optical system restores connectivity, data packets are transmitted
over this link only.
Switching and Synchronization: Switching itself can be done on almost all layers of the OSI
reference model, although it is done preferably on the lower layers 1-3. Hardware
multiplexers, virtual LAN switches or commercially available Ethernet routers are only a
few possible switches to name. Switching on the physical layer (PHY) has the advantage of
fast reconfigurations of the links, at least if the connected wireless systems are capable of
such quick changes. If both links are providing Ethernet compliant interfaces only, fast
physical layer switching may be problematic; connecting a CAT5 cable to an Ethernet device
always requires a so-called auto-negotiation phase where the devices determine link speed,
duplex mode and other transmission parameters. This negotiation takes some time, usually
in the order of 1-2 seconds. Consequently, during this time data cannot be transmitted,
resulting in a relatively long link loss time (LLT). As it was shown in (F. Nadeem et alt, 2009)
LLT is around 1.62 s for a self-made hardware multiplexer.
Moreover, it may occur that packets are cut in halves and lost during the switch-over process.
On the other hand, switching on the physical layer has the advantage of a completely
transparent link between the two networks to be connected via the hybrid system. Switching
on the medium access (MAC) layer also provides a transparent link, if the wireless connections
allow it. FSO equipment usually operates on the PHY layer and thus is protocol independent.
WLAN equipment can at least be configured in a way that it is transparent to the MAC layer
(bridging or WDS bridging modes), so that source and destination MAC addresses are
unchanged over the link. Switching on MAC layer, e.g. by reconfiguring a virtual LAN
(VLAN), has the advantage that no physical re-connections and subsequent auto-negotiation
phases are necessary, significantly reducing LLT. (F. Nadeem et alt, 2009) reports LLTs of 1.29
s for a commercially available VLAN device. In some cases, however, the VLAN device
requires a soft re-boot after configuration, slightly prolonging the LLT (F. Nadeem et alt, 2009).
However, more sophisticated devices use a store and forward algorithm which at least
guarantees that every packet arriving at the multiplexer is forwarded – if not over the active
link. Switching on network layer can be done via routers and can rely on both physical
measurements of the ORSS or interface statistics of the router ports, such as BER and packet
loss (J. Pacheco de Carvalho et alt, 2008). In the former case the router needs special hardware
extensions or interfaces to some measurement equipment, in the latter the router has to be
capable of running a custom-made program. Furthermore, switching based on physical signal
strengths has the advantage that it does not rely on actual errors, but allows for
implementation of a more or less generous link margin. Switching itself is usually done by
changing the route metrics (J. Pacheco de Carvalho et alt, 2008) of the different links according
to the measurements. Consequently, all packets are transmitted, although not all may reach its
destination. Moreover, one can expect that depending on the type of link status information
available reactions on changes in the ORSS are delayed. Links switched on the network layer
are usually not transparent to lower layers. In order to provide full connectivity,
synchronization between the different multiplexers has to be ensured. Assuming that the
channel is reciprocal one may state that both sides of the system will always measure the same
ORSS and therefore chose the same link as active automatically. However, usually one does
not want to base a hybrid system designed to achieve maximum reliability on that assumption.
Moreover, using a side channel for transmitting information about which link to take is
questionable as well, because that very channel has to be made reliable itself. Of course, one
can feed such information into both the FSO and the RF link, but that certainly adds to the
complexity of the system. Besides, it was the main intention of the SO system that at each time
instant only one link has to be active. Finally, an asymmetric scenario, where one of the
multiplexers chooses the link according to the ORSS would be possible. The other multiplexer
accepts packets from either link, but responds only over the very link from where the last
packet arrived. Such a self-synchronizing setup, as it was introduced in (F. Nadeem, 2009) has
the disadvantage that it inherently relies on quasi-continuous transmission from the network
on the side of the active multiplexer – a condition which is usually fulfilled by higher-layer
protocols, such as TCP. In any case, despite all considerations about synchronization and
switching, the overall hybrid network still has to be considered unreliable; the difference,
however, is that the availability is increased significantly.
Fig. 10. Comparison of discrete and continuous ORSS values. (with the permission of
B. Flecker, 2006) Fog event from October 25th, 2005, 03:00 to 11:00.
Switch-Over Algorithms: After deciding upon the layer on which switching is performed,
upon the data which is determining the active link, and the synchronization method, the
designer of the hybrid SO system has to take a multitude of switch-over algorithms into
consideration.
Remembering the advantages of physical measurements describing the link status we will
put special emphasis on SO systems switching based on those measurements. This,
however, rises an important question: Can it be guaranteed that the FSO system (which
purely determines the active link) provides link status information in an accurate, timely
manner? It can be taken as granted that the system itself measures and uses ORSS
information, but is this information accessible to the user over a certain interface? The
MultiLink 155/2 system for example indicates the ORSS continuously with a LED bar
(discrete amplitude as shown in Fig. 10), but provides this data over an RS232 interface on a
per-second basis only (discrete time). Somewhere within the system, however, the time and
amplitude continuous ORSS will be available (as also shown in Fig. 10). Thereby one can
assume that the previous assumption of link status information provided to the user can be
justified, even if hardware reconfiguration are necessary. As one can see in Fig. 10,
especially during the gradient from clear sky conditions to foggy weather there are many
variations in the ORSS. These variations, as one can expect, cause a certain threshold to be
crossed multiple times. If now the SO system is designed to employ a straightforward
threshold comparison (TC) algorithm, it would suffer from frequent switching between the
links due to these variations. Since after each switching operation a certain time is required
to restore the link completely (so called link loss time, or LLT), frequent switching would
cause reduced bandwidth and availability. Consequently, other algorithms coping with
these variations have to be evaluated – it is the purpose of this section to introduce some of
them and compare their performance.
a) Power Hysteresis (PH): A power hysteresis defines two thresholds and two states: a lower
and an upper threshold, WLAN and FSO operation. If during FSO operation the lower
threshold is crossed, WLAN is activated. If during WLAN operation the upper threshold is
crossed, FSO is activated – it’s as simple as that to prevent the system from switching back
and forth. The width of the hysteresis (i.e. the distance between upper and lower threshold)
depends on the amplitude of variations and has to be optimized with respect to actual
system measurements. To maximize availability, the lower threshold has to be set to values
equal to or greater than the receiver sensitivity of the FSO device. b) Time Hysteresis (TH):
Relating variations in the ORSS to bouncing of electrical contacts, one can also use
techniques called debouncing to cope with these variations. Such techniques usually employ
a wait period T during which the ORSS are evaluated and during which after every
threshold crossing the wait period is restarted. Consequently, only if the signal does not
cross the threshold for a certain time, an SO operation is performed. The duration of the
wait period in that case is determined by the frequency and the amplitude of the variations.
To maximize availability, the wait period can be set to different values for crossing the
threshold in either directions; in the limiting case, the wait period can even be omitted for
switching from FSO to WLAN. c) Filtering: Treating variations in the ORSS as noise,
methods for noise mitigation come into view. Most prominently, low-pass filtering can be
named as such a method. Different realizations of low-pass filters in the analog (RC-
network) and digital (moving average filter, raised-cosine filter, etc.) domain are
possibilities to cope with this unwanted noise. Filters are characterized by their order, pass
and stop band characteristics, and by their cut-off frequency. These characteristics have to be
designed with respect to the frequency of the variations. It is of vital importance that the
frequency of these variations is in the stop band, while the highest occurring frequency in
climatic changes still lies in the pass band to allow for a timely reaction on an emerging fog
event.
Moreover, combinations of these methods can be considered (e.g. filtering and hysteresis
methods, power hysteresis and debouncing, etc.). Unfortunately, seasonal and diurnal, as
well as geographic peculiarities make a general design or general optimization of
parameters impossible. The reader will understand that the design of a SO system in a
coastal area with dense fog conditions differs from a system in a metropolitan area, where
moderate fog can be expected. The following simulations therefore focus on the continental,
metropolitan climate of Graz, which is characterized by moderate, persistent fog events
during fall and winter and strong rainfall during summer.
Simulations and Results: For the simulation, measurement data was taken from (B. Flecker
et alt, 2006). For parameter optimization the particular fog event depicted in Fig. 10 was
used. The benefit of focusing on one fog event is based on the fact that the influence of LLT
after switching is increased compared to the influence on an all-year average availability.
Moreover, as it can be seen in Fig. 1, fog events mainly occur during fall and winter, making
a separate analysis of these seasons sensible. Assuming little or no unavailability during the
summer months, simulation data can be extrapolated. Receiver sensitivity of the FSO system
was set to -22 dBm. For this value, a significant number of threshold crossings occurred
which allows an optimization of the algorithms. Using this sensitivity, the fog event under
consideration yielded an FSO availability of only 67.43%. Link loss time after switching was
set to 3 s in order to include an additional margin to link re-establishment. For bandwidth
simulations, bandwidths of the FSO and WLAN link were set in accordance to (F. Nadeem
et alt, 2004) to 91.9 Mbps and 18.8 Mbps, respectively. The WLAN link was assumed to be
active whenever FSO was inactive. This assumption holds for fall and winter periods where
FSO outages are usually caused by fog only and where rain is rarely occurring
simultaneously. During summer months where strong rainfall in combination with severe
fog affects both links this assumption may not be valid anymore (E. Leitgeb et alt, 2004).
As one can imagine, finding the best algorithm parameters is related to finding a trade-off
between availability and bandwidth efficiency. While WLAN may be available throughout
the year, its bandwidth is prohibitively low. Consequently, the simulations are limited by a
minimum bandwidth of 60 Mbps. A more complete evaluation of simulation results and a
comprehensive discussion of this topic can be found in (F. Nadeem et alt, 2009).
a) Threshold Comparison: Pure threshold comparison (TC) is done by comparing the ORSS
to the RX sensitivity and switching based on the outcome of this comparison. TC yields an
increase in availability to 98.62% while achieving best bandwidth performance (see Table 3).
This can be explained by the fact that TC uses the FSO link whenever it is available, and the
outstanding bandwidth of this link compensates for relatively high unavailabilities due to
LLTs. However, for maximizing availability this may not be the best of all choices as Fig. 11
shows. b) Power Hysteresis: For all simulations, the width of the power hysteresis was set to
1 dB and the lower threshold was varied. Fig. 12 shows that availability can be increased
significantly by increasing the lower threshold to values much greater than the receiver
sensitivity. The only problem is that by increasing this threshold FSO underutilization
increases and, subsequently, bandwidth efficiency is low. E.g. to obtain a minimum
bandwidth of 60 Mbps, the lower threshold has to be below -21.7 dBm. In these regions, also
the beneficial effects of filtering cannot be exploited anymore, because the time to react on
changes in the ORSS due to fog increases.
Fig. 11. Bandwidth for different switch-over methods (with the permission of F. Nadeem et
alt, IET submitted 2009)
0.999
0.998
0.997
0.996
Availability
0.995
0.994
0.993 Pure PH
MA−PH (10th order)
0.992 MA−PH (20th order)
MA−PH (40th order)
0.991 MA−PH (60th order)
0.99
0 0.5 1 1.5 2
Lower Threshold above RX sensitivity in dBm
Fig. 12. Availability for pure PH and MA-PH (with the permission of F. Nadeem et alt, 2009)
0.995
0.99
Availability
0.985
Pure TH
MA−TH (10th order)
0.98
0.975
0 10 20 30 40 50 60
Wait period T in s
Fig. 13. Availability for pure TH and MA-TH (with the permission of F. Nadeem et alt, 2009)
0.998
0.996
0.994
0.992
Availability
0.99
0.988
0.986
MA (10th order)
0.984 MA (20th order))
MA (40th order)
0.982 MA (60th order)
0.98
0 0.5 1 1.5 2
Threshold above RX sensitivity in dBm
Fig. 14. Availability for different filter orders (with the permission of F. Nadeem et alt, 2009)
c) Time Hysteresis: For the time hysteresis the threshold was set to the receiver sensitivity,
and the wait period T was varied as a simulation parameter. As seen in Fig. 13, availability
can be increased significantly. Moreover, FSO underutilization is low, so the minimum
bandwidth of 60 Mbps is achieved for all depicted values of T. Filtering beforehand is
counter-productive. d) Filtering: For filtering, the threshold was increased in steps starting
from the receiver sensitivity. As the only filter type, a moving average (MA) filter was
considered, where the order N of the filter automatically determines its cut-off frequency
(other filter types are discussed in (F. Nadeem et alt, 2009). Fig. 14 shows that availability
increases with increasing thresholds. Interesting, though, might be the fact that higher
orders (i.e. lower cut-off frequencies) perform better than lower ones, as long as the
threshold is set to values high enough. High-order filters perform smoothing, but do not
allow timely reactions on critical changes in the ORSS. Consequently, high availability is
only achievable with a combination of smoothing and a large margin to the receiver
sensitivity.
This in turn leads to FSO underutilization and limits maximum threshold values to -21 dBm
to obtain a minimum bandwidth of 60 Mbps. In these regions, however, lower filter orders
outperform higher orders.
e) Combined Power and Time Hysteresis: For the combined power and time hysteresis the
lower threshold was set to the receiver sensitivity and the width of the hysteresis was 1 dB.
The wait period of the time hysteresis portion was varied. As it can be seen in Fig. 15 and
Table 3, pure PT delivers best results in term of availability. Furthermore, minimum
bandwidth of 60 Mbps can be achieved for wait periods below 40 s, where availability still
has values above 99.8%. Extending this simulation to the whole measurement campaign,
availabilities of 99.988% can be achieved, as it is shown in Table 3. Simulations proved that
by doubling this period to T = 80 s, availability could be increased to 99.997%. Taking these
values into consideration one can see that carrier class availability becomes a graspable goal,
even for hybrid switch-over systems.
0.998
0.996
Availability
0.994
0.992
Pure PT
MA−PT (10th order)
0.99 MA−PT (20th order)
0.988
0 10 20 30 40 50 60
Wait period in s
Fig. 15. Availability for pure PT and MA-PT (with the permission of F. Nadeem et alt, 2009)
Table 3. Performance comparison of different switch over methods (with the permission of
F. Nadeem et alt IET submitted 2009)
Fig. 16. Map of the campus of the Technical University of Graz
An Application: Interconnection of different sites of the campus Finally, to conclude

about hybrid switch-over systems, a possible application scenario shall be introduced,
where different sites of the campus of Graz University of Technology will be interconnected.
Such an application is widely evaluated in the literature (J. Pacheco de Carvalho et alt, 2008;
A. Akbulut et alt, 2005), but only (F. Nadeem et alt, 2009) considers not only availability of
the different links but also traffic demands of the different sites. In Fig. 16 one can see the
location of the main sites of the campus, which are known as “Alte Technik”, “Neue
Technik” and “Inffeldgründe”. The latter one is the largest, housing many offices and
student computer rooms. Moreover, one can see that the distance between the sites never
exceeds 2 km, making the use of directional WLAN links and FSO links possible. For
evaluation purposes, characteristics described in Table 2 were considered. Moreover, it was
assumed that the line-of-sight for the FSO and for the WLAN link is free. As already
mentioned availability of the FSO link is generally high during the summer months, and
during the winter months only during daytime (cf. Fig. 1 and Fig. 2) for the continental
climate of Graz. Traffic demands were recorded using the Multi-Router-Traffic-Grapher
(MRTG), where green bars indicate incoming and blue lines indicate outgoing traffic.
Fig. 17. Traffic data recorded for “Inffeldgasse” (with the permission of F. Nadeem et alt, 2009)
Traffic recordings were made during December 2008, assuming that the average availability
due to fog is similar as in Fig. 1. Fig. 17 shows the traffic demands for campus “Inffeldgasse”
on December 3rd, 2008. It can be seen that peak traffic demands are occurring between 10
am and 4 pm, medium traffic was caused from 8 am to 10 am and from 4 pm to 6 pm,
whereas traffic during the night time is low. Obviously, the major traffic requirements
coincide with office and lecture hours. These considerations do not only hold for a particular
day, but throughout the year – naturally, on holidays and weekends, traffic demands are
much lower. Moreover, one can see that this peak of incoming traffic at 8 pm occurs every
day, which is most likely related to an automatic backup. Scheduling such events more
properly, traffic demands can be distributed accordingly. Simulations were performed using
a set of measurements of the years 2000 and 2001 (J. Tanczos, 2002). Link bandwidth was set
to 155 Mbps for FSO and to 15 Mbps for WLAN, respectively. For the WLAN link a slightly
lower bandwidth was taken, assuming that the Fresnel zones may be partially blocked by
surrounding buildings and trees. Link loss time was neglected, since it affects the average
bandwidth only very little. A more complete discussion of these things can be found in
(F. Nadeem et alt, 2009). Comparing Fig. 18 with Fig. 17, the diurnal changes in the traffic
requirements are reflected in the average as well. Furthermore, one can see from Fig. 18 that
the proposed hybrid FSO/WLAN switch-over system can satisfy traffic demands on
average. Unfortunately, peak traffic demands of campus “Inffeldgasse” exceed even the
available bandwidth for FSO, let alone WLAN. However, by using multiple links of each
technology, or newer, more sophisticated equipment (such as Gigabit FSO equipment and
IEEE 802.11n standard compliant WLAN links), traffic demands could be satisfied in a
highly reliable manner over wireless links.
Fig. 18. Average required and achieved bandwidth (with the permission of F. Nadeem et alt,
2009)
5. Blue sky applications: inter satellite, inter planetary, under sea, chip to
chip FSO communications (G. Incerti – G.M. Tosi Beleffi)
The today increase of the networks complexity, with several devices and subsystems
intensively used, involves an aggressive use of the bandwidth management thus to
guarantee an high rate and security, especially in military scenarios. In fact, the military
applications requires more strictly features respect to the civilian applications. The
bandwidth offered by the optical cables is very high and for this reason the optical fibres are
also used in military area. Inside airplanes, UAVs, vessels, cars and so on. Several
informations and data can run through the same optical fibre and an high rate can be
transmitted and managed. We started with this introduction on military purposes because
this is the first market that boost the optical wireless, from the paper to the real
implementation. FSO communication, infact, is a valid solution especially in military
situations because of the previously mentioned ability to guarantee a confidential
transmission with a huge bandwidth.
Challenge Mitigation Description

approach
–RF system facilitates coarse acquisition and tracking
Pointing, RF/FSO hybrids
–An RF channel can serve as control channel for FSO data
acquisition, and link
Adaptive optics
tracking –Adaptive optics systems achieve very fine beam steering
and tracking
–Path redundancy and topology control are implemented
Weather/ Path redundancy
in an FSO network to counteract link obscuration
environment RF/FSO hybrids –Environmental obscuration for optical may be
permissible for RF or vice versa
–Adaptive optics correct beam distortion
Turbulence Adaptive optics
–Channel coding/diversity improve BER through forward
Channel error correction
coding/Diversity
–Infrared wavelengths such as 1550 nm are more eye safe
Eye safety Infrared wavelengths
than visible wavelengths
Eye safety Adaptive optics –Adaptive optics reduce the need for increased power by
correcting beam for improved SNR
–RF system provides channel for topology control, link
Networking RF/FSO hybrids
monitoring, and broadcasting network status
QoS techniques –Differentiated services protocols sort data by priority to
counter capacity changes
–Application layer QoS algorithms prioritize data
Table 4. FSO mitigation approach (Juan C. Juarez et al., 2006)
The limited scenario offered by the radio frequency (RF) spectrum available for military use,
contributed to the exploration of alternative systems able to convey the secret informations
generated by military devices and/or systems. RF based systems reach only hundreds of
Mbps per link and the RF beam cover an high area, in terms of spatial aperture, thus
increasing the eavesdropping percentage. On the contrary, FSO systems can guarantee
robust optical link with a very small beam size. Granting, at the same time, a huge
bandwidth in the order of Gbps. Confining the data flow in a small spatial portion represent
an advantage because becomes very difficult to detect the beam and subsequently drop
some information from one or more miscreants. Furthermore, several beams, close to each
other, can be used at the same time to transport the information without any kind of
interference and or interaction.
The precision in the pointing and tracking steps is still a challenge especially in complex and
variable scenarios like, for example, the sea one. Mounting FSO on vessels, infact, means,
first of all, that a fast tracking system should be implemented. Maintaining, of course, a
minimum power budget and the numeric aperture already set. In particular, for the military
applications, is required an high degree of accuracy to obtain an alignment of laser beam
with the receiver (Juan C. Juarez et al., 2006).
In order to win the challenges induced by the adoption of FSO systems in the
communication scenario, sometime is possible to discover the presence of RF backup lines,
as supporting elements of the optical counterpart. This is commonly referred as an hybrid
communication system. The hybrid system shares a common aperture and the use of FSO
with RF system permit to facilitate principal function like for example control signalling,
tracking, acquisition and signals discovery. The RF beam is used to search the other device
(neighbour discovery) or to start the acquisition step. The RF beacon is used also as control
signalling and to rehabilitate the communications in case of optical channels fades.
In order to minimize the dependence from a single link, reconfigurable FSO links can be
accomplished. Path redundancy and topology control are two ways to set up a fast and
smart network able to counteract in case of path obscuration. Moreover, to compensate the
environment effects without increase the total beam power, the adoption of adaptive optics
such as deformable mirrors, is considered a proper solution. In the following figure, are
illustrated several static and mobile nodes. This is a real scenario in which an FSO link can
be established in a few time.
Fig. 19. Topology of a battlefield scenario using FSO system to connect all partners (Juan C.
Juarez et al., 2006)
Referring to the picture, node C has two optical heads and must be able to manage the
optical beams. Moreover, it is near at the B, D, E nodes; thus node C must be able to take a
decision in which direction must point its optical beam. This kind of decision can be
managed considering operational aspects; distance from the other devices, rate and traffic
demand, distance between the end users, and environment measurement because weather
conditions could limit the optical link. In this case, node C is able to send the beam through
another path using another nodes. How shown in the picture, node D result isolated
because the dense fog does not allow to established a FSO link with the airborne node. Thus,
the only way result node C but its optical heads are both in use. The idea of the optical
reconfigurable network is to establish a link between nodes C and D instead of node B.
Node B will be reached from node.
The quickness in term of time to install, together with a small size, make this kind of
technology able to operate in different segments, like for example, the rehabilitation of link
in case of terroristic attacks or disaster recovery due to natural catastrophic events (E.
Leitgeb et al., 2005).
FSO technology is used also in several non conventional scenarios like:
 Aerospace communications: a laser beam can directionally guarantee a link

between two satellites, between the heart and a satellite platform, and vice versa,
without any kind of interference and, at the same time, achieving an high data rate.
Several experiments have been performed demonstrating that the FSO technology
is mature to accomplish such kind of challenges. On the inter satellite side. In 2008
has been demonstrate an optical link between two LEO orbiting satellites, Terra
SAR-X and NFIRE, at 5.5 Gbps on a total distance of 5500km and at a speed of
25000 km/h. On the downstream side. The KIODO (KIrari's Optical Downlink to
Oberpfaffenhofen) project demonstrated a downlink stream from the OICETS LEO
Japanese satellite. In 2006, 5 trials performed successfully achieving a BER of 10-6
with a modulated optical signal at 50 Mbps and 847nm (N. Perlot et alt, 2007). The
FP6 CAPANINA (Communications from Aerial Platform Networks delivering
Broadband for ALL) project regarded the downlink between a stratospheric ballon
at 22 km and a transportable ground station in Kiruna, Sweden at 1.25 Gbps (M.
Knapex et alt 2006). In 2008 a 2.5 Gbps experiments using a 1W laser at 1064nm
with a BPSK modulation format between a LEO satellite and a ground station has
been demonstrated (E. Leitgeb et alt, 2009).
FSO has been also used to establish links between satellites and aircrafts or
between aircrafts or even between satellite, aircrafts and ground stations realizing
an ad hoc optical broadband wireless airborne network. The LOLA (Liaison
Optique Laser Aéroportée) programme, for example, in 2006 demonstrated the first
two way FSO link between the ARTEMIS GEO satellite and a Myster 20 airplane
flying at 9000m. At the receiver side an accurate hemispherical broadband pointing
system and a CMOS sensor for detection and tracking, with a pointing accurancy
better than 1 micron rad, was used (Cazaubiel et alt, 2005). Real time data
communications, video and audio, demonstrated via a 50 Mbps transmission with
a link acquisition time under the second in 2006 and 2007.
 Deep Space Communications: ultra-long distance can be reached with the FSO
system, thanks to the recent developments in this field, in order to allow the link
with deep space. A great number of studies investigate about the beam divergence
and the geometrical loss to obtain the features to established high data rate FSO
link between Earth-Satellite, Earth-Moon, Earth-Mars and Earth to celestial bodies
within the solar system. To obtain a detector able to work with very low power,
new technologies propose devices such as low noise photon-counting detector tob e
placed on the planets in the form of fields array. The link budget description is
based on EIRP (effective isotropic radiated power), Space losses and PDE
(photodetection efficiency. Furthermore problems can arise from laser to optics
coupling and turbulence in high atmosphere if passed. Modulation formats ar
based on pulse pattern modulation (PPM), at wavelengths ranging from 1064nm
and 1550nm via YDFA technology (D. Caplan et alt, 2007). Dimensioning these
parameters, an FSO system can be used for deep space mission since the optical
beam can cover large distances and go through the space to reach the destination
(Harris Alan et al., 2006). In particular the MLCD project expected performances
are 1Mbit/s farthest Mars and 30 Mbit/s nearest Mars. Increased performances can
support data rate up to 1Gbit/s maximum Mars distance, 100 Mbit/s Jupiter and
10 Mbit/s Uranus (D. Boroson, C. Chen, B. Edwards, 2005).
 Undersea communications: FSO is used to transport information from fixed or

mobile sea-platforms to other stations without deploying any cable. Using a
particular wavelength it is possible to cover small distances (around 100 m) but
with transmission rate around 10Mb/s. The disadvantage is that the sea offers an
attenuation greater than the attenuations offered by the air. Submarine Laser
Communications (SLC) is implemented to achieve high data rate transmissions also
in case of emergency between two platforms or between a platform and an airborne
system. SLC is used for communication with deeply submerged submarine using
an FSO laser at particular wavelength. Often it is used green or blue wavelengths,
thus the radiation is placed in visible spectrum. For undersea application,
particular type of lasers are used such as xenon chloride (XeCl) laser shifted in the
visible spectrum. Each submarine receiver, has the ad-hoc detector to capture the
optical laser beam. Several tests was made to study the performances of undersea
laser link and also aircraft to submarine transmission system was implemented. An
aircraft flying at 40.000 feet was connected with a submarine using a double
wavelength: blue wavelength for uplink and green wavelength for the downlink
stream. There was also clouds between aircraft and submarine but the detector
installed on board of the submarine was able to detect the signal since a special
optical receiver was applied. In this way, the use of the blue-green optical
wavelength for undersea applications confirmed the use of these optical
frequencies in the sea field. The disadvantage of this kind of laser is that its time
operational life is not so long; but the technology, today permit to have solid-state
lasers. This kind of solution permit to have a longer operational life laser and also
with its efficiency is improved. The smaller cost respect to normal gas lasers,
permit also the use of these devices in deep space scenario.
 Air to Earth communications: to prolong the band of the RF technology used today
for several airplanes carrier to monitor, perform surveillance actions and for GIS
(Geographic Information System) applications (Juan C. Juarez et al., 2006). NASA
JPL, on this side, demonstrated a 2.5 Gbps FSO link between an UAV and a ground
station studying in particular the atmospheric fades and the problems related to the
pointing systems (G. Ortiz et alt, 2003).
 Inter island communications: With the DOLCE study, ESA funded project, an
inter island free space communication has been demonstrated covering 142 km
between La Palma and Tenerife at 10 Mbps with a 1W MOPA (Master Oscillator
Power Amplifier) using a 32 PPM (Pulse Position Modulation) and a simple Si-
APD as receiver placed on a OGS (Optical Ground Station) (G. Baister et alt, 2009).
On the same link, the ROSA project performed experiments to investigate an
optical telemetry system for the mars sample return mission (T. Dreischer, 2008).
 Inter optical communications subsystems: board to board and chip to chip optical
wireless interconnections become a reality in the last years (Hirabayashi K. et alt,
1997). The main reasons have been the need to compensate the board to board
bottlenecks and to increase the backplane interconnections speed. Especially in chip
to chip wireless interconnections, the main problem arise from design and package
issues. Other challenges are focused to the development of ultra low driving
devices for VCSEL arrays, commonly used in board to board interconnections, and
to the increase of alignment tolerance. Experiments demonstrated that is possible to
establish more than 1000 channels per printed circuit board using a 1mm pitch
optical beam array having at 1Gbps per channel, thus a throughput per board up to
1 Tbps (Hirabayashi, Yamamoto, Hino 2004).
6. Techno economic analysis (Tosi Beleffi – Forin)

A correct approach, in the techno economic analysis of a FSO based scenario, must focus on
the main market drivers that, today, are mainly related to the civil telecommunications.
Undersea, inter planetary, inter satellites or military based communications, infact, still are
in the research or prototyping field where, everyone knows, the costs are not in principle,
especially at the beginning, strictly taken into account for a mass production. This preamble
is important to justify the fact that this sub chapter will be mainly devoted to the civil
telecommunication applications where the competition of different actors is today
increasing the market portfolio thus lowering the overall costs.
The demand for broadband infrastructures, mainly driven today by the request for new
multimedia applications, is pushing the Operators to implement specific strategies
characterized by a continuous and slow migration to the so called FTTx family (Fiber to The
Curb/Cabinet/Building/Home) where the final step is constituted by the FTTH (Fiber To
The Home). In the future, infact, is expected that we will have an exact replica of the PSTN
network but with fibres instead of the copper. Each end user will have a single or a pair of
fibres directly connected with the CO (Central Office).
Today the main effort is, for what has been previously mentioned, devoted to the
development of new burying strategies to lower the CAPEX that, in the fibre optical based
infrastructures, are mainly due to the fibre installation. The installation costs of a 36 fibre
optical cable, in a typical urban area environment, are, for example, divided between the dig
(12%), the cable and the cable lying (14%) and the civil works for surface footway (74%)
(A.L.Harmer, 1999). For these reasons a tremendous proliferation of new techniques has
been experienced: trench, micro trench, dig, micro dig, one day dig, Teraspan, aerial cables.
Several Operators and Municipalities are today performing demo trials to demonstrate the
possibility to put the fibre cables in the sewer pipes, inside the urban lighting systems or
even in the gas pipes. All these different approaches can, in principle, reduce the installation
costs, depending of course on the particular case and/or situation, in average of a 30-40%
respect to a standard trench approach.
But what can happen if must be crossed a river, a railway or connected a neighbour island?
In this scenario, which can be the role of the FSO?
It should be considered that nothing is so simple as reported with the pen on the paper. In
the case of digs, still many problems are present especially under the regulatory point of
view and, most important, for the huge amount of authorizations that has to be requested to
the Regions, Provinces, Municipalities and Districts. Is not so simple, costless and fast,
infact, taking a excavator and start to dig along a street. On the other side, it must be pointed
out that gas pipes, urban lighting systems, sewer pipes, water systems can be in principle
used to host optical fibres but still remain critical infrastructures, under the security point of
view, and so the fully access to them is still difficult.
We start to understand that the development and the diffusion of the broadband to the end
user is not only a matter of digging the fibre. Is a more complex problem where mixed
wireless and wired infrastructures can and must coexist. This to limit the digital divide,
increase the broadband to all, taking care of the costs, both CAPEX and OPEX (Operational
Expenditures). Depending on the geography, infact, a particular technology can be better
than a different one, being wired or wireless. This open the way to the implementation of
mixed infrastructures and to the deployment of different technologies like: WiMax, WiFi,
SDH Radio, LMDS, WDCMA/UMTS, UTRAN, GSM, FSO, Satellite, xDLS, Fibre, Coax.
Considering the FSO technology as a point to point based system, we can start to define and
differentiate it respect to other possible competitors like the Fibre, the Microwave links, the
XDSL and the COAx (see table 5).
FSO Microwave Optical Coaxial xDSL

radio Fibre cable
Speed Gpbs Mbps Independent Mbps Mbps
Installation Moderate Difficult Difficult Moderate Difficult
Uses P2P/P2MP P2P short P2P/P2MP Campus, Phone and
short and reach short and multi data, access
long reach long reach drop telecom
short sector
reach
Advantages Price vs Speed vs Huge Better Low cost, is
performances, installation Available than already
security bandwidth, other present.
security copper
media.
Disadvantages Dependent on Can be Installation Costs Speed
the climatic intercepted costs limited by
conditions interference
and cable
quality
Security Good Poor Very good Good Good
Maintenance Low Low Low Moderate High
Skills Moderate High High Moderate Moderate
Table 5. Comparison between different P2P networks
From table 5 we can start to figure out, respect to other P2P technologies, the sector of
influence that can be covered by the FSO technology. In order to understand which
broadband technology can be the most efficient in terms of CAPEX and OPEX, we have to
go deeper in the problem considering also the following economic factors like: cost per line,
average return per user, mean time before failure (MTBF), mean time to repair (MTTR),
warranty by vendor, upgradable characteristics, operation and maintenance costs,
manufactured respect to which standard.
Going deeper and deeper in the analysis we can make a simple calculation pointing out the
main media/devices needed to set up a point to point link (see Table 6). In this case we can
see that the main characteristics of a point to point optical wireless link are to be wideband,
easy and fast to install as well as low cost respect to the other technologies. The mix made by
cost per bandwidth per easy to use/install is the winner. The main drawback is of course
due to the climatic conditions encountered that limit, in case, the maximum distance/bit rate
achievable. Adding a RF (Radio Frequency) link, increases the costs but increases also the
availability.
TECHNOLOGY MAIN COSTS LICENCES NOTES

FSO From 13k€ @ 155 Mbps to Is not needed a Fast installation both
19k€ @ 1.2 Gbps (up to 3.6 licence. An una indoor and outdoor.
km). From 22k€ @ 155 Mbps tantum per year is Radio backup is needed
(up to 5.7km) to 33k€ @ 1.2 due to the reference to increase the
Gbps (up to 5.3km) (source: / control PA. availability up to
vendor). 99.999%.
RF From 20k€ to 30k€ depending Is needed a licence. Fast installation, limited
on the length (1-10km) Example: For bandwdith. High time
working @ 18, 26 or 38 GHz 28MHz in the 7GHz window if is considered
can transmit up to 300 Mbps. bandwidth along the time needed to have
(source: vendor). 20km costs around the licence.
5k€. (source: PA)
FIBRE In the case of a P2MP system Dark fibres can be Slow installation, dig is
that is the most cost effective rented in average needed with or without
in terms of CAPEXs respect to from 5 to 15 €/m trench. The bandwidth
a pure FTTH. A standard with a 15 year is virtually unlimited.
GPON OLT (with 4 G- based contract Today the new dig
Ethernet ports), a 1:16 splitter, (source: consulting techniques can lower
an ONU serving 48 VDSL2 center). Free ducts the costs up to 40%
end users, 15 ONTs with 2 can be also rented depending on several
GEthernet ports and VOIP at around 12,3€/m factors. (source:
[Everything compliant with (source: company). operator)
the ITU G.984 standard] has a
total cost of around 26k€.
(Source: vendor).
Around 70€/m for the dig
and 10€/m for the cable in
urban environment. In river
basin the installation plus the
cable (120 fibres) can go down
at around 16€/m. (Source:
PA).
xDSL The costs can vary from 14k€ Copper line/client Maximum available
for an IP DSLAM serving 64 in unbundling bandwid is limited by
users (modems included) up regime is in intereferences. Major
to 40k€ for an IP DSLAM average at 7,5€ costs due to copper
serving up to 1024 end users (source: operator). maintenance Other
(modem included). In this last A twisted pair can voices (COapparatus
case the average bit/rate per vary from 80€ to rental, CO apparatus
client is at around 10-15 Mbps 225€ depending on maintenance, et alt).
over 1.2 km – access network the cable quality.
segment. (Source: consulting (source: operator).
center). Installation costs
are in average at
80-85€/m. (source:
consulting center).
Table 6. Main costs per technology. All these costs do not include installation&maintenance.
It can be also pointed out that, especially where the duct availability is limited, FSO can
provide an economically favourable alternative to both FTTH/FTTC scenarios (T. Rokkas et
alt, 2007). Using a particular tool originated from several EU projects (IST-TONIC TechnO-
ecoNomICs of IP optimised networks and services, and the CELTIC ECOSYS Techno-
ECOnomics of integrated communication SYStems and services), a geometric area model
has been obtained to calculate the number of network elements required (Cabinet, Local
Exchange, Central Office, Fiber Cables) to reach the end user premise. Based on this
geometric area, taking a MTBF of 10 years and a MTTR of 8h and considering the duct
availability, has been demonstrated a Net Present Value (NPV) for FSO Local Exchange (LE)
better than FTTC and FTTH, in case of no duct availability. The fibre based scenarios are
better that the FSO LE building alternative if the duct availability is greater than 70%
(T.Rokkas et Alt, 2007).
Last but not least, analizing the world market from the very beginning is interesting to see
that in principle the society consider, in general, communications not as a fad but as a
necessity. The global consumption of fiber optic components in communication networks
exploded from only 2.5$ million in 1975 to 15.8$ billion in 2000 (J.D. Montgomery, 1999).
Continued growth to 739$ billion in 2025 is expected. In particular, from 2000 to 2025, the
average annual growth rate has been predicted (J.D. Montgomery, 1999), not considering the
actual negative conjuncture, to 12% for fiber optic cable, 21% for active components, 19% for
passive components, 15% for other components. Analyzing data more close to the current
time period we can outline that the global fiber optic connector and mechanical splice
consumption in 2006 was $1.401 billion and by the year 2011, the worldwide consumption
value is forecasted to reach $3.462 billion (ElectroniCast, 2008). Fiber optic collimating lenses
market in Japan/pacific Rim lead the global consumption (number of units used) with
46.7%, 37.9% in North America. Europe with a market share close to 15% is forecasted to
experience a flat growth trend considering the manufacturing outsourcing strategies. Fiber
optic collimating lens assemblies market in North America lead the global consumption
volume (number of units used) with 40%, 35% in Japan Pacific Rim. Europe is forecasted to
increase due to the value-added building of sub assemblies and equipment (ElectroniCast,
2008). Furthermore, the laser product market continued to expand at a healthy rate in 2008
with fiber laser sales almost reached $300 million in revenue (Strategies Unlimited,
2009). Bank of America forecast 21% growth in 2010 for the semiconductor market mainly
driven by electronic equipment sales. Global electronic equipment revenue will rise by 4.9%
in 2010 after a 9.8% decline in 2009, respect to 2008, while the global semiconductor revenue
is set to loose 23% in 2009 respect to 2008 (D. Manners, 2009). It must be pointed out that, in
this case, the main electronic equipment categories are: automotive, data processing, wired
communications, wireless communications, consumer and industrial. The automotive
decline from 2008 to 2009 had a major impact on the overall electronic equipment sales.
Analyzing the data, it can be affirmed that despite the global crisis, the telecom market is
holding up the shock and is representing a real answer for the Governments. Focusing the
telecom sector, especially in the part related to FSO technologies/components, all the data
analyzed demonstrate a good vitality with both capital expenditures, consumes and
revenues. On this point it is expected from GE and 10GE technology to be the next driver for
the FSO companies. 10 Gbps Ethernet lines becomes reality not only within campus or high
rise networks but also to all that customers involved in applications like redundant data
centres, medical imaging and HDTV editing. The costs of 10G FSO systems will certainly
tied by the adoption of the 10GE technology and by it corresponding decrease in the cost
curve. The trade-off for the 10G FSO will be, for example, from the receiver side point of
view on the balance between the size/sensitivity of the InGaAs receivers (suitable at
1550nm) and the tracking/focusing system. FEC, DWDM, laser driver and preamplifiers at
the receiver will certainly must be taken into account to accomplish an economy of scale for
10G FSO (I. Kim, 2009).
7. Conclusion (All)
Communications and transmissions are two fundamental concepts that follow the human
being from the beginning. Free Space Optics is only one of the several choices that we have
today in the complex and mixed telecommunication environment (WiFi, WiMax, LTE, Fibre,
Coax, GSM, UMTS, et alt). With its inherent carachteristics like ease of installation, fast ROI,
low CAPEX, intrinsic security, broad band and with its wide applications range, from under
sea to deep space, free space optics represents today a solution that all the telecom actors
have to take into account. Especially now that 10/100 GEthernet technology is appearing at
the horizon.
8. References
F. Nadeem, B. Geiger, E. Leitgeb, M. S. Awan, S. S. Muhammad, M. Loeschnig, and G.
Kandus (2009) Comparison of link selection algorithms for fso/rf hybrid network, IET,
submitted 2009.
I. Kim (2009) 10G FSO systems position technology for the future, Ligthwave, PennWell July 2009.
L. Stotts, L. Andrews, P. Cherry, J. Foshee, P. Kolodzy, W. McIntire, M. Northcott, R.
Phillips, H. Pike, B. Stadler, and D. Young (2009) Hybrid optical rf airborne
communications, Proceedings of the IEEE, vol. 97, no. 6, pp. 1109–1127, June 2009.
E. Leitged, M.S. Awan, P. Brandl, T. Plank, C. Capsoni, R. Nebuloni, T. Javornik, G. Kandus,
S. Sheik Muhammad, F. Ghassenlooy, M. Loschnigg, F. Nadeem (2009) Current
Optical Technologies for wireless access, Proc. Of IEEE CONTEL Conference, Zagreb
Croatia 8-10 June 2009.
J. Wells (2009), Faster than fiber, IEEE Microwave Magazine, vol. 10, no. 3, pp. 104–112, May 2009.
F. Nadeem, B. Geiger, M. S. Awan, E. Leitgeb, and G. Kandus (2009) Evaluation of switch-over
methods for hybrid FSO-WLAN systems, in Wireless VITAE, Aalborg, May 2009, pp.
565–570.
F. Nadeem, B. Geiger, M. Henkel, E. Leitgeb, M. S. Awan, M. Gebhart, S. Hranilovic, and G.
Kandus (2009) Comparison of wireless optical communication availability data and traffic
data, in VTC Spring, Barcelona, April 2009, pp. 1–5.
F. Nadeem, B. Geiger, M. Henkel, E. Leitgeb, M. Löschnig, and G. Kandus (2009) Switch-over
implementation and analysis for hybrid wireless network of optical wireless and GHz links,
in WTS, Prag, April 2009, pp. 1–6.
F. Nadeem, M. Henkel, B. Geiger, E. Leitgeb, and G. Kandus (2009) Implementation and
analysis of load balancing switch over for hybrid wireless network, in WCNC, Budapest,
April 2009, pp. 1–6.
G. Baister, K. Kudielka, T. Dreischer, M. Tüchler (2009), “Results from the DOLCE (Deep Space
Optical Link Communications Experiment) Project”, Proc. of SPIE Vol. 7199 71990B-1.
D. Manners (2009) Semiconductor sales to grow 10% in Q3, says iSuppli, ElectroncsWeekly.com
Strategies Unlimited (2009) Fiber and industrial Laser Market review and forecast, OM-46 Third
Edition 2009.
S. Bloom andW. S. Hartley (2009), The last-mile solution: hybrid FSO radio,
http://www.systemsupportsolutions.com/whitepapers.htm.
ElectroniCast (2008) Fiber Optic Connector and Mechanical Splice Global Market & Technology
Forecast (2007-2012), Global Summary.
ElectroniCast (2008) Fiber Optic Communication Collimator Lens & Lens assemblies global market
forecast (2007-2012), Global Summary.
T. Dreischer, “ROSA Final Presentation (RF-Optical System Study for Aurora)”, ESA Contract
No. 20371/07/NL/EK Final Presentation, ESTEC, 2nd December 2008
S. Gurumani, H. Moradi, H. H. Refai, P. G. LoPresti, and M. Atiquzzaman (2008) Dynamic path
reconfiguration among hybrid fso/rf nodes, in GLOBECOM, November 2008, pp. 1–5.
M. S. Awan, E. Leitgeb, S. S. Muhammad, Marzuki, F. Nadeem, M. S. Khan, and C. Capsoni,
Distibution function for continental and maritime fog environments for optical wireless
communication, in CSNDSP, July 2008, pp. 260–264.
J. Pacheco de Carvalho, H. Veiga, P. Gomes, and A. Reis (2008) Experimental development and
study of Wi-Fi and FSO links, in CSNDSP, July 2008, pp. 137–141.
V. Kvicera, M. Grabner, and O. Fiser (2008) Visibility and attenuation due to hydrometeors at 850
nm measured on an 850 m path, in CSNDSP, July 2008, pp. 270–272.
A. Chaman Motlagh, V. Ahmadi, Z. Ghassemlooy, and K. Abedi (2008) The effect of
atmospheric turbulence on the performance of the free space optical communications, in
CSNDSP, July 2008, pp. 540–543.
R. Nebuloni and C. Capsoni (2008) Laser attenuation by falling snow in CSNDSP, July 2008, pp.
265–269.
H. Alma and W. Al-Khateeb (2008) Effect of weather conditions on quality of free space optics
links (with focus on malaysia) in ICCCE, May 2008, pp. 1206–1210.
A.K. Majumdar, J.C.Ricklin (2008) Free Space Laser Communications Principles and Advances,
Springer ISBN 978-0-387-28652-5.
S. Vangala and H. Pishro-Nik (2007) A highly reliable fso/rf communication system using efficient
codes, in Global Telecommunications Conference, 2007. GLOBECOM ’07. IEEE,
Nov. 2007, pp. 2232–2236.
D. Caplan, M. Stevens, B. Robinson, S. Constrantine, D. Boroson Ultra-Long Distance Free
Space Laser Communications, CLEO Conference 2007, Baltimore USA.
Perlot, N. et al, “Results of the Optical Downlink Experiment KIODO from OICETS Satellite to
Optical Ground Station Oberpfaffenhofen (OGSOP)”, Proceedings of SPIE, Vol. 6457,
2007.
T. Rokkas, T. Kamalikis, D. Katsianis, D. Varoutas and T. Sphicopoulos (2007), FBusiness
prospects of wide-scale deployment of free space optical technology as a last mile solution: a
techno-economic evaluation, Journal of Optical Networking Vol.6 No.7, July 2007.
S. Vangala and H. Pishro-Nik (2007) Optimal hybrid RF-wireless optical communication for
maximum efficiency and reliability, in CISS, March 2007, pp. 684–689.
Markus Knapek, Joachim Horwatha, Nicolas Perlota, Brandon Wilkersonb (2006) The DLR
Ground Station in the Optical Payload Experiment (STROPEX) - Results of the
Atmospheric Measurement Instruments, Proc of SPIE, ISSN 0277-786X
CODEN PSISDG.
Harris Alan, Sluss James J., Refai Hazem H., Lopresti Peter G., (2006). Analysis of beam
steering tolerances and divergence for various long range FSO communication links,
proceedings of SPIE, the international Society for Optical Engineering, ISSN 0277-
786X.
Juan C. Juarez, Anurag Dwivedi, A. Roger Hammons, Steven D. Jones, Vijitha Weerackody,
Robert A. Nichols, (2006). Free-Space Optical Communications for Next-Generation
Military Networks, IEEE Communications Magazine, November 2006.
M. Abtahi and L. Rusch (2006) Mitigating of scintillation noise in FSO communication links using
saturated optical amplifiers, in MILCOM, October 2006, pp. 1–5.
E. Leitgeb, S. S. Muhammad, B. Flecker, C. Chlestil, M. Gebhart, and T. Javornik (2006) The
influence of dense fog on optical wireless systems, analysed by measurements in Graz for
improving link-reliability, in ICTON, July 2006, pp. 154–159.
B. Flecker, M. Gebhart, E. Leitgeb, S. S. Muhammad, and C. Chlestil (2006) Results of
attenuation-measurements for optical wireless channel under dense fog conditions
regarding different wavelengths, in SPIE, 2006, pp. 1–11.
Cazaubiel, V. et al., “Le terminal optique aeroporte sur lola”, Proceedings of OPTRO 2005
Conference, Paris, France, May 2005
Don M Boroson, Chien-Chung Chen, Bernard Edwards Overview of the Mars Laser
Communications Deminstration Project, IEEE LEOS Vol 19 N.5 Oct. 2005.
D. Grace, M. Mohorcic, M. H. Capstick, M. Bobbio Pallavicini, M. Fitch (2005) Integrating
Users into the Wider Broadband Network via High Altitude Platforms, IEEE Wireless
Communications, Vol. 12, No. 5, October 2005.
J. Derenick, C. Thorne, and J. Spletzer (2005), On the deployment of a hybrid free-space
optic/radio frequency (FSO/RF) mobile ad-hoc network, in IROS, August 2005, pp. 3990–
3996.
T. Kamalakis, I. Neokosmidis, A. Tsipouras, T. Sphicopoulos, S. Pantazis, and I.
Andrikopoulos (2005) Hybrid free space optical/millimeter wave outdoor links for
broadband wireless access networks, in PIMRC, 2005, pp. 1–5.
CAPANINA project website www.capanina.org 2005
Arun K. Majumdar and Jennifer C. Ricklin, (2005). Free-Space laser communication performance,
principles and advantages, pp.9-56, Springer, New York, 2005.
E. Leitgeb, S. S. Muhammad, C. Chlestil, M. Gebhart, and U. Birnbacher (2005) Reliability of
FSO links in next generation optical networks, in ICTON, July 2005, pp. 394–401
Osorio, José Alberto Huanachín Osorio (2005) Simulação e desenvolvimento de um enlace de
„Free-Space Optics‟ no Rio de Janeiro e a relação com a ITU-T G826, Master Thesis,
PUC – Rio de Janeiro, 2005.
S. S. Muhammad, P. Köhldorfer, and E. Leitgeb (2005) Channel modeling for terrestrial free
space optical links, in ICTON, July 2005, pp. 407–410.
A. Akbulut, H. Ilk, and F. Ari (2005) Design, availability and reliability analysis on an
experimental FSO/RF communication system, in ICTON, July 2005, pp. 403–406.
ITU (2005) Attenuation due to clouds and fog, recommendation ITU-R P840-3.
E. Leitgeb, S. Sheikh Muhammad, M. Gebhrt, Ch. Chlestil, U. Birnbacher, O. Koudelka, P.
Schrotter, A. Merdonig, Gorazd Kandus, (2005). Hybrid wireless networks
combining WLAN, FSO and satellite technology for disaster recovery, Proceedings of
14th IST Mobile & Wireless Communications Summit, Dresden, Germany, 19-23 June
2005. (COBISS.SI-ID19153703).
S. Bloom (2005) The physics of free-space optics, May 2005, white Paper, Air Fiber.
H. Wu, B. Hamzeh, and M. Kavehrad (2004) Achieving carrier class availability of FSO link via a
complementary RF link, in ACSSC, November 2004, pp. 1483–1487.
S. Hashmi and H. Mouftah (2004) Integrated optical/wireless networking, in Electrical and
Computer Engineering, 2004. Canadian Conference on, vol. 4, May 2004, pp. 2095–
2098 Vol.4.
E. Leitgeb, M. Gebhart, U. Birnbacher, W. Kogler, and P. Schrotter (2004) High availability of
hybrid wireless networks, Proceedings of the SPIE, vol. 5465, pp. 238–249, 2004
Hirabayashi K., Yamamoto T., Hino S. (2004) Optical backplane with free-space optical
interconnections using tunable beam deflectors and a mirror for bookshelf-assembled terabit
per second class asynchronous transfer mode switch, Optical Engineering 37(2004), pp
1332-1342, Donal O’Shea Ed.
S. D. Milner and C. C. Davis (2004), Hybrid free space optical/RF networks for tactical operations,
in MILCOM, 2004, pp. 409–415.
H. Izadpanah, T. ElBatt, V. Kukshya, F. Dolezal, and B. K. Ryu (2003) High-availability free
space optical and RF hybrid wireless networks, IEEE Wireless Communications, pp. 45–
53, April 2003.
W. Kogler, P. Schrotter, U. Birnbacher, E. Leitgeb, and O. Koudelka (2003), Hybrid wireless
networks - high availability with combined optical / microwave links, in
Telecommunications and Mobile Computing, Mar. 2003, pp. 1–3.
Ortiz, G.G. et al, “Design and development of robust ATP subsystem for the Altair UAV-to-ground
lasercomm 2.5-Gbps demonstration”, Proceedings of the SPIE Vol. 4975, 2003.
A.A. Huurdeman, (2003). The worldwide history of telecommunications John Wiley & Sons ISBN
0-471-20500-2
A Paulo S. B. André, António L. J. Teixeira, João L. Pinto (2003) Simulation of Laser Beam
Propagation Through the Atmosphere for Optical Communications Proposes, Proceedings
of ConfTele 2003, Portugal.
Zhu, Xiaming Zhu, Joseph M. Kahn (2002) “Free-Space Optical Communication Through
Atmospheric Turbulance Channels”, IEEE Transactions on Communications, vol. 50,
pp. 1293-1300, August 2002.
J. Tanczos (2002) Untersuchungen der Verfügbarkeit optischer Freiraumübertragungsstrecken,
Master’s thesis, Technical University of Graz, 2002.
T. ElBatt and H. Izadpanah (2001) Design aspects of hybrid rf/free space optical wireless networks,
in Broadband Communications for the Internet Era Symposium digest, 2001 IEEE
Emerging Technologies Symposium on, 2001, pp. 157–161.
Almeida, Rui Sérgio Rainho de Almeida, José Luís de Jesus Ferreira Santo (2001) Sistemas de
Transmissão/Recepcção sem fios por Raios Infravermelhos, University of Aveiro, Aveiro,
Portugal, Graduation Project, 2001.
B. Bova, S. Rudnicki, (2001). The Story of Light, Sourcebook ISBN 1-4022-0009-9
I. I. Kim and E. Korevaar (2001) Availability of free space optics (FSO) and hybrid FSO/RF
systems, http://www.freespaceoptic.com/WhitePapers/SPIE2001b.pdf.
J.D. Montgomery (1999) Fifty year of fiber optics, Nasua, N.H. Lightwave/Penn Well Corp.
A.L.Harmer, (1999) Communication cables and related technologies, IOS Press, EuroCable 1999,
p.170.
Strickland, Brian R. Strickland, Michael J. Lavan, Eric Woodbridge, Victor Chan (1999)
Effects of fog on the bit-error rate of a free-space laser communication system, Applied
Optics, vol. 38, pp. 424-431, January 1999.
Kim, Isaac I. Kim, Eric Woodbridge, Victor Chan, Brian R. Strickland (1998) Proceedings of
SPIE, vol. 3266, pp. 209-220, 1998.
T. Carbonneau and D. Wisley (1998) Opportunities and challenges for optical wireless; the
competitive advantage of free space telecommunications links in today’s crowded market
place, in SPIE Conference on Optical Wireless Communications, 1998, pp. 119–128.
Hirabayashi K., Yamamoto T., Hino S.Kohama Y., Tateno K. (1997) Optical beam direction
compensating system for board-to-board free space optical interconnection in high-capacity
ATM switch, Journal of Lightwave technology Vol15, N.5, pp.874-882.
Kim, Isaac I. Kim, Harel Hakakha, Prasanna Adhikari, Eric Korevaar, Arun K. Majumbar
(1997) Scintillation reduction using multiple transmitters, Proceedings of SPIE, vol.
2990, pp. 102-113, 1997.
Wichel, Hugo Weichel (1990) Laser Beam Propagation in the Atmosphere, Washington: Tutorial
Texts, 1990.
Clifford, S. F. Clifford, R. J. Hill (1981) Relation between irradiance and log-amplitude variance for
optical scintillation described by the K distribution, Journal of the Optical Society of
America A, vol. 71, pp. 1112-114, January 1981.
Strohbehn, J. W. Strohbehn ed., (1978) Laser Beam Propagation in the Atmosphere New York:
Springer-Verlag, 1978.
G. Pottino, (1976). I Cartaginesi in Sicilia Palermo, Italy.
Unknown, (1968) Da Trastevere alla Colombo le telefonare del futuro, Column on “Il
Messaggero”, Italian National Newspaper, 11 June 1968 Rome, Italy.
E.Kube, (1968). Information Transmission by light beams through the atmosphere, Journal
Nachrichtentechnik 19 (1969) II.6
A.R.Michaelis, (1965). From Semaphore to Satellite, International Telecommunication Union,
Geneva, p.18.
G.Agnello, (1963). Le torri costiere di Siracusa nella lotta anticorsara Archivio storico di
Siracusa, Vol I-IV, p.25, Siracusa, Italy.
V. Castelli, (1700). Dizionario enciclopedico siculo, B.C. Palermo, Italy
Eschilo, (458 b.C.). Orestea (Agamennone-Coefore.Eumenidi), prol. 15-30.
Novel multiple access models and their probabilistic description 297
14
X
Novel multiple access models and their

probabilistic description
Dmitry Osipov
Institute for Information Transmission Problems Russian Academy of Science
(Kharkevich Institute)
Russia
1. Introduction
Nowadays as both the number of subscribers of the wireless services and the amount of data
transmitted via the aforesaid services are growing rapidly multiple access techniques are
becoming of special urgency for specialists in the field. Frequency Hopping technique
appears to be one of the most promising paradigms for the next-generation multiple access
system design.
In conventional frequency hopping the entire available frequency band is divided into N
subbands (following the terminology used in (Zigangirov, 2004) we shall further on refer to
the set of all subbands available to the user as a hopset.). Each user’s transmitter chooses
one of the N subbands in a pseudorandom manner and transmits a signal via the chosen
subband using a conventional modulation technique (for the most part FSK is used). In the
multiple access theory it is common to refer to each change of the subband in use as a hop.
The process of switching between the subbands (which can be also interpreted in terms of
assigning subbands to the users) is called frequency hopping. Frequency Hopping can be
easily combined with the OFDM technique, which enables to reduce the distortion
considered by both fadings and impulse noise (Note that OFDM is the fundamental tech-
nique for the most of the advanced telecommunications systems).
Frequency hopping (assigning subbands to the users) can be either coordinated (this means
that when choosing a code sequence for each user, we use knowledge about other users;
therefore, this scheme is mostly used for the downlink transmission) or uncoordinated. The
uncoordinated frequency hopping has a number of advantages (enables to realize random
multiple access, is well protected from eavesdropping or intentional jamming, and does not
require elaborated protocols, which are inevitable in a system with coordinated subcarrier
assignment) and is used, as a rule, for the uplink transmission. We shall consider uplink
transmission, and, hence, uncoordinated subcarrier assignment strategies. In turn, these
strategies can be divided into two groups. Methods of the first group exploit code sequences
specially designed for multiple access systems and satisfying certain requirements. The
second group includes methods where subcarrier numbers are assigned to users pseudo-
randomly. In what follows a FH CDMA system using a method from the second group will
be considered.
The proposed chapter is aimed at introducing to the readers a novel class of FH CDMA
systems; these are Dynamic Hopset Allocation FH CDMA (DHA FH CDMA) systems. This
class has been initially introduced in (Zyablov & Osipov, 2008). As can be seen from its
name the main difference between the conventional FH CDMA and the proposed class of
Dynamic Hopset Allocation FH CDMA is that the hopset allocated to each user varies in
time.
2. DHA FH: transmission technique

Let us now consider the basics of the DHA FH CDMA in more detail. Consider a multiple
access system where K active users transmit data to the base station through a channel di-
vided with the help of the OFDM technology into Q frequency subchannels; the
transmission is asynchronous and uncoordinated (the latter means that neither of the users
has information about the others). It is assumed that all the users transmit binary β -tuples.
In the course of the transmission of each consecutive tuple the subchannel number generator
assigned to the user under consideration chooses (in a random manner) q  2 β subchannels
out of Q subchannels. Each tuple (or a part of the tuple) that is to be transmitted by the
aforesaid user within the frame is mapped into the number of the subchannel, via which the
signal is transmitted.
In what follows we shall assume that in the system under consideration optimal power con-
trol is used. The latter means that the amplitudes of all the signals from distinct users are
equal at the receiver side. Furthermore we shall assume that in the system under considera-
tion optimal phase estimation and prediction mechanism is used (i. e. phases of all the sig-
nals from distinct users are known to the base station). Hereinafter we shall consider a
multiple access system, which uses an Additive White Gaussian Noise channel. In this case
near optimal power control and phase estimation can be done fairly easily since both trans-
mission ratio and phase of the signal from a certain user depend only on the distance
between the transmitter and the receiver. The latter can be measured either by using pilot
sequences or by analyzing information obtained from the downlink channel (which seems
preferable since it is more convenient for implementing open loop power control and
enables to avoid other users interference). Note, however, that in either case the transmitted
signal is affected at least by the background noise, thus even in this case only near optimal
power control and phase estimation can be maintained.
It is assumed that the base station is equipped with the subchannel numbers generator
synchronized with that of the active user. The latter means that within the scope of the
reception of the respective tuple subchannel numbers generator of the base station produces
the very same subchannel numbers vector that has been generated by the subchannel num-
bers generator of the user under consideration. Note that this assumption is not restrictive
since synchronized generators are an essential part of any conventional FH CDMA system.
Thus we simply replace a generator producing random numbers with a generator
producing random vectors.
Assume now that there is an eavesdropper, who intends to reconstruct the signals sent by
the active users. In the conventional FH CDMA system eavesdropping is considered to be a
complicated task since it is preassumed that the hopping sequences are not known to the
eavesdropper. However even if the hopping sequence is not known to the user it is still
possible to reconstruct the transmitted signal. Note that the DHA FH CDMA system is
much more eavesdropping proof than the conventional one (since to reconstruct a tuple an
eavesdropper needs not only to detect, which subchannel has actually been used to transmit
a signal, but also to detect its position in the hopset, the latter being known only to the active
user under consideration).
However, the introduction of the aforesaid new class of FH CDMA systems results in the
emergence of a whole complex of problems. The problem of giving a suitable probabilistic
description appears to be one of the most urgent ones. The present chapter is aimed at
solving this problem. The latter is to be done using characteristic function apparatus. Since
the probabilistic description depends on the reception strategy two different strategies, i. e.
threshold reception and MAXP reception, will be considered. For each of the above
mentioned reception strategies an analytical expression will be given. The results to be ob-
tained will enable to get new insights into the whole complex of problems of novel types of
CDMA systems study and development.
3.1 DHA FH OFDMA system with threshold reception:

system model and probabilistic description
A receiver, equipped with a subchannel number generator synchronized with the
subchannel number generator of this user, compares values of projections of signals
received through these subchannels onto a direction denoted by the phase of the signal
from the respective user at the receiver side with a ceratin threshold (in what follows, we
shall assume that the base station knows the phases of complex transmission
coefficients of all the subchannels; i. e., the system exploits an ideal estimation an
prediction mechanism for frequency characteristics of the channel). If threshold
crossing is detected in only one subchannel, a tuple corresponding to the subchannel
where the threshold crossing was registered, is accepted. Otherwise, an erasure decision
is made. If a symbol other than the transmitted one is accepted, we say that an error
has occurred. The block diagram of the DHA FH CDMA system with a threshold
receiver is shown in Figure 1.
Fig. 1. DHA FH OFDMA system with threshold reception: block diagram.
Due to the above-made assumption on the availability of an ideal power control in the
system, the signal of each user at the receiver end can be represented by a vector of a unit
amplitude with a random phase uniformly distributed on the circle [0, 2 π ] . Since all the
users transmit data asynchronously, the duration of interaction of a fixed user with other
active users (we call them "interfering" users) is a random variable uniformly distributed
on [0, T ] . Therefore, components of the noise from each "interfering" user can be represented
by vectors with amplitudes uniformly distributed on [0,1] and random phases uniformly
distributed on [0, 2 π ] .
Let us consider the projection of an output of the j-th subchannel onto a unit-amplitude
vector with phase  j (here  j is the phase of the signal from the user under consideration at
the receiver end.) The projection is of the form:
Kj
k 1
 
y j  a j   rjk  cos  jk  nj (3.1)
where a j is the amplitude of the signal transmitted by this user through this
subchannel
1 j  j
*
aj  
0 j  j
*
K j is the number of "interfering" users in the j-th subchannel; rjk is the amplitude of the
noise caused by the interaction with the k-th "interfering" user, which is uniformly
distributed on [0,1];  jk   j   k , where  k is the phase of the noise caused by the
interaction with the k-th "interfering" user, which is uniformly distributed on the circle
[0, 2 π ] ;  j is the phase of the transmitted symbol at the receiver end, which is uniformly
distributed on the circle [0, 2 π ] ; and nj is the projection of a vector corresponding to the
white Gaussian noise.
Since the components of the white noise have the same variance, the value of the projection
of the corresponding vector onto any direction is also normally distributed with the same
variance. Since the phases of both the noise and the transmitted symbol are uniformly
distributed on the circle [0, 2 π ] , the phase difference  jk is also uniformly distributed on
[0, 2 π ] .
Since for parameters that adequately describe modern multiple access systems, the
probability of a collision of multiplicity greater than two is much less than the probability of
a collision of multiplicity two (see Appendix A.) further on only collisions of multiplicity
two will be considered; the occurrence probability of such a collision is set to be equal to the
collision occurrence probability
p J  1  p0 . (3.2)
where p0 is the probability of no collision. Since transmission is asynchronous, during the

time when a user transmits a symbol, each of the other active users chooses two
subchannels. Therefore, the probability of no collision is given by
2( K 1)
 Q 1
p0    (3.3)
 Q 
and the projection is given by:
y j  a j  λ cos  α   nj . (3.4)
An expression for the density function of the value of the projection of the noise from the
k-th “interfering” user onto a given vector follows from a formula due to (Feller, 1971) and
is of the form:
1  1  1  z2 
f ( z)  ln  . (3.5)
π  z 

Finding an analytic expression for the density function of y j directly (for instance, using the
convolution theorem) presents considerable difficulties. By definition (Lukacs, 1987), the
characteristic function of a random variable Z  λ  cos  α  is the expectation of the
function y  φξ ( λ , α)  e iξZ ( λ , α ) . This expectation can be defined in a straightforward manner:
 
gZ  ξ     φ  λ , α   f  λ , α  dαdλ
ξ (3.6)
 
here
 1
 λ  [ 0,1]  α  [0 , 2 π ]
f  λ , α   2π is the joint probability density function.
0 otherwise
Taking into account the domain of definition of the function, the characteristic function can
be represented as follows:
1 2 i cos  
e
gZ       d d . (3.7)
0 0 2
Note that
2
1 i cos  
J o ( ) 
2 e
0
d . (3.8)
is nothing but the zero-order Bessel function of the first kind of the variable υ  ξλ (Watson,
1945). Integrating the series expansion of this function term by term, we obtain:
1 æ 2k ö
2 k (xl ) x 2k
÷÷dl = 1 + (i )2 k
¥ ¥
ç
g1 (x ) = ò çç1 + å (i ) 2 k 2÷÷ å . (3.9)
çè 2 ( k !) ÷ø (2k + 1) 22 k ( k !)
2
0 ç k =1 k =1
By a property of characteristic functions, the desired density function of the variable y is of

the form:

f  y , a,   e
 i y
gZ      , a,  d , (3.10)

where χ  ξ , a, σ  is the characteristic function of a normal distribution with mean a and
1 d η  y, a, σ 
 n
variance  . Taking into account  eiξy  ξ n  χ  ξ , a, σ  dξ  n n we obtain
  i  dy

1 d 2 k η  y , a, σ 
f s  y , a, σ    . (3.11)
 2k  1 2  k !
2
k 1
2k
dy 2 k
here   , a,  is the probability density function of a normal distribution with mean a
and standard deviation 
This function characterizes the density function of the value of the projection of a subchan-
nel output provided that a collision of multiplicity two occurred in the subchannel (i. e.,
there was one “interfering” user) and that the user under consideration transmitted a signal
of amplitude a through this subchannel.
Thus, the conditional density function of the variable at the output of a subchannel, through
which the user transmitted a symbol of amplitude a, is of the form:
μ  y , a , σ   η  y , a , σ  p0   f  y , a , σ 1  p0   . (3.12)
Below we are using the obtained probability density function to solve the detection problem
in the described system.
3.2 DHA FH OFDMA system with threshold reception:

threshold choice and analytical expressions for error and erasure probabilities
Symbol detection in the described system boils down to choosing between two hypotheses:
H0: the user did not transmit any symbol through this subchannel (a = 0);
H1: the user transmitted a symbol through this subchannel (a = 1).
The maximum likelihood condition implies that:
f s ( y 1) pa  0 
if   H1isaccepted
f s ( y 0) pa 1
(3.13)
f s ( y 1) pa  0 
if   H 0 isaccepted
f s ( y 0) pa 1
f s ( y 1) 1 q 1
where L  is the likelihood ratio, and pa 1  , pa  0   1  pa 1 
f s ( y 0) q q
a priori probabilities. Hence we obtain,
f s ( y 1)
 q 1 . (3.14)
f s ( y 0)
Our goal is to find the threshold value y. The value will be found by solving the nonlinear
equation (3.14) numerically.
Direct computation of the likelihood ratio on the left-hand side of (3.14) seems to be difficult.
At the same time, numerical values of the conditional density function at any point can be
computed using finitely many terms of the above-given expansion. To estimate these values to
a given degree of accuracy, it suffices to find an expansion term, whose absolute value at a
given point is less than the accuracy parameter and check that the same holds for the
subsequent term too. Thus, the threshold finding problem is reduced to solving a nonlinear
equation (3.14) which can be done numerically using specialized software packages. The
solution of the equation (3.14) ŷ is the optimal (in a maximum a posteriori sense) threshold
value.
Hereinbefore we have treated the process of assigning subchannels to the users in terms of
series of random trials. Strictly speaking, when considering a set of subchannels, we are to
consider the parameters of each consecutive test run as dependant on outcomes of the
previous test runs. Consider the i-th subchannel of the q subchannels chosen by the
subchannels number generator. Let us assume that k users were transmitting in i-1
previously considered subchannels. The probability of collision is to be estimated as
æ Q - i - 1ö÷ (
2 K -1)- k
p¢J = çç ÷÷ .
èç Q - i ø÷
However, since i < q  Q and k < K
ö(
2 K -1)- k
æ Q - 1÷ö (
2 K -1)
æ
çç Q - i -1÷÷ » çç ÷ .
çè Q - i ÷ø ÷ çè Q ÷÷ø
Thus, in what follows we shall assume that the probability of collision in a certain
subchannel does not depend on the situation in other subchannels.
The error probability is the probability that the threshold is not crossed in the subchannel,
through which the user under consideration has transmitted a symbol and at the same time
the threshold is crossed in one of the subchannels, where this user has not transmitted any
signal (and only in it). This probability can be written as:
yˆ q2

 yˆ 
perr   q  1   f ( y 1)dy   f s ( y 0)dy    f s ( y 0)dy  . (3.15)
s
 
 yˆ   
To find the erasure probability, we shall find the probability of correct reception, i. e., the
probability that the threshold crossing occurs only in the subchannel, through which the
user has transmitted a symbol. This probability is described by the relation:
q 1

 yˆ 
pcor   f s ( y 1)dy    f s ( y 0)dy 
 
. (3.16)
yˆ   
Since correct reception, error, and erasure form an exhaustive group of events, we may claim
that the erasure probability is
pers  1  pcor  perr (3.17)
Note that the obtained expressions are approximate (due to the aforementioned
assumptions). More exact expressions can be easily obtained using the same method.
However in this case the obtained expressions will have a much more cumbersome form.
4.1 DHA FH OFDMA system with a MAXP receiver:

system model and probabilistic description
In the previous section a threshold reception has been considered; i. e. a receiver, equipped
with a subchannel number generator synchronized with that of the user under
consideration, was to compare values of the projections of the signals received from the
subchannels with a certain threshold and take a decision on the symbol sent by the
user under consideration. Hereinafter a far more simple and intuitive reception strategy
will be used; i. e. the receiver is to compute the projections of the signals from the respective
subchannels and to choose the subchannel with a maximum projection. In what follows we
shall refer to this receiver as a MAXimum Projection receiver (a MAXP receiver). The block
diagram of the DHA FH CDMA system with a threshold receiver is shown in
Figure 2.
Fig. 2. DHA FH OFDMA system with a MAXP receiver: block diagram.

Note that in DHA FH CDMA with a threshold receiver an optimal threshold value is to be
precomputed in advance. Moreover, as can be seen from the previous paragraphs, the
optimal threshold value depends on the number of active users and SNR. In DHA FH
CDMA with a MAXP receiver, on the other hand, neither additional computation nor any
kind of side information is needed. It should be noted, however, that for DHA FH CDMA
system with a threshold receiver and DHA FH CDMA system with a MAXP receiver differ-
ent types of outer codes are to be used. In a DHA FH CDMA system with a threshold
receiver outer codes capable of correcting erasures (a number of low density codes suited for
the task were introduced recently) are to be used, whereas in a DHA FH CDMA system
with a MAXP receiver only error correction is needed (woven and woven turbo codes seem
to be the best solution in the case).
Let us consider a difference:
δ j* j = y j* - y j = a j* + ρ j* j + n, (4.1)
here y j* is the projection of the output of the j* -th subchannel (i. e. the subchannel, via
which the user under consideration has actually transmitted a symbol); y j is the projection
of the output of the j -th subchannel ( j  S, j  j* ).
Note that for parameters that adequately describe modern multiple access systems, the
probability of a collision of multiplicity greater than two is much less than the probability of
a collision of multiplicity two. Therefore, hereinafter we are considering the case of a
collision of multiplicity two only. Correspondingly we are to consider several distinct
situations:
a. A collision has occurred both in the j* -th subchannel and in the j -th subchannel
(see Figure 3).
Fig. 3. “case a.” (three interfering signals): diagram.
Then ρ j* j is given by:
r j* j (c * , s, c ) = l cos (a ) - m cos (b ) - k cos (j ). (4.2)

b. A collision has occurred in the j* -th subchannel and a user has transmitted a
symbol via the j -th subchannel. (see Figure 4).
Fig. 4. “case b.” (two interfering signals): diagram
r j* j (c * , s, c ) = l cos (a ) - m cos (b ). (4.3)
c. A collision has occurred in the j* -th subchannel and none of the users has
transmitted a symbol via the j -th subchannel (see Figure 5).
Fig. 5. “case c.” (one interfering signal): diagram.
r j* j (c * , s , c ) = l cos (a ). (4.4)
d. A user has transmitted a symbol via the j -th subchannel and no collision has
occurred either in the j -th subchannel or in the j* -th subchannel (see Figure 6).
Fig. 6. “case d.” (one interfering signal): diagram.
r j* j ( c * , s, c ) = -m cos (b ). (4.5)
e. A collision has occurred in the j -th subchannel and no collision has occurred in
the j* -th subchannel (see Figure 7).
Fig. 7. “case e.” (two interfering signals): diagram.
r j* j ( c * , s, c ) = -m cos (b ) - h cos (j ). (4.6)
f. No collision has occurred in the j* -th subchannel and none of the users has
transmitted a symbol via the j -th subchannel (see Figure 8).
Fig. 8. “case f.” (no interfering signals): diagram.

r j* j = 0.
Note that the distribution of ρ j* j does not depend on the sign of the component since all the
phases are distributed uniformly on  0, 2π  . Therefore ρ j* j (c * , s , c ) and ρ j* j ( c * , s, c ) have the
same distribution.
Moreover the probability density function of differences

d j* j (c * , s , c ) = a j* + l cos (a ) + n and d j* j ( c * , s, c ) = a j* - m cos (b ) + n
can be obtained using the same method we have used to obtain the projection y j* and is
given by
 y , a,    f 

1 
d 2 k y , a, 2 
c , s , c  
y , a,    f 
 c ,s, c  
y , a,   
1
f (4.7)
 2k  1 2  k !
2 2k
dy
* *
2k
j* j j* j k 1

here η ξ , a, σ 2  is the probability density function of a normal distribution with mean a
and standard deviation σ 2
Analogously r j* j (c * , s, c ) and r j* j ( c * , s, c ) have the same distribution. Since all the

components in (4.3) and (4.6) are independent the characteristic function is simply the prod-
uct of the characteristic functions of the components:
¥ ¥
2 k +2m x 2 k + 2m
g 2 (x ) = g1 (x ) ⋅ g1 (x ) = 1 + åå (i ) 2
m =1 k =1 (2k + 1)(2m + 1) 22 k + 2m ( k !m!)
t km
æ ö÷
¥
x 2t çç t 1 ÷÷.
g 2 (x ) = 1 + å 2 t çå
ç 2÷
t =1 2 ççè k =1 (2 k + 1)(2t - 2k + 1)( k !(t - k )!) ÷ø÷
The probability density function of differences d j* j (c * , s, c ) = a j* + l cos (a ) - m cos (b ) + n

and d j* j ( c * , s, c ) = a j* - m cos (b ) - h cos (j ) + n is given by:

f 2  y , a ,   f
j* j
c* ,s , c  
y , a ,   f
j* j
 c * ,s ,c  
y , a ,   e
 i y
 
g 2      , a , 2 d ,

¥ æ d h (y , a, s 2 ) æ ö÷÷ö
2t
ç 1 çç t 1 ÷÷÷÷.
f 2 ( y , a , s ) = å ççç 2t ç å (4.8)
t=0 ç dy 2t
2 ç k =1 (2 k + 1)(2t - 2k + 1)( k !(t - k)!) ÷÷÷÷÷
ç
2 ÷
è è øø
Similarly the characteristic function of d j * j (c * , s , c ) = a j * + l cos (a) - m cos (b ) - k cos (j) + n

is given by:
¥ ¥ ¥
x 2 k + 2 m+ 2 w
g 2 (x ) = g1 (x ) ⋅ g1 (x ) ⋅ g1 (x ) = 1 + ååå (i)
2 k + 2 m+ 2 w
2
.
k =1 m =1 w =1 (2 k + 1)(2m + 1)(2 w + 1) 22 k + 2 m+ 2 w ( k ! m ! w !)
And the probability density function of difference δ j* j (c * , s, c ) is
f 3 (y , a, s) = fd
(c * , s , c) (
y , a, s) =
j* j
¥ æ d h (y , a, s 2 ) æ ö÷ö÷ (4.9)
2t
ç 1 çç t k 1
= å ççç çåå
2t ç
÷÷÷÷.
2 ÷÷
ç
t =0 è dy 2t
2 èç k =1 m=1 (2 k + 1)(2m + 1)(2t - 2 k - 2m + 1)( k ! m !(t - k - m)!) ÷ø÷ø÷÷
Note that difference d j * j (c * , s , c ) = a j * + n is simply a Gaussian viable.
4.2 DHA FH OFDMA system with a MAXP reception: error probability

To find the error probability let us consider the probability of correct reception. Correct
reception is possible if (and only if ) δ j * j  0j  S , j  j * . As can be seen from what has been
said the distribution of the differences δ j * j depend on how many active users are there in the
respective subchannels. Moreover, as has already been mentioned, we shall confine
ourselves to the case of collisions of multiplicity two (in the subchannels chosen by the
subchannel number generator of the active user under consideration). Let us consider the
following situation: in the rest q  1 subchannels (i. e. in all the subchannels chosen by the
subchannel number generator of the active user under consideration but for the one in use)
there are exactly a subchannels where a collision has occurred (i. e. in each of the a
subchannels 2 users were transmitting) b subchannels where users were transmitting but
no collision had occurred (i. e. there was 1 active user in each subchannel ) and
c  q  1  a  b subchannels where no users were transmitting any data.
The probability of such a situation is given by:
 q  1 !
Q  q 
  2a b
a ! b ! q  1  a  b  !
p  a, b , q , Q ,     
. (4.10)
 2    2 a
 q  1 !
  a ! b ! q  1  a  b  ! Q  q 
2ab
a0 b0
where  is the total number of signals that can be transmitted via all the subchannels but
for the one, via which the user under consideration has transmitted (see Appendix B.)
As has been already mentioned if the total number of active users amounts to K the number
of signals that can interfere with the signal transmitted by the user under consideration is
Κ  2   K  1 due to the transmission asynchrony. Thus, we are to consider two distinct
cases:
I. A collision has occurred in the subchannel, via which the user has transmitted a
signal (this case obviously corresponds to the cases a.)-c.) from the previous section)
II. No collision has occurred in the subchannel, via which the user has transmitted a
signal (this case corresponds to the cases d.)-f.) from the previous section)
Since we are considering collisions of multiplicity two only in case II, the number of
interfering users is given by Ξ  2  K  1 and in case I by Ξ  2  K  1  1
To obtain the conditional probability note that in the abovementioned case, if a collision has
occurred in the subchannel in use, we obtain a differences with pdf f 3 , b differences with
pdf f 2 and c  q  1  a  b differences with pdf f 1 . Thus, the conditional probability of
error for the situation in question is given by:
    a
  
b
   
q 1 a  b

pe  a , b , q   1  1     f 3  y , a ,  dy    f 2  y , a ,  dy    f 1  y , a ,  dy   . (4.11)
  0      
  0 0 
If there is no collision in the subchannel in use we obtain a differences with pdf f 2 , b dif-
ferences with pdf f 1 and c  q  1  a  b differences with pdf f 0 . The conditional
probability is then given by:
    a
  
b
   
q 1 a  b

pe  a , b , q   0   1     f 2  y , a ,  dy    f 1  y , a ,  dy    f 0  y , a ,  dy   . (4.12)
  0  0  0  
  
Thus, the probability of error is given by:

K 2
 2   2 ( K 1 a ) 1
perr     p  a, b , q   1 p p  a , b , q , Q , 2   K  1  1  
e J
a0 b 0
for K  2 (4.13)
 K 1 
 2  2 ( K 1 a )
    p  a , b , q   0  1  p  p  a , b , q , Q , 2   K  1  
a0 b 0
e J
where p J is the probability of collision (see (3.2)).
5. Conclusions and future work

Hereinabove a novel class of FH CDMA systems has been introduced: these are Dynamic
Hopset Allocation FH CDMA systems. Two models with different reception strategies were
considered: the DHA FH CDMA system with a threshold receiver and the DHA FH CDMA
with a MAXP receiver. For both models a probabilistic description has been given using the
approach based on characteristic function apparatus (for the DHA FH CDMA system with a
threshold reception both a nonlinear equation for optimal threshold value and analytical
expressions for error and erasure probabilities have been given; for the DHA FH CDMA
with a MAXP receiver an analytical expression for the probability of error has been given).
However, even though several simplifying assumptions were used the obtained expressions
are still complicated. Thus, obtaining rougher but less cumbersome expressions for the
probability density functions of the random variables in question and respectively obtaining
upper and lower bounds for error and erasure probabilities is one of the most important
tasks. Moreover, it is to be mentioned that the expressions in question were obtained for the
AWGN channel. Generalizing the obtained results for fading channel models (say, Raleigh,
Nakagami ect.) is also a very important task. Obviously power control at the receiver side is
unacceptable in this case. However, even in case of open loop nearly ideal power control it
is to be taken into account that the power of a real life transmitter is bounded i. e. even in
this case expressions for error and erasure probabilities will be different from those
obtained for the AWGN channel.
6. Appendix A
Let us show that the probability of collisions of multiplicities greater than two is much less
than the probability of a multiplicity-two collision. To this end, consider the process of
choosing subchannels for transmission by users as a sequence of independent trials. Due to
asynchronicity, during the time when a user transmits a symbol, each other active user
chooses two subchannels. Thus, the process of choosing subchannels for transmission by
other active users is equivalent to a series of N = 2(K − 1) trials with “success” probability p
= 1/Q (in this case, it is the probability that there is chosen the subchannel which is used by
the user under consideration). Therefore, the probability that a collision of multiplicity
m = k+1 occurs in a given subchannel is the probability that in a series of N = 2(K − 1) trials
with “success” probability p = 1/Q there are precisely k successes. This probability is given
by the Bernoulli formula:
k Nk
1  1
p J (m)  C Nk     1   ,
Q
   Q 
where p J (1) is the probability of no collision. Consider the ratio:
p J (m  2)
κ .
pJ  2
It is seen from the aforesaid that:
k N k
N
1  1 N
p J ( m  2)   C  Q  1
N k
C Nk     1   k
k2  Q  Q N
κ   k 2
, (6.1)
pJ  2 1 
1
1
N 1
N   Q  1
N 1
C 1N     1  
Q  Q
hence we get:
 C  Q  1
k  k 1
N
κ k 2
. (6.2)
N
For each term in this expansion, we may write:
k 1
C Nk   Q  1  Q  1 N   N  1   N  k  1
 k 1  k 1
N!  N 1 
  
  Q  1  2 
. (6.3)
N  N  k !  k ! N N   Q  1
k 1
  k !  
thus, it is obvious that:
k 1
N 
N 1 
κ    . (6.4)

k  2   Q  1  2 
N 1
For any Q such that Q  (and in real-world systems we have Q >> N), the expression
2
on the right-hand side is a sum of a decreasing geometric progression starting with the
second term:
k 1 k 1
N 
N 1   
N 1  1
κ    1    1  1 (6.5)

k 1   Q  1  2 

k 1   Q  1  2 
N 1
1
Q  1  2
1 N
Note that 1  1 Q Q  .
N 1 2
1
Q  1  2
As we have already noted, in real-world systems we have Q  N . Therefore, we may claim
that for any parameters that adequately describe modern systems, the formulated statement
certainly holds.
7. Appendix B
To obtain the probability of such a situation let us find the number of all sets complying
with the abovementioned condition. To simplify the process let us treat the process of
random subchannel choice performed by each of the active users in terms of assigning the
interfering signals from other active users to the subchannels (the latter procedure is
analogous to the classical examples with balls randomly placed into boxes). Let us again
consider the subset of q  1 subchannels (these are all the subchannels chosen by the
subchannel numbers generator but for the one, via which the user under consideration has
transmitted) where there are exactly a subchannels, to which 2 signals are assigned, b
subchannels, to which 1 signal is assigned, and the rest of the subchannels are empty (i. e. no
signal has been assigned to none of these subchannels). The number of ways to perform
such an assignment is given by

N  a , b , q  1 
 q  1 ! (7.1)
a ! b ! q  1  a  b  !
Now let us consider the process of the symbol transmission by the user under consideration.
In what follows we shall assume that there are  interfering signals and that the
assignment for all the subchannels chosen by the subchannel numbers generator (but for the
one, via which the user under consideration has transmitted) is performed in the very way

that has been discussed above. Thus we still have to assign k  Ξ  2 a  b signals to
Θ  Q  q subchannels (these are all the subchannels but for those chosen by the subchannel
numbers generator of the user under consideration). However, from now on we needn’t
confine ourselves to the case of collisions of multiplicity two. The process of “assigning” the

remaining subchannels to the signals boils down to a series of k  Ξ  2 a  b trails, each of
which is a random choice of one of Θ  Q  q subchannels. Thus the number of all the
possible outcomes is

N  q , Q ,  a, b   Q  q 
  2a b
(7.2)
And the number of all the sets meeting the abovementioned requirements is given by
 
N  a , b , q , Q ,    N  a , b , q  1 N  q , Q ,  a , b  
 q  1 !
Q  q 
  2a b
(7.3)
a ! b ! q  1  a  b  !
Apparently the probability of the occurrence of the aforesaid situation is the fraction of all
the sets meeting the abovementioned requirements in the total number of sets. The latter is
to be computed by summing N  a, b, q, Q, Ξ  over all possible values of a and b :
 q  1 !
Q  q 
  2a b
a !b ! q  1  a  b !
p  a, b , q , Q ,     
. (7.4)
 2    2 a
 q  1 !
 Q  q 
2ab
a 0 b 0 a ! b ! q  1  a  b !
8. References
Feller W., (1971). An Introduction to Probability Theory and Its Applications vol.2, Wiley and
Songs, ISBN: 0471257095 / 0-471-25709-5, NY.
Lukacs E., (1987). Characteristic Functions, A Charles Griffin Book, ISBN: 0195205782,
London.
Watson G. N., (1945). Theory of Bessel Functions, Cambridge Press, ISBN: 0521483913,
Cambridge.
Zigangirov K. Sh. (2004), Theory of Code Division Multiple Access Communication, IEEE
Press, ISBN: 0471457124, Piscataway, New Jersey.
Zyablov V. V. & Osipov D. S., (2008). On the optimum choice of a threshold in a frequency
hopping OFDMA system. Problems of Information Transmission, Vol. 44, No. 2, 2008,
pp. 91-98, ISSN 0032-9460.
Performance analysis of multi-server
queueing system operating under control of a random environment 317
15
X

queueing system operating under
control of a random environment
1 2 2
Che Soong Kim , Alexander Dudin , Valentina Klimenok ,
2
and Valentina Khramova
1
Sangji University
Korea
2
Belarusian State University
Belarus
1. Introduction
Since the early 1900th the Erlang multi-server queueing systems with losses ( B -model or
M/M/N/ 0 system) and with an infinite size buffer ( C -model or M/M/N system) provided
good mathematical tools for capacity planning and performance evaluation in the classic
telephone networks for many years. Good quality of the loss probability forecasting in real
world networks based on the formulas obtained for the M/M/N/ 0 system and the delay
prediction based on the formula obtained for the M/M/N system was a rather surprising
because the requirement that inter-arrival and service times have an exponential
distribution, which is imposed in the M/M/N/ 0 and M/M/N models, seems to be too strict.
The interest of mathematicians to the fact of good matching of the calculated under
debatable assumptions characteristics to their measured value in real world systems have
lead to the following two results.
By efforts of many mathematicians (A. Ya. Khinchin and B. I. Grigelionis first of all), it was
proved that the superposition of a large number of independent flows having uniformly
small intensity approaches to the stationary Poisson input when the number of the
superposed inputs tends to infinity. It explains the fact that the flows in classic telephone
networks (where flows are composed by small individual flows from independent
subscribers) have the exponentially distributed inter-arrival times.
Concerning the service time distribution, situation was more complicated. The real-life
measurements have shown that the service (conversation) time can not be well
approximated by means of the exponentially distributed random variable. So, due to the
good matching of results obtained for the M/M/N/ 0 queue performance characteristics to
characteristics of real systems modelled by such a queue, the hypothesis has arisen that the
stationary state distribution in the M/M/N/ 0 queue is the same as the one in the M/G/N/ 0
queue conditional that the average service times in both models coincide. This property was
called the invariant (or insensitivity) property of the model with respect to the service time
distribution. The work [10] by B. A. Sevastjanov is the first one where this property was
proven strictly.
So, the question why the Erlang's models give very good results for a practice was
highlighted. The special books containing the tables for a loss probability under the given
values of the number N of channels and intensities of the input and service exist. Different
design problems (for a fixed value of permissible loss probability, to find the maximal
intensity of the flow, which can be served by the line consisting of a fixed number of
channels under the fixed average service time, or to find the necessary number of channels
sufficient for transmission of the flow with a fixed intensity, etc) are solved by means of
these tables.
However, the flows in the modern telecommunication networks have lost the nice
properties of their predecessors in the old classic networks. In opposite to the stationary
Poisson input (stationary ordinary input with no aftereffect), the modern real life flows are
non-stationary, group and correlated. The BMAP (Batch Markovian Arrival Process) arrival
process was introduced as a versatile Markovian point process ( VMPP ) by M.F. Neuts in
the 70th. The original development of VMPP contained extensive notations; however these
notations were simplified greatly in [7] and ever since this process bears the name BMAP .
The class of BMAP s includes many input flows considered previously, such as stationary
Poisson ( M ), Erlangian ( Ek ), Hyper-Markovian ( HM ), Phase-Type ( PH ), Interrupted
Poisson Process ( IPP ), Markov Modulated Poisson Process ( MMPP ). Generally speaking,
the BMAP is correlated, so it is ideal to model correlated and (or) bursty traffic in modern
telecommunication networks.
As it was mentioned above, the question why the inter-arrival times in the classical
networks have the exponential distribution was answered in literature. However, Erlang's
assumption that the service time distribution has the exponential distribution is not
supported by the real networks measurements. In the case of the M/M/N/ 0 system, good
fitting of performance measures of this system with the respective measures of real world
systems is easy explained by Sevastjanov's result. But in the case of the M/M/N system, it
was necessary to generalize results by Erlang to the cases of another, than exponential,
service time distributions. This work was started by Erlang who offered so called Erlang's
distribution. He introduced Erlangian of order k distribution as a distribution of a sum of k
independent identically exponentially distributed random variables (phases). Further, so
called phase type ( PH ) distribution was introduced into consideration as the
straightforward generalization of Erlangian distribution, see, e.g., [8]. PH distribution
includes as the special cases the exponential, Erlangian, Hyper-exponential, Coxian
distributions. In our chapter we assume that service times at the fixed operation mode of the
system have PH distribution.
It follows from discussion above that it is interesting to extend investigation of Erlang's
models to the case of the BMAP input and PH type service process. This work was started
by M. Combe in [1] and V. Klimenok in [5] where the BMAP/M/N and BMAP/M/N/ 0
models, respectively, were investigated.
In paper [2], we investigated the BMAP/PH/N/ 0 model having no buffer. It was shown
there that the stationary distribution of the system states essentially depends on the shape of
the service time distribution and so Sevastjanov's invariant property does not hold true in
the case of the general BMAP arrival process. Here we analyze the BMAP/PH/N/L system
with a finite buffer and the BMAP/PH/N system with infinite buffer. Simultaneously, we
make one more essential generalization of the model under study. Motivation of this
generalization is as follows.
Even if one will use such general models of the arrival and service process as the BMAP
and PH , he may fail in application to practical systems. The reason is the following.
Assumption that the input flow is described by the BMAP allows to take into consideration
a burstiness, an effect of correlation in the arrival process and variation of inter-arrival
times. Assumption that the service process is described by the PH distribution allows to
take into consideration variation of service times. But the BMAP arrival process and PH
service process are assumed to be stationary and independent of each other within the
borders of the models of BMAP/PH/N/L type, 0  L  . While in many real world systems
the input and service processes are not absolutely stable and may be mutually dependent.
They may be influenced by some external factors, e.g., the different level of the noise in the
transmission channel, hardware degradation and recovering, change of the distance by a
mobile user from the base station, parallel transmission of high priority information, etc.
Information transmission channel modeled by means of the BMAP/PH/N/L queueing
system can be a part of complex communication network. The rest of the network may
essentially vary characteristics of the arrival and service process in this system by means of:
(i) changing the bandwidth of the channel (due to reliability factors or the needs to provide
good quality of service in another parts of the network when congestion occurs); (ii)
changing the mean arrival rate due breakdowns, overflow or underflow of alternative
information transmission channels. Thus, to get the mathematical tool for adequate
modeling such information transmission channels, more complicated queues than the
BMAP/PH/N/L queueing system should be analyzed. These queues, in addition to the
account of complicated internal structure of the arrival and service processes by means of
considering the BMAP and PH , must take into account the influence of random external
factors. In some extent, it can be done by means of analyzing the models of queues
operating in a random environment. Such an analysis is the topic of this chapter.
Importance of investigation of the queues operating in a random environment ( RE )
drastically increased in the last years due to the following reason. The flows of information
in the modern communication networks are essentially heterogeneous. Some types of
information are very sensitive with respect to a delay and an jitter but tolerant with respect
to losses. Another ones are tolerant with respect to the delay but very sensitive with respect
to the loss of the packets. So, different schemes of the dynamic bandwidth sharing among
these types exist and are developing. They assume that, in the case of congestion,
transmission of the delay tolerant flows is temporarily postponed to provide better
conditions for transmission of the delay sensitive flows. Analysis of such schemes requires
the probabilistic analysis of the multi-dimensional processes describing transmission
process of the different flows. This analysis is often impossible due to the mathematical
complexity. In such a case, it is reasonable to decompose a simultaneous consideration of all
flows to separate analysis of the processes of transmission of the delay sensitive and the
delay tolerant flows. To this end, we model transmission of the delay sensitive flows in
terms of the queues with the controlled service or (and) arrival rate where the service or the
arrival rate can be changed depending on the queue length or the waiting time.
Redistribution of a bandwidth to avoid congestion for the delay sensitive flows causes a
variation, at random moments, of an available bandwidth for the delay tolerant flows.
Correspondingly, the queues operating in a random environment naturally arise as the
mathematical model for the delay tolerant flows transmission.
Mention that the BMAP/PH/N/ 0 model operating in the RE was recently investigated in
[4]. Short overview of the recent research of queues operating in the RE can be found there.
In this chapter, we consider the models BMAP/PH/N/L and BMAP/PH/N operating in the
RE.
2. The Mathematical Model

We consider the queueing system having N identical servers. The system behavior
depends on the state of the stochastic process (random environment) rt  t  0, which is
assumed to be an irreducible continuous time Markov chain with the state space {1,  , R} ,
R  2 , and the infinitesimal generator Q.
The input flow into the system is the following modification of the BMAP . In this input
flow, the arrival of batches is directed by the process  t  t  0, (the underlying process) with
the state space {0 , 1,… , W }. Under the fixed state r of the RE, this process behaves as an
irreducible continuous time Markov chain. Intensities of transitions of the chain  t  t  0,
which are accompanied by arrival of k -size batch, are described by the matrices Dk( r )  k  0 ,

r  1, R , with the generating function D( r ) ( z)   Dk( r ) z k  z  1. The matrix D( r ) (1) is an
k 0
irreducible generator for all r  1, R. Under the fixed state r of the random environment,
the average intensity (r ) (fundamental rate) of the BMAP is defined as
 (r )
 (r )
 D ( z)
(r )
z 1 e, and the intensity  (r )
b of batch arrivals is defined as
b( r )   ( r )  D0( r )  e. Here the row vector  (r ) is the solution to the equations
 D (1)  0 ,  e  1, e is a column vector of appropriate size consisting of 1's. The
(r ) (r ) (r )
r)
variation coefficient c (var of intervals between batch arrivals is given by
c 
(r) 2
var  2b( r ) ( r ) ( D0( r ) )1 e  1,
(r )
while the correlation coefficient ccor of intervals between successive batch arrivals is
calculated as
  b( r ) ( r ) ( D0( r ) )1 (D( r ) (1)  D0( r ) )( D0( r ) )1 e - 1    c var 
(r ) (r )2
ccor
At the epochs of the process rt  t  0, transitions, the state of the process  t  t  0, is not
changed, but the intensities of its transitions are immediately changed.
The service process is defined by the modification of the PH -type service time distribution.
Service time is interpreted as the time until the irreducible continuous time Markov chain
mt  t  0, with the state space {0 , 1,… , M  1} reaches the absorbing state M  1. Under the
fixed value r of the random environment, transitions of the chain mt  t  0, within the state
space {1,… , M} are defined by an irreducible sub-generator S( r ) while the intensities of
transition into the absorbing state are defined by the vector S(0r )  S( r )e. At the service
beginning epoch, the state of the process mt  t  0, is chosen according to the probabilistic
row vector  ( r ) , r  1, R. It is assumed that the state of the process mt  t  0, is not changed
at the epoch of the process rt  t  0, transitions. Just the exponentially distributed sojourn
time of the process mt  t  0, in the current state is re-started with a new intensity defined
by the sub-generator corresponding to the new state of the random environment rt  t  0.
The system under consideration has L, 0  L   , waiting positions. In the case of an
infinite buffer ( L =  ) all customers are always admitted to the system. In the case of a
finite L, the system behaves as follows. If the system has all servers being busy at a batch
arrival epoch, the batch looks for the available waiting position, and occupies it in case of
success. If the system has all servers and all waiting positions being busy, the batch leaves
the system forever and is considered to be lost. Due to a possibility of the batch arrivals, it
can occur that there are free servers or waiting positions in the system at an arrival epoch,
however the number of these positions is less than the number of the customers in an
arriving batch. In such situation the acceptance of the customers to the system is realized
according to the partial admission ( PA ) discipline (only a part of the batch corresponding
to the number of free servers is allowed to enter the system while the rest of the batch is
lost), the complete rejection ( CR ) discipline (a whole batch leaves the system if the number
of free servers is less than the number of customers in the batch), complete admission ( CA )
discipline (a part of the batch corresponding to a number of free servers starts the service
immediately while the rest of the batch waits for a service in the system in some special
waiting space). All these disciplines are popular in the real life systems and got a lot of
attention in the literature. Here, we consider all these disciplines.
Our aim is to calculate the stationary state distribution and main performance measures of
the described queueing model.
For the use in the sequel, let us introduce the following notation:
• en (0 n ) is a column (row) vector of size n, consisting of 1's (0's). Suffix may be

omitted if the dimension of the vector is clear from context;
•I O  is an identity (zero) matrix of appropriate dimension (when needed the
dimension of this matrix is identified with a suffix);
 
• diag ak , k  1, K is a diagonal matrix with diagonal entries or blocks ak ;
•  and  are symbols of the Kronecker product and sum of matrices;
l 1
•  l       l  1, 0  1,  l   I nm    I nlm1 , l  1, where n is the

l m0
dimension of square matrix  ;


• D( z)   diag Dk( r ) , r  1, R z k ;
k 0


• k( n )  diag Dk( r )  I M n , r  1, R , n  0, N , k  0;

•  ( n ) ( z)   k( n ) z k , n  0, N ;
k 0
• l
(n)

 diag I W  I M n    ( r )  , r  1, R , n  1, N , W  W  1;
l

•  (n)  diag I W
 S 
( r ) n

, r  1, R , n  1, N ;

•   diag S( r ) , r  1, R ; 

• 0( n )  diag I W   S(0r ) 
n

, r  1, R , n  1, N ;
 diag I  S  
(r ) N

(N ) (r )
• 0
W 0 , r  1, R ;
(n) (n) (n)
•  Q  IW  I Mn   0  , n  0, N ;

(n)
•  Q  I W  I M min{n,N}  0( min{n,N} )  
k N Ln1
k( min{n,N} )   ( min{n,N} ) , n  0, N ;

(N L)
•  Q  I W  I M N   k( N )   ( N ) .
k 0
3. Process of the System States

It is easy to see that operation of the considered queueing model is described in terms of the
regular irreducible continuous-time Markov chain
t  {nt , rt , t , mt(1) , , mt(min{ n , N }) }, t  0,

t
where
• nt is the number of customers in the system, where nt  0, N  L in case of PA
and CR disciplines and nt  0 in case of CA discipline;
• rt is the state of the random environment, rt  1, R ;
•  t is the state of the BMAP underlying process,  t  0, W ;
• mt( n ) is the phase of PH service process in the n th busy server, mt( n )  1, M ,
nt  1, N , (we assume here that the busy servers are numerated in order of their occupying,
i.e. the server, which begins the service, is appointed the maximal number among all busy
servers; when some server finishes the service, the servers are correspondingly enumerated)
at epoch t , t  0.
Let us enumerate the states of the chain t , t  0, in the lexicographic order and form the
row vectors pn of probabilities corresponding to the state n of the first component of the
process t , t  0. Denote also p   p0 , p1 , p2 , .
It is well known that the vector p satisfies the system of the linear algebraic equations (so
called equilibrium equations or Chapman-Kolmogorov equations) of the form:
p A  0 , pe  1, (1)
where A is the infinitesimal generator of the Markov chain t , t  0.

Structure of this generator and methods of system (1) solution vary depending on the
admission discipline.
3.1. The Case of Partial Admission Discipline

Lemma 1. Infinitesimal generator A of the Markov chain t , t  0, in the case of partial
admission discipline has the following block structure:
A   An , n n , n 0 , N  L 
  (0)  (0)   (0)  (0)  (0)   (0) ˆ (0) 

1,1 N  1, N  1 N ,N N  1, N N  L  1, N N L ,N
 (1) 
 0  (1)   (1)  (1)  (1)   (1) ˆ (1)
 N  L  1, N  1 
N  2 ,N 2 N  1, N  1 N ,N 1 N  L  2 ,N 1
 ( 2) (2) (2) ( 2) ˆ (2) 
 O 0   N  3, N  3  N  2 ,N  2  N  1, N  2   (N2) L  3, N  2  N L 2 ,N 2 
          
 
 O O   ( N  1)
 ( N  1)
 ( N  1)
  (LN,1 1)  ˆ ( N  1) 
1,1 2 ,1 L  1,1
 
 O O   (N )
0  (N )
 (N )
1  L( N 1) ˆL( N ) 
 
ˆL( N 1) 
(N )
 O O  O 0  (N )  L( N 2)
 
          
 (N ) (N L) 
 O O  O O O  0  
where
 (mk,)m  m( k )m( k ) , k  0, N  1, m  1, m  1, N ,
 
ˆ (k) 
 m1 , m  (k)
m , m , ˆm( k1 )   (k)
m .
m  m1 m  m1
Proof of the Lemma follows from analysis of Markov chain t , t  0, transitions during an
infinitesimal interval. Block entries of the generator have the following meaning. The non-
diagonal entries of the matrix  ( k ) define intensity of transition of the components
{rt , t , mt(1) , , mt( nt ) } of the Markov chain t , t  0, which do not lead to the change of the
number k of busy servers. The diagonal entries of the matrix  ( k ) are negative and define,
up to the sign, intensity of leaving the corresponding states of the Markov chain t , t  0.
The entries of the matrix  (mk,)m  m( k )m( k ) define intensity of transitions of the components
{rt , t , mt(1) , , mt( nt ) } of the Markov chain t , t  0, which are accompanied by arrival of m
customers and occupying m servers conditional the number of busy servers is k. The
entries of the matrix 0( k ) define intensity of transitions, which are accompanied by a
departure of a customer, conditional the number of busy servers is k.
To solve system (1) with the matrix A defined by Lemma 1, we use the effective
numerically stable procedure developed in [2] that exploits the special structure of the
matrix A (it is upper block Hessenberg) and probabilistic meaning of the unknown vector
p. This procedure is given by the following statement.
Theorem 1. In case of partial admission, the stationary probability vectors pi , i  0, N  L ,
are computed as follows:
pl  p0Fl , l  1, N  L ,
where the matrices Fl are calculated recurrently:
 l 1

 
1
Fl   A0 , l   Fi Ai , l   Al , l , l  1, N  L  1,
 i 1 
 N L1

FN  L   A0 , N  L   Fi Ai , N  L    AN  L , N  L  ,
1
 i 1 
the matrices AI , N  L are calculated from the backward recursions:
AI , N  L  Ai , N  L , i  0, N  L ,
Ai , l  Ai , l  Ai , l  1Gl , i  0, l , l  N  L  1, N  L  2, ,0,
the matrices Gi , i  0, N  L  1, are calculated from the backward recursion:
1
 N Li 1

Gi    Ai  1, i  1   Ai  1, i  1  lGi  lGi  l  1 Gi  1  Ai  1, i ,
 l 1 
i  N  L  1, N  L  2, ,0,
the vector p0 is calculated as the unique solution to the following system of linear algebraic
equations:
 N L 
p0 A0 ,0  0, p0   Fl e  e   1.
 l 1 
3.2. The Case of Complete Rejection Discipline

Lemma 2. Infinitesimal generator A of the Markov chain t , t  0, in the case of complete
rejection discipline has the following block structure:
A   An , n n , n 0, N  L 
  (0)  (0)   (0)  (0)  (0)   (0)  (0) 

 1,1 N  1, N  1 N ,N N  1, N N  L  1, N N L,N

  (1) 
(1)
  (1)  (1)  (1)   (1)  (1) 
 0 N 2 ,N 2 N  1, N  1 N ,N 1 N L2 ,N 1 N  L  1, N  1

 O 0( 2 )   (N2) 3, N  3  (N2) 2 , N  2  N  1, N  2   (N2) L  3, N  2
(2) ( 2)
 N L2 ,N 2 
 
          
 ( N  1) N  1) 
 O O    (1,1  (2N,1 1)   (LN,1 1)  (LN 1,1
 1)
.
 (N ) 
 O O  0( N )  1( N )  L( N 1) L (N )

 (N ) (N ) 
 O O  O 0   L( N 2) L( N 1) 
          
 
 O O  O O O  0
(N )

(N L)

 
The proof of the Lemma is analogous to the proof of the previous Lemma and takes into
account the fact that the number of customers in the system does not change when the
number of customers in an arriving batch exceeds the number of free servers.
To solve system (1) with the matrix A defined by Lemma 2, we also use the procedure
described by Theorem 1.
3.3. The Case of Complete Admission Discipline

Lemma 3. Infinitesimal generator A of the Markov chain t , t  0, in the case of complete
admission discipline has the following block structure:
A   An , n n , n 0 
  (0)  (0)
1,1   (0)
N ,N  (0)
N  1, N   (0)
N  L  1, N  (0)
N L ,N  (0)
N  L  1, N 
 (1) (1) (1) (1) (1) (1) (1) 
 0    N  1, N  1  N ,N 1   N  L  2 ,N 1  N  L  1, N  1  N L ,N 1 
 O 0( 2 )   (2)
 ( 2)
  (2)
 (2)
 ( 2)

N 2 ,N 2 N  1, N  2 N  L  3, N  2 N L 2 ,N 2 N  L  1, N  2
 
          
 O O  N  1)
 (1,1  (2N,1 1)   (LN,1 1)  (LN 1,1
 1)
 (LN 2,11)  

 O O   (N ) 1( N )  L( N 1) L( N ) L( N 1) 
 (N )
.
 O O  0  (N )  L( N 2) L( N 1) L( N ) 
 
          
 O O  O O   (N ) 1( N ) 2( N ) 
 
(N ) (N L)
 O O  O O  0  O 
 
(N ) (N L)
 O O  O O  O  0  
 
           
Essential difference of complete admission discipline is that the state space of the Markov
chain t , t  0, is infinite and this makes its analysis more complicated. However, the block
rows, except the first N  L boundary block rows, have only two non-zero blocks and this
Markov chain behaves as Quasi-Death process when the state of the first component nt of
the Markov chain t , t  0, if greater than N  L. It allows to construct effective stable
algorithm for calculation of the stationary distribution of this Markov chain. Note, that
although the state space of the Markov chain t , t  0, is infinite, this Markov chain is
ergodic under the standard assumptions about the parameters of the BMAP input, the PH
type service and the random environment. The algorithm for calculation of the stationary
distribution is given in the following statement.
Theorem 2. In case of complete admission discipline, the stationary probability vectors
pl , l  0, are calculated as follows:
pl  p0 Fl , l  1,
where the matrices Fl are calculated recurrently
 l 1

 
1
Fl   A0 , l   Fi Ai , l   Al , l , l  1,
 i 1 
the matrices Ai , I are calculated as:

Ai , l  Ai , l   Ai , l  k G max{0 , l  k  N  L }Gmin{ N  L , l  k }  1Gmin{ N  L , l  k }  2 Gl , i  0, , l , l  1,
k 1
the matrix G has a form

G   
( N  L ) 1 (N )
0 ,
the matrices Gi , i  0, N  L  1, are calculated from the backward recursion
1
 

Gi    Ai  1, i  1   Ai  1, lG max{0 , l  N  L }  Gmin{ N  L , l }  1  Gmin{ N  L , l }  2    Gi  1  Ai  1, i ,
 li 2 
the vector p0 is the unique solution of the system:
  
p0 A0 ,0  0, p0   Fl e  e   1.
 l1 
The proof of the Theorem follows from the theory of multi-dimensional Markov chains with
continuous time, see [6]. It is worth to note that Neuts' matrix G , which is usually found
numerically as solution to matrix equation, see [9], here is obtained in the explicit form.
3.4. The Case of an Infinite Size of a Buffer  L= 

The system under consideration in this section has an infinite waiting space. If an arriving
batch of customers sees idle servers, a part of the batch corresponding to the number of free
servers occupy these servers while the rest of the batch joins the queue. If the system has all
servers being busy at a batch arrival epoch, all customer of the batch go to the queue.
Lemma 3. Infinitesimal generator A of the Markov chain t , t  0, has the following block
structure:
A   An , n n , n 0  (2)
  (0 )  (0
1,1
)
 (02 ,2)   (0 )
N  1, N  1  (0 )
N ,N  (0)
N  1, N  (0 )
N  2 ,N 
 (1) (1) 
 0   (1)
1,1
(1)
  N  2 ,N 2 (1)
 N  1, N  1  (1)
N ,N 1
(1)
 N  1, N  1 
 O 0( 2 )  (2)   (N2 ) 3, N  3  (N2 ) 2 , N  2  (N2 ) 1, N  2  (N2 ,)N  2 
 
         
  ( N  1)  (1,1 N  1)
 (2N,1 1) N  1)
 (3,1   .
 O O O 
 O O O  0( N ) (N ) 1( N ) 2( N ) 
 (N )

 O O O  O  0  (N ) 1( N ) 
 (N )

 O O O  O O 0  (N ) 
 
         
In what follows we perform the steady state analysis of the Markov chain having generator
of form (2). To this end, we use the results for continuous time multi-dimensional Markov
chain ( QTMC ) presented in [6].
Theorem 3. The necessary and sufficient condition for existence of the Markov chain
t , t  0, stationary distribution is the fulfillment of the inequality
   /   1, (3)
where
  x 1D' ( z) z1 e, (4)
  x 2 diag  S 0 
(r ) N
, r  1, R e , 
the vectors x n , n  1, 2, are the unique solutions to the following systems of linear algebraic
equations:
x 1  Q  I W  D(1)  0 , x 1e  1, (5)
 
x 2  Q  I M N  diag S( r )  S(0r )  ( r ) 
N

, r  1, R   0 , x 2 e  1.

(6)
Proof. Using the results of [6], we directly obtain the desired condition in the form of
inequality
x  ( N ) z  z ( N ) ( z) e  0, (7)

z1
where x is the unique solution to the system

x Q  I W M N   ( N ) (1)   ( N )   0
(N )
  0, xe  1. (8)
It is easy to show that inequality (7) is reduced to the following inequality:
x 1D' ( z) z1 
e  x 2 diag S(0r ) 
N
, r  1, R e ,  (9)
where x 1  x  I R W  e M N  , x 2  x  I R  e W  I M N  .
To get the equations for the row vectors x 1 and x 2 , we multiply equation (8) by the
matrices I RW  e M N and I R  e W  I M N respectively. After multiplication and some algebra
we obtain equations (5), (6) for the vectors x 1 and x 2 . So, inequality (9) is equivalent to
inequality (3) and the theorem is proved.
The value  has a meaning of the system load. In what follows we assume inequality (3) be
fulfilled.
To solve system (1) with the matrix A defined by (2), we use the effective numerically
stable procedure [6] based on the account special structure of the matrix A , notion of the
censored Markov chain and probabilistic meaning of the unknown vector p. For more
detail see [6]. This procedure is given by the following statement.
Theorem 4. The stationary probability vectors pl , l  0, are calculated as follows:
pl  p0 Fl , l  1,
where the matrices Fl are calculated recurrently:
 l 1

 
1
Fl   A0 , l   Fi Ai , l   Al , l , l  1,
 i 1 
the matrices Ai , I are calculated as:
Ai , l  Ai , l  Ai , l  1Gl , 0  i  l  N ,
Ai , l  Ai , l  Ai , l  1G , l  max i , N , i  0,
the matrix G is calculated from the equation
1
  
G    AN  1, N  lG l  1  AN  1, N ,
 l1 
the matrices Gi , i  0, N  1, are calculated from the backward recursion:
1
 

Gi    Ai  1, i  1   Ai  1, lG max{0 , l  N }  Gmin{ N , l }  1  Gmin{ N , l }  2    Gi  1  Ai  1, i ,
 li 2 
the vector p0 is calculated as the unique solution to the following system of linear algebraic
equations:
  
p0 A0 ,0  0 , p0   Fl e  e   1.
 l1 
4. Performance Measures
Having the probability vector p been computed, we are able to calculate performance
measures of the considered model. The main performance measure in the case of a finite
buffer is the probability Ploss that an arbitrary customer will be lost (the loss probability).
Theorem 5. The loss probability Ploss is calculated as follows
(i) in the case of PA discipline
1 N L1 N Li
Ploss  1 

i 0
pi 
k 0
( k  i  N  L )k( i )e , (10)
(ii) in the case of CR discipline
1 N L 1 N Li
Ploss  1 

i 0
pi 
k 0
kk( i )e , (11)
(iii) in the case of CA discipline
1 N L 1 
Ploss  1 

 p  k
i 0
i
k 0
(i )
k e. (12)
Proofs of formulae (10) - (12) are analogous. So, we will prove only formula (10). According
to a formula of the total probability, the probability Ploss is calculated as
N L1 
Ploss  1   P P
i 0 k 1
k i
(k)
R( i , k ) (13)
where Pk is a probability that an arbitrary customer arrives in a batch consisting of k

customers; Pi( k ) is a probability to see i servers being busy at the epoch of the k  size batch
arrival; R( i , k ) is a probability that an arbitrary customer will not be lost conditional it arrives
in a batch consisting of k customers and i servers are busy at the arrival epoch.
It can be shown that

pi k( i )e
Pi( k )  , i  0, N  L  1, k  1, (14)
x 1k(0)e
kx 1 k(0)e x 1 k(0)e
Pk   k , k  1, (15)
x 1D( z) 'z  1 e 
1, k  N  L  i,

R( i , k )   N  L  i (16)
 , k  N  L  i , i  0, N  L  1.
k
By substituting (14)-(16) into (13) after some algebra we get (10). 

Some performance measures for the case L   are presented below.
• The probability to see i customers in the system
pi  pi e , i  0;
• The mean number of customers in the system

Lqueue   ipi e ;
i 0
• The probability to see n busy servers

pn  pn e , n  0, N  1, pN   p e;
nN
n
• The mean number of busy servers
N 
N busy   npn e  N  pn e ;
n1 nN
• The mean number N idle of idle servers
N idle  N  N busy ;
 
•The vector pˆ n whose W (r  1)    1 -th entry is the joint probability to see n
busy servers, the random environment in the state r and the process  t in the state 

pˆ n  pn  I RW  e M n  , n  0, N  1, pˆ N   p In RW
 e Mn ;
nN
•The vector of conditional means of the number of busy servers under the fixed
states of the random environment
 

n   min n , N pˆ n  I R  e W  diag qr1 , r  1, R ;
n1
 
•The vector p( a ) (n) whose W (r  1)    1 -th entry is the joint probability that
an arbitrary arriving call sees n busy servers and the random environment in the state r
and the state of the process  t becomes  after the arrival epoch
p( a ) (n)   1pˆ nD' ( z) z1 , n  0, N ;
• The probability p( a ) ( n) that an arbitrary arriving call sees n busy servers
p( a ) ( n)  p( a ) (n) e , n  0, N ;
 
•The vector p(ba ) (n) whose W (r  1)    1 -th entry is the joint probability that
an arbitrary arriving batch of size k sees n busy servers and the random environment in
the state r and the state of the process  t becomes  after the arrival epoch
 
p(ba ) ( k , n)  b1pˆ n diag Dk( r ) , r  1, R , n  0, N , k  1,
where b  x  D(1)  D(0) e ;
• The probability pb( a ) ( k , n) that an arbitrary arriving batch of size k sees n busy
servers
pb( a ) ( k , n)  p(ba ) ( k , n) e , n  0, N , k  1;
 
•The vector p(ba ) (n) whose W (r  1)    1 -th entry is the joint probability that
an arbitrary arriving batch sees n busy servers and the random environment in the state r
and the state of the process  t becomes  after the arrival epoch
p(ba ) (n)  b1pˆ n  D(1)  D(0)  , n  0, N ;
• The probability pb( a ) ( n) that an arbitrary arriving batch sees n busy servers
pb( a ) ( n)  p(ba ) (n) e , n  0, N ;
•The probability Pimm that an arbitrary customer will enter the service
immediately upon arrival (without visiting a buffer)
N 1 N i
Pimm   1  pi  ( k  i  N )k( i )e.
i 0 k 0
5. Actual Sojourn Time

Let  a (s ), Re s  0, be the Laplace-Stieltjes transform ( LST ) of the sojourn time distribution
and  a be the mean sojourn time of the arbitrary customer in the system.
Theorem 6. The Laplace-Stieltjes transform  a (s ) is calculated as follows
1 N  1 
 a (s )    pi  min k , N  i k  I R  e Wmi  
(i )
(17)
  i 0 k 1
  k 
  pi  k(min{ i , N }) (max{0 , N  i })    (s)
iN l
I R  e M N   (s )e R ,
i 0 k  max{1, N  i  1} l  max{1, N  i  1} 
where
 
(s )   diag  ( r ) , r  1, R  sI  (Q  I M   ) diag S(0r ) , r  1, R ,
1
 
    
1
 (s )    sI  Q  I M N  diag  S( r )  , r  1, R  diag  S(0r )  ( r )  , r  1, R ,
N N
 

 ( n )  diag e W  I M N n    ( r ) 
n

, r  1, R , n  0, N ,

  diag S( r ) , r  1, R . 
Proof. We derive the expression for the LST  a (s ) by means of the method of collective
marks (method of additional event, method of catastrophes) for references see, e.g. [3], [11].
To this end, we interpret the variable s as the intensity of some virtual stationary Poisson
flow of catastrophes. So,  a (s ) has the meaning of probability that no one catastrophe
arrives during the sojourn time of an arbitrary customer. Then, the proof of the theorem
follows from the formula of total probability if we analyze the states of the system at an
arbitrary customer arrival epoch and take into account the probabilistic meaning of the
involved matrices. The matrix (s ) is the matrix LST of an arbitrary customer service time
distribution. It is the R -size square matrix whose ( r , r ) entry is a probability that during
the service time of a customer a catastrophe does not arrive and the process rt  t  0, transits
from the state r to the state r , r , r   1, R. It is defined by the formula:

(s )  diag  ( r ) , r  1, R e  st ( Q  I M  diag { S( r ) , r  1, R }) t
e 
dt diag S(0r ) , r  1, R . 
0
Analogously, the entries of the matrix LST  (s ) are the probabilities of no catastrophe
arrival and corresponding transitions of the process {rt , mt(1) , , mt( N ) }, t  0, during the time
interval from an arbitrary moment when all N servers are busy till the first epoch when one
of these servers finishes the service of a customer. This matrix is defined by the formula:
 


(Q  I  diag {( S( r ) ) N , r  1, R })t N
 (s )   e  st e MN
dt diag S(0r )  ( r ) , r  1, R .
0
Theorem 7. The mean sojourn time  a of an arbitrary customer in the system is calculated
by
1   k i lN 1
 a     pi      (0)  (0) I R  e M 
m
k(min{ i , N }) (max{0 , N  i }) N
  i  0 k  max{1, N  i  1} l  max{1, N  i  1} m  0
N 1 
min{ k , N  i} k( i )  I R  e W M i  (0) 
  pi  � (18)
i 0 k 1
  k 
  pi  k(min{ i , N }) (max{0, N  i })    (0)
i N l
�
I R  e M   (0) e R

i 0 k  max{1, N  i  1} l  max{1, N  i  1}
where
 
(0)   diag  ( r ) , r  1, R Q  I M    diag S(0r ) , r  1, R ,
2
 
   
2
 (0)   Q  I M N  diag  S( r )  , r  1, R  diag  S(0r )  ( r ) 
N N
, r  1, R .
 
Proof. To get expression (18) for  a we differentiate (17) at the point s  0 and use the
'
formula  a   a (0).
6. Numerical Examples
The goal of the numerical experiments is to demonstrate the feasibility of the proposed
algorithms for computing the stationary distributions of the number of customers and the
sojourn time in the system and to give some insight into behavior of the considered
queueing systems. In particular, the following issues are addressed:
• Comparison of the mean sojourn time of an arbitrary customer and the
probability of immediate access to the servers in the systems with varying traffic intensities
and different coefficients of correlation in the BMAP s (experiment #1 );
• Comparison of the mean sojourn time of an arbitrary customer and the
probability of immediate access to the servers in the original system in a RE and in more
simple queueing systems for different system loads (experiments #2 );
• Demonstration of possible positive effect of redistribution of traffic between
the peak traffic periods and normal traffic periods (experiment #3 );
• Comparison of the exact value of performance measures of the system in a RE
and their simple engineering approximations in cases of slowly and quickly varying RE
(experiment # 4 );
• Investigation of the rate of convergence of the mean sojourn time and the
probability of immediate access in the system with the finite buffer to corresponding
performance measures of the system with an infinite buffer when the buffer size increases
(experiment #5 );
• Demonstration of the possibility to apply the presented results for
optimization of the number of servers in the system (experiment #6 ).
In numerical examples, we consider the systems operating in the RE which has two states
 5 5 
( R  2 ). The generator of the random environment is Q    . The stationary
 15 15 
distribution of the RE states is defined by the vector q   0.75, 0.25  . The number of servers
is N  3.
In the presented examples, we will use several different MAP s and BMAP s for description
of the arrival process and two PH type distributions for description of the service processes
under the fixed value of the RE. For the use in the sequel, let us define these processes.
We consider four arrival processes MAPr , r  1, 4. MAPr is defined by the matrices D0( r ) ,
D1( r ) , r  1, 4, where
 3.9 0.15 0.15   3.5 0.08 0.02 
   
D0(1)   0.13 0.6 0.1  , D1(1)   0.03 0.3 0.04  ;
 0.15 0.14 0.5   0.02 0.06 0.13 
 
 6.4 0.1 0.1   6.06 0.12 0.02 
   
D0( 2 )   0.04 0.6 0.1  , D1( 2)   0.01 0.4 0.05  ;
 0.07 0.1 0.44   0.01 0.06 0.2 
 
 2.9 0.73 0.77   0.68 0.45 0.27 
( 3)   ( 3)  
D0   0.87 3.06 0.53  , D1   0.48 1.08 0.1  ;
 0.85 0.5 2    
  0.35 0.05 0.25 
 1.3 0.21 0.17   0.46 0.32 0.14 
( 4)   ( 4)  
D0   0.16 2.04 0.21  , D1   0.13 1.34 0.2  .
 0.01 0.16 1.3   0.02 0.01 1.1 
 
All these MAP s have fundamental rate  ( r )  1.25. The MAP1 has the squared variation
coefficient  c (1)
var   4 and the coefficient of correlation of the lengths of successive inter-
2
(1)
arrival times ccor  0.2. For the rest of the MAP s, the corresponding parameters are:
c 
( 2) 2
var
(2)
 4, ccor  0.3; c 
( 3) 2
var
( 3)
 1.097, ccor  0.0052; c 
( 4) 2
var
( 4)
 1.037, ccor  0.0065.
Based on these MAP s, we construct batch flows BMAP s as follows. If the MAP is defined
by the matrices D0( r ) , and D1( r ) , r  1, 4, then the BMAP having the maximal size of a batch
equal to K is defined by the matrices D0( r ) , Dk( r )  D1( r )q k  1 (1  q )/(1  q K ), k  1, K , r  1, R ,
where q  0.9.
Following this way, we construct the BMAP1 , BMAP2 , BMAP3 , BMAP4 flows based on
the MAP1 , MAP2 , MAP3 , MAP4 correspondingly, with K  5. Note that the coefficients of
variation and correlation of all BMAP s are the same as these coefficients for the
(r )
corresponding MAP s. Fundamental rate  ( r ) and the mean batch size k of the BMAP s
(1) ( 2) ( 3) ( 4)
are the following:  (1)
 ( 2)
 ( 3)
 ( 4)
 3.488, k k k k  1.989.
The PH r , r  1, 2, service processes are defined by the vectors  (1)
  1, 0  ,  ( 2)   0.2, 0.8 
and the matrices
 4 4   10 2 
S(1)    , S( 2)   .
 0 4   2 20 
The mean rates of service are  (1)  2,  ( 2 )  14. The coefficients of variation of the service
time distribution are defined by  c (1)
var   0.5,  c var   1.24.
(2) 2 2
In the first experiment, we compare the dependence of  a and Pimm on the system load 
for the BMAP s with different correlations.
In the experiment we use service processes defined by PH 1 and PH 2 and four different
input flows which are described by BMAP1 , BMAP2 , BMAP3 and BMAP4 having the same
mean fundamental rate equal to 3.488 but different correlation coefficients.
We consider three queueing systems which have different combinations of the BMAP s
under the first and second states of the RE.
The input flow in the first system is defined by BMAP1 and BMAP2 . These BMAP s have
(1) (2)
large coefficients of correlation ccor  0.2 and ccor  0.3.
The input flow in the second system is defined by BMAP3 and BMAP4 . These BMAP s
( 3) ( 4)
have small coefficients of correlation ccor  0.0052 and ccor  0.0065.
In the third system the input is defined by BMAP1 and BMAP4 . The correlation coefficients
of these BMAP s differ significantly.
Figures 1 and 2 show the dependence of the mean sojourn time  a and the probability Pimm
on the system load  for all these systems. Variation of the value of  in all experiments is
performed by means of multiplying the entries of the matrices, which define the
corresponding BMAP , by some varying factor  . This implies the increase of the
fundamental rate of all the BMAP by a factor  . Service time distributions are not
modified. It is clear from Figure 1 that correlation in BMAP has a great impact on the
sojourn time in the system. An increase of correlation at least in one of the BMAP s
describing input in the system implies an increase of the sojourn time in the system in all
range of the system load.
Fig. 1. Mean sojourn time in the system as a function of the system load for different
correlations in the BMAP s
Fig. 2. Probability of immediate access to the servers in the system as a function of the
system load for different correlations in the BMAP s
In the second experiment we compare the values  a and Pimm in the BMAP / PH / N
system operating in the RE and in more simple queueing systems which can be considered
as its simplified analogs. The first type analog is the M X / PH / N system in the RE where,
under the fixed value of the RE , the input flow is a group stationary Poisson with the same
batch size distribution and intensity equal to fundamental rate of the corresponding BMAP
in the original system. The second type analog is the system M X / M / N with parameters of
arrival and service processes which are obtained by means of averaging, according to
stationary distribution of the RE , parameters of the original system.
Input flow is described by BMAP1 and BMAP2 . Service processes are PH 1 and PH 2 .
Figures 3 and 4 show the dependence of the the mean sojourn time  a and the probability
Pimm on the value of  .
Fig. 3. Mean sojourn time in original system and more simple queueing systems
Fig. 4. Probability of immediate access to the servers in original system and more simple
queueing systems
It can be seen from Figures 3 and 4 that an approximation of the mean sojourn time and the
probability that an arbitrary call reaches the server immediately by means of their values in
some specially constructed more simple queueing system can be rather bad.
The idea of the third experiment is the following. Let us assume that the RE has two states.
One state corresponds to the peak traffic periods, the second one corresponds to the normal
traffic periods. Service times during these periods are defined by PH 1 and PH 2
distributions. Arrivals during these periods are defined by the stationary Poisson flow with
the rates 1 and 2 correspondingly and initially we assume that 1  2 . It is intuitively
clear that if it is possible to redistribute the arrival processes (i.e., to reduce the arrival rate
during the peak periods and to increase it correspondingly during the normal traffic
periods) without changing the total average arrival rate, the mean sojourn time in the
system can be reduced. In real life system such a redistribution is sometimes possible, e.g.,
by means of controlling tariffs during the peak traffic periods. The goal of this experiment is
to show that this intuitive consideration is correct and to illustrate the effect of the
redistribution.
We assume that the averaged arrival rate  should be 12.5 and consider four different
situations: a huge difference of arrival rates 1  502 , a very big difference 1  102 , a big
difference 1  32 and equal arrival rates 1  2 . The generator of the random
 15 15 
environment is Q   .
 5 5 
It can be seen from Figures 5 and 6 that the smoothing of the peak rates can cause essential
decrease of the mean sojourn time and the increase of the probability that an arbitrary call
reaches the server immediately upon arrival in the system.
Fig. 5. Mean sojourn time as a function of system load for different relations of arrival rates
Fig. 6. Probability of immediate access to the servers as a function of system load for
different relations of arrival rates
In the second experiment, we have seen that an approximation of the system performance
measures by means of their values in more simple queueing system can be bad. However, it
is intuitively clear the following. If the random environment is "very slow" (the rate of the
RE is much less then the rates of the input flow and the service processes), an
approximation called below as "mixed system" can be applied successfully. This
approximation consists of calculation of the system characteristics under the fixed states of
the RE and their averaging by the RE distribution. If the random environment is "very fast",
approximation called below as "mixed parameters" can be successfully applied. This
approximation consists of averaging parameters of the arrival and service processes by the
distribution of the RE and calculation of performance measures in BMAP / PH / N system
with the averaged arrival and service rates.
In the fourth experiment, we show numerically that sometimes the described
approximations make sense. However, in situations when environment is neither "very
slow" nor "very fast", these approximations can be very poor. We consider the RE s with
 5 5 
different rate which are characterized by the generators of the form Q( k )   k
  10 .
 15 15 
We vary the parameter k from -7 to 4 what corresponds to the variation of the RE rate
from "very slow" to "very fast". In this and further experiments, the input flow is described
by the BMAP1 and BMAP2 and the service process is defined by the PH 1 and PH 2 . The
results are presented in Figures 7, 8 and 9. In application of "mixed system" approximation,
the averaged arrival rates under both states of the RE are equal to 3.488. The averaged
service rate is equal to 2 at the first state of the RE and is equal to 14 at the second state. The
mean sojourn times of an arbitrary customer at these states are equal to 3.5297 and 0.0998,
respectively; the probabilities of immediate access to the servers are equal to 0.2021 and
0.81399; the mean numbers of customers in the system are equal to 12.311 and 0.3482. The
averaged, according to the stationary distribution of the RE, mean sojourn time of an
arbitrary customer is equal to 2.6722 and the probability that an arbitrary arriving customer
sees an idle server is equal to 0.355. In application of "mixed parameters" approximation, the
averaged, according to the stationary distribution of the RE, arrival rate is equal to 3.778
while the averaged service rate is equal to 4.0625. The value of the mean sojourn time of an
arbitrary customer in the system with averaged arrival and service rates is equal to 0.7541,
the probability of immediate access to the servers is equal to 0.4168, the mean number of
customers in the system is equal to 2.849.
Figures 7, 8 and 9 confirm the hypothesis that the first type approximation ("mixed system")
is good in case of "very slow" RE and the second one ("mixed parameters") can be applied
to case of "very fast" RE. But sometimes the second type approximation is not very good
(see Figures 8) because it is not quite clear how to make averaging of service intensity.
Simple averaging of service rates under the different states of the RE may be not correct
when the load of the system is not high because there are time intervals when the system is
empty and no service is provided. It is worth to note also that there is an interval for RE
rate (interval k   3, 0  ) where one should not use the values of the system performance
measures calculated based on the considered approximating models. The use of these values
can lead to the large relative error. Thus, Figures 7, 8 and 9 confirm the importance of
investigation implemented in this chapter. Simple engineering approximations can lead to
unsatisfactory performance evaluation and capacity planning in real world systems.
Fig. 7. Mean sojourn time of an arbitrary customer as a function of the RE rate
Fig. 8. Probability of immediate access to the servers as a function of the RE rate
Fig. 9. The mean number of customers in the system Lqueue as a function of the RE rate
In the fifth experiment we compare the mean number of customers, probability of

immediate access to the servers and loss probability in the BMAP / PH / N and
BMAP / PH / N /L systems operating in the RE for different values L of the buffer

capacity and different customers admission discipline.
Fig. 10. Mean number of customers in the system as a function of the buffer capacity L
Fig. 11. Probability of immediate access to the servers as a function of the buffer capacity L
Fig. 12. Loss probability as a function of the buffer capacity L

Looking at Figures 10-12, it should be noted that the rate of convergence of the curves
corresponding to the disciplines PA and CR to their limits defined by the system with an
infinite buffer is not very high. When we further increase the value L , we discover that even
for the buffer capacity L about 5000, the difference is not negligible. So, estimation of
performance measures of the system with an infinite buffer by the respective measures of
the system with a finite buffer can be not very good. This explains why we made the
separate analysis of the system with an infinite buffer.
Finally, in the sixth experiment we consider the next optimization problem:
J  N   c1 a  c 2 N  max , (19)

N
where
•  a is the mean sojourn time in the system,
• N is the number of servers,
• c1 is the charge for an unit of customer sojourn time in the system,
• c 2 is the cost of a server maintenance per unit of time.
It is clear that this problem is not trivial. When the number of servers is small, the cost of
servers maintenance is also small, but the mean sojourn time is large. If we increase the
number of servers, the mean sojourn time decreases while the cost of servers maintenance
increases.
Let us assume that the cost coefficients be fixed as c1  5 and c 2  3. Service time
distribution at both states of the RE is exponential with intensities  (1)  1,  ( 2 )  7.
On Figure 13, dependence of the cost criterion J  N  on N is presented along with the
dependences of the summands c 1 a and c 2 N .
Fig. 13. Criterion J  N  as a function of number of servers in the system

Based on Figure 13, one can conclude that our analysis allows effectively solve the problems
of the system design and that the optimal value of the cost criterion (in this example it is
provided by N  4 ) can be significantly smaller than the values of the cost criterion for
other values on N .
7. Conclusion
The BMAP / PH / N /L system operating in a finite state space Markovian random
environment is investigated for the finite and infinite buffer capacity. The joint stationary
distribution of the number of the customers in the system, the state of the random
environment, and the states of the underlying processes of arrival and service processes is
calculated. The analytic formulas for performance measures of the system are derived. The
Laplace-Stieltjes transform of sojourn time distribution is derived and the mean sojourn time
is calculated. Selected results of numerical study are presented. They show an impact of the
correlation in arrival process, illustrate the poor quality of the system characteristics
approximation by means of more simple models, confirm the positive effect of the traffic
redistribution between the peak and normal operation periods. The results can be used for
the optimal design, capacity planning, and performance evaluation of real world systems in
which operation of the system can be changed depending on some external factors.
Acknowledgement
This work was supported by the Korea Research Foundation Grant Funded by the Korean
Government (MOEHRD)(KRF-2008-313-D01211).
8. References
Combe, M. (1994) Queueing models with dependence structure. Amsterdam: CWI, p. 165.
Kim, C.S., Klimenok, V.I., Orlovsky, D.S., Dudin, A.N. (2005) Lack of invariant property of
the Erlang loss model in case of MAP input. Queueing Systems, Vol. 49. pp. 187-213.
Kasten, H., Runnenburg, J.Th. (1956) Priority in waiting line problems. Mathematisch
Centrum, Amsterdam, Holland. Dec. 1956.
Kim, C.S., Dudin, A.N, Klimenok, V.I., Khramova, V.V. (2009) Erlang loss queueing system
with batch arrivals operating in a random environment. Computers and Operations
Research, Vol. 36, pp. 674-967.
Klimenok, V.I. (1999) Characteristics calculation for multi-server queue with losses and
bursty traffic. Automatic Control and Computer Sciences, Vol. 12, pp. 393-415.
Klimenok, V.I., Dudin, A.N. (2006) Multi-dimensional asymptotically quasi-Toeplitz
Markov chains and their application in queueing theory. Queueing Systems, Vol. 54
pp. 245-259.
Lucantoni, D. (1991) New results on the single server queue with a batch Markovian arrival
process. Comm. Stat. Stochastic Models, Vol. 7: pp. 1-46.
Nuets, M.F. (1981) Matrix-Geometric Solutions in Stochastic Models. John Hopkins Univ.
Press, Baltimore, 1981.
Neuts, M.F. (1989) Structured Stochastic Matrices of M/G/1 Type and Their
Applications. Marcel Dekker, New York, 1989.
Sevastjanov, B.A. (1959) Erlang formula in telephone systems under the arbitrary
distribution of conversations duration. In: Proc. Of 3rd All-Union Mathematical
Congress (Academy of Science, Moscow), 1959, Vol. 4, pp. 68-70.
Van Danzig, D. (1955) Chaines de Markof dans les ensembles abstraits applications aux
processus avec regions absorbantes et au problem des boucles. Ann. De l’Inst. H.
Pioncare 14 (fasc. 3), pp. 145-199.
Interdomain QoS paths finding based on overlay topology and QoS negotiation approach 345
16
X
Interdomain QoS paths finding based on

overlay topology and QoS negotiation approach
Şerban Georgică Obreja and Eugen Borcoci
University Politehnica Bucharest
Romania
1. Introduction
The real time multimedia services, delivered on Internet networks, raised new challenges for
the network regarding the end to end (E2E) quality of services (QoS) control in order to
ensure the proper delivery of the services from content provider (source) to content
consumer (destination). But, despite a lot of studies and research done, the actual traffic
processing in real Internet deployments is still mostly best effort. Several approaches have
been proposed, focused on provisioning aspects – usually solved in the management plane -
and then performing monitoring and adjustments in the control plane: e.g., well known
dynamic techniques have been standardized, like IntServ, Diffserv, or combinations.
Offering multimedia services in multi-domain heterogeneous environments is an additional
challenge at network/ transport level. Service management is important here for
provisioning, offering, handling, and fulfilling variety of services. Appropriate means are
needed to enable a large number of providers in order to extend their QoS offerings over
multiple domains. To this aim, an integrated management system can be a solution to
preserve each domain independency while offering integration at a higher (overlay) layer in
order to achieve E2E controllable behaviour.
This chapter deals with the problem of establishing QoS enabled aggregated multi-domain
paths, to be later used for many individual streams. A general framework is described
exposing the ideas of overlay topologies solutions. Then a simple but extendable procedure
is proposed, running at management level, to find (through communication between
domain managers) several potential inter-domain end to end paths. Then, using a resource
negotiation process performed also in the management plane, QoS enabled aggregated
pipes, spanning several IP domains, are established. All these functions are performed at an
overlay level, based on abstract characterization of intra and inter-domain capabilities
delivered by an intra-domain resource manager. This is important in the sense that each
domain (or, Autonomous System – AS) can preserve its own independency in terms of
resource management. The subsystem is part of an integrated management multi-domain
system, dedicated to end to end distribution of multimedia streams.
The QoS path finding solution presented here is not like a traditional routing process: it is
not implemented on routers, and it does not choose a route between network devices, but
between two or more nodes of an overlay virtual topology described at inter-domain level.
Together with the intra-domain QoS routing available inside each network domain we will
obtain an E2E QoS routing solution.
We recall that in our context, QoS enabled aggregated pipes are established (at request of a
Service Provider entity), in advanced to the real traffic flow transportation. These are mid-
long term virtual links. Related to this, the main advantage of the proposed solution is that,
by separating the process of path finding from the QoS negotiation, the path searching
process doesn’t need to work real time. So, one can find several paths in very complex
overlay topologies. Also, the overlay topology is made simple by virtualisation: each
domain (including its manager) is considered as an abstract node in the virtual topology.
Therefore the solution is scalable and capable to work in cases of large topologies, being no
need for a hierarchical approach.
This Chapter is organized as follows: the Section 2 contains the state of the art in QoS inter-
domain routing; the Section 3 shortly describes the general Enthrone architecture focusing
on the service management at the network level. The Section 4 introduces the proposed QoS
inter-domain path finding solution. Section 5 presents details about the implementation and
Section 6 contains conclusions, possibilities of extensions and open issues.
2. State of the art

Because our approach deals with QoS path finding and routing, a short overview of the
available approaches for QoS routing is presented below [13][14][15][17][18]. We distinguish
between intra - and inter-domain QoS problems.
The intra-domain QoS routing solutions could be divided in two major approaches.
Classically, intra-domain QoS routing protocols run on the routers and find paths with QoS
constraints from source to destination. While having the advantage of being an Internet
philosophy compliant solution , i.e., completely distributed and dynamic, this approach
does not offer at the domain level an image of the available resources. For mid-long term
paths with QoS guarantees a centralized solution is better. This introduces a domain central
manager having knowledge of the total resource allocation inside the domain. To find the
routes it could use an algorithm to determine QoS routes between source and destination.
In this case the QoS routing process would be run by a dedicated module of the domain
manager. Note that such a solution would centralize completely the routing and would not
benefit from Internet intra-domain routing protocols. Other approach is that the manager
can collect information from routers which are capable to compute QoS constrained paths.
The main thing is that the resulted routes are installed on the network equipments at the
initiative of the manager which commands such actions to a network controller. Usually the
QoS routing process is triggered by a new request addressed to the manager for a QoS path
through the domain.
For inter-domain QoS routing also we can distinguish between two kinds of approaches.
The first one proposes enhancements for the BGP protocol in order to support QoS features.
The BGP advertises QoS related information between autonomous systems (ASes), and the
routing table is build taking into consideration this additional QoS information. The Q-BGP
protocol, proposed in MESCAL project [20], is such an example.
Another category of inter-domain QoS routing solutions are based on the overlay network
idea [13][14]. An overlay network is built, which abstracts each domain with a node,
represented by the domain service manager, or with several nodes represented by the egress
routers from that domain. Then protocols are defined between nodes for exchanging QoS
information and, based on this information, QoS routing algorithms are used to choose the
QoS capable path. In [13] a Virtual Topology (VT) solution is proposed. The VT is formed by
a set of virtual links that map the current link state of the domain without showing internal
details of the physical network topology. Then a Push and a Pull model for building the VT
at each node are considered and analyzed. In the Push model each AS advertise their VT to
their neighbor ASes. This model is suited for small topologies. In the Pull model the VT is
requested when needed, and only from the ASes situated along the path between source
and destinations, path which is determined using BGP routing information. If BGP kept
several routes between source and destination than the VTs for each domain situated along
the founded paths are requested. Based on this VTs information the QoS route from source
to destination is calculated. After that an end to end QoS negotiation protocol is used to
negotiate the QoS resources along the path.
One problem with these solutions is that they suppose that the ASes make available for
others their virtual resource topology information. This requirement could be not accepted
by the actual network providers, due to their confidentiality policy regarding their resource
availability.
Also, these solutions are based on an end to end QoS negotiation process. After the QoS
path is found, the negotiation process is started. The QoS routing process previously
performed in advance would increasing the chance of negotiation success, but the overall
process implies two QoS –related searching processes: building the QoS topology and
secondly negotiation in order to reserve resources.
This chapter proposes a simpler approach by separating the process of path searching in a
virtual topology (built by abstracting each domain with a node) from the process of QoS
negotiation (QoS searching path). By combining these two processes we will obtain a QoS
inter-domain routing solution.
This solution has been developed and integrated in an E2E QoS management system
[2][8][9][10]. The system was proposed and implemented by an European consortium in the
FP6 European project ENTHRONE [2][3][4][5], and continued with ENTHRONE II [6][7][8].
The ENTHRONE project developed an integrated management solution to solve the end-to-
end QoS – enabled transportation of multimedia flows over heterogeneous networks, from
content sources to terminals. It proposes an integrated management solution that covers the
entire audio-visual service distribution chain, including protected content handling,
distribution across networks and reception at user terminals.
The overlay QoS path finding solution is based on the overlay network topology abstracting
each pair (IP domain + manager) with a node. The overlay network graph in this case is
only a connectivity one, with no information about the resources available intra and inter-
domain. Several alternative inter-domain paths are computed, at overlay level, for each
destination domain. Then, the end to end QoS negotiation mechanism is used to ask for and
to reserve resources. Together they will act as a QoS inter-domain routing algorithm.
3. Enthrone End to End QoS Management System

As mentioned before the ENTHRONE project, IST 507637 (continued with ENTHRONE II,
IST 038463) European project, had as main objective the delivery of real time multimedia
flows with end to end quality of services (QoS) guarantees, over IP based networks. To
achieve this goal, a complex architecture has been proposed, which cover the entire audio-
visual service distribution chain, including content generation, protection, distribution
across QoS-enabled heterogeneous networks, and delivery of content at user terminals
[2][3][4][5][6][7]. A complete business model has been considered, containing actors
(entities) such as: Service Providers (SP), Content Providers (CP), Network Providers (NP),
Customers (Content Consumers – CC), etc.
3.1 Enthrone basic concepts

ENTHRONE has defined an E2E QoS multi-domain Enthrone Integrated Management
Supervisor (EIMS). It considers all actors mentioned above and their contractual service
related relationships, Service Level Agreements (SLA) and Service Level Specifications
(SLS), as defined in [2][3][4][5][6][7]. One of the main EIMS components is the service
management (SM). It is independent of particular management systems used by different
NPs in their domains, and it is implemented in a distributed way, each network domain
containing Service Management entities. They are present in different amounts in SP, CP,
NP, CC entities, depending on the entity role in the E2E chain. The SM located in NPs
should cooperate with each IP domain manager and also with other actors in the E2E chain.
Figure 1 shows the general architecture and emphasizes the cascaded model for pSLS
negotiation.
Management and control

EIMS@SP
Service
Provider TM
EIMS@NP EIMS@NP RM@AANP
EIMS@CP
omer
Content Core Network Core Network
Provider 1 3 Access
Aggregation
… Network (1) Customer
(sTVM)
Content
server
pSLS-link = IP multi-domain
QoS enabled Data pipe
CC
cSLS links in AAN
Fig. 1. Forwarded cascaded model for pSLS negotiation
Legend:
EIMS@CP, EIMS@SP, EIMS@NP – ENTHRONE Integrated Management Supervisor at: Content
Provider (CP) , Service Provider (SP) and Network Provider (NP)
sTVM –Source TV and Multimedia Processor ( Content server)
AAN – Access Aggregation Network; AANP – AAN Provider
RM@AANP – Resource Manager of AAN (it is ENTHRONE compliant), TM Terminal Manager
ENTHRONE supposes a multi-domain network composed of several IP domains and access

networks (AN) at the edges. The CPs, SP, CCs, etc., are connected to these networks. The
QoS transport concepts of ENTHRONE are shortly described below.
First, QoS enabled aggregated pipes, to be negotiated based on forecasted data (and later
installed in the network), span the core network, which is part of the multi-domain network.
They are mid-long term logical pipes built by the Service Management entities. The
aggregated QoS enabled pipe, called pSLS pipe (we also will use the term pSLS-link), is
identified by the associated pSLS agreement (Provider SLS) established between the
Network Providers, in order to reserve the requested resources. Each pSLS-link belongs to a
given QoS class, [20].
Then, slices/tracks of pSLS-links are used for individual flows based on individual
cSLA/SLS contracts, concluded after CC requests addressed to the SP. An individual QoS
enabled pipe is identified by a cSLS agreement, which is established between the manager of
the Service Provider (EIMS@SP) and a CC for reserving the necessary resources for the
requested quality of service. Several cSLSs pipes are included at the core network level by
an aggregated pSLS pipe, belonging to the same QoS class.
In the data plane of core IP domains, Diffserv or MPLS can be used to enforce service
differentiation corresponding to the QoS class defined. In the ANs, the traffic streams
addressed to the users (Content Consumers) is treated similar to the intserv, i.e. individual
resource reservations and invocations are made for each user.
3.2 Service Management at Network Provider

The EIMS architecture at NP (EIMS@NP) contains four functional planes: the Service Plane
(SPl) establishes appropriate SLAs/SLSs among the operators/ providers/customers. The
Management Plane (MPl) performs long term actions related to resource and traffic
management. The Control Plane (CPl) performs the short term actions for resource and
traffic engineering and control, including routing. In a multi-domain environment the MPl
and CPl are logically divided in two sub-planes: inter-domain and intra-domain. Therefore,
each domain may have its own management and control policies and mechanisms. The Data
Plane (DPl) is responsible to transfer the multimedia data and to set the DiffServ traffic
control mechanisms to assure the desired level of QoS.
One main task of the EIMS@NP is to find, negotiate and establish QoS enabled pipes, from a
Content Server (CS), belonging to a Content Provider, to a region where potential clients are
located. Each pipe is established and identified by a chain of pSLS agreements, between
successive NP managers. The forwarded cascaded model is used to build the pSLS pipes [5].
The pipes are unidirectional ones. An E2E negotiation protocol is used to negotiate the pSLS
pipe construction across multiple network domains [5].
The process of establishing a pSLS–link/pipe is triggered by the SP. It decides, based on
market analyses and users recorded requirements, to build a set of QoS enabled pipes, with
QoS parameters described by a pSLS agreement. It starts a new negotiation session for each
pSLS pipe establishment. It sends a pSLS_Subscribe_request to the EIMS@NP manager of
the Content Consumer network domain. The EIMS@NP manager performs the QoS specific
tasks such as admission control (AC), routing and service provisioning. To this aim it splits
the pSLS request into intra-domain respectively inter-domain pSLS request. It also performs
intra-domain routing to find the intra-domain route for the requested pSLS, and then it
performs intra-domain AC. If these actions are successfully accomplished, and if the pSLS
pipe is an inter-domain one, then the manager uses the routing agent to find the ingress
point in the next domain, does inter-domain AC and then send a pSLS Subscribe request
towards the next domain. This negotiation is continued in the chain and up to the
destination domain, i.e., the domain of the CC access network. If the negotiation ends
successfully the QoS enabled pipe is considered logically established along the path from
source to destination.
The Figure 2 shows the signalling message sequence associated to the pSLS-link negotiation.
The actual installation and configuration of routers is considered in ENTHRONE a separate
action and is done in invocation phase in a similar signalling way, plus the “vertical”
commands given by EIMS@NP to the intra-domain resource manager.
After the pSLS pipe is active (i.e. subscribed and invoked) the Service Provider is ready to
offer the new service to the users from the access network situated at the end of the pipe.
Now the process of cSLS individual agreements establishment, for this new pSLS pipe,
could be started.
pSLS
EIMS@SP pSLS
pSLS split adminision
provisioning control (AC) EIMS@NP2
pslsSubscribeRequest(psls)
Transit pSLS ?
transit pSLS
(EIMS@NP2,
pSLS intra/interdomain
AC?pSLS admitted
Local
processin
pslsSubscribeResponse(psls, accepted)
pSLS intra/interdomain
AC?pSLS admitted
Install pSLS
Update resources
pslsSubscribeResponse(psls)
Fig. 2. pSLS negotiation for QoS enabled path establishment (negotiation and installation)
4. Finding an end to end path with guarantied quality

4.1 General considerations
The main concepts of ENTRONE as stated in [8] are:
• E2E QoS over multiple domains is a main target of EIMS.
• But each AS has complete autonomy regarding the network resources, including off-line
traffic engineering (TE), network dimensioning and dynamic routing.
Each Network Service Manager (cooperating with Intra-domain network resources

manager) is supposed to know about its network resources in terms of QoS capabilities.
ENTHRONE assumes that each AS manager has an abstract view of its network and output
links towards neighbours in a form of a set of virtual pipes (called Traffic Trunks in
ENTHRONE I, see [5][6]) each such pipe belonging to a given QoS class.
A solution to the route finding problem is to define/use routing protocols with QoS
constraints called QoS routing protocols. They can find a path between source and
destination satisfying QoS constraints.
While finding the QoS path is only a first step, then maintaining the QoS with a given level
of guarantees during the data transfer requires additional actions of resource management,
including AC applied to new calls.
EIMS@NP management system performs these tasks. It is a centralized manager knowing
the topology and resources of a domain: otherwise, if using Diffserv only, in the core one
cannot hope to have guaranteed QoS. Being a central management node for a network
domain, a centralized QoS routing solution is appropriate inside the domain.
On the other side the multiple domain pSLS-links should also belong to some QoS classes
and therefore inter-domain QoS aware routing information is necessary to increase the
chances of successful pSLS establishment when negotiating the pSLSes. Several approaches
are possible for inter-domain provisioning of QoS-enabled routing for ENTHRONE system,
and they will be presented in the next sections. These solutions are based on the overlay
topology approach.
4.2 Overlay QoS virtual topologies

The overlay solutions come out of the ideas expressed in the section above. Also this
approach has been considered in [13] [14] which propose a Virtual Topology Service (VTS)
offering multiple domains QoS enabled virtual pipes. The VTS abstracts the physical
network details of each AS and can be integrated with BGP. Note that the pSLS links already
proposed in the ENTHRONE project are similar.
Therefore one can define two separate types of network services providers: NP itself owning
and responsible of the network infrastructure, which is actually an Infrastructure Provider;
the overlay (virtual) network services provider (ONSP), which in our case is represented by
EIMS@NP, which establishes agreements with other similar providers the final target being
to offer QoS enabled pSLS-links.
In the ENTHRONE system each AS can assure QoS enabled paths towards some destination
network prefixes while implementing its own network technology: DiffServ, MPLS, etc.
Each AS is seen in an abstract way as an Overlay Network Topology (ONT) expressed in terms
of TTs (traffic trunks) characterized by the bandwidth, latency, jitter, etc. The Overlay
Network Service (ONS) is responsible for getting the ONT of each AS on a path in order to
give to the source AS information related to QoS towards a given destination. The End to
End Negotiation Service, which is supported by the EIMS@NP, will then negotiate the pSLS
contracts with the chosen domains in order to reserve resources and then to invoke them.
The ONS can be modeled in two ways, [13][14]: a proactive (Push) model and a reactive one
(also called pull or on demand) model in order to obtaining the overlay (virtual) topologies
of other ASes.
In the proactive case every AS advertises its ONT to other ASes without being requested for,
while the proactive model assumes that overlay topologies are obtained on demand by an
AS which is interested to reach a given destination prefix.
The proactive (push) model has the advantage offered by traditional IP proactive routing
protocols: the ONTs of other ASes are already available at a given AS because they are
periodically advertised among AS managers; therefore latency in offering a route to a new
request is small. The advertisement can also be done at each AS manager initiative, so this
model allows promotion of some routes to other domains. This can be subject of policies.
The dynamicity is high (advertisements can be event driven). But the proactive model is
more complex than the reactive model. Scalability problems may exist, because of high
control traffic volume and also flooding the neighbor ASes with (maybe) not needed
information. The managers will keep information on some routes that are of no interest for
them (yet).
The reactive (on-demand) model is simpler than the proactive model, because an AS will
query each domain of a given path to get the ONTs. No advertising mechanism is necessary.
The scalability is higher because only the ONTs of the chosen routes will be obtained.
Studies show that the mean E2E communication in the Internet traverses between 3 and 4
domains. Therefore the number of domains to be queried to obtain the ONTs is small. The
pull model latency when finding a path is higher (need time for queries and calculations).
The updates of ONT knowledge is not event driven because the lack of an advertisement
mechanism. In [13], a notification message is proposed to solve this problem, i.e., to allow a
domain to notify other domains about local events. Then the source domain can invoke the
VTS to obtain a new set of ONTs.
Traffic trunks for

class QCi
NSMk NSMn
B=6, D=2 Egress

Ingress
ASk
B=6, D=2 ASn
B=3, D=1
B=5, D=4
ONT(ASk)
Ingress
ASm
NSMn
Fig. 3. Overlay network topology of ASk – example
4.3 Proactive approach

Each AS Manager (e.g. NSM@NP manager in ENTHRONE case) knows its ONT. Figure 3
presents a graphic diagram showing several ASes. The ONT of ASk includes all
unidirectional TTs of ASk. One TT can be internal to ASk or external, linking the ASk with
other neighbor domains. One external TT is defined from an egress point (router) of the
domain up to the input of an ingress router of a neighbor domain. Each TT belongs to a
given QoS class and may be characterized by parameters like bandwidth (B), delay (D), etc.
It is the responsibility of the AS manager to find out these values by using internal
mechanisms. Two styles, different in terms of flexibility, can be applied in the proactive
solution, as described below:
1) Maximum flexibility: Proactive ONT advertisements

Each AS Manager advertises its ONT to the neighbors, and also the ONTs learned from
other domains. In such a way (similar to the procedures used in the link-state routing
protocols) each AS manager will become aware of the inter-domain overlay virtual topology
and, applying some constrained routing algorithm, can select its paths to given destinations.
The Figure 4 shows this process.
This solution is based on flooding so it exposes scalability problems. It can be useful for
“regional” scenarios in which the number of ASes is not high. In [13] it is mentioned that a
regional scenario may be formed by “condominiums of domains”; group of domains which
agreed on advertising overlay topologies to each other. All the ASes making part of the
same condominium will eventually know the overlay topologies of other ASes. In such
regions of domains one could apply different business policies/rules and create new
relationships to make the interactions more customer-oriented.
1. ONT(ASl, ASp, ..) 2. ONT( ASk, ASl, ASp, ..) 1. ONT(ASu)
NSMk NSMn
2. ONT( ASn, ASu )
ASk 2. ONT(ASw , ASw)

ASn
2. ONT( ASk, Asl, Asp, ..) 2. ONT(ASw , ASw)
2. ONT( ASn, ASu )
NSMm
1. ONT(ASw)
ASm
Fig. 4. ONT advertisement
2) Minimum flexibility: Proactive vector paths advertisements

At the other extreme there is a solution in which each AS knows its ONT and some path
vectors (in the BGP meaning) reported by other ASes. In such case the advertisements do
not contains ONTs but vector paths (in the sense of BGP, but having additional QoS
information). Each AS wanting to reach a given destination will compute the best path(s)
using its ONT and paths reported by other ASes (based on constrained routing algorithm).
The degree of freedom in path selection is minimum in the sense that a given AS manager
can only select among paths proposed by the other domains and maybe benefit from its own
ONT. But the scalability is better. The Figure 6 shows this solution in a simplified manner.
Advertisments of ASk
1. M[Path(n-x)]
NSMk NSMn
Egress
Ingress
ASn
ASk
Path(n-x)
1. M[Path(m1-x, m2-x)]
ONT(ASk) NSMm Dstx

Path(m1-x)
ASz Rj
Advertisments of ASm Path(m2-x)
for a path offered by Destination AS
ASm
ASm to the destination
Dstx=ASz.Rj Ingress
Fig. 5. Phase 1: Path advertisments from ASm, ASn sent to ASk for destination ASz.Rj
M[Path(k1-x, k2-x)] NSMk NSMn
Egress
Ingress
ASn
ASk
Path(k1-x)
Path(k2-x)
ONT(ASk)
NSMm Dstx
Advertisments of ASk
for paths offered by ASz Rj
ASk to the destination Path(m2-x)
Dstx=ASz.Rj Destination AS
ASm
Ingress
Fig. 6. Phase 2: Paths computed and selected by Ask
We distinguish two phases:

• ASk receives from the ASn advertisements messages having the generic form
M[Path(n-x)], indicating a path vector to destination – ASz domain, Router Rj. In a
similar way, other messages are received from other domains. Note that actually these
messages are not synchronized.
ASk uses its own ONT values and computes one or more “best” or “equivalent” paths for
each QoS class, Figure 6 shows two such paths computed and selected as acceptable ( to be
also advertised to other domains)
4.4 On demand approach

The domains do not advertise their overlay topologies to the neighbouring domains. The
ONT is obtained by each domain at request if it wants to know the ONTs of other domains.
When a given AS needs to find an E2E QoS-enabled inter-domain route, it queries its BGP
local table and determines the possible routes towards the destination. BGP delivers a list of
ASes to follow for a path to destination. Then the initiator domain can query each domain
on the path chain towards the destination and gets the ONT of the domains specifically for
that route. Based on this the initiator domain can get more routes to the destination than
BGP offers.
BGP path vector:

ASp, ASq, ASr, ASz Querry of ASk for ONT of
each domain interrogated
In order to finally reach
Dstx=ASz.Rj
ASm
NSMk
BGP
ASk
ONT
ONT
ONT Dstx
ONT
ASz Rj
ASr
ASq Destination AS
ASp
Alternatives paths through the set

of ONTs towards destination Rj
Fig. 7. Hub model to obtain the ONTs in the on-demand model
Figure 7 shows an example in which the domain ASk needs to reach a destination ASz with
a required QoS set of values for this path.
The manager of the domain ASk queries its local BGP table and finds the BGP route to the
destination ASz, through ASp, ASq, ASr. The ASk manager invokes the overlay topology
service (OTS) to obtain the ONT of each domain. Figure 7 presents for this a hub model.
Note that there can be more than one BGP path to the destination ASz. Therefore ASk can
recursively query each domain in each path and find the best path towards the destination.
After obtaining all the ONTs of each possible route towards domain ASz, the source domain
ASk can use a Constraint Shortest Path (CSP) algorithm to find the best route that fits the
QoS requirements. The path calculation can be done using only one attribute or more than
one (bandwidth, latency, loss). It may happen that after obtaining the ONTs of each route
towards a destination, the ASk can realize that there is no route that satisfies the QoS
requirements. A solution to this is proposed in [13]: to make use of the Internet hierarchy to
collect more alternative routes towards a given destination. Taking into account the
hierarchy of ASes in the world, going upwards this hierarchy may produce several routes
which have been not advertised initially by BGP.
Suppose that the stub domain ASk is multi-homed with domains ASq and ASp and it has
received two BGP routes from its providers to reach prefixes at domain ASz. The first route
is ASq , ASr, ASz and the second is ASp, ASz. Then, to increase the number of paths to
query for ONTs , the OTS can invoke its providers and asks other BGP routes that were not
initially advertised. In this case, domain ASp would return to domain ASk the paths ASu,
ASz and ASz towards ASz ,
So, ASk will have one more route available. In general this procedure can be used when a
domain is multi-homed
4.5 The proposed overlay inter-domain QoS path finding solution

We proposed a simplified version [1], which takes into account the following assumption
regarding the specific characteristics of the Enthrone system:
• The number of E2E QoS enabled pipes is not very large because they are long term
aggregated pipes.
• The number of NP entities is much lower than the number of routers.
• The EIMS@NPs are implemented on powerful and reliable machines, having enough
computing and storage capabilities.
• The inter-domain core IP topology is stable during the inter-domain routing process; this
is fulfilled within the assumptions of this work.
This solution is also based on the idea of Overlay Virtual Network (OVN) similar as in [13],
but the OVN consists only of network domains (autonomous systems) abstracted as nodes.
Each node will be represented by an EIMS@NP in this Overlay Virtual Network. This virtual
network contains only information on connectivity between the domains, represented by the
EIMS@NP nodes, or additionally static information regarding the inter-domain QoS
parameters: links bandwidth, maximum jitter and delay, mean jitter and delay, etc.
This overlay virtual connectivity topology (OVCT) can be learned statically (offline) or
dynamically.
The statically approach considers that the OVCT is built on a dedicated server – a topology
server, like in the Domain Name Service (DNS). When a Network Provider wants to enter in
the Enthrone system, then its EIMS@NP should register on this topology server. The
topology server will return the Overlay Virtual Connectivity Topology. So we will consider
that each EIMS@NP has the knowledge of this connectivity topology.
In the dynamic case each EIMS@NP, if it wants to build the OVCT, will query its directly
linked (at data plane level) neighbour domains’ managers. It is supposed that it has the
knowledge of such neighbours.
Each queried EIMS@NP returns only the list of its neighbours. At receipt of such
information, the queerer EIMS@NP updates its topology database (note that this process is
not a flooding one as in OSPF). Then it queries the new nodes learned and so on. The
process continues until the queerer node EIMS@NP learns the whole graph of
“international” topology.
As we mentioned above the graph contains as nodes the EIMS@NPs which means that is
made from the Network Service Managers of Enthrone capable domains, as shown in Figure 8.
EIMS@NP2
EIMS@NP3
EIMS@NP1
EIMS@NP6
EIMS@NP7
EIMS@NP4
EIMS@NP5
Fig. 8. Overlay Virtual Connectivity Network
If the Enthrone system will be implemented at large scale the number of nodes in the graph
will be large, which means that the time required calculating the routing table will be also
large. But because the topology structure changes events (adding new EIMS domains) are
not frequent (it might happen at weeks, months), the topology construction process could
run at large time intervals (once a day for example). In this case the routes calculation is
triggered also at large time intervals, which means that there are no real time constraints.
Another consequence is that the messages used to build the OVCT will not overload
significantly the network. Enthrone capable domains can be separated by normal domains,
with no Enthrone capabilities. In this case we consider that static initially QoS enabled pipes
are built between Enthrone capable domains, pipes crossing the Enthrone non capable
domains. The domains (Enthrone non capable) will be transparent for the Enthrone
domains.
On the graph learned each EIMS@NP can compute several paths between different sources
and destinations, thus being capable to offer alternative routes to the negotiation function.
The number of hops is used as a primary metric for the path choosing process. By the “hop”
term we refer to a node in the Overlay Virtual Topology.
The process of route selection is as follows:
• When a request for a new pSLS arrived at one EIMS@NP, this will select the best path to
the destination (the next EIMS@NP node that belong to this path), based on the overlay
routing table.
• After the next hop is selected, the EIMS@NP will check if it has an intra-domain QoS
enabled path for this route, i.e., between an appropriate ingress router and an egress router
to the chosen next hop domain. If there is no such QoS enabled route, the next hop
EIMS@NP node is selected from the overlay routing table.
• In case that a QoS enabled route intra-domain is found, the EIMS@NP, based on
mechanisms defined in Enthrone, triggers a request for a new pSLS or a modified pSLS
negotiation to the chosen EIMS@NP neighbor.
• This process continues until the destination is reached. If the negotiation ends with
success than the pSLS pipe with guaranteed QoS parameters is found. If the process fails
then the EIMS@NP will choose another overlay path to the destination and starts a new
negotiation.
In Figure 9 the messages sequence for pSLS negotiation process in the case of multiple paths
towards the destination is shown. The Service Provider decides to build a pSLS enabled pipe
between a source located in the NP1 domain and a destination located in NP5 domain. We
consider for this example that the working overlay topology is the one given in Figure 8.
One can see that we have four possible routes between NP1 and NP5 domains. The first two
of them, in terms of cost value, are the routes through NP6 and NP7 respectively. In Figure 9
it is illustrated the case when the pSLS negotiation along the route NP1-NP6-NP5 fails, due
to admission control rejection in NP6 domain, either on intra-domain pipe inside NP6
domain, or on the inter-domain pipe between the NP6 and NP5 domains.
When it receives the rejection response at the pSLS subscription request the NP1 domain
checks for an alternate route towards the NP5 domain. It finds the route through the NP7
domain and starts a new negotiation using this new route. This negotiation ends
successfully, so the QoS enabled pipe between NP1 and NP5 will follow the route NP1-NP7-
NP5.
This solution has the advantage of being simple and that it not require at an AS the
knowledge of current traffic trunks for the other network domains as in [13].
A drawback of our solution (proposed above) is a larger failure probability in negotiating a
segment (therefore a longer mean time for negotiation process), if comparing with solutions
which calculate the QoS path before the negotiation process. The latter approach increases
the probability that the negotiation finished with success at the first try.
The path finding process described above is not based on BGP information at all. BGP is
used only for best effort traffic. The process of QoS routing takes place at service
management level. But it is possible in principle to use such BGP information.
5. Routing Tables
5.1 Design Details
As mentioned before this solution is based on the knowledge of the overlay network
connectivity topology. The topology can be kept in a form of a square matrix. The
dimension M is equal to the number of nodes in the overlay topology network. Each entry
rij, has an integer value. A zero value means that there is no direct connectivity between the
nodes i and j. A value different from zero, value 1 for example, implies that there is a direct
connection between the two nodes: Lij represents the link between nodes i and j (more
precisely, between the two domains there are some linked routers). Because the matrix is a
sparse one it can be easily compress in order to be stored in case that the dimension M is
large.
Based on this overlay topology each EIMS@NP builds a routing table which contains, for
each destination node in the network, several possible paths to this destination node, and
the costs associated with each of these paths. Because in the routing table several entries will
exists for each destination, the QoS negotiation process will be able to be carried on
successively on multiple paths, increasing the probability that a path fulfilling the QoS
requirements is found.
pSLS pSLS pSLS

EIMS@SP split adminision EIMS@NP EIMS@NP EIMS@NP
prov control
pslsSubReq
Transit Next hop
EIMS@NP6
transit
pSLS intra/interdomain AC?
pSLS admitted
pslsSubscribeRequest(psls
Reject
pslsSubscribeResponse(reject) Request
Next route ?
Next hope
pSLS processing Local
pslsSubscribeRequest(psls) processing
Accept
Request
pslsSubscribeResponse(psls,acc)
pslsSubscribeResponse(psls,acc) pSLS processing
pSLS intra/interdomain AC?
pSLS admitted
pSLS pipe Install pSLS
crosses NP1, Update resources
pslsSubscribeResponse(psls, acc)
Fig. 9. MSC for QoS negotiation in case of multiple paths towards a given destination
Because the number of possible paths from source to a certain destination could be high we
have limited it to the first four ones, with the lowest costs. If the number of neighbours are
less than four, than the number of possible routes towards a destination is limited to this
number. It is used the same principle as in the case of distance vector protocols. In the case
when there are several paths to the same destination EIMS@NP node, using as first next hop
the same node, in the routing table we will store the best cost of all the possible paths going
through that node.
This is not a limitation because in our case the routing decision is taken hop by hop so the
source node has no idea what route to the destination will be chosen at the node where the
paths are splitting. An EIMS@NP does not need to keep the whole path information (but the
total cost only) because it cannot influence the route chosen decision at the next hops along
the path.
Let’s suppose that the EIMS@NPk node has the neighbour nodes EIMS@NPm, EIMS@NPn,
EIMS@NPp. The routing table from EIMS@NPk node to EIMS@NPl node will be:
Destination EIMS@NPl EIMS@NPl EIMS@NPl

Nex Hop EIMS@NPm EIMS@NPp EIMS@NPn
Cost
5 0 3
(Nb of hops)
Table 1. Routing table at node K for node L destination
The EIMS@NP at node k builds such a record for each node in the overlay network. This
process, of searching several possible paths for each possible destination, in this overlay
network topology, is an expensive one in terms of calculation. But based on the assumptions
presented above, which are realistic ones, if such a management system will be
implemented in the network domains, this routing table building process will be run only
on topology updates, which means at very long time intervals. Such a process will put low
computing overhead on the Service Manager. Also, it could be scheduled to run on intervals
with low management activity [5]. Taking this in consideration, it could be considered that
the routing table is a static one, and the route search process reduces to a simple database
search one. We do not need to run the searching algorithm for each pSLS subscription
request. It is enough to search, in the routing table, the route to the destination with the
smallest cost, and forward the request to the chosen next node. If the negotiation for QoS
parameters along this path failed, then we will chose the next path, in terms of cost, from the
routing table.
5.2 Possible Improvements

It is said that the solution did not take into consideration any QoS parameters, in the first
phase, for path building process. This task left for the QoS negotiation process.
A possible improvement is to take into account some general data about the QoS
parameters, in the path finding phase. For example, based on agreements with Service
Managers of some domains, or based on some general QoS parameters of the domains, the
Policy Based Management module could associate different costs for the links in the
topology matrix. It is supposed that domains agreed to share these parameters, such as: the
min/mean/max delay and jitter, introduced by the domain. In such a way the Policy
module could influence the routing decision process. In this case the matrix element rij
could have the value Cij if the link Lij exists. The value Cij is the cost for the link Lij, and
could be established by weighting appropriately the general QoS parameters mentioned
above. These weights could be established by the domain administrator and transmitted to
the Policy module.
The cost of a link could be also modified based on statistics regarding the acceptance or
rejection rate of previous negotiated pSLS pipes. For example, if some domain with a good
link cost rejects several times the requesting domain could modify the costs of the links
crossing that domain.
When the path cost is computed it could be taken into account the existence of resource
price agreements between some domains. These agreements could be negotiated using pull
model, based on some statistics. For example an EIMS@NP node has two different paths
towards a destination with similar path costs. It chose the path with a better cost, but it also
could periodically request resource price information from both neighbour nodes crossed by
the two paths. If the second node has available resources and is interested to carry traffic
from the source domain, it will propose a better resource price as a response to resource
price requests. So the EIMS@NP source node could modify the routing table by improving
the path cost for the second path, and the future pSLS pipe requests will be routed through
the second path. Such a resource price communication could be easily implemented because
the EIMS@NP managers are built as web-services, which implies very flexible
communication capabilities.
5.3 Overlay topology building

For our solution we have chosen to build the overlay topology by means of successive
interrogations of all the available nodes. The node which decides to build/refresh overlay
the topology starts to interrogate all the other overlay nodes about their neighbours. It starts
with its direct connected neighbours and then continues interrogating the new found
neighbours, and so on.
For the EIMS@NP implementation we have used the web service technology. The interfaces
between the EIMS@NP modules are implemented using WSDL language. The interdomain
path finding WSDL interface it is used by EIMS@NP to interact with other EIMS@NPs in
order to build the overlay topology used to search for interdomain overlay paths.
The interdomain path finding WSDL interface has defined the following messages:
• getEimsNeighborsRequest ()
• getEimsNeighborsResponse(EimsNeighborsArray eimsNeighbors)
• getDomainQoSRequest ()
• getDomainQoSResponse(DomainQoS qos)
The first two messages are used by the Overlay Path Building module from EIMS@NP
subsystem to build the overlay topology. The response message contains an array with all
the neighbours of the interrogated domain, and their associated data about the web services
addresses, identification, IP addresses.
The next two messages are used to get general information about the QoS parameters of the
domain: min/max/mean delay and jitter, mean transit cost, max bandwidth. These values
refer to the transit parameters for the domain. We have considered that such information
could be offered by each domain without affecting its confidentiality policy. These
parameters are used to establish the cost associated with a link between two neighbour
domains. For establishing the cost we have weighted the normalized values for these
parameters. The weights were chosen arbitrarily, such as their sum to be one. No studies
have been done to find the optimal values.
The format of messages parameters are given in table 2.
<wsdl:types>
<xsd:schema xmlns:xsd="http://www.w3.org/2001/XMLSchema"
targetNamespace="http://webservice.enthrone.org/eims/
/InterdomainPath/datatype”>
<xsd:complexType name="EndPoint">
<xsd:sequence>
<xsd:element name="IPAddress" type="xsd:string"/>
<xsd:element name="NetMask" type="xsd:string"/>
</xsd:sequence>
</xsd:complexType>
<xsd:complexType name="Neighbor">
<xsd:sequence>
<xsd:element name="id" type="xsd:string"/>
<xsd:element name="OverlayPathWebserv" type="xsd:string"/>
<xsd:element name="Address" type="tns:EndPoint "/>
</xsd:sequence>
</xsd:complexType>
<xsd:complexType name="DomainQoS">
<xsd:sequence>
<xsd:element name="minDelay" type="xsd:int"/>
<xsd:element name="maxDelay" type="xsd:int"/>
<xsd:element name="meanDelay" type="xsd:int"/>
<xsd:element name="minJitter" type="xsd:int"/>
<xsd:element name="maxJitter" type="xsd:int"/>
<xsd:element name="meanDelay" type="xsd:int"/>
<xsd:element name="meanCost" type="xsd:int"/>
<xsd:element name="maxBandwidth" type="xsd:int"/>
</xsd:sequence>
</xsd:complexType>
</xsd:schema>
</wsdl:types>
Table 2. Data type section for the interdomain path finding WSDL interface
In order to be able to perform the pSLS negotiation and to obtain the overlay topology we
have defined several database tables used to store the data required by the above mentioned
operations. These tables are shortly described next:
• Overlay_topology table – it contains data about each EIMS node in the topology, such as
the addresses of the web-services available, the IP address, the domain identifier, and the
QoS parameters. It is updated by the Inter-domain Overlay Path module at each overlay
topology building cycle. It is used by the overlay routing process to build the overlay
topology matrix used in the overlay route searching process.
• Eims_neighbors table – stores information about the neighbours for each EIMS node
contained in the overlay_topology table. It is also updated by the Inter-domain Overlay Path
module at each overlay topology building cycle.
• Overlay_interdomain_routes table – is used to store several alternative routes towards a

destination overlay node. The number of alternative routes is limited to four. It is managed
by the overlay routing process.
• Local_eims table – stores information about the local NetSrvMngr@NP such as: IP
address, web services ports, domain Id. It is managed by the system administrator.
• Border_routers table – stores information about the local domains border routers. It
contains the border routers IP address and neighbour EIMS reached through this border
router. It is managed by the system administrator.
• Access_networks table – stores information about the access networks for the local
domain. It contains the access network IP address and the border router IP address. It is
managed by the system administrator.
• Local Eims_neighbors table - stores information about the EIMS neighbours for the local
domain. It contains information about the border routers used to connect the local domains
and the neighbors, border router IP address, web service port addresses, etc. It is managed
by the system administrator.
• Domain_qos_parameters table – it is used to store global QoS parameters about the

domain. It is managed also by the system administrator.
5.4 Functionality tests

This solution was implemented on the test-bed built at our university in the Enthrone
project framework [21] [22]. The test-bed consists of three Autonomous Systems, each
managed by a Network Service Manager (EIMS@NP). The EIMS@NP managers are
implemented using web services technology. Between domains the BGP protocol is used to
route the best effort traffic. A Network Manager is used to install the pSLS pipes on network
devices. Also the test-bed has a Service Provider EIMS Manager, and the other modules
required by the Enthrone system. The connectivity tests involved only the Network
Provider managers and Service Provider manager.
The EIMS@SP was used to trigger pSLS subscribe requests, between a Content Provider and
one of the available Access Networks, until the resources on the lowest cost path between
the chosen source and destination, were exhausted. Then it has been triggered additional
requests between the same source and destination. These new requests were admitted but
the pSLS pipes were built along the next cheapest path between the chosen end points.
Because the testbed is a small one, is difficult to evaluate the performances of the proposed
solution for a large number of domains. We have measured how fast, a request for getting
the neighbours EIMS from a network domain, is served. We have obtained a mean time less
than 0.1s per request. If we take for example a topology consisting of 1000 domains then,
because we can consider that the total processing time is increasing linearly with the
number of domains, the total processing time required to obtain the overlay topology is
about 100s. We can increase it with 50% to take into account that at a large number of
domains the local processing time, between two interrogations could be higher. So we could
consider that for 1000 domains the topology building process takes about 150s, which is an
acceptable value. Also, the solution used to build the overlay topology implies a large
number of messages to be exchanged in order to build the topology. Each node should
communicate with the other nodes. But the messages exchanged are small, because each of
them contains only a few data about the neighbors of the interrogated node. If it have been
adopted a link state like protocol to build the topology, then the messages would have been
very big in case of large number of domains, so the amount of signaling data in the network
would have been bigger. Also in our case we don’t have convergence problems.
It has not been evaluated till now the time needed to compute several paths towards all the
destinations nodes in the overlay topology.
The testbed used is not appropriate to test the scalability for the path finding process
performed in the first phase. It could only be used to see that routing table is built correctly,
containing several paths towards each destination domain in the topology. Then, several
requests for QoS enabled pSLS pipes have been triggered. These pipes were built along the
first path specified in the routing table. When the resources on this path were exhausted,
during the negotiation process, the next route was used for the following pSLS pipe. These
tests proved that the solution is able to find QoS enabled pipes, in a multi domain
environment.
6. Conclusion
This chapter has proposed a simple solution for solving the problem of QoS enabled inter-
domain path finding, applicable when Network Service Management systems exist in each
domain capable of constructing mid-long term pSLS pipes with imposed QoS parameters.
Because the solution does not require at a domain the knowledge of other domain resources,
it could be attractive and accepted by the real life network providers. Another advantage is
that it does not burden a given domain manager with the need of knowing the available
traffic trunks of other network domains. Also, by separating the process of path finding
from the QoS negotiation, the path searching process doesn’t need to work real time. So we
can find several paths in very complex overlay topologies. Also, by simplifying the overlay
topology, considering only the domain managers as topology nodes, the solution might
work for very complex topologies, being no need for an hierarchical approach.
The solution has as a main disadvantage that it does work only in the presence of a QoS
negotiation system capable. It is based on this feature to check the QoS constraints on the
paths founded in the overlay topology. Another disadvantage is that, it may not find the
best QoS enabled path, as could be the case with other solutions.
The solution has proved to be simple to implement and is well suited for ENTHRONE
Integrated Management System. It is also naturally extensible for more sophisticated
techniques in QoS capable paths finding.
Further studies and simulations will be done in order to validate this solution for a real
network environment. Also, it has been supposed that, because the path finding process
could be run offline and the topology is a simplified one, a non hierarchical solution could
be adopted for Internet. Simulations should be done to establish the amount of resources
need by such a process.
7. References
[1] S.G. Obreja, E. Borcoci, Overlay Topology Based Inter-domain QoS Paths Building. AICT
apos;08. Fourth Advanced International Conference on Telecommunications Proceedings,
Volume , Issue , 8-13 June 2008 Page(s):64 - 70
[2] ENTHRONE I Deliverable 05 IMS Architecture Definition and Specification, June 2004.
[3] A. Kourtis, H. Asgari, A. Mehaoua, E. Borcoci, S. Eccles, E. Le Doeuff, P. Bretillon, J.
Lauterjung, M. Stiemerling, Overall Network Architecture, D21 ENTHRONE
Deliverable, May 2004.
[4] T.Ahmed - ed. Et al., End-to-end QoS Signal-ling & Policy-based Management
Architectures, ENTHRONE IST Project Public Deliverable D23F, September 2005,
http://www.enthrone.org.
[5] H. Asgari, ed., et.al., Specification of protocols, algorithm, and components, the
architecture, and design of SLS Management, ENTHRONE IST Project Public
Deliverable D24F, July 2005, http://www.enthrone.org
[6] P.Bretillon ed., et. al, Overall system architecture, ENTHRONE II Deliverable D01,
February 2007
[7] P.Souto ed., et al, EIMS for ENTHRONE, ENTHRONE II Deliverable D03f, March 2007
[8] E. Borcoci, Ş. G. Obreja eds., et al, Service Management and QoS provisioning,
ENTHRONE II Deliverable D18f, March 2008.
[9] Project P1008, Inter-operator interfaces for ensuring end-to-end IP QoS, Selected Scenarios
and requirements for end-to-end IP QoS management, Deliverable 2, January 2001.
[10] P.Trimintzios, I.Andrikopoulos, G.Pavlou, P.Flegkas, D. Griffin, P.Georgatsos,
D.Goderis, Y.T’Joens, L.Georgiadis, C.Jacquenet, R.Egan, A Management and
Control Architecture for Providing IP Differentiated Ser-vices in MPLS-Based
Networks, IEEE Comm. Magazine, May 2001, pp. 80-88.
[11] E.Marilly et. al, SLAs: A Main Challenge for Next Generation Networks, 2nd European
Conference on Universal Multiservice Networks Proceedings, ECUMN'2002 April 8-10,
2002.
[12] T.Engel, H.Granzer, B.F. Koch, M.Winter, P.Sampatakos I.S. Venieris, H.Hussmann,
F.Ricciato, S.Salsano, AQUILA: Adaptive Resource Control for QoS Using an IP-
Based Layered Architecture, IEEE Communications Magazine, January 2003, pp. 46-
53. See also http://www-st.inf.tu-dresden.de/aquila/
[13] Fabio L. Verdi, Maurcio F. Magalhaes Using Virtualization to Provide Interdomain
QoS-enabled Routing, Journal of Networks, April 2007.
[14] Z. Li, P. Mohapatra, and C. Chuah, Virtual Multi-Homing: On the Feasibility of
Combining Overlay Routing with BGP Routing, University of California at Davis
Technical Report: CSE-2005-2, 2005.
[15] Z. Wang and J. Crowcroft, Quality of Service Routing for supporting multimedia
applications, IEEE Journal of Selected Areas in Communication (JSAC), 14 (7) (1996),
pp. 1228-1234.
[16] D. Eppstein, “Finding k-shortest paths”, SIAM Journal on Computing, 28 (2) (1998), pp.
652-673.
[17] D. Griffin, J. Spencer, J. Griem, M. Boucadair, P. Morand, M. Howarth, N. Wang, G.
Pavlou, A. Asgari, P. Georgatso, Interdomain routing through QoS-class planes,
Communications Magazine, IEEE,Feb.2007
[18] S.P. Romano, ed., Resource Management in SLA Networks, D2.3 CADENUS Deliverable,
May 2003.
[19] T.Ahmed, A.Asgari, A.Mehaoua, E.Borcoci, L.Berti-Équille, G.Kormentzas, End-to-End
QoS Provisioning Through an Integrated Management System for Multimedia
Content Delivery, Computer Communication Journal, May 2005.
[20] E.Borcoci, A.Asgari, N.Butler, T.Ahmed, A.Mehaoua, G.Kourmentzas, S.Eccles, Service
Management for End-to-End QoS Multimedia Content Delivery in Heterogeneous
Environment, AICT Conference Proceedings, July 2005, Lisbon.
[21] T. Ahmed, ed. et al., Pilot and services integration and tests, ENTHRONE II Deliverable
D27 , March 2008
[22] M.Mushtaq, T. Ahmed ed. et al., Trials and evaluation, ENTHRONE II Deliverable D28,
November 2008
Dual Linearly Polarized Microstrip Array Antenna 367
17
X
Dual Linearly Polarized

Microstrip Array Antenna
M. S. R Mohd Shah, M. Z. A Abdul Aziz and M. K. Suaidi
Faculty of Electronic and Computer Engineering,
Universiti Teknikal Malaysia Melaka,
Hang Tuah Jaya, Ayer keroh, 75450, Melaka.
Malaysia
M. K. A Rahim,
Radio Communication
engineering,
Faculty of Electrical Engineering,
Universiti Teknologi Malaysia,
81300 Skudai, Malaysia.
1. Introduction
The wireless communications systems have been greatly expand to the high performance
applications. Nowadays, most of the wireless communications systems offers high data rate
transmission and keep growing for higher data rates technology. Then, the communication
devices were design to be small in size, low power consumption, low profile and practical.
2. Important
Recently, Multiple Input Multiple Output (MIMO) has become popular research topic
among researchers for development of a new wireless communications technology. The
system capacity can be increase with deployment of MIMO technique in the
communications system. Thus, the used of high frequency bandwidth can be avoid since
this method required high cost implementation. High transmitted power also is not required
because all transmitted branch will transmit same power in MIMO system. There are three
major studies in MIMO which are research on array antenna and adaptive signal processing,
research on information theory and coding algorithm and research on MIMO channel
propagation (Nirmal et al., 2004).
MIMO channel capacity can be increase with the increase of number of transmitter and
receiver. When the number of the antennas used is fixed, the channel capacity is related to
the spatial correlation and the diversity gain from antenna spacing configuration at
transmitter or receiver. The spatial correlation in MIMO system is always exploited by using
diversity technique such as frequency diversity, space diversity, time diversity and
polarization diversity. Polarization diversity can be achieved by deploying two or more
different polarized antenna at transmitter or receiver. The transmitted signal with different
polarized in MIMO channel will improved the un-correlation channel between transmitter
and receiver (Collins, Brain. S, 2000)(Manoj. N et al.,2006)(Byoungsun. L, 2006).
A few technique have been introduce to obtain dual polarized antenna such as aperture-
coupled microstrip antenna, two port corporate feed network and two or more probe feeds
technique. The aperture coupled microstrip antenna was developed by using cross slot
aperture at the plane between feed line plane and ground plane. Each aperture excite the
patch in single direction and two orthogonal modes can be excited from the cross aperture
(Ghorbanifar &Waterhouse., 2004)(S. B. Chakraby et al., 2000). Besides, the used of T, H and
U slot configuration can offer better isolation between the two ports (Sami Hienonen et al.,
1999)(S. Gao et al., 2003)(S. Gao & A. Sambell, 2005)(B. Lee, S. Kwon& J. Choi, 2001). A good
isolation between ports will lead to good axial ratio if the circular polarized is used. Thus,
the combination of the slots and slots modifications has been widely investigated by the
researchers as report in (S. K. Padhi et al., 2003)(A. A. Serra et al., 2007)( Kin-Lu Wong et al.,
2002) (B. Lindmark, 1997). Higher gain for these technique can be achieve by using number
of patch and array feed network (M. Arezoomand et al., 2005)(J. Choi & T. Kim, 2000). This
technique requires relatively complicated feed arrangement or multilayer construction in
order to reduce the coupling between two feed lines (W.-C. Liu et al., 2004).
Two port feed network technique will excite two independent dominant mode from the
patch with fed at the dual central point. Thus, the patches mode will degenerates at the far
fields and produce the orthogonal and linear polarized at angles of designed (LJ du Toit &
JH Cloete, 1987). A patches with corner fed also can excite two orthogonal polarized with
equal amplitude and in phase. The corner fed method produce higher isolation as compared
to edge centre fed method (Shun-Shi Zhong et al., 2002) (ShiChang Gao & Shunsui Zhong,
1998)(S. C. Gao et al., 2001).
Dual linear polarized antenna can also develop by using square patch with two feed probes.
Each feed probe will generate one polarized signal primarily such as horizontally and
vertically polarization (K. Woelders & Johan Granholm, 1997). The cross polarization at far
field will cause the field generate by the patches is not purely orthogonal. These problem
can be reduce by integrate bend slots in the square patch and reducing the antenna size as
well (W.-C. Liu et al., 2004)(Keyoor Gosalia & Gianluca Lazzi, 2003).
Most of the dual polarized microstrip antenna was design to generate signals with vertical
and horizontal polarized or +45 and -45 polarized. Vertical and horizontal polarized can be
excite from patch with vertical and horizontal in position. However, +45 and -45 polarized
signal excite from the patch which are slant at the angle of +45 and -45 from the principle
plane. This topic will discussed the design of ±45 dual polarized microstrip antenna with a
single port at the single layer substrate. The further investigate also will be done to
investigate the dual polarized signal excitation for array technique.
All the design will used 1.6 mm FR4 substrate with εr = 4.7 and tanδ = 0.019. First, the design
simulation and measurement of single patch slant at ±45 will be presented. Then, further
investigation for array implementation also will be discussed later. The Computer
Simulation Technology (CST) Studio 2006 was used as CAD tools and fabrication was done
by using chemical etching technique.
3. Design specification
As this design was intended to confirm the basic concept, it was decided to build the
antenna using a best and successful approach. The specification such as the dielectric
substrate and impedance matching will be meeting and find. Appropriate components will
choose including the SMA/coaxial connector and FR4 board. A single element of square
geometry +45º and -45º slanted polarized as shown in Figure 3.2 and Figure 3.3 can be
designed for the lowest resonant frequency using transmission line model.
The substrate used is FR4 with a dielectric constant of 4.7 and a thickness of 1.6 mm. The
loss tangent of the substrate is 0.019. After all dimensions have been calculated, the design
would then be simulated in CST Studio Suite 2006 software to obtain the return loss,
radiation pattern, and VWSR.
3.1 Transmission line model

The method used that allows the design of square microstrip patch antenna is the
transmission line model. A square microstrip antenna fed to excite only one dominant mode
(TM10 or TM01) has a single resonance which may be modeled as this method. These values
are designated Ra, La, Ca as shown in Fig 1. This figure represents the inset fed patch antenna
which the arrangement of feed is shown in Figure 2. At resonance the relationship between
the resonant frequency f0 and the patch model values La and Ca are;
�
fo2 = ��
Equation 3.1
When the patch is resonant the inductive and capacitive reactance of La and Ca cancel each
other, and the maximum value of resistance occurs. If the patch is probe fed and thick, the
impedance at resonance will have a series inductive reactance term Ls;
�� Equation 3.2
In order to obtain the values of La and Ca from measured or computed data one must
subtract the series inductive reactance from the impedance. The value of two points either
side of resonant frequency is obtained.
f1 = fo - ∆ f1 Equation 3.3
f2 = fo + ∆ f2 Equation 3.4
With the subtraction of the series inductance, the reactance now changes sign either side of
fo. The admittance at each frequency may be expressed as;
� �
�� Equation 3.5
��
� �
�� Equation 3.6
��
The conductance G1 and G2 in the equivalent circuit of the patch antenna will account for the
losses through radiated power, and the susceptance B1 and B2 will give a measure of the
reactive power store in neighborhood of the radiating slots. Since the slots are identical G1 =
G2 = G, the expression of B1 and B2 is;
ଵ
‫ܤ‬ଵ ൌ ݂ଵ ‫ܥ‬௔ െ Equation 3.7
௙భ ௅ೌ
ଵ
‫ܤ‬ଶ ൌ ݂ଶ ‫ܥ‬௔ െ Equation 3.8
௙మ ௅ೌ
Solving the equations for C the expression can be obtained as;
௙భ ஻భ ି௙మ ஻మ
‫ܥ‬௔ ൌ Equation 3.9
௙భ మ ି௙మ మ
The susceptance, B can be obtained by equation below;
ඥఌ೐೑೑
‫ ܤ‬ൌ ݇଴ ο݈ Equation 3.10
௓బ
Where; ∆݈ = Extended incremental length

εeff =Effective dielectric constant
Fig. 3.1. Equivalent circuit for proposed microstrip patch antenna
3.2 Microstrip patch design
3.2.1 Square Patch

The design of the square shape patch follows the equation for designing the rectangular
shape patch. The same length and width of the patch of the antenna was made to ease the
design steps. Inset feeding is introduced into the design to offset the feeding location to the
point where matched impedance can be achieved. The design calculation for the square
patch has been discussed in this section. The parameters that needed to be calculated are the
length of the patch, the inset feed and the feed line’s length as shown in Fig 3.2.
Fig. 3.2. Layout of the square patch.
The calculated parameters of the patch have been calculated as shown in Table 3.1. The
input impedance level of the patch can be control by adjusting the length of the inset.
Variations in the inset length do not produce any change in resonant frequency, but a
variation in the inset width will result in a change in resonant frequency (M. Ramesh & K. B.
Yip, 2003). The feed line is made to be a quarter wavelength of the operating frequency. The
width of patch can be determined using the equation 3.11.
ଵ ଶ
ܹ ൌ ටఌ Equation 3.11
ଶ௙ೝ ඥఓబ ఌబ ೝ ାଵ
The ε0 and the μ0 are the permittivity and the permeability in free space respectively.The
1
equation can also be interpreted as the speed of light, c which is 3×108 m/s. The symbol
ඥμ0 ε0
f is the resonant frequency that the antenna intended to be operating and εr is the
permittivity of the dielectric.The patch’s length can be calculated using the equations 3.12.
The length’s extension, ΔL and the effective permittivity, εreff have to be calculated before
calculating the length of the microstrip patch as shown in equation 3.13 and 3.14. The h is
the height of the substrate while the W is the width of the patch as calculated before.
ଵ
‫ܮ‬ൌ െ ʹο‫ܮ‬ Equation 3.12
ଶ௙ೝ ඥఌೝ೐೑೑ ඥఓబ ఌబ
W
(ε reff +0.3)( +0.264)
∆L= 0.412h h
W Equation 3.13
(εreff -0.258)( +0.8)
h
1
εr +1 εr -1 h -2
εreff = + ቂ1+12 ቃ Equation 3.14
2 2 W
where:
f = Operating frequency μ0 = Permeability in free space
εr = Permittivity of the dielectric W = Patch’s width
ε0 = Permittivity in free space h = Thickness of the dielectric
εreff = Effective permittivity of the dielectric
The type of feeding technique that will be used is the inset feed technique. It is one of the
best feeding techniques and it is also easy to control the input impedance of the antenna.
The input impedance level of the patch can be control by adjusting the length of the inset.
The calculation of the inset fed is shown in the equations 3.19 which show the resonant
input resistance for the microstrip patch.
�
�� Equation 3.15
�
��
��
� �
��
where:
��
��
So, for resonant input resistance, Rin

� �
�� ℓ� � �� Equation 3.19
��
L is the length of the patch, ℓ is the length of the inset, and G1 is the conductance of the
microstrip radiator. As reported in frequency (M. Ramesh & K. B. Yip, 2003), the calculations
for finding the inset length can be simplified as shown in the equation 3.20. This equation is
valid for εr from 2 to 10. Using the equation below helps to ease the calculation for the inset
length of the microstrip antenna.
��

ℓ � � �� Equation 3.20
��
where: εr = Permittivity of the dielectric

L = Length of the microstrip patch
The summary of the calculated characteristics of the designed patch antenna is shown on
Table 3.1. All calculation for square patch dimension is applied onto CST Studio Suite 2006.
Patch characteristics Dimension (mm)

Microstrip line width (w0) 3.00
Patch width (W) 37.00
Effective dielectric constant (εeff) 4.35
Extended incremental length (∆L)= 0.732
Patch effective length (Leff)= 29.94
Patch actual length (L) 28.48
Table 3.1. summary of patch characteristics
Figures 3.3 show simulation result of return loss for single element obtained by using CST
Studio Suite software. According to this figure, the result of the return loss of a single patch
design has a good result at frequency of 2.4GHz which is-31.88dB which could be
considered as a good result. Where at the resonant frequency of 2.4GHz which is the
intended design frequency has a value of -10dB. The bandwidth obtained from the
simulation of this microstrip antenna is 108.7 MHz which in percentage value is 4.05%.
Fig. 3.3. Return loss simulation results of a single patch design.
Fig. 3.4. E-plane and H-plane for single patch design
From the radiation pattern as shown in Fig 3, the normalized value of the radiation pattern
which 50Ω input impedance will give half power beamwidth value. Half power beamwidth
is a measurement of angular spread of the radiated energy. From this radiation pattern, the
values at 3 dB for E-plane and H-plane are 94.9°and 99.6° respectively. The summary of the
simulation results for single element patch design is shown in Table 3.2. Half power
beamwidth for both E and H-Plane, directivity and gain that has extracted from radiation
pattern are also shown in this table.
Type Single patch

Return loss -31.88 dB
Bandwidth 108.7 MHz (4.05%)
Directivity 6.11 dBi
Gain 2.56 dB
HPBW (E-Plane) 83.6°
HPBW (H-Plane) 80.0°
Table 3.2. Summary of simulation results for single patch antenna.
3.2.2 Square patch slanted +45° and -45° polarized

To gain insight into the behavior of dual polarized antenna, a single inset feed was designed
for geometry slanted at +45º and -45º linear polarized. As indicated in the introduction, all
work was carried out at 2.4 GHz which is implementing onto WLAN application.
(a) (b)
Fig. 3.5. (a) Layout of the +45° slanted polarization patch antenna
(b)Layout of the -45° slanted polarization patch antenna
The basic single linear +45° and -45° polarized microstrip antenna configuration is a shown
in Fig 3.5. The baseline configuration uses a square patch inset-feed technique on the top
layer. All dimension of a single patch +45° and -45° polarized microstrip antenna such as
length, width and inset are calculated exactly using equation 3.11-3.20. Then, a single
element patch is rotated at 45° for antenna slanted at +45° and 45° to produce polarized
needed.
Hence, the width and length of single patch used in slant 45° and -45° are the same which its
width, W and length, L equal to 27.67 mm. However, the inset length, ℓ is changed due to
the band element connected to the square patch. Since slant 45° and -45° have perpendicular
polarizations, the antennas not have much effect on each other and give similar results in
terms of return loss and bandwith.The simulation of return loss and bandwidth of the
design single 45° and -45° polarization are shown in Fig 3.6. All plots contain impedance
data that has been normalized to 50 Ω. The resonant frequency was 2.4 GHz with return loss
of -12.84 dB for single 45° and -16.24 dB for single -45°.
Fig. 3.6. Return Loss [dB] for 45º and -45º polarized antenna
(a) (b)
Fig. 3.7. (a) E and H-plane of the +45° slanted polarization patch antenna
(b) E and H-plane of the -45° slanted polarization patch antenna
From the radiation pattern as shown in Fig 3.7, the normalized value of the radiation pattern
will give half power beamwidth value. The summary of the simulation results for single
element patch design is shown in Table 3.3. Half power beamwidth for both E and H-Plane,
directivity and gain that has extracted from radiation pattern are also shown in the table.
Type Single 45° Single -45°

Return loss -16.84 dB -16.8 dB
Bandwidth 87 MHz (3.7%) 86 MHz (3.6%)
Directivity 5.69 dBi 5.71 dBi
Gain 2.56 dB 2.61 dB
HPBW (E-Plane) 83.4° 89.8°
HPBW (H-Plane) 89.8° 82.5°
Table 3.3: summary of simulation results for single 45° and -45° patch antenna.
3.3 Dual Polarized Array Antenna
3.3.1 1x2 Dual Polarized Array Antenna

After designed the slanted polarized for each +45° and -45°, the combination for both
layouts can give the dual polarized radiation in term of array. A parallel or corporate feed
configuration was used to build up the array. In parallel feed, the patch elements were fed
in parallel by using transmission lines. The transmission lines were divided into two
branches according to the number of patch elements. The impedances of the line were
translated into length and width by using AWR Simulator. Fig 3.8, Fig 3.9, and show the
circuit layout of the 1x2 array antennas with different position of the patch. In this project,
the position of the patch is considered at 45º and -45º to obtain dual linearly polarized.
Fig. 3.8. Layout of the 1x2 +45º polarized array antenna
In Fig 3.8 a single +45° polarized was combined using corporate feed network to produce an
array antenna. The comparison result between single element and 1x2 array antenna was
describe clearly in terms of return loss, radiation pattern and gain. Same like Fig 3.9, this
structure was built using single -45° polarized and combines with two elements to achieve
polarization slant at -45°.
Fig. 3.9. Layout of the 1x2 -45º polarized array antenna
The simulation results for 1x2 array antennas slanted at 45º polarization were 103 MHz and
–28.11 dB for bandwidth and return loss respectively. While, the simulation result for 1x2
array antennas slanted at -45º polarizations were 103 MHz and -31.82 dB for bandwidth and
return loss respectively. Fig 3.10 show simulation result for 45º and -45º polarized 1x2 array
antenna.
Fig. 3.10. Return Loss for 45º and -45º polarized 1x2 array antenna.
(a) (b)
Fig. 3.11. (a) E and H-plane of the +45° slanted polarization patch antenna
(b) E and H-plane of the -45° slanted polarization patch antenna
The resulting radiation pattern of the E-plane and the H-plane of the two element antenna
array is shown in Figure 3.11 (a) and (b), respectively. It is clear from these figures that the
array antenna demonstrates a more directive pattern with better half power beamwidth and
gain compared to that of individual patch.
Fig. 3.12. Layout of the dual polarized 1x2 array antenna

Using built single patch slant at 45° and -45° polarization; 2-element array patch had
designed and simulated in CST Studio Suite 2006 as shown in Fig 3.12. The array network is
used to combine the 2 element of single patch antennas. A microstrip feed line has
connected to the patch from the edge of the substrate.
An array of 1x2 dual polarized array antenna is build from combination of slant +45° and
slant -45°. In order to combine, corporate feed again is involved to connect a single +45° and
-45° polarized. According to the layout in figure 3.12, the antenna exhibits to have radiation
of dual polarization pattern. The simulated return loss of the 1x2 dual polarization array
antennas are shown in Fig 3.13. The simulation results for 1x2 dual polarization array
antennas were 82.5 MHz and –21.31 dB for bandwidth and return loss respectively.
Fig. 3.13. Return Loss for dual polarization 1x2 array antenna.
Fig. 3.14. Simulation radiation pattern of 1x2 dual polarization array antennas.
Fig 3.14 show the radiation pattern of the 1x2 dual polarization array antennas for E-plane
and H-plane respectively. Overall, this design give better gain and directivity compared 1x2
array at slant 45° and -45° polarization antennas. The simulation of HPBW for E-plane is
about 61.1°; while at H plane is about 89.9°.All simulation data for 1x2 array antenna
designs are tabulated in Table 3.4.
Design Return BW HPBW HPBW

Gain Directivity
Loss (dB) (%) (E-Plane) (H-Plane)
45º 4.29
-28.11 2.98 7.82 57.5 89.8
polarized (102.3MHz)
-45 º 4.29
-31.82 2.96 7.71 54.9 90
polarized (102.3MHz)
Dual-
-17.72 4.42 3.09 8.18 61.1 89.9
polarized
Table 3.4. Simulation results for 1x2 array antennas

Based on the pervious design of 1x2 dual linear polarized a 1x4, 2x2 and 2x4 arrays was
designed and simulated. The initial dimensions for dual linear polarization are the same as
the single polarization element. The patch and feed dimensions were maintained from the
1x2 dual linear polarized designs when designing 1x4 arrays antenna. 1x4 array antennas
had designed and simulated in CST Studio Suite 2006. A microstrip feed line has connected
to the patch from the edge of the substrate. As mention before, the design center frequency
is 2.4 GHz applied for WLAN application. The most important results of the array design
that should be achieved are the return loss result, bandwidth result, radiation pattern results
and gain result. The much element used for designing dual polarized the higher gain and
performance can be achieved.
Fig. 3.15. Layout of the 1x4 +45º polarized array antenna
In Fig 3.15, two set of 1x2 array antenna slant at +45° polarized was combined using
corporate feed network to produce an array antenna. The comparison result between single
element and 1x2 array antenna was describe clearly in terms of return loss, radiation pattern
and gain. Same like Fig 3.16, this structure was built using single -45° polarized and
combines with two elements to achieve polarization slant at -45°.
Fig
g. 3.16. Layout of the 1x4 -45º polaarized array anten
nna
An n array of 1x4 duaal polarized arrayy antenna is build

d from combinatio on of 1x2 array an
ntenna
slaant +45° and slannt -45°. According g to the layout in
n Fig 3.17, the an
ntenna exhibits to
o have
bettter radiation patttern and return lo
oss compared to 1xx2 dual polarized
d array antennas.
g. 3.17. Layout of the 1x4 dual lineer polarized array

Fig y antenna.
he simulated retu
Th urn loss of the 1xx4 microstrip arra
ay is shown in Fiig 3.18. The simu
ulation
ressults for 1x4 array antennas were 79.4 MHz and -2 25.74 dB for banddwidth and returrn loss
resspectively. Fig 3.19
3 shows the radiation
r pattern
n for 1x4 array antenna. Note in i this
raddiation pattern is has consist of muutual coupling beetween the radiatting elements.
Fig
g. 3.18. Return Lo
oss for dual polariization 1x4 array antenna.
Fig. 3.19. Simulation radiation pattern of 1x4 dual polarization array antennas.
The simulation radiation pattern of the 1x4 dual polarization array antennas for E-plane and
H-plane are shwon, respectively. The HPBW achieved for the E-plane and the H plane is
about 524.6° and 89° respectively. The HPBW show that at H-Plane cut is better compared to
E-Plane cut. Moreover, there is a null appears in E-Plane pattern result of 1x4 array patch
design which decrease the HPBW lower than 2x2 dual polarization array antenna. At 2.4
GHz as shown in figure 4.24, the antenna directivity is about 8.673 dBi while antenna gain is
about 5.01 dB.

As seen in Fig 3.20, the 2x2 dual linear polarized designs are feed by coax probe. This was
integrated with 1x2 dual polarized array antenna and feed at centre of the quarter wave
transmission line using coaxial technique. Compared with the expected result for a single
element design, this result can be considered as a better result where a single microstrip
element produces a very low gain. The most important results of the array design that
should be achieved are the return loss, bandwidth, radiation pattern and gain result.
Fig. 3.20. Layout of the 2x2 dual linear polarized array antenna.
Thhe simulated retu

urn loss of the 2x
x2 microstrip arra
ay is shown in F
Fig 3.21. As menttion in
pervious chapter th
he design was useed coax probe coompare to other d
design use transmmission
lin
ne technique. The square patch dimmension was maiintained from thee single element design.
d
Thhe simulation results for 2x2 array
y antennas weree 89 MHz and -37.45 dB for band dwidth
and return loss.
g. 3.21. Return Lo
Fig oss for dual polariization 1x4 array antenna.
Fig
g. 3.22. Simulation
n radiation patterrn of 2x2 dual polarization array aantennas.
Acccording to Fig 3.22, the antenn na gain for this design is betterr comparing 1x2 array
antennas which 1.22 dB higher. Thiss radiation patternn show the E-Plane and H-Plane for f 2x2
duual polarization array antenna. Thhe HPBW show th hat at H-Plane cuut is better compaared to
E-PPlane cut. Moreoover, there is a nu
ull appears in E--Plane pattern result of 2x2 arrayy patch
design. This may due to mutual cou upling occurred inn arrays, beside tthat each four eleements
in the array design configuration is facing
f the back of
o each other, whiich also influencee in the
ull that appeared in
nu i the radiation pattern
p results.
3.4 Measurement result
3.4.1 Dual Polarized 1x2 Array Antenna measurement result
Fig. 3.23. Return Loss [dB] for 1x2 dual linear polarized array antenna.
The comparison between simulated and measured result was shown in Fig 3.23. The
measured of return loss slightly different at desired frequency compare to simulated result.
This because due to error on fabrication process. Since, the simulation result of the return
loss has a value of -17.72dB at resonant frequency of 2.4GHz. While the fabrication results of
the return loss has a value of -18.28dB at resonant frequency of 2.53GHz.
Fig. 3.24. 1x2 array antenna radiation pattern fabrication results
The radiation pattern for this antenna is presented in Fig 3.24, where it can be seen that the
pattern seem like radiating in slant 45° and -45°. The gain of this antenna is 2.83 dB, which is
lower than 0.26 dB from simulation result.
According to Fig 3.25, the result of the return loss of the 4-elemnt array patch design has a
good result at frequency of 2.5 GHz which is-23 dB. This result could be considered as a
good result. Where at the resonant frequency of 2.45GHz which is the intended design
frequency has a value of -9.8dB. However, the bandwidth of measurement value is lower
than simulation which is only 3.03%.
Fig 3.26 show the measurement radiation pattern of the 1x4 dual polarization array
antennas. The HPBW achieved for the antenna is about 54.6°. At 2.4 GHz as shown in this
pattern, the antenna gain is about 4.37 dB. From the measurement result, one can considered
that there is a variation in the resonant frequency which shift to 2.5 GHz compared to the
simulation result. According to this variation, the other measurement method like radiation
pattern of both the electrical field and magnetic field, gain and directivity will be applied
using the resonant frequency of the return loss fabrication result. Since, the resonant
frequency of 2.5 GHz has the best value compared to the intended resonant frequency of the
design which is 2.4 GHz.

The measurement result of return loss for 1x4 microstrip array is shown in Fig 3.27. The
measurement results for 1x4 array antennas were 3.615% and -23.74 dB for bandwidth and
return loss respectively. The resonant frequency for fabrication result has shifted by 2.49
GHz which is 5.4% from the simulation resonant frequency. The root cause of the shift is
could be due to the FR4 board has εr that varies from 4.0 to 4.8. In practical world, a material
which has varying εr along a length/width/height, will affect resonant frequency to shift.
The other factors affecting etching accuracy such as chemical used, surface finish and
metallization thickness also could be the reason for the resonant frequency shifting.
According to Fig 3.28, the beam pattern for 2x2 dual-polarizations has lower sidelobe level
compared to 1x2 and 1x4 antennas, but the bandwidth at resonant frequency was very
narrow. The narrow bandwidth characteristic of 2x2 antennas can be improved by adjusting
the distance of array network, which is quarter wavelength between the patches. This
enhancement was achieved without any significant degradation of the beam patterns and
bandwidths. The HPBW achieved for the antenna is about 87°. At 2.4 GHz as shown in Fig
3.28, the antenna gain is about 3.57 dB.
3.4.5 Comparison of the simulation and measuremet result

Table 3.5 shows a comparison between simulation and fabrication results of the radiation
pattern. According to the variation that occurred in the return loss result, the radiation
pattern results were measured by adjusting the resonant frequency at 2.53GHz instead of
2.44 GHz. From this table, one can notice that the HPBW for simulation and fabrication
results are in a good agreement.
The gain of the single element antenna was almost 2.21 dBi, and the gain of 1x2 arrays was
2.83 dBi. By designing more patches, which were 2x2 and 1x4 array antennas, the
enhancement of gain achieved were 3.57 dBi and 4.37 dBi, respectively. The radiation
pattern for 2x2 dual-polarizations has lower sidelobe level compared to 1x2 and 1x4
antennas, but the bandwidth at resonant frequency was very narrow. The narrow
bandwidth characteristic of 2x2 antennas can be improved by adjusting the distance of
radiation, which is quarter wavelength between the patches. This enhancement was
achieved without any significant degradation of the radiation patterns and bandwidths.
1x2 1x4 2x2
Sim Meas Sim Meas Sim Meas
Resonant 2.4 2.54 2.4 2.51 2.4 2.48

Freq(GHz)
Return loss -17.6 -17.3 -21.1 -18.19 -19.4 -21.03

(dB)
VSWR 1.35 1.18 1.37 1.16 1.24 1.17
BW (%) 4.42 3.45 4.41 4.77 5.46 3.61
Gain 3.09 2.83 5.01 4.37 4.29 3.57

Table 3.5. A comparison of the radiation pattern results for simulation and fabrication
4. Conclusion
A high gain of 3 design microstrip patch antennas oriented at 45º and -45º was proposed to
obtain dual polarization. The antennas were operated at resonant frequency, around 2.4GHz
with low VSWR. The return loss, radiation pattern and antenna gain have been observed
forsingle, 1x2, 1x4 and 2x2 dual-polarization microstrip patches array antennas. It can be
concluded that the responses from the 2x2 and 1x4 patches were better compared to the 1x2
array antenna and single patches antenna. Although the results from the measurement were
not exactly the same as in the simulation, there were still acceptable since the percentage
error was very small due to the manual fabrication process.
5. References
Nirmal Kumar Das, Takashi Inoue, Tetsuki Taniguchi and Yoshio Karasawa (2004). An
experiment on MIMO system having three orthogonal polarization diversity
branches in multipath-rich environment, Proceeding of Vehicular Technology
Conferenc , pp.1528-1532.vol2, ISBN: 0-7803-8521-7, Tokyo, 26-29 Sept 2004, Japan.
Collins Brian S. (2000). Polarization-diversity antennas for compact base stations. Microwave
journal, January, 2000.
Narayanan Manoj, Amir Faizal & S. Santosh Palai.(2006). Multi Polarization Antenna
Structures for Efficient MIMO Systems Employing Software Radio Design
Methodology, Proceeding of RF & Microwave Conf, pp 390-392, ISBN 0-7803-9745-2,
Putrajaya, 12-14 Sept, Malaysia
Byoungsun. L, Sewoong Kwon, Hyun.Y, Jewoo. L, Jeho. S.(2006). Modeling the indoor
channel for the MIMO system using Dual Polarization Antenna, Proceeding of the 9th
EuMA, pp.334-337, ISBN 2-9600551-5-5, September, Manchester, U.K
K. Ghorbani and R. B. Waterhouse. (2004). Dual Polarized Wide-Band Aperture Stacked
Patch Antennas. IEEE Transactions on Antennas and Propagation, pp.2171-2174.vol.
52, no. 8, ISSN: 0018-926X, August 2004.
S. B. Chakrabarty, F. Klefenz, A. Dreher. (2000). Dual Polarized Wide-Band Stacked
Microstrip Antenna with Aperture coupling for SAR application. Proceeding of
Antennas and Propagation Society International Symposium, pp 2216-2219, ISBN: 0-
7803-6369-8, Salt Lake City, 16-21 July 2000, UT.
S. Hienonen, A. Letho, Annti V.R (1999). Simple broadband dual-polarized aperture-
coupled microstrip antenna. Proceeding of Antennas and Propagation Society
International Symposium, pp 1228-1231, ISBN: 0-7803-5639-x, Orlando, 11-16 July,
Florida
S. Gao, L. W. Li, M. S. Leong, and T. S. Yeo (2003). A broad-band dual-polarized microstrip
patch antenna with aperture coupling. IEEE Transactions on Antennas and
Propagation, pp. 898-900.vol. 51, no. 4, ISSN: 0018-926X, April 2003.
S. Gao & A. Sambell (2005). Dual polarized Broad- Band Microstrip Antenna fed by
Proximity coupling. IEEE Transactions on Antennas and Propagation, pp. 526-530.vol.
53, no. 1, ISSN: 0018-926X, January 2005.
B. Lee, S. Kwon & J. Choi (2001). Polarisation diversity microstrip base station antenna at
2 GHz using T-shaped aperture- coupled feeds. Microwaves, Antennas and
Propagation, IEE Proceedings, pp. 334-338. Vol 148, no. 5, ISSN: 1350-2417, October
2001
S. K. Padhi, N. C. Karmakar, Sr., C. L. Law & S. Aditya, Sr. (2003). A dual Polarized aperture
coupled circular patch antenna using a C- shaped coupling slot. IEEE Transactions
on Antennas and Propagation, pp. 3295-3298.vol. 51, no. 12, ISSN: 0018-926X,
December 2003.
A. A. Serra, P. Nepa, G. Manara, G. Tribellini & S. Cioci (2007). A wide-band dual polarized
stacked patch antenna. IEEE Antennas and Wireless Propagation Letters, pp. 141-
143.vol. 6, no. 1, ISSN: 1536-1225, 2005.
Kin-Lu Wong, Hao-Chun Tung & Tzung-Wern Chiou (2002). Broadband dual-polarized
aperture-coupled patch antennas with modified H-shaped coupling slots. IEEE
Transactions on Antennas and Propagation, pp. 188-191.vol. 50, no. 2, ISSN: 0018-926X,
February 2002.
B. Lindmark (1997). A novel dual polarized aperture coupled patch element with a single
layer feed network and high isolation. Proceeding of Antennas and Propagation Society
International Symposium, pp. 2190-2193. vol.4, ISBN: 0-7803-4178-3, 13-18 July, 1997.
M. Arezoomand, H. Aziminezhad, F. Hakamizadeh & S. Sadat (2005). A dual polarized
aperture coupled microstrip antenna for GSM 900 MHz systems. Proceeding of
international Symposium on Telecommunications, pp. 539-543, 10-11 September, 2005,
Iran.
J. Choi & T. Kim (2000). Microstrip array antenna for PCS and IMT-2000 base station.
Proceeding of Microwave Conference, 2000 Asia-Pacific, pp. 25-28. 2000.
W.-C. Liu, C.-C. Huang and C.-M. Wu (2004). Dual-polarised single-layer slotted patch
antenna. IEE Electronics Letters, pp. 717-718. vol. 40. no.12, 10 June 2004.
LJ du Toit & JH Cloete (1987). Dual polarized linear microstrip patch array. Proceeding of
Antennas and Propagation Society International Symposium, pp. 810- 813. Vol.25, Jun
1987.
Shun-Shi Zhong, Xue-Xia Yang, Shi-Chang Gao & Jun-Hai Cui (2002). Corner-fed micostrip
antenna element and arrays for dual-polarization operation. IEEE Transactions on
Antennas and Propagation, pp. 1473-1480.vol. 50, no. 10, ISSN: 0018-926X, October
2002.
Shi-Chang Gao & Shun-Shi Zhong (1998). A dual-polarized microstrip antenna array with
high isolation fed by coplanar network. Microwave and Optical Technology Letters.
pp. 214-216, vol.19, no.3, 20 October 1998.
S. C. Gao, L. W. Li, T. S. Yeo & M . S. Leong (2001). Low cost, dual linearly polarized
microstrip patch array. IEE Proceeding of Microwave Antenna and Propogation,
pp. 21-24. vol. 148, no.1, ISSN: 1350-2417, February 2001.
Kim Woelders & Johan Granholm (1997). Cross-Polarization and sidelobe suppression in
dual linear polarization antenna arrays, IEEE Transactions on Antennas and
Propagation, pp. 1727-1740.vol. 45, no. 12, ISSN: 0018-926X, December 1997.
Keyoor Gosalia & Gianluca Lazzi (2003), Reduced size, dual-polarized microstrip patch
antenna for wireless communications. IEEE Transactions on Antennas and
Propagation, pp. 2182-2186.vol. 51, no. 9, ISSN: 0018-926X, September 2003.
M. Ramesh, K. B. Yip (2003), Design Inset Fed Microstrip Antenna, Microwaves and RF
Letters, vol.42, no. 12, Dec. 2003.
A Time-Delay Suppression Technique for Digital PWM Control Circuit 389
18
X
A Time-Delay Suppression Technique

for Digital PWM Control Circuit
Yoichi Ishizuka
Nagasaki University
Japan
1. Introduction
Recently, power management has been introduced to improve the power efficiency of Micro
Processing Unit (MPUs), Field Programmable Gate Array (FPGAs) and Digital Signal
Processor (DSPs). The power management system includes a full operation mode, standby
mode, and sleep mode. The clock frequency, core voltage and/or core current are changed
in each mode accordingly. As a result, the output current of the point-of-load (POL) DC-DC
converters is intermittent and has a high slew rate. A low output voltage, a large output
current and a high speed response are required for the POL. In such a condition for the
control circuit, highly accurate and high-speed control demands that the tolerance of the
output voltage becomes internally severe, advanced by speed-up and lowering of the
voltage of the MPUs, FPGAs and DSPs. A general control method is pulse width
modulation (PWM) control with PID. Generally, such control circuits are composed with
analog circuits and/or simple combination digital circuits.
In these days, robustness or flexible controls for versatile conditions are demanded which
cannot accomplished with analog control circuit. For the control purpose, DPWM control is
a one of appropriate technique (Edward Lam, Robert Bell, and Donald Ashley (2003),
A.V.Peterchev and S.R.Sanders (2003), B.J.Patella, A.Prodic, A.Zirger and D.Maksimovic
iL iO
SW1
Ei SW2 CO eO RO
PWM
Drive Digaital ADC
Circuit Controller
Fig. 1. Common digital control DC-DC converter.
(2004), D.Maksimovic, R.Zane, and R. Erickson (2004), S.Saggini, D.Trevisan and

P.Mattavelli (2007), S.Saggini, E.Orietti, P.Mattavelli，A.Pizzutelli and Bianco(2008)).
Digital control or DPWM can accomplish robust and flexible power control with soft-tuned
parameters and will become popular control technique.
Although, there are some disadvantages in cost and speed, against analog control circuit.
Especially, A/D converter circuit, which doesn’t need for analog control, is the one of the
key circuits which effects on cost and speed. Generally, A/D converter is located in front of
digital controller as shown in Figure 1. Therefore, the transition speed of A/D converter
directly effects on the response speed of the control circuit. And, the cost and speed are
always trade off problem. This problem is especially serious in POL DC-DC converter which
is required to design the control circuit in relatively low-cost and high speed control
response. Moreover, generally, there is sample-hold circuit in front of A/D converter which
degrades the response speed.
A delay in any feedback system degrades the stability and damping of the system.
Especially, in DPWM, if a total of the delays described in above become larger than on-term
of one switching period, a factor of A/D converter becomes Vq/Z shown in Figure 2 where
Vq is a coefficient constant.
An objective of this paper is to design high speed and low cost voltage sensing circuit for
DPWM control circuit for DC-DC converter. And, also real-time PID control method is
proposed. In Sec. II, the details of proposed system are described. In Sec. III, the some
characteristics of the system are confirmed with experimental results. Finally, in Sec. IV, the
summary is described.
r(k) e(k) u(k) d y1

C(z) DPWM G(s)
-
+
y2(k)
H(z) Vq/Z
y1(k)
Fig. 2. Control System.
2. Proposed System
We propose a scheme of a digital control and DPWM circuit for DC-DC converter without
A/D converter shown in Figs.3 and 4, respectively. In this proposed control circuit, most
components are digital components. Analog components for the control circuit are
essentially only D/A converter and analog comparator. Theoretical waveforms of each part
are shown in Figure 5. The control circuit is composed of three major blocks.
2.1 Analog-Timing Converter (ATC)

The first block is ATC block which detects the output voltage eo and outputs the detected
signal to latch register. The maximum output value of D/A converter DAC is set as a sum of
the output reference voltage of DC-DC converter Vref and margin . A digital staircase
waveform data, pre-stored in memory Memory1, is output to DAC synchronized with a
system clock, and converted analog staircase waveforms Vref’ is compared with eo. As soon
as Vref’> eo, the comparator outputs high.
2.2 PID Control with Look-up Table
iL iO
SW1
Ei SW2 CO eO RO
Vcomp
PWM
Drive Digital DAC
Vref’
Circuit Controller
(a) Total Systems.
eo Latch Vg
ATC
Resister
Digital
Controller
DPWM
※ATC:Analog-Tim
(b) Control blocks.
Fig. 3. Proposed DPWM Control DC-DC converter
Digital
Controller eo
ATC
fS Vref +
Vcomp
System address memory c(m)
Up DAC
CLK Counter 8 1 8 Vref ’
8 y2(k)
Dff Q
8 memory 11
a-b nI(k) Dff Q nI(k-1) a
3
+1
PC 8 memory
initial y2(k) y2(k-1) 11 b
Dff Q
10 4
Look-up Table
address’
memory 9
Dff Q
2
DPWM
u(k)
fS’ Up 9 Digital PWM
PLL
Counter Comparator
Fig. 4. Proposed digital control circuit.

Fig. 5. Theoretical waveforms.
u(k) which is output from Memory2 is calculated by general PID digital control laws as
u  k   u Ref  K P e  k   K I nI  k   K D  e  k  - e  k - 1  (1)
where uRef is a reference value of u(k), e(k) is an digitalized error value between r which is
digitalized reference voltage Vref in switching term k, and nI(k)=nI(k-1)+e(k)( G. F. Franklin，
J. D. Powell and M. L. Workman (1997)). KP, KI and KD are a proportional gain, an integral
gain and an derivative gain, respectively.
Equation (1) can be transformed to
u  k   uRef -  K P  K I  r
KI K
 A{ y2  k   nI  k -1 - D y2  k -1} (2)
A A
where A = K P + K I + K D and y2(k) is digitalized output voltage eo in switching period k.

In Figure 4,
KI
a nI  k -1 (3)
A ,
K
b  D y2  k -1 (4)
A .
Memory3 and Memory4 store a and b, respectively.

In Eq. (2), a - b in the term k is pre-calculated in the term k-1 and the obtained value
becomes the initial value of programmable counter PC of the term k. And, address’ which
indicates address of Memory2 is incremented with system clock and u(k) is called from
Memory2, simultaneously.
address ' =y2 (k)+a - b . (5)

From (2) and (5),
u  k   uRef -  K P  K I  r  A{adress '} (6)
.
Therefore, u(k) is determined as soon as y2 (k) is detected.
2.3 DPWM
In this system, on-term Ton(k) of DPWM signal is decided by u(k), which is normalized Ton(k),
and system clock frequency fS’as
Ton(k)=u(k)/fs’ (7)
u(k) is decided by latched value of Memory2.

In parallel with the processing of ATC block, the u(k) is called with system clock and latched
by ATC output as trigger.
3. Sensing Resolution
3.1 Resolution Increasing
As described in previous section, all blocks are synchronized with only one clock source.
From this advantage, all blocks are modified in easy way.
In this paper, voltage sense resolution increase of output voltage is proposed.
In proposed system, R-2R ladder type D/A converter is used. The output voltage is set
between Vref +(=Vref +) and Vref -. Therefore,
c(m) + (2.8)
Vref ' = (Vref - Vref- ) + Vref
2n
where n is bits. Also, least significant bit (LSB) voltage aLSB becomes
1 + (2.9)
a LSB = (Vref - Vref- )
2n
In this paper, Vref- is set to Vref / 2. Almost n+1 bits resolution can be realized by n bits digital
system as shown in Table 1.
3.2 Sensing Time Delay

With this method, sensing time is increased. To avoid the time delay, the reference voltage
waveform data pre-set in memory1 Vref’ is modified as shown in Figure 6.
LSB
bits Reference voltage voltage
value
n[bit] Vref+[V] Vref-[V] aLSB[mV]
[3] 8 1.7 0 6.64
[3] with 9bits 9 1.7 0 3.32
Table 1. Relations between parameter and quantization error of D/A Converter
Vref+
eo
Vref’[V] Vref’
eo [V] Vref
-
0
t
Vcomp[V]
0 t
Fig. 6. Modified reference voltage waveforms
4. Prototype Circuit Experiments

4.1 Experimental Conditions
Some experiments are performed to verify the scheme. The proposed controller with
prototype circuit is shown in Figure 7. The digital controller part is designed in FPGA Altera
Stratix with Quartus II. 149 logic elements and 1 PLL block are used. All mem-ory blocks,
Memory1, Memory2, Memory3 and Memory4, are including in the logic elements.Intersil
CA3338MZ is used as 8bit DAC. National Semiconductor LMV7219 is used as an analog
comparator.
The DC-DC converter topology is basically same as Figure 3. The experimental conditions
are shown in Table 2.
c(m) Vref’ Analog Vcomp

FPGA DAC
Comparator
Intersil
National
CA3338MZ Semiconductor
eo
Altera: Stratix LM360N
Vcomp
(EP1S10F780C7ES)
Driver
PWM SW1
Texas
Instruments
SW2
UCC27222
Fig. 7. Prototype proposed control circuit.
The static experimental waveforms are shown in Figure 8.
From this result, it is able to confirm that the output voltage sensing is done within the on-
width of PWM signal.
Ei[V]
0 V:5V/div
eo[V]
V:1V/div
0
1.7
V:500mV/div
0.75
Vref
0
Vcomp[V]
V:1V/div
0
PWM[V]
V:1V/div
0
H:2 s/div
Fig. 8. Experimental waveforms
4.3 Dynamic Characteristics

Figure 9 and 10 show dynamic characteristics with load current io changing between 0.5A
and 2.5A, respectively. The mixed-signal oscilloscope Textronix MSO4034 is used to
measure analog and digital signal, coinstantaneously. The load current chang-ing is
(a) H: 20s/div.
(b) H: 2s/div.
Fig. 9. Dynamic characteristics (from 0.5A to 2.5A)

(a) H: 20s/div.
(b) H: 2s/div.
Fig. 10. Dynamic characteristics (from 2.5A to 0.5A).

performed with 1kHz driven power MOSFET parallely-connected to load resistance. Yellow
and Blue line shows the output voltage and the output current, respectively. The 9 bits pulse
waveforms shown at the bottom of Figure 9 are calculated DPWM of FPGA.
Figure 9 shows the sudden load current increasing results. From these results, after the 1s
voltage drop, the output voltage immediately recovers to the reference voltage.
Figure 10 shows the sudden load current decreasing results. From these results, after the 1s
voltage rising, the output voltage immediately recovers to the reference voltage.
5. Conclusion
This paper describes a digital PWM controller IC without A/D converters. The analog
timing converter (ATC) is proposed for output voltage sensing. In this system, analog circuit
are realized with an comparator and an D/A converter.
6. References
A. Ichinose, Y. Ishizuka, and H. Matsuo (2006). A Fast Response DC-DC Converter with
DPWM Control, Technical Report of IEICE, vol. 105, no. 538, EE2005-58, 67-71
Y. Ishizuka, M. Ueno, I. Nishikawa, A. Ichinose, and H. Matsuo (2007). A Low-Delay Digital
PWM Control Circuit for DC-DC Converters, IEEE Applied Power Electronics
Conference (APEC’07), 579-584
M. Nishi, Y. Asako, Y. Ishizuka, and H. Matsuo (2008). A control circuit composition and
several characteristics of the proposed DPWM controlled POL, Technical Report of
IEICE, vol. 107, no. 430, EE2007-46, 13-18
Edward Lam, Robert Bell, and Donald Ashley (2003). Revolutionary Advances in
Distributed Power Systems, Proc. IEEE APEC ’03, 1.5
A.V.Peterchev，and S.R.Sanders (2003): “Quantization Resolution and Limit Cycling in
Digitally Controlled PWM”，IEEE Trans. on Power Electronics, Vol. 18, No. 1, 301-
308
B.J.Patella, A.Prodic, A.Zirger, and D.Maksimovic (2004). High-frequency digital PWM
controller IC for DC-DC converters, IEEE Transactions on Power Electronics, Vol.
18 D.Maksimovic, R.Zane, and R. Erickson: “Impact of Digital Control in Power
Electronics”, IEEE International Symposium on Power Semiconductor Devices & ICs,
Kitakyushu, Japan, 13-22
Kaiwei Yao (2004). High-Frequency and High-Performance VRM Design for the Next
Generation of Processors, Doctor thesis of Virginia Polytechnic Institute and State
University
S.Saggini, D.Trevisan, and P.Mattavelli (2007). Hysteresis-Based Mixed-Signal Voltage-
Mode Control for dc-dc Converters, IEEE Power Electronics Conference (PESC’07),
Orlando, Florida
S.Saggini, E.Orietti, P.Mattavelli，A.Pizzutelli and Bianco (2008). Fully-Digital Hysteretic
Voltage-Mode Control for dc-dc Converters based on Asynchronous Sampling,
IEEE Applied Power Electronics Conference (APEC’08)，Issue , 24-28, 503 - 509
G.F.Franklin, J.D.Powell, and M.L.Workman (1997). Digital Control of Dynamic Systems,
Addison Wesley Longman Press，Menlo Park. CA
Layer 2 Quality of Service Architectures 399
19
X
Layer 2 Quality of Service Architectures

Christos Bouras1,2, Vaggelis Kapoulas1, Vassilis Papapanagiotou1,2,
Leonidas Poulopoulos2,3, Dimitris Primpas1,2 and Kostas Stamos1,2
1Research
Academic Computer Technology Institute, Greece
2Computer Engineering and Informatics Dept., University of Patras, Greece
3Greek Research and Academic Network - GRNET, Greece
1. Introduction
Quality of Service (QoS) provisioning has become indispensable in today’s networks. Most
existing QoS solutions are deployed in Layer 3 (network layer). In order to provide end-to-
end QoS guarantees in these networks, the need for Layer 2 QoS deployment as well as the
cooperation between any existing Layer 3 QoS deployment must be studied. QoS
provisioning in Layer 2 is very important to networks that are primarily based on Layer 2
infrastructure as it is the only way to provide QoS on the network. Furthermore, networks
based on both Layer 2 and Layer 3 network devices could benefit from a more integrated
approach in end-to-end QoS provisioning that includes both Layer 2 and Layer 3.
In today’s broadband networks, congestion does not necessarily occur at the edge of the
network (the link interconnecting the subscriber to the network core): congestion is equally
likely to occur at the edge and in the core of the network. A common congestion cause of in
broadband networks is the capacity mismatch in different parts of the network core. This
calls for protection measures at the network perimeter and unified or interoperable QoS
schema across all network (both layer 2 and layer 3).
Moreover, Layer 2 QoS is lightweight, easily implemented and independent of Layer 3.
Because of its independency, it can also be applied to non-IP networks where any QoS
provisioning was impossible or very difficult. In this chapter, we examine the cooperation
between Layer 2 and Layer 3 QoS in IP networks. When discussing Layer 2 devices and
procedures in this chapter, we are specifically referring to Ethernet technology switches,
which have become the dominating Layer 2 technology during the past years and have
largely substituted older technologies at the same layer, such as ATM and Frame Relay.
Layer 2 Ethernet switches rely on 802.1p standard to provide QoS. The standard 802.1p is
part of the IEEE 802.1Q (IEEE, 2005) which defines the architecture of virtual bridged LANs
(VLANs). This architecture uses tagged frames inserted in Ethernet frames after the source
address field. One of the tag fields, the Tag Control Information, is used by 802.1p in order
to differentiate between the classes of service. More specifically, the 3 most significant bits of
the Tag Control Information field known as Priority Code Point (PCP) are used to define
frame priority. Taking advantage of PCP, QoS in Layer 2 can be applied.
Layer 2 QoS experiments with Ethernet switches have been conducted and described in
(Ubik & Vojtech, 2003). In (Ubik & Vojtech, 2003) 4 Layer 2 QoS experiments are conducted
and effects on link throughput and packet loss are shown. Other researchers such as
(Liotopoulos & Guizani, 2002) have dealt with Layer 2 QoS in ATM networks. An
interesting application of L2 Ethernet QoS has been studied in the field of avionics networks
with the demand for low latency and jitter in (Wernicke, 2006) and (Jacobs et al., 2004),
while 802.1p has been studied as an approach for the improvement of traffic performance
originating from collaborative systems applications in (Perez et al., 2006).
In the next sections of this chapter, we discuss the issue of Layer 2 QoS deployment, and in
particular we present in detail:
 The cooperation of Layer 2 QoS with possibly pre-existing Layer 3 QoS architectures in
MAN broadband networks.
 The architecture for Layer 2 QoS deployments, with analysis of the authors’ experience
at GRNET as a case study.
 The status of Layer 2 QoS support in various vendors’ equipment according to our
experience.
2. Integration of Layer 2 and Layer 3 architectures

Quality of Service support was initially enabled on routing devices using the available fields
in the IP header. Therefore, a comprehensive architecture has to take this into account and
be able to accommodate the addition of more devices, which do not route packets, in the
overall QoS architecture. For example, Layer 2 Ethernet switches rely on 802.1p standard to
provide QoS. An example of a network where the need for integrated Layer 2 and Layer 3
QoS provisioning has been investigated is the Greek Research Network (GRNET, 2009a),
which is described in detail in the next section of this chapter.
In general, the integration of Layer 2 and Layer 3 QoS architectures can take several forms,
depending on the purpose for which the switching and/or routing devices are used (access,
core), their functionality (since several devices have capabilities that tend to blur the L2/L3
distinction, for example switches that can inspect the IP header), the policies of the network
domain and other factors. A basic distinction of the most common choices is provided
below:
 Layer 2 devices at the edge (access) of the network and routing devices (possibly using
MPLS) in the core: A very common case, this scenario is studied in detail in the next
sections of this chapter, as it largely describes the structure of GRNET.
 A combination of Layer 2 and Layer 3 across the network, with switching devices
comprising L2 MANs: Sometimes it is convenient to build “islands” where traffic is
simply switched and not routed. This approach is usually accompanied with extensive
VLAN usage for traffic management purposes. Part of the GRNET network has been
built with this philosophy, and its practical implications are described in the relative
section of the chapter.
 Switched-based network, with Layer 3 devices at the periphery: A growing tendency in
high speed modern networks has been towards connectivity at lower layers, bypassing
expensive routing functions. In this case, L2 QoS takes center stage in the planning of a
QoS architecture.
In all cases, a proper QoS architecture has to include the basic steps of classification,
policing, marking and scheduling. The design of the network and the specific mix of devices
and requirements will determine whether both Layer 2 and Layer 3 devices perform all of
these functions, or whether these functions are distributed.
A network that serves several clients at its periphery will generally have to consider the
traffic sources untrustworthy, in terms of proper packet classification and marking.
Therefore, the edge devices of the network will have to take up this role. Classification
typically requires packet inspection and is therefore more suited for higher-layer devices,
although several switching devices do support some form of inspection of encapsulated
higher layer headers. If that is not the case, then traffic might have to be classified when it
first traverses suitable equipment. This means that incoming traffic might not be subjected
to prioritized treatment immediately upon arriving at the borders of the network.
3. Layer 2 QoS case studies

GRNET is the Greek National Research and Education Network (NREN) (GRNET, 2009a).
GRNET is a mixed IP- and Ethernet-based network, operating at Gigabit speeds. Together
with the high-speed LANs of its subscribers (universities and research institutes) and the
European academic and research backbone, GEANT, GRNET forms a set of hierarchically-
federated networks.
However, because part of its backbone consists of switch based MANs, this architecture had
to be extended in order to encompass Layer 2 (Ethernet) devices, which should
appropriately implement QoS policies and QoS signaling/metering as well.
The GRNET backbone consists of network nodes in 8 major Greek cities, namely, Athens (2
PoPs), Thessaloniki, Patras, Ioannina, Xanthi, Heraklion, Larisa and Syros as shown in Fig. 1, a
screenshot from the GRNET’s GoogleMaps Topology Visualization (GRNET, 2009b).
The WAN network is built on DWDM links with 2.5Gbps capacity (STM-16 lambdas). The
access interfaces of the routers are using Gigabit Ethernet technology and connect the 130
subscribers of GRNET which consist of universities, technological educational institutes,
research institutes, two content providers (the Greek National Television and the Greek
Parliament) and the school network. During the last few months the major Greek
Universities upgraded their connection speed to GRNET from 1-Gbps to 10-Gbps. In
addition to the WAN, GRNET also has 2 distinct MAN networks. The Athens MAN is
router-based (Fig. 2), whereas the Crete MAN is switch-based (Fig. 3), with a router in the
main aggregation site (Heraklio).
The Greek Research and Academic Network (GRNET, 2009a) has deployed for several years
a Layer 3 QoS service based upon the features provided by the MPLS technology deployed
in the core of the GRNET network, and DiffServ architecture. This architecture allows the
support of multiple classes of service. The focus is on three separate classes of service,
namely IP Premium for absolute performance guarantees, best effort for the usual treatment
of traffic packets and Less than Best Effort (LBE) for non-critical traffic that can be dropped
first in case of congestion. IP Premium service is a circuit-like subscriber-to-subscriber
service, where both subscriber end-networks and the necessary bandwidth allocation are
known at request time. IP Premium service is provided using a provisioning tool called
ANStool (Varvitsiotis et al., 2005; GRNET, 2009c). LBE is provided unprovisioned, which
means that each subscriber decides on its own and uses this service simply by marking the
packets appropriately. In order to provide the QoS service, the Layer 3 network equipment
(routers) has to perform traffic marking, classification, policing and shaping. Per-flow
functions are performed at the edge routers of GRNET network, while core routers only
perform per-traffic class functions, based on the MPLS Exp field.
Fig. 1. GRNET’s Layer 3 Country Network Topology
The above service design has several implications for traffic between two GRNET clients
(such as institutions, universities or other research organizations). It means that traffic
coming out of GRNET network (“output” for GRNET edge routers) has been subjected to
the specified QoS mechanisms. However, traffic coming into the GRNET network (“input”
for GRNET edge routers) receives no treatment up to the point of reaching the edge Layer 3
device (router) of the GRNET network.
Fig. 2. Athens’ MAN
In the most common case (except Crete’s MAN), traffic between the GRNET client and the
GRNET edge router will go through one or more Layer 2 devices (Ethernet switches). For
the simple case where only one Layer2 device is located between GRNET and the
subscriber, we use scripting to query the speed and bandwidth settings at each L-2 border
interface. We then reflect the speed setting of the border interface into a traffic shaping
queue for the respective VLAN at the L-3 border. Using this technique, we make sure that
the congestion points occur only at the L-3 border.
With the advent of hybrid networks and the tendency to carry high speed network traffic at
the lowest layer possible (in order to avoid handling it with costly Layer 3 equipment), this
part of current and future network is bound to expand. Whether this Layer 2 part of the
network forms multiple paths between the connected Layer 3 devices (in which case the
need for spanning tree algorithms arises in the common Ethernet case) determines in large
part the complexity of the Layer 2 QoS solution that will have to be adopted.
Therefore, in designing and implementing the service described in this chapter, we took into
account the current need for controlling traffic behavior at the edge of the GRNET network
(where it slips from current Layer 3 QoS model) and we also considered the increasing
importance of that part of the network to the overall network architecture in the future.
Fig. 3. L2 Crete’s MAN
Fig. 4. Schematic of GRNET core/edge/L2-only edge network parts

3.1 Implementation Issues

IEEE 802.1Q (also known as VLAN tagging) defines a 3-bit field called Class of Service
(CoS), which can be used in order to differentiate traffic. Table 1 shows the 8 possible values
of the CoS field and their original purpose.
CoS Acronym Purpose

0 BE Best effort
1 BK Background
2 - Spare
3 EE Excellent Effort
4 CL Controlled Load
5 VI “Video” < 100 ms latency and jitter
6 VO “Voice” < 10 ms latency and jitter
7 NC Network control
Table 1. CoS field values
For the purposes of our deployment, we have adopted the usage of CoS value 5 for marking
premium traffic (which requires quality of service), CoS 0 for best-effort traffic and CoS 1 for
less than best effort traffic. Traffic is marked as less than best effort when it is of minor
importance, and is allowed to occupy at most 1% of the total bandwidth. The usage of CoS
value 5, indicates that the default DSCP-to-CoS mapping scheme is followed, bearing in mind
that in GRNET IP Premium is marked with DSCP 46 as denoted in (Varvitsiotis et al., 2005).
In the case of the GRNET (GRNET, 2009a) network, end to end traffic between client
network interconnected through GRNET will traverse a combination of Layer 2 (switches)
and Layer 3 devices (routers). To this end, the policies of the edge routers of the GRNET
network must be adapted so that ethernet frames belonging to premium traffic are marked
with CoS 5 at the output. Additionally, the port of the subscriber’s switch which is
connected on the edge router has to be configured in order to trust the values of CoS of the
received traffic streams. Because CoS is part of the standard 802.1Q (IEEE, 2005), the port on
which the edge router is connected must be in trunk mode. When a port is in trunk mode it
uses the tagged frames of 802.1Q (IEEE, 2005) to communicate, which contain CoS and other
information about virtual bridged local area networks (VLANs).
Fig. 5. Schematic of L3 and L2 QoS actions

The procedure of deploying Layer 2 Quality of Service is quite similar to the one of Layer 3
QoS. Classification procedure is applied in incoming packets along with policing functions.
Next, if traffic is in profile it is marked accordingly, else the packet is marked down or
dropped. Next, the packets enter the switch’s queues according to their markings.
Queue management and scheduling are the most important issues in configuring Layer 2
Quality of Service. L2 Ethernet switches support a number of ingress and egress queues
(switches in our testbed support 2 ingress queues and 4 egress queues). Scheduling in our
equipment (Cisco Systems devices) is performed using the Shaped Round Robin (SRR)
algorithm. The ingress queues can only be shared whereas the egress queues can also be
shaped. When queues are shared their bandwidth is guaranteed to configured weights but
is not limited to it. When a queue is empty, the other queues in shared mode share its
unused bandwidth. When a queue is shaped it is guaranteed a percentage of bandwidth but
it is rate limited to that amount. By default, from the ingress queues the second one is used
to handle high priority traffic, and from the egress queues the first one is the high priority
queue and it cannot be changed. Additionally, the high priority egress queue is by default
shaped to occupy 1/25 of total bandwidth, and when a queue is shaped any sharing settings
are overridden. When the expedited output queue is enabled (as in our experiments, using
the command priority queue-out), the expedited queue is serviced first until it is empty and
then the other queues are serviced in a round-robin manner. More information can be found
in (Cisco, 2009). In the GRNET network the edge routers shape the traffic on the output, so
there is no need to shape the queues on the switches, however in our experiments, we use
policies to limit the bandwidth when needed. Additionally, in the GRNET network the
switch trusts the CoS of the packets coming from a GRNET edge router. By contrast, in our
experiments traffic was classified by the switch and the DSCP field (46 for premium traffic, 0
for best-effort) was set, as in testing equipment policies that set CoS are not supported.
In order to verify the proper configuration and operation of the Layer 2 QoS service,
extensive experimentation was performed in both laboratory and production environments
(Bouras et al., 2008). The conducted experiments acknowledged and proved that the
activation of L2 QoS does benefit the overall result that was previously produced by only L3
QoS in GRNET’s network.
An additional step is the enhancement of the GRNET QoS provisioning tool (GRNET, 2009c)
with the necessary functionality and features in order to manage the L2 QoS service as well.
In this direction, a module was developed and integrated into GRNET’s QoS provisioning
tool. This service is unprovisioned and does not require any type of interaction of GRNET’s
customers with the Layer 2 module. Of course, for the proper operation of the end-to-end
QoS provisioning, GRNET’s clients must submit a Layer 3 request in the first place.
The Layer 2 module of the QoS provisioning tool provides the network administrator with
the appropriate vendor-specific configuration, which in turn is applied to the network
switches. A view of the switches’ list is presented in Fig. 6.
When a switch is selected, the administrator must activate QoS at the switch by selecting
“Standard CPE QoS configuration” as shown in Fig. 7, an option that provides the
appropriate configuration. Finally, the requested interfaces are selected. Once the Layer 3
configuration is applied to router and the Layer 2 configuration is applied to the switch, the
customer is given end-to-end QoS.
Fig. 6. View of network switches
Fig. 7. Switch QoS configuration options

The interoperability between Layer 3 QoS and Layer 2 QoS takes actual place at the border
router and in particular, at the interface which is connected to the Layer 2 device. As the
majority of GRNET’s switches are multilayer (mls), meaning that it is possible to classify
based either on DSCP or CoS, there are two options concerning the marking of packets or
frames performed by the border router at the egress:
 CoS marking: For each packet with a given DSCP value, mark the frame with the
corresponding CoS value as indicated in Table 2.
 DSCP marking: For each packet with a given DSCP value keep this value intact.
Regarding the switch, at the ingres of its interfaces, the DSCP and CoS values of the packets
and frames received respectively are trusted as marked by the router. At this point it should
be mentioned that the majority of vendors, by default, disable the trusting of DSCP and CoS
values at the ingress interfaces of Layer 2 devices. In order to achieve integration between
Layer 3 and Layer 2 for QoS, trusting of DSCP and/or CoS values should be enabled. Thus
internal DSCP-to-CoS mapping at the switch is avoided. However, this should be performed
in combination to very strict policies so as to avoid abuse of the QoS service.
GRNET’s switches provides the flexibility to classify incoming traffic based either on DSCP
or CoS. CoS classification is preferred as it can provide backwards compatibility with some
no-mls capable switches.
DSCP CoS Description

46,47,40 5 IP Premium
0 0 Best Effort
8 1 Less than Best Effort
Table 2. DSCP-to-CoS mapping
3.2 Multiple L2 paths in Crete’s MAN

An exception to the more common structure of the GRNET network described above is the
part of the GRNET network at the island of Crete, which forms the Crete’s MAN. It consists
exclusively of L2 Ethernet switches which are aggregated to the only L3 device, a router at
the city of Heraklio connected to the rest of GRNET (Fig. 3). Some of the L2 interfaces are
therefore considered part of the GRNET core network (the ones which form the MAN itself),
while the rest connect to client networks, similarly to the common case discussed in
previous sections. Therefore, for the latter case, the existing L2 approach can be still utilized.
The core L2 devices form a ring consisting of 3 Ethernet switches (Cisco 3750), with several
client networks connected on each one of them. Traffic between the client networks in Crete
and towards the rest of the GRNET network is carried in VLANs in order to form isolated
VPNs. A related limitation of the current Cisco L2 equipment is that it does not support QoS
classification of traffic on VLAN ports, but only on physical ports.
Each VLAN has its own spanning tree which directs the traffic accordingly, and which can
be quickly adjusted using Rapid Spanning Tree Protocol (RSTP) for link failure recovery and
load balancing. In the case of a link failure, VLAN traffic using the failed link will be
redirected due to the corresponding spanning tree protocol switching a blocking link’s state
to forwarding. This means that assuming the worst case scenario, a core L2 link will have to
be able to carry the whole of the traffic traversing the core of Crete’s L2 MAN. Under such
an assumption, the worst-case dimensioning algorithm will have to allow premium traffic
reservations up to the specified allocated percentage for the whole of the L2 MAN
(conversely this can be expressed as the requirement that the allocated percentage should be
calculated by adding all allowed traffic reservations through the MAN). The premium
allocated percentage can follow the guidelines set by L3 allocations for L3 links of similar
bandwidth. The symmetry of Crete’ MAN regarding link capacity simplifies this calculation.
The worst case assumption has also been the selected approach for premium reservations at
the L3 part of the network, and is therefore a natural extension for this case.
4. Status of Layer 2 QoS support

Many vendors have presented Layer 2 devices (switches) with increased capabilities, which
are able to inspect Ethernet frames and support CoS or DSCP based differentiation. In this
chapter we discuss the approaches chosen by each of the main switching equipment
vendors where we have enabled Layer 2 QoS capabilities.
Cisco is possibly the most important vendor of network equipment and, as discussed above,
GRNET has traditionally based its network infrastructure largely on Cisco equipment and in
particular Cisco Catalyst switches. L2 QoS has been extensively tested on Catalyst 2970 and
Catalyst 3750 series that comprise a large part of GRNET’s access network and some parts of
its core MAN networks as described above (Bouras et al., 2008).
GRNET includes many switches from Extreme Networks, and namely SummitX450 and
SummitX350 type switches, which also support L2 QoS (Extreme Networks, 2008). In
particular, Policy-based Quality of Service (QoS) is implemented in ExtremeXOS, the
operating system used by Extreme Networks switches, and it allows the user to specify
different service levels for traffic traversing the switch. The hardware implementation varies
depending on the platform, for example some Extreme Networks BlackDiamond series
switches contain separate hardware queues on every physical port, while other switches
such as the Summit series contain two default queues and several more configurable queues
on a switch-wide level. When two or more queues are contending for transmission on the
same physical port, the switch makes sure to prioritize usage of the port with regard to the
respective queue management parameters. Extreme Networks consider Layer 2 QoS to be
applicable in a number of traffic requirements, and provide specific guidelines for
applications such as voice, video, critical databases, web browsing and file server
applications. Configuration is based on the concept of QoS profiles, which encompass a list
of parameters (depending on the specific hardware implementation, this list generally varies
between different families of products), such as the maximum amount of packet buffer
memory available and the relative weight assigned, or the maximum bandwidth that can be
transmitted, the minimum bandwidth reserved and the level of priority.
L2 QoS support by Extreme Networks switches has been extensively tested by GRNET in
both laboratory and production environments. The GRNET network is in a position to
successfully integrate Extreme Networks switches in a production network comprised of
devices from multiple vendors and has verified their interoperability.
Moreover, during the last few months there have been conducted a series of tests concerning
the interoperability between Layer 2 and Layer 3 devices. The majority of GRNET’s Layer 3
devices is CISCO GSR (12xxx) Series Routers (IOS), while there has been a Juniper T1600
(JunOS 9.4) in production for the last 6 months. Juniper seemed to be much more flexible
and granular than Cisco concerning QoS provisioning and setup (Juniper Networks, 2009).
There have been tests with all the combination of equipment showing that GRNET is
capable of providing end-to-end QoS regardless of vendor.
5. Conclusions
In this chapter we have provided a wide overview of the existing solutions and
deployments for QoS provisioning at Layer 2, with an emphasis on Ethernet-based
deployments, which is the dominant Layer 2 technology. We have discussed the integration
of existing Layer 3 QoS deployments with the introduction of Layer 2 devices (switches)
with relevant capabilities, the implementation issues from a case study implemented in
Greece at the GRNET network, and we have presented the current status of Layer 2 QoS
support for various equipment vendors.
Our future work includes extensive interoperability testing, including Layer 2 QoS solutions
by Cisco, Extreme Networks, Juniper Networks and more vendors such as Nortel. Such
interoperability testing has to include all combinations of vendor equipment and Layer 2 –
Layer 3 interactions. Furthermore, large scale testing and results from production
availability of the services are going to be conducted and analyzed for further service
refinement.
6. References
Bouras C., Kapoulas V., Papapanagiotou V., Poulopoulos L., Primpas D., Stamos K., (2008).
Extending QoS support from Layer 3 to Layer 2, Proceedings of 15th International
Conference on Telecommunications, St. Petersburg, Russia, 16 - 19 June 2008
Cisco (2009). Catalyst 2970 Switch Software Configuration Guide. Chapter 27:
Understanding QoS.
http://www.cisco.com/en/US/docs/switches/lan/catalyst2970/software/release/12.1_14_ea1/
configuration/guide/2970SCG.pdf
Ek Niclas (1999) Department of Electrical Engineering, Helsinki University of Technology.
IEEE 802.1 P,Q - QoS on the MAC level, http://www.tml.tkk.fi/Opinnot/Tik-
110.551/1999/papers/08IEEE802.1QosInMAC/qos.html
Extreme Networks, (2008). ExtremeXOS Concepts Guide, Software Version 12.1, May 2008
GRNET (2009a). Greek Research Network (GRNET) www.grnet.gr
GRNET (2009b) http://netmon.grnet.gr/networkmap/gmindex.php
GRNET (2009c). GRNET’s Advanced Network Services Provisioning Tool
http://anstool2.grnet.gr
IEEE (2005). IEEE Standard for Local and Metropolitan area networks,Virtual Bridged Local
Area Networks 802.1Q. http://standards.ieee.org/getieee802/download/802.1Q-2005.pdf
Jacobs A.; Wernicke J.; Oral S.; Gordon B.; George A., (2004). Experimental characterization
of QoS in commercial Ethernet switches for statistically bounded latency in aircraft
networks, Proceedings of 29th Annual IEEE International Conference on Local Computer
Networks, 2004, 16-18 Nov. 2004 Page(s): 190 – 197
Juniper Networks (2009). JUNOS Configuration Guide, Class of Service
http://www.juniper.net/techpubs/software/junos/junos93/swconfig-cos/swconfig-cos.pdf
Liotopoulos F.K., & Guizani M. (2002). Implementing layer-2, connection-oriented QoS on a

3-stage Clos switch architecture., Proceedings of Global Telecommunications Conference,
2002, GLOBECOM ’02. IEEE Volume 3, Issue , 17-21 Nov. 2002 Page(s): 2741 - 2746
vol.3
Perez, J.A., Zarate, V.H., Cabrera, C., and Janecek, J., (2006). A Network and Data Link
Layer Infrastructure Design to Improve QoS for Real Time Collaborative Systems,
Proceedings of International Conference on Internet and Web Applications and
Services/Advanced International Conference on Telecommunications, 2006,. AICT-ICIW
apos;06. 19-25 Feb. 2006 Page(s): 19 – 19
Ubik S. & Vojtech J. (2003). QoS in Layer 2 Networks with Cisco Catalyst 3550, CESNET
Technical Report 3/2003
Varvitsiotis A., Siris V., Primpas D., Fotiadis G., Liakopoulos A., & Bouras C., (2005).
Techniques for DiffServ-based QoS in Hierarchically Federated MAN Networks –
the GRNET Case, Proceedings of The 14th IEEE Workshop on Local and Metropolitan
Area Networks (LANMAN 2005), Chania. Island of Crete, Greece, , 18 - 21 September
2005.
Wernicke John, (2006). Simulative Analysis of QoS in Avionics Networks for Reliably Low
Latency, Journal of Undergraduate Research,. Volume 7, Issue 2 - January/February
2006.
Secrecy on the Physical Layer in Wireless Networks 413
0
20
Secrecy on the Physical Layer

in Wireless Networks
Eduard A. Jorswieck, Anne Wolf, and Sabrina Gerbracht*
Technische Universität Dresden
Germany
1. Introduction
This chapter provides a comprehensive state-of-the-art description of the emerging field of
physical layer security. We will consider wireless security from an information theoretic view,
which allows us to talk about provable secrecy and to derive ultimate secrecy limits. Our
main focus is on the optimization of transmit strategies and resource allocation schemes under
secrecy constraints.
We will consider the following scenario, which is illustrated in Figure 1: Alice wants to send a
private message to Bob, which should be kept perfectly secret from Eve. Eve listens and tries
to decode the message that Alice sends to Bob.
Bob
communication
Alice
network
Eve
Fig. 1. Communication system with a transmitter (Alice), a legitimate receiver (Bob) and an
eavesdropper (Eve).
In this communication system, Alice is the transmitter, Bob is the intended or legitimate re-
ceiver, and Eve is the eavesdropper. We assume that Bob and Eve perfectly know their in-
dividual channel realization and that Alice has full channel state information (CSI), i.e., she
knows all channel realizations perfectly. This assumption, which is essential for our further
discussion, seems to be unrealistic in the wiretap setting in which Eve probably only listens.
However, this assumption will be justified, if Bob and Eve are both users in a cellular environ-
ment using up- and downlink transmission.
Within this chapter, we give an overview on the research problems and current results con-
cerning secrecy on the physical layer. In the first section, we describe the attacker model and
some conventional cryptographic methods. Afterwards, we introduce the wiretap channel
and define the secrecy on the physical layer. In the second part of the chapter, we present
results for the achievable secrecy rates or the secrecy capacity in various single-user systems
including single-antenna, multi-antenna and multi-carrier systems and provide power alloca-
tion strategies for secrecy rate optimizations. In the third section, we extend these results to
* Part of this work is supported by DFG under grant Jo 801/2-1.
multi-user systems. We study basic elements that can be used to model more complex net-
works and give an overview on current research results on the secrecy capacity regions or
the secrecy rate regions. The chapter is completed with a discussion of the results and open
research problems.
1.1 Attacker Model

We consider a wireless communication system and focus on a cellular system. The transmitter
has perfect CSI for the channels to all potential receivers, irrespective of the fact, whether
the receiver is a legitimate receiver or an eavesdropper. The receivers only know their own
channels perfectly using channel estimation based on pilot signals. Every user of the system
has knowledge of the structure of the system, including all technical details, e.g., codebooks
and transmit strategies.
system • wireless communication (cellular system)

transmitter
knowledge? • perfect CSI for the channels to both, the legitimate receiver and the
eavesdropper
• structure of the system (including all technical details, e.g., code-
books and transmit strategies)
legitimate receiver
knowledge? • only perfect CSI for his own channel
eavesdropper
who? • member of the system
objective? • passive attack, eavesdrops the communication between transmitter
and legitimate receiver, undermines confidentiality of communica-
tion (without interfering)
how? • within range of transmitter
• tries to decode the intercepted message
knowledge? • only perfect CSI for his own channel
Table 1. Attacker model at a glance.
The attacker is a passive attacker. He wants to undermine the confidentiality by eavesdrop-

ping the communication of one or more legitimate users of the system without interfering the
communication between transmitter and receivers. For this reason, we use the terms attacker
and eavesdropper synonymously. The attacker himself is also a user of the system. He is in
reach of the transmitter and tries to decode the intercepted message. He has perfect CSI for
the channel from the transmitter to himself, but he does not know the channel between the
transmitter and the legitimate receiver. Since the eavesdropper is a user of the system, the
transmitter knows the channel to the attacker and is able to fend the attack.
An overview of all important facts of the attacker model can be found in Table 1.
1.2 Cryptography
Currently, the mostly used method to ensure confidentiality in communication systems is the
end-to-end cryptography (Schneier, 1996). What all cryptographic algorithms have in com-
mon is the fundamental attacker model. The sender, namely Alice, wants to send a message to
the receiver, called Bob. Eve, the eavesdropper, should not obtain any knowledge of the mes-
sage content. In order to achieve this, Alice performs a number of mathematical operations
on the original message, predetermined by the cryptographic algorithm and the encryption
key. Bob, who knows which algorithm was used, decrypts the cipher message with his key.
Eve may know the algorithm, but as long as she does not know the key, it is difficult for her
to decipher the message.
There are two basic concepts in the field of cryptography, the symmetric and the asymmet-
ric cryptography. One of the main differences between both concepts is located in the key
management.
secret area
k r
key
generation
key random number
key k
w w
encryption decryption
e(w,k) d( ,k)
plaintext ciphertext = d(e(w,k),k)
Fig. 2. The basic concept of symmetric cryptography, including key generation, key manage-
ment, encryption and decryption.
The classical method of symmetric cryptography requires identical keys at sender and re-
ceiver, as it can be seen in Figure 2. The difficulty is to transmit the key from Alice to Bob
(or vice versa) in a secure and secret way before the communication. In order to avoid this
problem, Whitfield Diffie and Martin Hellman invented in 1976 the basic principles of asym-
metric cryptography, also called public-key cryptography (Diffie & Hellman, 1976). Diffie and
Hellman implemented a key exchange protocol. The first real cryptographic algorithm was
designed by Ronald L. Rivest, Adi Shamir, and Leonard Adleman in 1977 at the Massachusetts
Institute of Technology (MIT), named RSA by the initial letters of the three inventors (Rivest
et al., 1978).
secret area
r
c key
public key generation
random number
secret key c-1
w w
encryption decryption
e(w,c) d( ,c-1)
plaintext ciphertext = d(e(w,c),c-1)
Fig. 3. The basic concept of asymmetric cryptography, including key generation, key manage-
ment, encryption and decryption.
As the name public-key cryptography suggests, Alice und Bob do not share the same key
anymore. Figure 3 shows that Bob initially generates a key pair consisting of a public key and
a private key. Alice, and everyone else, can now encrypt a message, which she would like to
send to Bob, with the published key, whereas only Bob is able to decrypt the ciphertext with
his private key that he has kept secret. Because of this concept, it is not longer necessary to
exchange the encryption key secretly.
Today, we mostly use a mixture of these both concepts. The main message is encrypted by
a symmetric encryption algorithm, whereas the symmetric key is enciphered by public-key
cryptography. By this method, we can combine the advantages of both concepts: the fast
computable encryption and decryption of the symmetric cryptography with the very simple
key management of the asymmetric cryptography.
Since all cryptographic algorithms are assigned to the application layer, it is in the user’s hand
to ensure the secrecy of his data. In this chapter we want to present a possibility to enhance the
security of the transmitted information without the requirement of cryptographic protocols
and the engagement of the user. This type of secrecy is realized on the physical layer.
1.3 Notation
We use the following mathematical notations throughout the chapter:
• [·]+ = max(0, ·).
• A† is the adjoint matrix of the matrix A, i.e., the conjugate transpose matrix of A.
• A 0 means that the matrix A ∈ Cn×n is positive semidefinite, where we use the
following definition for positive semidefiniteness, which automatically implies that A
is Hermitian: z † Az is real and z † Az ≥ 0 for all complex vectors z ∈ Cn .
• | x | is the absolute value of a (complex) variable x. √
x is the Euclidean norm of a (real or complex) vector x with x =
• x† x =
∑in=1 | xi |2 , if we assume a vector of length n.
• Vectors and matrices are denoted by lower and upper case bold symbols, respectively.
• Vectors are column vectors if not stated otherwise.
1.4 The Wiretap Channel
X Y X Y
Alice channel 1 Bob Alice channel 1 Bob
Z Z
channel 2 Eve channel 2 Eve
(a) The degraded wiretap channel. (b) The non-degraded wiretap channel.
Fig. 4. Two models for the wiretap channel.
The first important results in this research area were presented by Wyner and by Csiszár and
Körner. They provided the theoretical basis and introduced two basic system models that are
still used today: the wiretap channel (Wyner, 1975), which was later referred to as degraded
wiretap channel, and the non-degraded wiretap channel (Csiszár & Körner, 1978). Both sys-
tem models are depicted in Figure 4.
From a system theoretic view, the models are characterized by random variables at the channel
inputs and channel outputs. For the channel input of channel 1 (Alice), we use the random
variable X. The channel output of channel 1 (Bob) and channel 2 (Eve) are referred to as Y and
Z, respectively. The corresponding channel input or output alphabets are written as X , Y and
Z.
In Wyner’s degraded wiretap channel, Bob receives a signal that was transmitted over chan-
nel 1, the so-called main channel, whereas Eve observes a signal that was additionally sent
over channel 2, the so-called wiretapper channel. Therefore, Eve’s received signal is always a
degraded or noisier version of Bob’s received signal, i.e., the random variables form a markov
chain X → Y → Z. This fact simplifies the analysis and derivation of ultimate secrecy limits
in Wyner’s model compared to the model of Csiszár and Körner. In the non-degraded wiretap
channel of Csiszár and Körner, the channels to Bob and Eve are supposed to be independent
from each other. In principle, this model does not allow a statement, which channel is the bet-
ter one. However, it is more suitable for the discussion of secrecy in mobile communication
systems.
1.5 Secrecy on the Physical Layer from an Information Theoretic View

From an information theoretic view, the system can be characterized as follows. A message
from the message set W = {1, 2, . . . , M } with M = 2n RS is to be transmitted in n channel uses
while ensuring information theoretic security. The messages are chosen at random and thus
are modeled by a random variable W with alphabet W . Then, the message is encoded by the
encoding function
f enc : W → X n, w → x( n ) ,
which takes the channel state information at the transmitter into account. Since the messages
are random, the input to the channel is random too, and is modeled by the random vector
X (n) . The output of the channel at the legitimate receiver is denoted by Y (n) . It is decoded
by the decoding function
f dec : Yn → W, y (n) → ŵ,
which takes the channel state information at the receiver into account. An ( M, n)-code com-
prises a message set W , an encoding function f enc and a decoding function f dec .
(n)
The average decoding error probability Pe of such a code is defined as
(n) 1 M
Pr( f dec (Y (n) ) = w | X (n) = f enc (w)),
M w∑
Pe =
=1
which is the real decoding error probability, if the messages are uniformly distributed.
The level of secrecy is measured by the uncertainty of Eve about the message W, which was
sent by Alice, under the condition that Eve receives Z (n) . This measure is called equivocation
rate and is given with the conditional entropy function H by
(n) 1
Re = H (W | Z ( n ) ) . (1)
n
We are interested in secure data transmissions with an achievable secrecy rate RS . A secrecy
rate RS is said to be achievable over the wiretap channel if for any > 0, there exists an integer
n() and a sequence of ( M, n)-codes of rate
1
RS = log2 M, (2)
n
such that for all n ≥ n(), the average decoding error probability becomes arbitrarily small,
i.e.,
(n)
Pe ≤ , (3)
and the security constraint
1
H (W | Z ( n ) ) ≥ R S − (4)
n
is fulfilled.
For perfect secrecy, i.e., = 0, the secrecy capacity CS is the supremum of all achievable rates
that guarantee the secrecy of the transmitted data. This means, it can be proven that it is the
tightest upper bound on the amount of information that can be reliably transmitted to the
receiver and perfectly kept secret from the eavesdropper.
By now, we only focus on Gaussian wiretap channels and Gaussian wiretap channels with
an additional attenuation of the transmit signal. For the degraded Gaussian wiretap channel,
which was introduced by (Leung-Yan-Cheong & Hellman, 1978) and whose structure is equal
to that of Wyner’s wiretap channel (cf. Figure 4), the secrecy capacity is given by the maximum
difference of mutual informations:

CS = max I ( X; Y ) − I ( X; Z ) , (5)
f X ∈F
where F is the set of all probability density functions (pdfs) at the channel input under power
constraint at the transmitter. Since Eve always receives a degraded version of Bob’s signal,
the secrecy capacity in (5) is always non-negative. For the non-degraded Gaussian wiretap
channel, which is structured like the model of Csiszár and Körner (cf. Figure 4), the secrecy
capacity is given by a slightly modified term:
+
CS = max I ( X; Y ) − I ( X; Z ) , (6)
f X ∈F
i.e., the secrecy capacity CS is set to zero, if Eve has a better channel realization than Bob. In
the following, we will use the non-degraded system model, if it is not stated otherwise. The
mutual information terms I ( X; Y ) and I ( X; Z ) are concave in f X . This allows us to formulate
a lower bound RS for the secrecy capacity CS :
+ +
CS = max I ( X; Y ) − I ( X; Z ) ≥ max ( I ( X; Y )) − max ( I ( X; Z )) = RS . (7)
f X ∈F f X ∈F f X ∈F

channel capacity channel capacity
from Alice to Bob from Alice to Eve
Note that the secrecy rate RS is defined with the difference of the channel capacities from Alice
to Bob and from Alice to Eve. This lower bound RS is often used for a simplified calculation of
achievable secrecy rates since it is known how to maximize the mutual information terms. For
some scenarios, it has already been proven that the secrecy rate RS equals the secrecy capacity
CS , e.g., for the single-user system with multiple antennas (see Section 2.3) or for the MISO
and MIMO broadcast channel (see Section 3.1).
1.6 The Basic System Model and Preliminaries

Now, we consider Gaussian channels with an additional attenuation of the transmit signal.
As a basis for all system models, which are used throughout this chapter, we introduce the
following system model for each channel use:
y = h·x+φ and
z = g·x+ψ (8)
with Alice’ transmit signal x, channel coefficients h and g to model the signal attenuation for
the channels from Alice to Bob and from Alice to Eve, additive white Gaussian noise φ and ψ
and signals y and z at the receivers of Bob and Eve, respectively. Figure 5 illustrates how the
basic system model is independently used n times to transmit the codeword of length n that
is chosen for the message.
Alice Bob
h
W encoder decoder W
X (n) Y(n)
Eve
g H(W|Z (n))
decoder ?
Z(n)
Fig. 5. The basic system model.
For the system model, we make the following assumptions:

• The variables φ, ψ and x are stochastically independent.
• The noise variables φ and ψ are circular symmetric complex Gaussian distributed with
zero mean and variance σ2 . We write φ, ψ ∼ CN (0, σ2 ).
• At the transmitter, we have a power constraint P, i.e., for a codeword x of length n
1 n
| xi |2 ≤ P.
n i∑
(9)
=1
In order to achieve the channel capacities in (7), the variable x has to be circular
symmetric complex Gaussian distributed with zero mean and variance P. We write
x ∼ CN (0, P).
• We define the so called channel gains α and β by
α = | h |2 and β = | g |2 . (10)
βP
In this model, Bob’s and Eve’s signal-to-noise ratio (SNR) is given by αP
σ2
and σ2 , respectively.
Therewith, the secrecy rate RS can be quantified in bit per complex symbol (bpcs) and ex-
pressed as a function of the transmit power constraint P:

αP βP +
RS ( P) = log2 1 + 2 − log2 1 + 2 [bpcs], (11)
σ σ
where a Gaussian codebook maximizes both mutual information terms in (7). This is the
secrecy rate of the non-degraded Gaussian wiretap channel with channel gains α and β.
Based on the model above, we define a system with slow quasi-static block flat fading. In
order to model flat fading, the channel coefficients h and g in (8) become random variables,
which we call channel states. We assume slow quasi-static block fading, i.e., the channel states
are random but remain constant for a sufficiently long time to transmit a whole codeword.
The next channel state is independent of all other channel states before and is identically
distributed.
For every channel state, the secrecy rate can be calculated according to (11). Therefore, the
secrecy rate is called instantaneous secrecy rate. Depending on the statistics assumed for the
channel coefficients, we can calculate average or outage secrecy rates as defined in (Bloch
et al., 2008). In the following sections, we present instantaneous secrecy rates for given chan-
nel coefficients and average secrecy rates, where we make the following assumptions for the
distribution of the channel states: The channel coefficients h and g are stochastically indepen-
dent of each other, the transmit signal, and the noise variables. They are circular symmetric
complex Gaussian distributed with zero mean and variance 1. We write h, g ∼ CN (0, 1). This
means, we add Rayleigh fading to the Gaussian wiretap channel.
The interesting observation in (Bloch et al., 2008) for wiretapped fading channels is that even if
the average channel quality between transmitter and eavesdropper is better than the average
channel between transmitter and intended receiver, the average secrecy capacity can still be
positive.
1.7 Extension to a Multi-Carrier or a Multi-Antenna System Model

The system model in Section 1.6 can be extended to a multi-carrier or a multi-antenna system
model:
• For an ideal multi-carrier system with L carriers, the system model in (8) is used L
times in parallel. For each carrier , we have the same assumptions and the same re-
lations between the variables as listed in Section 1.6. Besides, the transmit signals and
the noise variables are assumed to be independent between the L carriers. For corre-
sponding variables, we assume an identical distribution. If we assume random channel
coefficients, the L parallel channel coefficients for the channels from Alice to Bob are
correlated in general. The same applies to the L parallel channel coefficients for the
channels from Alice to Eve. The power constraint P at the transmitter becomes a sum
L
power constraint over all L carriers, i.e., ∑= 1 P = P.
• In multi-antenna (muliple-input multiple-output, MIMO) systems, we assume that
Alice has mA transmit antennas, Bob has mB receive antennas, and Eve has mE receive
antennas. Then, the system model in Section 1.6 is expanded by using vectors and
matrices instead of scalars. The assumptions and relations between the variables men-
tioned in this context in Section 1.6 are still valid or can be formulated analogously. Each
noise vector consists of independent and identically distributed components. However,
the channels from Alice to Bob can be spatially correlated. The same applies to the chan-
nels from Alice to Eve. The power constraint P at the transmitter becomes a sum power
constraint over all antennas, which is written as trace(Q) = P with the covariance ma-
trix Q of the transmit signal vector x.
In the multi-carrier or multi-antenna scenario, Alice has more degrees of freedom than before.
Now, she can variate the power allocation (under the sum power constraint) over L carriers
or mA antennas to achieve a high secrecy rate for the data transmission to Bob. Note that
both models can be combined to have a MIMO multi-carrier system. In the following parts of
Section 2, we will derive an optimal power allocation that maximizes the achievable secrecy
rate to Bob for multi-carrier or multi-antenna scenarios.
The channel capacities, which we use in the secrecy rate formula in (11) or in secrecy rate ex-
pressions derived from it, are concave functions in P or Q. But the difference of two concave
functions generally is neither convex nor concave. Therefore, finding the optimal power allo-
cation over L carriers or mA antennas under a sum power constraint is a difficult, non-convex
optimization problem.
1.8 Extension to a Multi-User Scenario

So far, we have considered a single-user scenario, where Alice wants to transmit a private
message to Bob, and Eve is a passive eavesdropper who wants to decode this message. Now,
we want to introduce a multi-user scenario with one transmitter (Alice) and K receivers. Alice
wants to transmit private messages to each of the K users and to keep these messages secret
from all other users. In such a system, we have K secrecy rates or a K-dimensional secrecy rate
region. In this chapter, we will confine ourselves to the 2-user scenario with the receivers Bob
and Eve, who are now both: legitimate receiver of one message and potential eavesdropper
of the other.
In some multi-user scenarios that we present in Section 3, the signals for the different users
can interfere. For the evaluation of the achievable secrecy rates for the 2-user case, we slightly
modify the definition of the secrecy rate in (7). For the individual secrecy rates, we use the
signal-to-interference-and-noise ratio (SINR) for the legitimate user, where the complete in-
terference from the other user’s signal is simply treated as additional noise, and the signal-to-
noise ratio (SNR) for the eavesdropper. Under power constraint PB + PE = P, Alice allocates
power PB and PE for the data transmission to Bob and Eve, respectively. This results in the
following expression for the achievable secrecy rate RS B for the transmission to Bob:
+
αPB βP
RS B ( PB , PE ) = log2 1 + − log2 1 + 2B [bpcs]. (12)
σ2 + αPE σ
This is a worst-case assumption since we assume that Eve performs successive interference
cancellation (SIC), i.e., first, she is able to detect her own data, afterwards she subtracts it from
her received signal and tries to decode the message for Bob.
The achievable secrecy rate for the transmission to Eve can be formulated in the same way:
+
βPE αP
RS E ( PB , PE ) = log2 1 + − log2 1 + 2E [bpcs]. (13)
σ2 + βPB σ
2. Secrecy Capacity in Single-User Systems

2.1 Single-Antenna Systems
For a single-user single-antenna system, we have already presented the secrecy rate in Section
1.6. For this single-input single-output (SISO) system, the secrecy rate given in (11) is exactly
the secrecy capacity given in (6). In this scenario, Alice only has the choice to transmit the
message to Bob or not, according to the channel coefficients for the channels to Bob and Eve.
In a completely static system, this would result in a constant secrecy rate that is either positive
or zero all the time. But in a time-varying system where we assume slow quasi-static block flat
fading, the situation changes from block to block: we have instantaneous channel realizations
and thus instantaneous secrecy rates, which can be averaged in time.
2.2 Multi-Carrier Systems

In this section, we extend the basic model from Section 1.6 to the multi-carrier wiretap channel,
where Alice wants to send a private message to Bob in a system with L parallel carriers. This
message should be kept secret from the eavesdropper Eve. This is a single-antenna scenario
since every member of the system has only one transmit or receive antenna. We study the
resource allocation under the secrecy constraint and a sum power constraint over all carriers.
Alice Bob
h1
W1 encoder decoder W1
X1(n) Y1(n)
Alice Eve Bob
g1 hL 1|Z1(n))
H(W
WL encoder decoder ? decoder WL
carrier 1 XL(n) Z1(n) YL(n)
Eve
gL H(WL|ZL (n))
decoder ?
carrier L ZL (n)
Fig. 6. The multi-carrier wiretap channel with L carriers.
The system model is modified as described in Section 1.7 and illustrated in Figure 6. On carrier
with 1 ≤ ≤ L, Bob and Eve observe the received signals y and z , respectively:
y = h x + φ and
z = g x + ψ (14)
with Alice’ transmit signal x , channel coefficients h , g , and noise variables φ and ψ . The
assumptions listed for the basic system model in Section 1.6 also apply to this model. The
channel gains α , β are defined according to equation (10).
In this multi-carrier system, the secrecy rate is the sum over all secrecy rates per carrier, which
can be computed according to (7), and is given by
L +
α P β P
RS (PB ) = ∑ log2 1 + 2B − log2 1 + 2B [bpcs], (15)
=1 σ σ
where PB is the power that Alice allocates to carrier in order to transmit the message to Bob.
The power allocation over all carriers can be written in a vector PB = ( PB 1 , . . . , PB L ).
We derive the single-user optimal power allocation for maximizing the secrecy rate in this
multi-carrier system under sum power constraint P over all carriers:
L
max RS (PB )
PB
subject to ∑ PB ≤ P and PB ≥ 0. (16)
=1
This is a non-convex optimization problem with objective function RS .

The optimal power allocation that solves (16) is to allocate zero power to all carriers with
α ≤ β :
∀ ∈ {1, 2, . . . , L} : α ≤ β =⇒ PB = 0. (17)
The proof is based on the necessary Karush-Kuhn-Tucker (KKT) optimality conditions (Jors-
wieck & Wolf, 2008). Furthermore, it was shown that the remaining optimization problem
L
max RS (PB )
PB
subject to ∑ PB ≤ P, PB ≥ 0 and PB = 0 for α ≤ β (18)
=1
is convex (see (Boyd & Vandenberghe, 2004) for general convex optimization theory).
The optimal power allocation is a type of waterfilling (see (Cover & Thomas, 2006) for stan-
dard waterfilling). We give the solution in implicit form with

0
 if α ≤ β
+
PB = 2
σ (α + β ) 4
σ (α − β ) 2 2
σ (α − β ) (19)

 − 2α β + 4( α β )2
+ µ1 ln(2)α β otherwise

and µ > 0 such that

L
∑ PB = P. (20)
=1
However, the typical order of the channels is different from standard waterfilling. For small
SNR, the carriers are ordered according to (α − β ), i.e., the carrier with largest (α − β ) is
supported first, whereas for high SNR, the carriers are ordered according to βα .

100 500
Average Channel Capacity Average Channel Capacity
90 Average Achievable Secrecy Rate 450 Average Achievable Secrecy Rate
Rates from Alice to Bob in bpcs
80 400
70 350
60 300
50 250
40 200
30 150
20 100
10 50
0 0
−10 −5 0 5 10 15 20 25 30 −10 −5 0 5 10 15 20 25 30
Power Constraint at the Transmitter in dB Power Constraint at the Transmitter in dB
(a) 20 carriers (b) 200 carriers
Fig. 7. Average channel capacities and average achievable secrecy rates for transmission from
Alice to Bob in a multi-carrier system with 20 and 200 independent carriers.
In Figure 7, we compare the average achievable secrecy rate with the average channel capacity
of the single-user multi-carrier channel for different numbers of carriers. The main observa-
tion is that the high SNR channel capacity grows without bound whereas the secrecy rate
is bounded because the mutual information between the transmitter and the eavesdropper
is subtracted from the rate. If the number of carriers is increased, the asymptotic behavior
will remain the same. However, the high SNR bound is shifted to the right. We see that in
multi-carrier systems with a large number of carriers (and corresponding multipath fading),
the costs of security are decreased, i.e., the high SNR bound is increased.
2.3 Multi-Antenna Systems

In this section, we extend the basic model from Section 1.6 to the multi-antenna wiretap chan-
nel. As in the scenarios above, Alice wants to send a private message to Bob, which should
be kept secret from the eavesdropper Eve. Now, we consider a multi-antenna system, where
Alice has mA transmit antennas, Bob and Eve have mB and mE receive antennas, respectively.
We study the resource allocation under the secrecy constraint and a sum power constraint
over all transmit antennas.
Alice Bob
H
...
W encoder ... decoder W
X (n) Y(n)
Eve
G H(W|Z (n))
...
decoder ?
Z(n)
Fig. 8. The multi-antenna wiretap channel.
h11  
Alice Bob h11 h12 ... h1mA
h mB 1
1 1  h21
 h22 ... h2mA  
2 h1mA 2 H= . .. .. .. 
 .. . . . 
mA h mB mA mB h mB 1 h mB 2 ... h mB mA
Fig. 9. The structure of a channel matrix using the example of channel matrix H for the
channels from Alice to Bob.
The system model, which is depicted in Figure 8, is modified as described in Section 1.7. It
can be described by
y = H ·x+φ and
z = G · x + ψ. (21)
The complex channel coefficients are written as the components of H and G, which are chan-
nel matrices of dimension [mB × mA ] and [mE × mA ], respectively. Figure 9 illustrates the
structure of such a channel matrix. Alice’ transmit signals are written in a column vector x of
dimension [mA × 1]. The noise variables φ and ψ are column vectors of dimension [mB × 1]
and [mE × 1], respectively, with independent components. Bob’s and Eve’s received signals y
and z are column vectors of dimension [mB × 1] and [mE × 1], respectively. The assumptions
listed for the basic system model in Section 1.6 analogously apply to this model. The transmit
vector x, and the noise vectors φ and ψ are stochastically independent, i.e., the components
of one vector are stochastically independent of the components of the other vectors. The noise
vectors are composed of independent and circular symmetric complex Gaussian distributed
components. Their covariance matrices are normalized to the identity matrix. For fading sce-
narios, the channel matrices H and G are assumed to be stochastically independent of each
other, the transmit vector x, and the noise vectors φ and ψ.
In this multi-antenna system, the secrecy rate, which is the secrecy capacity for this scenario
(Oggier & Hassibi, 2008), is given by
+
RS (Q) = log2 det(ImB + HQH † ) − log2 det(ImE + GQG† ) [bpcs]. (22)
ImB and ImE are identity matrices of dimension [mB × mB ] and [mE × mE], respectively.
Q is
the covariance matrix of the input signal vector x, i.e., Q = Cov (x) = E xx† .
We derive the single-user optimal power allocation for maximizing the secrecy rate in this
multi-antenna system under sum power constraint P over all transmit antennas:
max RS (Q) subject to trace(Q) ≤ P and Q 0. (23)

Q
This is a non-convex optimization problem, which we analyze for some special cases.
Multiple-Input Single-Output (MISO) Systems

In the MISO case, where Bob and Eve have only one receive antenna each, the channel matrices
H and G in (21) reduce to row vectors h and g of dimension [1 × mA ]:
h = ( h1 , . . . , h mA ) and g = ( g1 , . . . , gmA ). (24)
The secrecy rate in (22) can be written as

+
RS (Q) = log2 (1 + hQh† ) − log2 (1 + gQg † ) [bpcs]. (25)
This scenario was analytically solved in (Li et al., 2007). The authors used an invertible coor-
dinate transformation with a unitary transformation matrix
 
h † (g − h
g

ζh)† gh†
T = , , further (mA − 2) columns with ζ = , (26)
h g 1 − ζ † ζ g h
where the last (mA − 2) columns are an orthonormal basis for the (mA − 2) dimensions and
orthogonal to the first two columns. Therewith, the transformed channel vectors h T and g T
have zeros in the subspace spanned by the last (mA − 2) columns of T . Focussing only on the
subspace spanned by the first two columns of T the transformed channel vectors h T and g T
can be represented by h̄ and ḡ with

h̄ = h (1, 0) and ḡ = g (ζ, 1 − ζ † ζ ). (27)
In the transformed space, the covariance matrix with the optimal power allocation for the
optimization problem derived from (23) for the MISO scenario is
Q̄ = Pq̄ q̄ † , (28)
where q̄ is the generalized eigenvector corresponding to the largest generalized eigenvalue

of the two matrices (I2 + Ph̄† h̄) and (I2 + Pḡ † ḡ ). The covariance matrix Q̄ has unit-rank,
which means that only one data stream is supported at the transmitter and beamforming can
be applied with vector q̄. Finally, the optimal covariance matrix Q in the orginal space is
obtained by adding zeros for the subspace spanned by the last (mA − 2) columns of T and the
inverse coordinate transformation.
In Figure 10, the difference between the average achievable secrecy rate and the average chan-
nel capacity of the single-user MISO scenario is illustrated for different numbers of transmit
antennas.
12 12
Average Channel Capacity Average Channel Capacity
Average Achievable Secrecy Rate Average Achievable Secrecy Rate
10 10

8 8
6 6
4 4
2 2
0 0
−10 −5 0 5 10 15 20 25 30 −10 −5 0 5 10 15 20 25 30
Power Constraint at the Transmitter in dB Power Constraint at the Transmitter in dB
(a) Two transmit antennas (b) Four transmit antennas
Fig. 10. Average channel capacities and average achievable secrecy rates in a MISO system
with two and four transmit antennas and uncorrelated channels.
Single-Input Multiple-Output (SIMO) Systems

In the SIMO case, where Alice has only one transmit antenna and Bob and Eve have an ar-
bitrary number of receive antennas, the channel matrices H and G in (21) reduce to column
vectors of dimension [mB × 1] and [mE × 1], respectively:
h = ( h1 , . . . , h mB )T and g = ( g1 , . . . , gmE )T . (29)
Similar to the single-antenna (SISO) case in Section 2.1, Alice has only the choice either to
transmit the message to Bob with power P or not. This SIMO scenario can be transformed in
an equivalent SISO scenario with modified channel statistics. Bob and Eve can apply matched
h2 P
filters at the receivers. In the equivalent SISO scenario, Bob’s SNR is σ2
and Eve’s SNR is
g 2 P
σ2
,
where P is the transmit power constraint and σ2 is the noise variance for each receive
antenna.
Some Special Multiple-Input Multiple-Output (MIMO) Systems

In the MIMO 2-2-1 scenario, where Alice has two transmit antennas, Bob has two receive
antennas, whereas Eve has only one single receive antenna, the channel matrices H and G
in (21) reduce to a matrix H of dimension [2 × 2] and a row vector g of dimension [1 × 2].
The optimization problem derived from (23) for the MIMO 2-2-1 scenario was analytically
solved in (Shafiee et al., 2008). The authors transformed the problem into a Rayleigh quotient
problem, whose solution is the optimal covariance matrix Q:
Q = Pqq † , (30)
where q is the eigenvector that corresponds to the largest eigenvalue of the matrix (I2 +
Pg † g )−1/2 (I2 + PH † H )(I2 + Pg † g )−1/2 .
For the general MIMO scenario, where each user can have an arbitrary number of antennas, it
has been proven in (Oggier & Hassibi, 2008) that the secrecy rate in (22) is equal to the secrecy
capacity of the system in (21).
In (Liu, Hou & Sherali, 2009), the authors presented a global optimization algorithm called
branch-and-bound with reformulation and linearization technique (BB/RLT). This method
guarantees finding a global optimal solution for the non-convex optimization problem in (23).
Another characterization of the optimal transmit covariance matrix Q is derived in (Liu, Liu,
Poor & Shamai (Shitz), 2009). This approach is discussed in the multi-user context in Section
3.1.
3. Secrecy Rate Region in Multi-User Systems

In this section, we extend some of the previously presented results to the multi-user case. Due
to the fact that there is more than one user, we will not use anymore the terms secrecy rate and
secrecy capacity, but secrecy rate region and secrecy capacity region.
In the literature, the case of one confidential (private) and one public message is often dis-
cussed. We focus on the case, where only confidential messages are sent. For convenience, we
confine ourselves to systems with only two users. The extension to more than two users can
be done straightforward.
3.1 Broadcast Channels

The broadcast channel (BC) is the logical extension of the basic system presented in Section
1.6 to the multi-user scenario. In this channel model, Alice additionally sends a message to
Eve that should be concealed from Bob. This new system setting is shown in Figure 11.
Alice Bob
h H(WE|Y(n))
WB WB
encoder decoder
WE ?
X (n) Y(n)
Eve
g H(WB|Z(n))
?
decoder
WE
Z(n)
Fig. 11. The basic model of the broadcast channel with two confidential messages.
The extensions of the basic model discussed in Section 2 can also be applied to the broadcast
channel. Based on the results of the single-user multi-carrier scenario in Section 2.2, we con-
sider now a cellular broadcast channel with two users, namely Bob and Eve and reuse the
system model shown in Figure 6.
The system model is equivalent to the model given in (14)
y = h x + φ and
z = g x + ψ , (31)
but now Bob and Eve eavesdrop each other. The assumptions listed in Section 1.6 also apply
to this model. The channel gains α and β are defined according to (10).
On carrier , Alice allocates power PB for data transmission to Bob and PE for data trans-
mission to Eve. The sum power constraint translates to
L
∑ ( PB + PE ) ≤ P. (32)
=1
We collect the power allocation for Bob and Eve in appropriate vectors, i.e., PB =
( PB 1 , . . . , PB L ) and PE = ( PE 1 , . . . , PE L ).
The achievable secrecy rates per carrier are modified according to the explanations in Section
1.7 and 1.8. The achievable secrecy rates for data transmission to Bob and Eve are the sum
over all secrecy rates per carrier and given by
L +
α PB
β P
R S B ( P B , PE ) = ∑ log2 1 + − log2 1 + 2B
2
[bpcs] and
=1 σ + α PE σ
L +
β PE α PE
RS E (PB , PE ) = ∑ log2 1 + 2 − log2 1 + [bpcs]. (33)
=1 σ + β PB σ2
The system operator might be interested in the sum of the individual secrecy rates in (33). The
sum secrecy rate is defined as
(sum)
RS ( P B , P E ) = R S B ( P B , P E ) + R S E ( P B , PE ) . (34)
The corresponding programming problem maximizes the sum secrecy rate in (34):
L
(sum)
max RS
PB ,PE
(PB , PE ) subject to ∑ ( PB + PE ) ≤ P, PB ≥ 0 and PE ≥ 0. (35)
=1
In (Jorswieck & Wolf, 2008), it was shown that it is optimal to support only the best user per
carrier. From that fact and the power constraint PA = PB + PE per carrier follows the user
allocation per carrier, which is

PA if α > β 0 if α ≥ β
PB = and PE = . (36)
0 otherwise PA otherwise
Then, the power allocation per carrier is derived from equation (19) by replacing (α − β ) by
(max(α , β ) − min(α , β )).
Note that the case α = β can be ignored in the fading scenario, if we assume a continuous
distribution for the channel coefficients and hence the channel gains, since Pr (α = β ) = 0.
Moreover, for the spectral power allocation in (19), it is all the same, which user is assumed to
be supported. The algorithm allocates zero power to this carrier and therefore the secrecy rate
on this carrier will be zero.
The previously described sum secrecy rate maximization for the broadcast channel can be
easily extended to the weighted sum secrecy rate maximization as discussed in (Jorswieck &
Gerbracht, 2009). The weighted sum secrecy rate is given by
(wsum)
RS (PB , PE , λ) = λRS B (PB , PE ) + (1 − λ) RS E (PB , PE ) (37)
with 0 ≤ λ ≤ 1. Herewith, the system operator is able to fulfill certain Quality of Service
(QoS) constraints.
The programming problem that maximizes the weighted sum secrecy rate is given by
L
(wsum)
max RS
PB ,PE ,λ
(PB , PE , λ) subject to ∑ ( PB + PE ) ≤ P, PB ≥ 0 and PE ≥ 0.
=1
The user allocation is equivalent to the case without weighting factor in (36). It is optimal to
support only the best user per carrier.
Furthermore, the spectral power allocation, which is similar to the one in the single-user multi-
carrier scenario in Section 2.2, is a kind of waterfilling. The optimal power allocation is given
by
 +

 σ2 ( α + β ) σ 4 ( α − β )2 λ σ2 ( α − β )
 − 2α β + 4α2 β2
+ µ ln(2)α β if ∈ L1
PA = + ,

 2 4 2 2
 − σ (α + β ) + σ ( β 2−α2 ) + 1−λ σ ( β −α ) if ∈ L 2
2α β 4α β µ ln(2)α β
where L1 = { ∈ {1, . . . , L} : α > β }, L2 = {1, . . . , L} \ L1 and µ > 0 such that

L
∑ PA = P. (38)
=1
2.5
Average Channel Capacity Region 10
Average Achievable Secrecy Rate Region Average Channel Capacity Region
9 Average Achievable Secrecy Rate Region
2
8
Rates from Alice to Eve in bpcs
Rates from Alice to Eve in bpcs
1.5
6
1 4
0.5 2
0 0
0 0.5 1 1.5 2 2.5 0 1 2 3 4 5 6 7 8 9 10
Rates from Alice to Bob in bpcs Rates from Alice to Bob in bpcs
(a) SNR = 0 dB (b) SNR = 10 dB
Fig. 12. The average channel capacity region and the average achievable secrecy rate region
for the multi-carrier broadcast channel with eight carriers and two users.
Figure 12 shows the achievable average secrecy rate region for the broadcast channel with
eight carriers compared to the average channel capacity region, which was found by exhaus-
tive search. We observe that the gap between the achievable secrecy rate region and the ca-
pacity region grows with increasing SNR. For SNR → ∞, we know that the secrecy rate region
does not grow without bound. It is limited by the second term in the equations in (33).
Now, we present the secrecy capacity region for the real-valued MIMO broadcast channel,
which can be found in (Liu, Liu, Poor & Shamai (Shitz), 2009). The system model is given by
Y = Hx + Φ and
Z = Gx + Ψ, (39)
where H and G are real channel matrices of size [mB × mA ] and [mE × mA ], respectively.
The noise is modeled by vectors of dimension [mB × 1] and [mE × 1]. For the distribution of
the noise vectors, we assume Φ, Ψ ∼ N (0, Imk ) with k ∈ {B, E}. The channel input x is a
vector
of the size [mA × 1]. Furthermore, we have an average power constraint, defined by
E x2 ≤ P.
The achievable secrecy rates are given by
R S B ( QB )
+
1 ImB + HQB H T
= log2 det [bps] and
2 ImE + GQB GT
R S E ( Q B , QE )
+
1 I m E + G ( QB + QE ) G T 1 I m B + H ( QB + QE ) H T
= log2 det − log2 det [bps],
2 ImE + GQB GT 2 ImB + HQB H T
(40)
where QB and QE are the covariance matrices for the transmission to Bob and to Eve, respec-
tively. They are positive semidefinite matrices with trace(QB + QE ) ≤ P.
For this system model, it has been shown in (Liu, Liu, Poor & Shamai (Shitz), 2009) that the
secrecy capacity region is given by

R= ( RS B (QB ), RS E (QB , QE )) . (41)
0≤trace(QB +QE )≤ P
Even though the secrecy capacity region has been proven for the MIMO and the MISO broad-
cast channel (Liang et al., 2009), it is still an open problem to find the secrecy capacity region
for the single-antenna case. So far, there are no results known about the optimal transmit
strategies in MIMO and MISO broadcast channels.
3.2 Multiple Access Channels
Alice
XA(n)
W1 encoder h
Bob
? decoder
H(W2|ZA(n)) Z A(n) W1
g g decoder
W2
Charly Y(n)
XC(n) h
W2 encoder
? decoder
H(W1|ZC(n)) ZC(n)
Fig. 13. The multiple access channel (one example).
The multiple access channel (MAC) is difficult to analyze in a system setting concerning se-
crecy on the physical layer. The conventional channel model for the MAC consists of two
or more transmitters, e.g., mobile devices, and only one receiver, e.g., the base station. In
this model, there is nobody who could eavesdrop the sent messages in accordance with the
attacker model in Section 1.1. However, if the uplink transmission (MAC) and the down-
link transmission (BC) are studied together, every user in the system can eavesdrop all other
users. But from the transmitter’s point of view, the channel model would always be a broad-
cast channel, where all other mobile devices and the base station serve as receivers.
There are currently a lot of research activities concerning the MAC in the secrecy context. One
of the models assumed for the MAC in this case is depicted in Figure 13 and studied in (Liang
et al., 2009). Another channel model is described in (Tekin & Yener, 2006). It deals with the
degraded MAC, where the eavedropper obtains a degraded version of the receiver’s signal.
3.3 Interference Channels

In this section, we will present two results for the interference channel (IFC). The first one will
be a weak interference, single-antenna channel, whereas the second one is a multi-antenna
interference channel. For both channel models, we need an additional sender, called Charly,
as it can be seen in Figure 14.
Alice wants to send a private message to Bob, which should be kept secret from Eve. Further-
more, Charly wants to send a confidential message to Eve, which should be concealed from
Bob. These communication channels have the channel coefficients h and g. The interference
or eavesdropper channels from Alice to Eve and from Charlie to Bob have the coefficients g̃
and h̃, respectively.
Alice Bob
h H(WE|Y(n))
WB
WB encoder g decoder
?
XA(n) Y(n)
Charly Eve
h H(WB|Z(n))
?
WE encoder g decoder
(n) WE
XC Z(n)
Fig. 14. The interference channel.
The basic system model from Section 1.6 has to be modified to be suitable for the interfer-
ence channel. Nevertheless, the assumptions listed for the basic system model also apply to
this model. For the interference channel, the system model, which was studied in (Zhang &
Gursoy, 2009), is given by
y = hxA + h̃xC + φ and

z = gxC + g̃xA + ψ, (42)
where h, g, h̃ and g̃ are deterministic channel coefficients. xA and xC are the channel inputs at
the transmitters. φ and ψ are independent and circular symmetric complex Gaussian random
variables with CN (0, σ2 ). The channel causes weak interference, i.e., α̃ β̃
α < 1 and β < 1,
where the channel gains α̃ and β̃ of the eavesdropper channels are defined according to (10)
by α̃ = | h̃|2 and β̃ = | g̃|2 . The individual power constraint at the transmitters are given by

E | xA |2 ≤ PA and E | xC |2 ≤ PC . (43)
The system model has to be modified according to Section 1.8. The consideration of interfer-
ence results in the achievable secrecy rates given by
+
αPB β̃P
RS B ( PB , PE ) = log2 1 + − log2 1 + 2B [bpcs] and
σ2 + α̃PE σ
+
βPE α̃PE
RS E ( PB , PE ) = log2 1 + 2 − log2 1 + 2 [bpcs], (44)
σ + β̃PB σ
where PB is the power allocated by Alice for data transmission to Bob and PE is the power
allocated by Charly for data transmission to Eve.
The achievable secrecy rate region is given by

R= ( RS B ( PB , PE ), RS E ( PB , PE )). (45)
0≤ PB ≤ PA ,
0≤ PE ≤ PC
Now, we present some results for the multi-antenna interference channel, which are discussed
in (Jorswieck & Mochaourab, 2009) in a game-theoretic context. Both transmitters use mA and
mC antennas, whereas Eve and Bob receive the messages with only one antenna each. The
system model is modified according to Sections 1.7 and 1.8 and is described by
y = h · xA + h̃ · xC + φ and
z = g · xC + g̃ · xA + ψ, (46)
where h and g̃ are row vectors of dimension [1 × mA ] and h̃ and g are row vectors of di-
mension [1 × mC ] with complex channel coefficients. xA and xC are vectors of dimension
[mA × 1] and [mC × 1] and are independent, circular symmetric, and complex Gaussian dis-
† ) and x ∼ CN (0, v v † ). The beamforming vectors v and
tributed, i.e., xA ∼ CN (0, vA vA C C C A
vC are of dimensions [mA × 1] and [mC × 1] with vA 2 = vC 2 = 1. φ and ψ are indepen-
dent white Gaussian noise with variance σ2 , i.e., φ, ψ ∼ CN (0, σ2 ). Both transmitters have a
power constraint P.
The achievable secrecy rate pair for the Gaussian MISO IFC is given by
+
| h · vA | 2 P |g̃ · vA |2 P
RS B (vA , vC ) = log2 1 + − log2 1 + [bpcs] and
σ + |h̃ · vC |2 P
2 σ2
+
| g · vC | 2 P |h̃ · vC |2 P
RS E (vA , vC ) = log2 1 + 2 − log2 1 + [bpcs]. (47)
σ + |g̃ · vA |2 P σ2
The efficient beamforming vectors are described in the following. According to (Jorswieck &
Mochaourab, 2009), we denote the maximum ratio transmission beamforming vector of user
( MRT ) ( ZF )
k as vk and the zero-forcing beamforming vector as vk , where k ∈ {A, C}. We obtain
(MRT) (ZF)
λ A · vA + ( 1 − λ A ) · vA
vA ( λ A ) =

and
(MRT) (ZF)
λA · vA + ( 1 − λ A ) · vA
(MRT) (ZF)
λ C · vC + ( 1 − λ C ) · vC
vC ( λ C ) =

(48)
(MRT)
λC · vC + (1 − λC ) · vC(ZF)
with transmit strategies 0 ≤ λA , λC ≤ 1.

The maximization of the secrecy rate from Alice to Bob depends on the given interference
caused by Charly. The rate can be described as the best response, if λC is given:
∗
λA (λC ) = arg max RS B (λA , λC ). (49)
0≤ λA ≤1
Equivalently, Charly’s best response to Alice’ transmit strategy is given by

∗
λC (λA ) = arg max RS E (λA , λC ). (50)
0≤ λC ≤1
From Alice’ and Charly’s point of view, the interference channel equals the broadcast channel.
Because of this fact, they do not have the possibility to influence, but to react to the interference
generated by each other. The optimal solution for the maximization problems in (49) and (50)
can be found by an iterative algorithm described in (Jorswieck & Mochaourab, 2009). This
optimum is not the best solution, which is possible in this scenario. It is an achievable and
stable point, the so-called Nash Equilibrium, that will be reached, if Alice and Charly do not
cooperate. Another approach to solve these non-convex optimization problems is to use a
monotonic optimization framework. This has been proven useful for the MISO interference
channel in (Jorswieck & Larsson, 2009) in order to optimize the transmit strategies.
4. Discussion and Open Problems

The information theoretic description of secrecy capacities and secrecy capacity regions (Liang
et al., 2009) is an important research area to support a better understanding of security on
the physical layer. Based on the secrecy capacity expressions or achievable secrecy rates, the
transmit strategies, including power allocation, beamforming and subcarrier allocation, are
optimized in order to choose a certain operating point. In this chapter, we focus on the opti-
mization of the physical layer transmit strategies for typical wireless communication scenar-
ios.
In single-user scenarios, the system design is more complicated with additional secrecy con-
straints, since the secrecy capacity expressions are in general not concave or convex in the
transmit strategies. The secrecy rate terms usually consist of a difference of two parts. The
first one corresponds to the amount of data that can be reliably transmitted to the intended
user and it is therefore concave in the transmit strategies. The second one corresponds to the
amount of data that is overheard by the eavesdropper and it is thus also concave in the trans-
mit strategies. The resulting transmit optimization problems are non-convex optimization
problems since the difference of two concave functions is not necessarily convex or concave.
However, in the multi-carrier case, the problem can be reduced to a convex optimization prob-
lem that can be efficiently computed. We conjecture that also the multiple antenna (MIMO)
scenario will be completely solved in the very next future.
In multi-user scenarios, the secrecy rate regions of all four elements of network information
theory, the broadcast, the multiple access, the relay, and the interference channel were recently
studied. The attacker models of the MAC and the relay case are more difficult than the well-
motivated ones of the broadcast and the interference channel. Therefore, we focus on the
broadcast and interference channel. The resource allocation for the parallel broadcast channel
without secrecy is involved due to a hard combinatorial problem – the matching of carriers
to users. Interestingly, with secrecy constraints, the resulting programming problem is much
simpler and the optimal power and resource allocation can be solved efficiently. A similar
observation in the context of interference channels with beamforming and without coopera-
tion shows that the secrecy constraint leads to a more altruistic and less selfish behavior. In
both cases, the additional term in the utility functions simplifies and improves the resulting
transmit optimization.
In addition to the resource allocation and transmit optimization problems discussed in this
chapter, there are many important practical issues to be solved. The assumption to have
perfect CSI at the transmitter(s) and receiver(s) is idealistic. The impact of channel estima-
tion errors and limited feedback on the achievable secrecy rates needs to be analyzed. The
assumption to apply Gaussian codebooks is idealistic, too. Finite modulation and coding
schemes lead to more difficult bit and power allocation problems at the transmitter. Recent
results in the development of channel codes for secure communications are not discussed in
this chapter due to length constraints. However, there is interesting current work on the anal-
ysis and development of channel codes that are able to achieve the secrecy capacity. Finally,
the attacker model studied in this chapter is important but not the only one possible. Future
work will also consider malicious user behavior as well as byzantine attacks. There are many
interesting open problems in the broad area of physical layer security in wireless communica-
tions.
Acknowledgement
The authors want to thank Martin Mittelbach for his review of this chapter, critical comments
and productive discussions.
5. References
Bloch, M., Barros, J., Rodrigues, M. R. D. & Laughlin, S. W. M. (2008). Wireless Information-
Theoretic Security, IEEE Transactions on Information Theory 54(6): 2515–2534.
Boyd, S. & Vandenberghe, L. (2004). Convex Optimization, Cambridge University Press.
URL: http://www.stanford.edu/ boyd/cvxbook/
Cover, T. M. & Thomas, J. A. (2006). Elements of Information Theory, Wiley & Sons.
Csiszár, I. & Körner, J. (1978). Broadcast Channels with Confidential Messages, IEEE Transac-
tions on Information Theory 24(3): 339–348.
Diffie, W. & Hellman, M. E. (1976). New directions in cryptography, IEEE Transactions on
Information Theory 22(6): 644–654.
Jorswieck, E. A. & Gerbracht, S. (2009). Secrecy Rate Region of Downlink OFDM Systems:
Efficient Resource Allocation, 14th International OFDM-Workshop (InOWo), Hamburg,
Germany.
Jorswieck, E. A. & Larsson, E. (2009). Monotonic Optimization Framework for the MISO
IFC, IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP),
pp. 3633–3636.
Jorswieck, E. A. & Mochaourab, R. (2009). Secrecy Rate Region of MISO Interference Channel:
Pareto Boundary and Non-Cooperative Games, International ITG Workshop on Smart
Antennas, Berlin, Germany.
Jorswieck, E. A. & Wolf, A. (2008). Resource Allocation for the Wire-tap Multi-carrier Broad-
cast Channel, Proceedings of International Workshop on Multiple Access Communications
(MACOM), Saint Petersburg, Russia.
Leung-Yan-Cheong, S. & Hellman, M. (1978). The Gaussian wire-tap channel, IEEE Transac-
tions on Information Theory 24(4): 451–456.
Li, Z., Trappe, W. & Yates, R. (2007). Secret Communication via Multi-antenna Transmission,
41st Annual Conference on Information Sciences and Systems (CISS), pp. 905–910.
Liang, Y., Poor, H. V. & Shamai (Shitz), S. (2009). Information Theoretic Security, Vol. 5 of Foun-
dations and Trends in Communications and Information Theory, now publishers, pp. 355–
580.
Liu, J., Hou, Y. T. & Sherali, H. D. (2009). Optimal power allocation for achieving perfect
secrecy capacity in MIMO wire-tap channels, 43rd Annual Conference on Information
Sciences and Systems (CISS), pp. 606–611.
Liu, R., Liu, T., Poor, H. V. & Shamai (Shitz), S. (2009). Multiple-Input Multiple-Output Gaus-
sian Broadcast Channels with Confidential Messages, CoRR abs/0903.3786. submit-
ted.
Oggier, F. & Hassibi, B. (2008). The Secrecy Capacity of the MIMO Wiretap Channel, IEEE
International Symposium on Information Theory (ISIT), pp. 524–528.
Rivest, R. L., Shamir, A. & Adleman, L. (1978). A method for obtaining digital signatures and
public-key cryptosystems, Communications of the ACM 21(2): 120–126.
Schneier, B. (1996). Applied Cryptography, Wiley & Sons.
Shafiee, S., Liu, N. & Ulukus, S. (2008). Secrecy Capacity of the 2-2-1 Gaussian MIMO Wire-tap
Channel, 3rd International Symposium on Communications, Control and Signal Processing
(ISCCSP), pp. 207–212.
Tekin, E. & Yener, A. (2006). The Gaussian Multiple Access Wire-Tap Channel with Collec-
tive Secrecy Constraints, IEEE International Symposium on Information Theory (ISIT),
pp. 1164–1168.
Wyner, A. D. (1975). The Wire-tap Channel, Bell System Technical Journal 54(8): 1355–1387.
Zhang, J. & Gursoy, M. C. (2009). Low-SNR Analysis of Interference Channels under Secrecy
Constraints, CoRR abs/0901.3132. submitted.
Performance Analysis of Time-of-Arrival Mobile Positioning in Wireless Cellular CDMA Networks 437
21
X
Performance Analysis of Time-of-Arrival Mobile

Positioning in Wireless Cellular CDMA Networks
M. A.Landolsi, A. H. Muqaibel, A. S. Al-Ahmari, H.-R. Khan
and R. A. Al-Nimnim
EE Department, King Fahd University of Petroleum & Minerals
KFUPM, P.O. Box 1413, Dhahran 31261, Saudi Arabia
1. Introduction
Recently, wireless mobile communication systems have experienced a tremendous growth
and became an integral part of people’s daily life worldwide. This global predominance of
wireless communications has been ever more pronounced with the success of new
generations of wireless communication standards that support a rich set of value-added
features in addition to basic phone services. Among these features is the possibility to offer
radiolocation services whereby the mobile system (MS) position is determined by
combining relevant information (such as signal time of arrival or angle of arrival) from
different base stations (BSs) having radio links with the intended mobile. This positioning
capability can support many services ranging from medical emergency help, security and
law enforcement, on-the-road assistance, location-dependent commercial advertisement, etc.
As such, mobile radiolocation has been mandated by several of the recently introduced
wireless standards, and is being widely deployed by cellular network operators worldwide
(Rappaport et al., 1996).
With the deployment of the 3rd generation wireless cellular standards such as the Universal
Mobile Telecommunication System (UMTS) (Dahlman et al., 2000) , the use of wideband
code division multiple access (CDMA) signals is poised to offer highly accurate positioning
capabilities with time-of-arrival (TOA) information owing to the fine timing resolution of
the high chip rate wideband spread-spectrum waveforms used. There are however some
impairments such as multi-path fading, multi-user interference and noise, that can affect the
performance of mobile positioning schemes. In particular, because of the need to use signal
detection at several base stations in mobile positioning, the problem of near-far interference
at remote base stations (whereby a far-away mobile weak signal can be overwhelmed by
strong signals from close-in mobiles) can constitute a major limiting factor. In CDMA
networks, this is further exacerbated by power control (Viterbi, 1995). Indeed, power control
loops operate in such a way to maintain the received power from different users at the same
level at their respective serving base station. However, at other non-serving base stations
(not actively involved in a call with the intended mobile), the mobile received power level
can be extremely low, thereby giving rise to a problem of signal hearability, which, in turn,
will affect the accuracy of the mobile positioning algorithms (Gosh & Love, 1998). A
thorough analysis of these different aspects under realistic channel modeling and network
traffic loading conditions is therefore necessary in order to obtain an accurate assessment of
the achievable positioning performance.
In this work, we study the accuracy of mobile radiolocation under near-far interference, and
show that it can vary considerably depending on the mobile link quality with the base
stations involved in position determination. We first present a detailed performance analysis
of mobile TOA estimation in broadband CDMA wireless cellular networks at the different
base stations surrounding the mobile terminal. In most previous works, the proposed
radiolocation algorithms have simplistically assumed that the time-of-arrival measurements
are Gaussian-distributed, with a fixed known variance that is commonly set irrespective of
the actual RF channel and interference conditions or other system parameters. Moreover, the
same TOA variance is typically assigned for all base stations involved in positioning, which
is an inaccurate and overly optimistic assumption, as our results will show. Instead, our
approach is the use of complete statistics for the mobile timing estimation errors, derived by
taking into account realistic system parameters at the different base stations of interest.
Since precise TOA estimation in CDMA receivers is typically implemented by delay-locked

loop (DLL) tracking systems (Viterbi, 1995), a detailed study of the DLL TOA tracking is
introduced under fading and multi-user interference conditions assuming a cellular model
with multiple tiers of base stations, and it is shown that the DLL tracking performance can
vary widely depending on the level of multiple-access interference and RF propagation
conditions of the links between the mobile and base stations. Using the TOA data collected
at the base stations involved in mobile positioning, a numerically-efficient, quasi-optimum
algorithm, based on Approximate Maximum Likelihood (AML) estimation (Chan et al.,
2006), is presented, and a generalization of this algorithm is also derived under the realistic
assumption of un-equal TOA estimation errors at the different base stations.
Based on our analysis, the impact of mobile link condition and its relative position with
respect to the base stations involved in its positioning is fully quantified for a number of
scenarios depending on the near-far interference environment and the level of Soft Handoff
(SHO) connectivity of the mobile with the base stations. Our results show that positioning
accuracy is greatly improved when the mobile station is in 2-way, and particularly in 3-way
SHO (i.e., with two and three base stations, respectively), compared to single connectivity
with only the home serving base station.
The rest of the chapter is organized as follows. First, in Section 2, we present an overview of
radiolocation techniques, focusing in particular on network-based approaches with TOA
processing suitable for cellular systems. The system modeling and analysis of near-far
interference in CDMA networks is presented in Section 3, followed by a detailed analysis of
the performance of delay-locked loops for time-of-arrival tracking in Section 4. Then, in
Section 5, the approximate maximum likelihood positioning algorithm is presented, and
various illustrative examples and numerical results are discussed in Section 6, followed by
summary and final conclusions in Section 7.
2. Wireless Network-Based Mobile Radiolocation

2.1 Overview
The concept of wireless radiolocation refers to the determination of the geographic position
information of a mobile user in terms of its geographic coordinates with respect to a
reference point. Wireless location is also commonly referred to as mobile positioning,
radiolocation, or localization. Position location techniques can be classified into two main
categories: handset-based and network-based. A well known example of handset-based
radiolocation is the Global Positioning System (GPS) and other similar systems (Kaplan,
1996). The other category consists of network-based techniques that utilize the existing
wireless cellular networks to obtain location information (Sayed, et al., 2005). In this work,
we mainly focus on the network-based approach which integrates seamlessly with the
widely deployed mobile cellular infrastructures throughout the world. The basic
architecture of such systems is illustrated in Figure 1.
Fig. 1. Network-based wireless mobile radiolocation
The network-based positioning schemes rely on data collected by several base stations
surrounding the mobile station of interest, and can be based on the mobile’s signal strength
(SS), angle of arrival (AOA), time of arrivals (TOA) or time difference of arrival (TDOA)
measurements. Using these measurements, specific geometric and/or statistical signal
processing algorithms are used to determine the mobile location. Hybrid methods involving
more than one of type of measurements can also be used (Sayed, et al., 2005). In general,
locating a mobile in two-dimensions requires a minimum of three sets of measurements
from corresponding base stations, although for AOA methods, two base stations are
sufficient. However, in the presence of noisy measurements, statistical signal processing

algorithms using data collected from multiple base stations are preferred in order to resolve
the ambiguities arising from multiple crossings of the lines of position, and to improve the
positioning accuracy.
2.2 Time-of-Arrival Mobile Radiolocation

TOA-based techniques offer several advantages compared to the other methods, including
low cost and ease of use. The TOA data is readily available from timing synchronization
mechanisms at the different base stations, without requiring complex hardware as with the
Angle-of-Arrival methods. In particular, with the widespread deployment of the latest 3G
CDMA-based wireless cellular networks, the spread-spectrum signaling waveforms offer
high time resolution and good robustness vis-à-vis the radio channel impairments (fading,
shadowing, and near-far interference), and are therefore well-suited to aid in achieving the
required accuracy in position location services.
In the TOA-based mobile radiolocation approach, the distance between an MS and a BS is

measured by finding the one-way propagation delay under direct line-of-sight (LOS)
propagation conditions. The TOA measurements at the different base stations are therefore
directly proportional to the mobile-base distance separation. The involved base stations are
assumed to have a common timing reference, with a known mobile transmission time.
Geometrically, the mobile position will trace a circle centered at the base stations. By using
three base stations to resolve ambiguities, the mobile position is given by the intersection of
these circles as illustrated in Figure 2. It should be pointed out that, in the presence of noisy
TOA data due to interference and synchronization errors, the three circles may not intersect
at a single point. Therefore, the geometric approach will not produce a single intersection
point, and “statistical” techniques are typically adopted to process the noisy data, as will be
further discussed subsequently when we introduce the AML algorithm.
Fig. 2. TOA radiolocation based on intersection of base station centered circles

3. Near-Far Interference Analysis

3.1 Signal and System Models
Consider a cellular CDMA network employing signaling schemes with quaternary phase-
shift keying (QPSK) modulation and complex spreading (conforming to 3G/UTMS
standards (Dahlman et al., 2000). At the transmitter, the complex baseband signal for a given
mobile user is given by

s(t )  s
m 
m h(t  mTc ) (1)
where {sm} are the complex spreading chip symbols, and Tc the chip duration. The
information data bits are omitted here for simplicity as this will be applicable for a pure
pilot signal that can be used for positioning purposes. The impulse response h(t) of the
pulse-shaping filter is assumed to be a root-raised cosine filter with roll-off factor 22%, as
recommended in the UMTS standard. As will be seen next, the Fourier transform H(f) of the
filter impulse response h(t) will have a major impact on the other-user interference in
cellular CDMA networks.
For wireless fading channels, the received signal at the output of the radio frequency (RF)
receiver filter is written as
M 
r (t )  P Re  s  t   i (t )  ai (t )e j 2 fc t   n(t ) (2)
 i 1 
where P is the received signal power, M is the number of resolvable multipath signals, with
ai and i denoting the i-th path complex Gaussian tap factor (with a Rayleigh-fading
magnitude) and its propagation delay, respectively, and fc is the carrier frequency. The noise
signal n(t) models the total noise-plus-interference terms and is assumed to be zero-mean
Gaussian random process. For the different resolvable multipath signal epochs, we mainly
focus on the first arriving signal which will be tracked to estimate its TOA. The combined
multiple-access interference (MAI) terms from both same-cell and other-cell users can be
modeled as being Gaussian distributed, which is a valid assumption for a large number of
users. In this case, it can be shown that the total composite power spectral density (PSD), I0,
that captures the effect of both thermal noise and MAI terms (assumed to be statistically
independent) is given by (Viterbi, 1995)
  
2 o 4 o 4
I0  N0 

H ( f ) df 
Tc 

H ( f ) df  N 0 
Tc 

H ( f ) df (3)
where No is due to the thermal noise component and the factor  represents both same-cell
and other-cell MAI terms. The function H(f) denotes the Fourier transfer of the chip shaping
filter impulse response h(t). The MAI PSD term is typically the dominant factor in CDMA
systems, and will depend on network loading, fraction of inter-cell to intra-cell MAI,
channel pathloss and shadowing models. For typical system parameters of interest, it is
found that the o MAI factor is approximately 1.6(Ku-1), where Ku is the number of users per
cell (Viterbi, 1995).
3.2 Near-Far Interference and Soft Handoff Impact

We consider a cellular network with a central cell and two tiers of surrounding cells as
illustrated in Figure 3, where mobiles are assumed to be uniformly distributed across the
coverage area.
BS8
BS19 BS9
BS18 BS2 BS10

d2
BS7 BS3
d3
d1
BS17 BS1 BS11
BS6 BS4
BS16 BS5 BS12
BS15 BS13
BS14
Fig. 3. Cellular network model, with center cell and two tiers of interfering cells.
The mobile received power at a given base station, BSi, is multiplied by an attenuation
factor,  , that reflects distance path loss, p(d), and log-normal shadowing factor, ,
according to (Viterbi, 1995):
BSi /10
 ( dBS ,  BS )  p( dBS )10 (4)
i i i
The path-loss factor is assumed to follow the model:
p( d )  10 n log 10  d  (5)
where n is the path loss exponent and d the mobile’s relative position with respect to the
base station. For a given mobile location, shadowing vis-à-vis the different base stations is
typically assumed to be partially correlated, log-normally-distributed, and given by (Viterbi,
1995):
 BS  ac  b i
i
(6)
where c andi denote the common and independent terms, respectively, and a2+b2=1.
With closed-loop CDMA power control, a given mobile will have the same received power
at its home (serving) base station compared with other intra-cell mobiles. On the other hand,
at the neighboring base stations, the mobile will be received at much lower power due to the
so-called near-far problem (which cannot be mitigated due to the lack of “inter-cell” power
control). However, when the mobile is in soft-handoff with other neighboring base stations,
its received power is relatively close to that at the home cell, and this greatly improve
radiolocation accuracy, as will be discussed next.
Since TOA estimation accuracy depends on the timing synchronization mechanism, which is
in turn affected by the received interference levels at the different base stations, the near-far
interference at the non-serving base stations will have a major impact on the final mobile
positioning accuracy. To illustrate this point, we consider a system where the mobile is
served by the center base station BS1 and is radio-located using TOA data from three or
more base stations BS1, BS2,…, BS7 (sorted in a descending order from BS1 that receives the
highest average received power). The near-far interference impact can be conveniently
assessed by defining the ratio of its average received power at BSi compared to BS1 (for
which the power will be normalized to 1, and used as a reference value). We then define the
following:
 i  Pi P1 (7)
where Pi is the received power at BSi and ≥  ≥  ≥  ≥ . A wide fluctuation in the 
factors is expected depending on the mobile position relative to the base stations of interest.
It should be noted that, due to power control, all mobiles will be received at equal power
(=1) at their respective home serving base stations, but much lower values of  are
expected at far-away base stations (because of the near-far problem). This however will
depend on the relative position of the mobile with respect to the other base stations (i.e., its
proximity to the cell border). Soft Handoff (SHO) is one of the salient features of CDMA
cellular systems which allow the mobile to be simultaneously connected to more than one
serving base station. In fact, the possibility of SHO calls enables a stronger signal reception
at multiple base stations, and this will in turn improve positioning accuracy.
To further investigate this point, we consider different scenarios denoted by Cases 1, 2, and
3, respectively. Case 1 refers to a mobile in close proximity (within half the cell radius, R) of
its serving BS1, with a signal at least 10dB above that at the other nearest two base stations.
Case 2 represents a two-way soft handover scenario, with the mobile power at BS2 within
3dB from that at BS1. Finally, Case 3 corresponds to a 3-way soft handover situation where
the mobile signal at both BS2 and BS3 is within 3dB compared to BS1. The numbers chosen
here merely serve to illustrate the variability in received signal power (and its subsequent
impact on TOA estimation accuracy), but do not attempt to model the specific thresholds
and soft handover mechanisms used in CDMA standards.
Table 1 and Table 2 give the different  factors for the three cases of interest, and for
different values for the pathloss and shadowing models, where a typical 50% correlation
factor is assumed (a=b). For the numerical results, we assumed typical parameters for the
radio propagation channel model. A two-segment pathloss model with breakpoint at
distance do=200m and exponents n=2 and 4, respectively, was used, with a load of 20 users
per cell, and a cell radius of 2Km. The relative powerfactors given in Table 1 and Table 2
clearly show that large variations occur across the base stations depending on the mobile
soft handoff conditions and its proximity to one or more base station, and this will impact
the accuracy of signal estimation at the different base stations as will be discussed next.
β1 β2 β3 β4 β5 β6 β7
Case 1 (mobile
1 0.0216 0.0113 0.0069 0.0045 0.0031 0.0021
close to home BS)
Case 2 (mobile in
1 0.6982 0.2215 0.1202 0.0735 0.0485 0.0331
two-way SHO)
Case 3 (mobile in
1 0.7922 0.6353 0.2993 0.1701 0.1065 0.0706
three-way SHO)
Table 1. Relative received power factors for various mobile soft-handoff link conditions.
Shadowing s.t.d 8dB.
β1 β2 β3 β4 β5 β6 β7
Case 1 (mobile
1 0.0248 0.0125 0.0072 0.0045 0.0030 0.0020
close to home BS)
Case 2 (mobile in
1 0.7000 0.2281 0.1258 0.0761 0.0486 0.0322
two-way SHO)
Case 3 (mobile in
1 0.7985 0.6443 0.3403 0.1953 0.1252 0.0808
three-way SHO)
Table 2. Relative received power factors for various mobile soft-handoff link conditions.
Shadowing s.t.d 12dB.
4. Signal Time-of-Arrival Estimation

4.1. Delay-Locked Loop Time Tracking
Timing synchronization for CDMA signals is typically implemented in two steps consisting
of an initial coarse timing acquisition (within an uncertainty range on the order of one-chip
interval), followed by fine time tracking achieved by a delay-locked loop (DLL) mechanism
(Viterbi, 1995). In this work, we assume that the initial timing acquisition has been achieved,
and focus on the more accurate DLL tracking as the main signal timing estimation
mechanism. For each base station involved in mobile positioning, a DLL device
continuously attempts to bring the local code timing estimate in perfect alignment with the
incoming mobile signal. However, this timing estimation will be subject to error due to
noise, fading and multiple-access interference. In the following, we consider TOA
estimation based on a non-coherent DLL (NC-DLL) scheme. In practice, the NC-DLL is
preferred over a coherent structure because of its insensitivity to carrier phase error and
data modulation. Figure 4 depicts a block diagram showing the different processing stages
of a NC-DLL code tracking loop. Because QPSK spreading is used, the NC-DLL employs
both I & Q branches where the I & Q received signals, after down conversion and chip
matched filtering, are fed to two early & late branches which correlate the spread-spectrum
waveforms with advanced and delayed code sequence replicas. The outputs obtained at the
I & Q channels of the early & late branches can be obtained as (Viterbi, 1995)
YI   AN PTC R(   ) cos  (8)
YQ   AN PTC R(   ) cos  (9)
where A is the fading signal envelope modeled as a Rayleigh random variable, and θ is its
uniform phase.  is the early & late timing offset (typically set to Tc/2), P is the signal
power, and N is the number of accumulated chips. The function R() is a correlation function
given by the convolution of the impulse responses of the pulse-shaping filter and its
matched filter (Viterbi, 1995):
 2
R( )  h( )  h(  )   H ( f ) cos(2 fc )df (10)

The DLL I & Q correlator outputs are combined as shown in Figure 4, and a resulting
discriminator metric Z is low-pass filtered to form an error signal used to drive a
numerically-controlled oscillator (NCO) that controls the code timing adjustment. The DLL
discriminator output can be obtained as (Viterbi, 1995):
Z   AS( )   (11)
where  is the power reduction factor reflecting the impact of DLL operation at different
base stations (compared to the main serving base station, as discussed previously). The term
 represents the combined Gaussian noise term with variance I0. The function S() is known
as the normalized DLL S-curve, and is given by
S( )  R 2 (   )  R 2 (   ) (12)
where  is the normalized timing error given by
  (  ˆ )/Tc (13)
with  denoting the correct TOA and ˆ the estimated one.

A common figure of merit for assessing the DLL performance is based on the tracking jitter
variance. For additive white Gaussian (AWGN) channels, a simple upper-bound
approximation for the tracking jitter variance, valid for linearized first-order DLL models, is
derived in (Viterbi, 1995) as
2 I o2  4 NEc I o
  2 2 2 (14)
N Ec 
where Ec is the received chip energy, is the slope of the S-curve at the origin, and I0 is the
variance of the thermal noise & MAI terms. However, there is no simple equivalent result
for the case of frequency-selective multipath fading channels considered in this work, and
the above approximation is only valid for high mobile received Signal-to-Noise Ratio (SNR)
conditions, which is not necessarily the case at the remote base stations (other than the home
serving one), as discussed previously. In this work, we resort to a more accurate
performance analysis valid for all SNR conditions and based on the computation of the full
statistics (probability density function) of the DLL tracking loop, as discussed next.
Fig. 4. DLL system used for CDMA signal timing synchronization and TOA estimation
4.2. DLL Tracking Error Statistics

For the purpose of analyzing the impact of TOA estimation error on mobile positioning
accuracy, we need to obtain accurate statistics for the TOA measurements at each of the base
stations involved in tracking the given mobile TOA. Hypothetical distributions (Gaussian
models) are commonly assumed for the TOA timing estimation error. The same variance is
usually assigned to the TOA error at all base stations involved in mobile positioning, and
this variance is sometimes set rather arbitrarily. In this work, we use results based on a more
rigorous analysis with full derivation of the TOA error probability density function (PDF).
The results are obtained following the approach presented in (Su & Yen, 1997) and extended
to quadrature-spread CDMA signals in (Khan, 2009). Assuming a discrete-time model, the
analysis is based on a discrete-time Markov model for the residual DLL error, according to
the following equation:
 k   k  1  K NCO   Ak2 1S( k  1 )   k  1  (15)
where KNCO is the NCO gain, and k is the additive Gaussian noise term. It can be seen that
the residual tracking error follows a discrete-time Markov process for which the first order
probability distribution can be obtained using the Kolmogorov-Chapman equation (Su &
Yen, 1997):

pk ( )   f k  1   |x  pk  1  x| 0  dx (16)

where 0 is the initial timing error, pk-1(x|0) is the PDF of x given 0, and fk-1(|x) is the
transition pdf of k given x. Through detailed analysis, one can calculate the exact
expressions fk-1(|x) under assumption of a Rayleigh fading channel. Using numerical
integration, it is then possible to iterate the Kolmogorov-Chapman equation to get the PDF
of the TOA estimation error. The lengthy details of this derivation are not included here, but
can be found in (Khan, 2009).
The procedure outlined above can be done for different scenarios reflecting mobile TOA
estimation at a given base station of interest. For our purpose, we analyze the performance
of TOA DLL tracking at the different base stations for each one of the three cases described
in Section 3. The results are shown in Figures 5 through 7, where it is clearly seen that the
TOA residual error behavior can vary widely depending on the mobile position vis-à-vis the
tracking base station. For example, in Figure 5 which corresponds to a mobile very close to
its home serving base station, the residual error at the home base station is well-confined
and nearly Gaussian-distributed (with a standard deviation found to be on the order of
0.15Tc), whereas for the other two base stations, the timing errors remain nearly uniformly-
distributed with a standard deviation of 0.29Tc (which shows that the DLL loops at those
base stations are not effectively tracking the mobile signal TOA). On the other hand, in
Figure 6 and Figure 7 corresponding to 2-way and 3-way SHO, respectively, the mobile time
tracking performance at BS2 and BS3 is markedly better, with distributions approaching that
of the home BS1. It is to be noted that the number of users per cell (assumed the same for all
cells, for simplicity) can also have a major impact on the timing estimation accuracy,
regardless of the mobile position scenario (i.e., for all different cases discussed previously).
Indeed, as shown in Figure 8, when the number of users increases, the tracking error PDFs
are more wide-spread, and have an increasing error s.t.d (given with a normalization factor
of 1/Tc).
Fig. 5. Mobile TOA Tracking error PDF at BS1, BS2, BS3 for Case 1 (MS without handoff)
Fig. 6. Mobile TOA tracking error PDF at BS1, BS2, BS3 for Case 2 (MS in 2-way soft handoff)
Fig. 7. Mobile TOA time tracking error PDF at BS1, BS2, BS3 for Case 3 (MS in 3-way soft
handoff)
Tracking Error Distribution as Function of Number of Users in a Cell

5
10 users, std=0.0924
4.5 20 users, std=0.152
30 users, std=0.214
4
3.5
3
PDF
2.5
1.5
0.5
0
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
normalized tracking error
Fig. 8. Illustration of the impact of the number of users on the TOA tracking error
probability density function.
5. TOA Processing for Mobile Positioning

5.1. Approximate Maximum Likelihood Algorithm
In time-of-arrival radiolocation techniques, the distance is calculated as the propagation
time multiplied by the speed of light c. Line-of-sight (LOS) propagation is assumed,
whereby the mobile signal travels on a direct path at the constant speed of light in free
space. It is further assumed that the base stations are synchronized and the mobile
transmission time is known (set to zero for simplicity). The TOA measurements, produced
at each base station, are therefore directly proportional to the mobile-base distance
separation. Geometrically, circles centered at the base stations can be drawn with the
calculated distance as the radius. With the help of three base stations, the mobile location
can be found geometrically as the intersection of the corresponding circles. However, in the
presence of noise and interference, the three circles may not intersect at a single point.
Therefore, the geometric approach is not suitable, and several other “statistical” techniques
have been proposed (Caffrey, 1999), (Sayed et al., 2005) to process the noisy data. Many of
these are based on iterative algorithms using least-squares or gradient search minimization.
On the other hand, since the positioning equations involved are typically nonlinear, some
traditional approaches based on linearization followed by a gradient search were proposed
(Niezgoda & Ho, 1994). However, these approaches suffer from sensitivity to initialization
errors and convergence problems. More recently, some researchers introduced new closed-
form linear techniques (Caffrey & Stuber, 2000), (Chan & Ho, 1994), but the drawback of
these methods is that optimum location estimates can only be found at high SNR values,
which may not always be the case in practice.
Another interesting solution was proposed by (Chan et al., 2006) and shown to have near-
optimum performance, with the added advantage of reduced complexity. Their method
uses an Approximate Maximum Likelihood (AML) algorithm, which we adopt in this study.
In the following, we give a general overview of the AML algorithm processing steps, and
introduce a modification tailored towards the case of unequal TOA variances at the different
base stations, and relevant to mobile radiolocation under near-far interference conditions
(which is the focus of this work), as will be outlined subsequently. Based on the assumed
cellular geometry, the true distances between the BSs and MS are given by
 x  xi    y  yi 
2 2
Ri  i=1,2,……,NBS (17)
where NBS is the number of involved BSs. The measured distances, li, is given by
li  Ri   i i=1,2,……,NBS (18)
where εi is the DLL timing error. In matrix form, this is written as
l = R+ε (19)
Dividing by c to get the measured TOA vector T, we obtain

R ε
T   T0 + e (20)
c c
where
R   R1  RN   R(Θ)
t
(21)
and
e   e1 ....eN 
t
(22)
is the vector of additive measurement noise, and T0 is the vector of true TOAs. The original
AML algorithm proposed by (Chan et al., 2006) assumes that all BSs have an equal error
variance. Hence, the elements of e are assumed to be independent, zero-mean Gaussian
random variables with covariance matrix
Q  E eet    2 I (23)
The AML algorithm can be generalized to account for different values of the error variance
at different BSs. If the variance of the error can be estimated at each BS, one can use this
information to improve the localization performance by giving more trust to BSs with lower
error variance. In this case the elements of e are assumed to be independent, zero mean
Gaussian random variables with covariance matrix:
Q  E eet   diag  12  N2  (24)
The conditional probability density function (PDF) of T given Θ is given by
 
N BS 1
  J
f ( T Θ)  (2 ) 2
(det Q ) 2
exp  (25)
2
where
1
t
 12  0   e1  N
 R (Θ )  1  R (Θ)        BS 1 
J  T   Q T    e1 ....e N          (26)
i 1  i
2
 c   c 
 2 
e N 
 0  N 
The ML estimate of the MS position (x, y) is the Θ that minimizes J (Trees, 1968). Minimizing
J is done by setting its gradient with respect to Θ to zero. Considering first the derivative
with respect to the position variable x, we have
J NBS 1 ei2 NBS 1 e

 2   2 2 ei  i (27)
x i  1  i x i  1  i x
By expressing the timing error in terms of the difference between the true TOA and
measured one, this gives
ei   Ri  1 Ri
 T    (28)
x x  c  c x
Utilizing Equation (17), we may write

Ri 1   x  xi    x  xi 
  2  (29)
x 2  Ri  Ri
Substituting the result in Equation (28) yields

ei 1  x  xi 
  (30)
x c Ri
Substituting in Equation (27), we get
J 1 N BS 1  x  xi 
    2  2 ei 
x c i 1  i Ri
(31)
2 1 N BS
 Ri  li    x  xi 
 2  2 
c i 1  i Ri
The above steps starting with Equation (27) can be repeated for ∂J/∂y in a straightforward
manner. Finally, by setting the gradient of J with respect to Θ to zero, we get the two ML
equations
N BS
1  Ri  li    x  xi 

i 1
2

Ri
0
i
(32)
N BS
1  Ri  li    y  yi 

i 1
2

Ri
0
i
The above equations can be expressed in matrix notation as
  gi xi
2
 g y   x    g  s  K
i i i i  li2  
 (33)
  hi xi  h x   y    h  s  K
i i i i  li2  
where
s  x2  y2 (34)
K i  xi2  y i2 (35)
x  xi
gi  (36)
  Ri   Ri  li 
i
2
y  yi
hi  (37)
 i2  Ri   Ri  li 
In a more compact form, Equation (34) can be re-written as
AΘ = b (38)
with the matrix A and vector b being functions of Θ. A suboptimal solution based on a
linear model (Chan & Ho, 1994) can be used as a first initial estimate of Θ, which will in turn
give starting values of A and b. Then, solving Equation (33) produces a new value of Θ to
update A and b, and subsequently Θ. This iterative procedure first gives an approximate
maximum likelihood (AML) estimator, which can then be iterated a number of times to
obtain a final solution. The final solution takes the Θ that gives the smallest J in Equation
(26). This ensures that the AML will not diverge. In fact, simulation results presented in
(Chan et al., 2006) show that the AML can nearly achieve the Cramer–Rao lower bound
(CRLB) with a small number of iterations (typically on the order of five updates are found to
be sufficient). It should be noted that, for the special case of equal measurement variance
assumption (as in the original AML), the common  term drops from Equation (32).
However, with the modified AML in Equation (33), the quantities gi and hi will be different
for different BSs, and this is found to yield some improvement in performance as will be
discussed next.
5.2. Positioning Accuracy with the Modified AML Algorithm

To illustrate the impact of the modified AML that takes into account the unequal error
variances, we consider an example of three BSs with different measurement error statistics
for mobiles located in three different regions, as described in the different cases of Section 3.
Figure 9 shows the geometry of the layout used for this purpose, where distances are shown
in meters. The dense scatter points represents a total of 105 noisy mobile locations generated
according to the TOA statistics for the different cases of interest which are referred to as
MS1, MS2, MS3, corresponding to mobiles classified in Section 3 as Case 1, 2, and 3,
respectively. For each mobile location, the AML and modified AML algorithms are executed
to obtain the estimated mobile coordinates, and by comparing with the known mobile
position, the resulting positioning error can be computed. The different parameters for the
radio channel and relative received power factors used are based on the results given in
Table 1 of Section 3.
Fig. 9. Layout of the mobile localization geometry used to test the accuracy of the AML
algorithm.
Various aspects of mobile positioning performance are illustrated in Figure 10 based on the
cumulative distribution function (CDF) of the radiolocation error. First, it can be seen that
the modified AML which takes into account the differences in TOA noise statistics at the
various base stations (as proposed in this work) outperforms, albeit slightly, the
conventional AML which assumes equal TOA noise variance at all base stations. This is
more pronounced for mobiles categorized as MS1 (i.e., which are in close proximity to their
home base stations). The other major observation from Figure 10 is that a large difference in
radiolocation accuracy is present depending on the mobile relative position with respect to
the base stations. It is clearly seen that MS2 and particularly MS3 mobiles, corresponding to
mobiles in 2-way and 3-way soft handoff, respectively, achieve much better performance as
opposed to MS1. This is due to the poor TOA accuracy at BS2 and BS3 for the latter case,
owing to the overwhelming near-far interference experienced by the mobile at the remote
bases stations, as discussed in Section 4. As an example, the probability of the residual
positioning error being less than 20m is almost one for MS3 mobiles, while it is on the order
of 70% for MS2, and only 50% for MS1–type mobiles. Similar observations also hold for
other distances. Therefore, as highlighted throughout this study, the accuracy of mobile
positioning is best when the mobile is in close proximity to a border cell region where soft
handoff connectivity is established with one or two more cells involved in its positioning in
addition to its home serving cell.
Fig. 10. Illustration of mobile positioning accuracy with the Approximate Maximum
Likelihood Algorithm.
6. Conclusion
This study dealt with the performance analysis of time-of-arrival (TOA) techniques for
mobile positioning in CDMA wireless cellular networks. Since several base stations are
typically needed for mobile radiolocation, the problem of weak signal hearability at remote
base stations is a major challenge, and a detailed analysis of this issue was presented by
taking into account the near-far interference usually present in CDMA cellular networks.
TOA-based positioning methods are well-suited for wide deployment of radiolocation

services since the synchronization circuits that can extract timing information are readily
available at the base stations receiving the mobile signal. For CDMA signals, the delay-
locked loop (DLL) is commonly used as a TOA estimation device, and a detailed analysis of
DLL-based TOA tracking was presented taking into account the cellular network layout, cell
loading and other-user interference, as well as RF channel shadowing and pathloss that
affect received signal strength at the different base stations. In particular, it was shown that
the TOA tracking error statistics can vary considerably depending on the mobile link
conditions with respect to the base stations involved in its positioning, and soft handoff
(SHO) links with two or three base stations for mobiles close to border-cell regions were
found to improve the precision of mobile TOA tracking.
Using TOA information, a mobile radiolocation positioning algorithm based on a

computationally-efficient, near-optimum approximate maximum likelihood (AML)
processing was presented, and a generalization to the case of unequal timing error variances
at the different base stations was also derived. Numerical results were presented to quantify
the achievable positioning accuracy of the AML algorithms, and it was also shown that SHO
radio links with two or particularly three base stations have a clear impact on the precision
of mobile radiolocation. Finally, it should be noted that the problem of non-line-of-sight
(NLOS) propagation, which constitutes another major challenge to mobile positioning
accuracy, was not included in this study, and would be addressed in future work.
Acknowledgment
This work was gratefully supported by King Abdulaziz City for Science & Technology and
King Fahd University of Petroleum & Minerals, Saudi Arabia.
7. References
Caffrey, J. J. (1999). Wireless Location in CDMA Cellular Radio Systems. Kluwer Academic
Publishers.
Caffrey, J. J. & Stuber, G. L. (2000). Effects of multiple access interference on the non-
coherent delay lock loop. IEEE Transactions on Communications, Vol. 48, (December
2000) 2109-2119.
Chan, Y-T.; Hang, H. & Ching, P-C. (2006). Exact and approximate maximum likelihood
localization algorithms. IEEE Transactions on Vehicular Technology, vol. 55, No. 1,
(January 2006) 10-16.
Chan, Y. T. & Ho, K. C. (1994). A simple and efficient estimator for hyperbolic location. IEEE
Transactions on Signal Processing, Vol. 42, No. 8, (August 1994) 1905–1915.
Chan, Y. T.; Yau, C. H. & Ching P. C. (2006). Exact and approximate maximum likelihood
localization algorithms. IEEE Transactions on Vehicular Technology, Vol. 55, No. 1,
(January 2006) 10-16.
Dahlman, E.; Gudmundson, B.; Nilsson, M. & Skold, A. (2000). UMTS/IMT-2000 based on
wideband CDMA. IEEE Communications Magazine, (September 2000) 70-80.
Gosh, A. & Love, R. (1998). Mobile station location in a DS-CDMA system, Proceedings of
IEEE VTC, Vol. 1, pp. 254-258.
Kaplan, E. (1996). Understanding GPS: Principles and Applications. Norwood, MA: Artech
House.
Khan, H. R. (2009). DLL Code Tracking for CDMA Signals under Fading and Multiple
Access Interference, M.S. Thesis, King Fahd University of Petroleum & Minerals,
February 2009.
Niezgoda, G. H. & Ho, K. C. (1994). Geolocation by combined range difference and range
rate measurements, Proceedings of International Conference Acoustics, Speech, Signal
Process (ICASSP’94), Adelaide, Australia, Vol. 2, 1994, pp. 357–360.
Rappaport, T.; Reed, J. & Worner, B. (1996). Position location using wireless communications
on highways of the future. IEEE Communications Magazine, (October 1996) 33-41.
Sayed, A. H.; Tarighat, A. & Khajehnouri, N. (2005). Network-Based Wireless Location. IEEE
Signal Processing Magazine, (July 2005) 24-40.
Su, S. L. & Yen, N. Y. (1997). Performance of digital code tracking loops for direct-sequence
spread-spectrum signals in mobile radio channels. IEEE Transactions on
Communications, Vol. 45, (May 1997) 596-604.
Trees, H. L. V. (1968). Detection, Estimation, and Modulation Theory: Part 1, Wiley, New York.
Viterbi, A. J. (1995). CDMA: Principles of Spread Spectrum Communication, Addison-Wesley.
Mobility and Handoff Management in Wireless Networks 457
22
X
Mobility and Handoff Management

in Wireless Networks
Jaydip Sen
Tata Consultancy Services
INDIA
1. Introduction
With the increasing demands for new data and real-time services, wireless networks should
support calls with different traffic characteristics and different Quality of Service (QoS)
guarantees. In addition, various wireless technologies and networks exist currently that can
satisfy different needs and requirements of mobile users. Since these different wireless
networks act as complementary to each other in terms of their capabilities and suitability for
different applications, integration of these networks will enable the mobile users to be
always connected to the best available access network depending on their requirements.
This integration of heterogeneous networks will, however, lead to heterogeneities in access
technologies and network protocols. To meet the requirements of mobile users under this
heterogeneous environment, a common infrastructure to interconnect multiple access
networks will be needed. Although IP has been recognized to be the de facto protocol for
next-generation integrated wireless, for inter-operation between different communication
protocols, an adaptive protocol stack is also required to be developed that will adapt itself to
the different characteristics and properties of the networks (Akyildiz et al., 2004a). Finally,
adaptive and intelligent terminal devices and smart base stations (BSs) with multiple air
interfaces will enable users to seamlessly switch between different access technologies.
For efficient delivery of services to the mobile users, the next-generation wireless networks
require new mechanisms of mobility management where the location of every user is
proactively determined before the service is delivered. Moreover, for designing an adaptive
communication protocol, various existing mobility management schemes are to be
seamlessly integrated. In this chapter, the design issues of a number of mobility
management schemes have been presented. Each of these schemes utilizes IP-based
technologies to enable efficient roaming in heterogeneous network (Chiussi et al., 2002).
Efficient handoff mechanisms are essential for ensuring seamless connectivity and
uninterrupted service delivery. A number of handoff schemes in a heterogeneous
networking environment are also presented in this chapter.
The chapter is organized as follows. Section 2 introduces the concept of mobility
management and its two important components- location management and handoff
management. Section 3 presents various network layer protocols for macro-mobility and
micro-mobility. Section 4 discusses various link layer protocols for location management.
Section 5 introduces the concept of handoff. Different types of handoff mechanisms are
classified, and the delays associated with a handoff procedure are identified. Some
important cross-layer handoff mechanisms are also discussed in detail. Section 6 presents
media independent handover (MIH) services as proposed in IEEE 802.21 standards. It also
discusses how MIH services can be utilized for designing seamless mobility protocols in
next-generation heterogeneous wireless networks. Section 7 discusses security issues in
handover protocols. Section 8 identifies some open areas of research in mobility
management. Section 9 concludes the chapter.
2. Mobility Management
With the convergence of the Internet and wireless mobile communications and with the
rapid growth in the number of mobile subscribers, mobility management emerges as one of
the most important and challenging problems for wireless mobile communication over the
Internet. Mobility management enables the serving networks to locate a mobile subscriber’s
point of attachment for delivering data packets (i.e. location management), and maintain a
mobile subscriber’s connection as it continues to change its point of attachment (i.e. handoff
management). The issues and functionalities of these activities are discussed in this section.
2.1 Location management

Location management enables the networks to track the locations of mobile nodes. Location
management has two major sub-tasks: (i) location registration, and (ii) call delivery or paging.
In location registration procedure, the mobile node periodically sends specific signals to
inform the network of its current location so that the location database is kept updated. The
call delivery procedure is invoked after the completion of the location registration. Based on
the information that has been registered in the network during the location registration, the
call delivery procedure queries the network about the exact location of the mobile device so
that a call may be delivered successfully. The design of a location management scheme must
address the following issues: (i) minimization of signaling overhead and latency in the
service delivery, (ii) meeting the guaranteed quality of service (QoS) of applications, and (iii)
in a fully overlapping area where several wireless networks co-exist, an efficient and robust
algorithm must be designed so as to select the network through which a mobile device
should perform registration, deciding on where and how frequently the location
information should be stored, and how to determine the exact location of a mobile device
within a specific time frame.
2.2 Handoff management

Handoff management is the process by which a mobile node keeps its connection active
when it moves from one access point to another. There are three stages in a handoff process.
First, the initiation of handoff is triggered by either the mobile device, or a network agent, or
the changing network conditions. The second stage is for a new connection generation,
where the network must find new resources for the handoff connection and perform any
additional routing operations. Finally, data-flow control needs to maintain the delivery of
the data from the old connection path to the new connection path according to the agreed-
upon QoS guarantees. Depending on the movement of the mobile device, it may undergo
various types of handoff. In a broad sense, handoffs may be of two types: (i) intra-system
handoff (horizontal handoff) and (ii) inter-system handoff (vertical handoff). Handoffs in
homogeneous networks are referred to as intra-system handoffs. This type of handoff occurs
when the signal strength of the serving BS goes below a certain threshold value. An
inter-system handoff between heterogeneous networks may arise in the following scenarios
(Mohanty, 2006) - (i) when a user moves out of the serving network and enters an overlying
network, (ii) when a user connected to a network chooses to handoff to an underlying or
overlaid network for his/her service requirements, (iii) when the overall load on the
network is required to be distributed among different systems.
The design of handoff management techniques in all-IP based next-generation wireless
networks must address the following issues: (i) signaling overhead and power requirement
for processing handoff messages should be minimized, (ii) QoS guarantees must be made,
(iii) network resources should be efficiently used, and (iv) the handoff mechanism should be
scalable, reliable and robust.
2.3 Mobility management at different layers

A number of mobility management mechanisms in homogeneous networks have been
presented and discussed in (Akyildiz et al., 1999). Mobility management in heterogeneous
networks is a much more complex issue and usually involves different layers of the TCP/IP
protocol stack. Several mobility management protocols have been proposed in the literature
for next-generation all-IP wireless networks. Depending on the layers of communication
protocol they primarily use, these mechanisms can be classified into three categories –
protocols at the networks layer, protocols at the link layer and the cross-layer protocols.
Network layer mobility protocols use messages at the IP layer, and are agnostic of the
underlying wireless access technologies (Misra et al., 2002). Link layer mobility mechanisms
provide mobility-related features in the underlying radio systems. Additional gateways are
usually required to be deployed to handle the inter-operating issues when roaming across
heterogeneous access networks. In link layer protocols, handoff signals are transmitted
through wireless links, and therefore, these protocols are tightly-coupled with specific
wireless technologies. Mobility supported at the link layer is also called access mobility or link
layer mobility (Chiussi et al., 2002). The cross-layer protocols are more common for handoff
management. These protocols aim to achieve network layer handoff with the help of
communication and signaling from the link layer. By receiving and analyzing, in advance,
the signal strength reports and the information regarding the direction of movement of the
mobile node from the link layer, the system gets ready for a network layer handoff so that
packet loss is minimized and latency is reduced.
3. Network Layer Mobility Management Mechanisms

Over the past several years, a number of IP mobility management protocols have been
proposed. Different mobility management frameworks can be broadly distinguished into
two categories - device mobility management protocol for localized or micro-mobility and
protocols for inter-domain or macro mobility. The movement of a mobile node (MN) between
two subnets within one domain is referred to as micro-mobility. For example, the movement
of MN from subnet B to subnet C in Figure 1 is an example of micro-mobility. An example of
micro-mobility in UMTS Terrestrial Radio Access Networks (UTRAN) is movement of an
MN from one BS to another, both BSs belonging to the same random access network (RAN),
while in WLAN it is a node movement between two access points (APs). The movement of
devices between two network domains is referred to as macro-mobility. For example, the
movement of MN from domain 1 to domain 2 in Figure 1 is an example of macro-mobility.
A domain represents an administrative body, which may include different access networks,
such as WLAN, second-generation (2G), and third-generation (3G) networks (Akyildiz et al.,
2004b). Next-generation all-IP wireless network will include various heterogeneous
networks, each of them using possibly different access technologies. Therefore, satisfactory
macro-mobility solution supporting all these technologies is needed.
Fig. 1. Mobile IP Architecture [Source: (Akyildiz et al., 2004)]
3.1 Macro-mobility protocols

Mobile IP is the most widely used protocol for macro-mobility management. In addition to
Mobile IP, three macro-mobility architectures are discussed in the section. These protocols
are: Session Initiation Protocol (SIP)-based mobility management, multi-tier hybrid SIP and
Mobile IP protocol, and network inter-working agent-based mobility protocol.
Mobile IP: Mobile IP (Perkins, 2008) is the most well-known macro mobility scheme that
solves the problem of node mobility by redirecting the packets for the MN to its current
location. It introduces seven elements: (i) Mobile node (MN) – a device or a router that can
change its point of attachment to the Internet, (ii) Correspondent node (CN) – the partner with
which MN communicates, (iii) Home network (HN) – the subnet to which MN belongs, (iv)
Foreign network (FN) – the current subnet in which the MN is visiting, (v) Home agent (HA) –
provides services for the MN and is located in the HN, (vi) Foreign agent (FA) – provides
services to the MN while it visits in the FN, (vii) Care-of-address (CoA) – defines the current
location of the MN; all packets sent to the MN are delivered to the CoA. Mobile IP protocol
has three steps: (i) agent discovery, (ii) registration, and (iii) routing and tunneling.
Agent discovery: An MN is able to detect whether it has moved into a new subnet by two
methods – agent advertisement and agent solicitation. In the agent advertisement method,
FAs and HAs advertise their presence periodically using agent advertisement messages.
These advertisement messages can be seen as beacon broadcasts into the subnets. An MN in
a subnet can receive agent advertisements. If no agent advertisement messages are found or
the inter-arrival time is too high, the MN may send agent solicitations. After the step of
agent advertisement or solicitation, the MN receives a CoA. The CoA may be either an FA or
a co-located CoA (Perkins, 2008). A co-located CoA is found by using Dynamic Host
Configuration Protocol (DHCP) or Point-to-Point Protocol (PPP).
Registration: After the MN receives its CoA, it registers it with the HA. The main objective of
the registration is to inform the HA about the current location of MN. The registration may
be done in two ways depending on the location of the CoA. If the CoA is the FA, the MN
sends its registration request to the FA which in turn forwards it to the HA. If the CoA is co-
located, the MN may send the request directly to the HA.
Routing and tunneling: When a CN sends an IP packet to the MN, the packet is intercepted by
the HA. The HA encapsulates the packet and tunnels it to the MN’s CoA. With FA CoA, the
encapsulated packet reaches the FA serving the MN. The FA decapsulates the packet and
forwards it to the MN. With co-located CoA, the encapsulated packets reach the MN, which
decapsulates them. In Figure 1, the tunneling (step b) ends at the MN instead of at the FA.
Paging Extension for Mobile IP: For saving battery power at MNs, IP paging mechanism has
been proposed (Haverinen & Malinen, 2000). Paging typically includes transmitting a request
for an MN to a set of locations, in one of which the MN is expected to be present. The set of
locations is called a paging area and it consists of a set of neighboring base stations. A network
that supports paging allows the MNs to operate in two different states – an active state and a
standby state. In an active state, the MN is tracked at the finest granularity such as its current
base station (resulting in no need for paging). In the standby state, the MN is tracked at a much
coarser granularity such as a paging area. The MN updates the network less frequently in
stand by mode (every paging area change) than in active state (every base station). The cost of
paging, however, is the complexity of the algorithms and the protocols required to implement
the procedures, and the delay incurred for locating an MN.
Drawbacks of Mobile IP: The Mobile IP has the following shortcomings:
 The packets sent from a CN to an MN are received by the HA before being
tunneled to the MN. However, packets from the MN are sent directly to the CN.
This inefficient mechanism of non-optimized Mobile IP is called triangular routing.
It results in longer routes and more delay in packet delivery.
 When an MN moves across two different subnets, the new CoA cannot inform the
old CoA about MN’s current location. Packets tunneled to the old CoA are lost.
 Mobile IP is not an efficient mechanism in a highly mobile scenario as it requires an
MN to send a location update to the HA whenever it changes its subnet. The
signaling cost for location updates and the associated delay may be very high if the
distance between the visited network and the home network is large.
Optimization in Mobile IP: In (Perkins & Johnson, 2001), an optimization technique has been
proposed to solve the problem of triangular routing. The idea is to inform the CN about the
current location of the MN so as to bypass the HA. The CN can learn the location of the
CoAs of the MN by caching them in a binding cache in the CN. When a CN sends packets to
an MN, it first checks if it has a binding cache entry for the MN. If there is an entry, the CN
tunnels the packets directly to the CoA. If no binding cache entry is available, the CN sends
the packets to the HA, which in turn tunnels them to the CoA. In optimized Mobile IP, the
packets tunneled by the HA to the old CoA are not lost in transit. When an MN registers
with a new FA, it requests the new FA to notify the previous FA about its movement. As the
old FA now knows the location of the current FA, it can forward the packets to the new FA.
SIP-Based Mobility Management: In (Salsano et al., 2008), a Session Initiation Protocol (SIP)-
based solution, called mobility management using SIP extension (MMUSE), has been proposed
that supports vertical handoffs in next-generation wireless networks. SIP has been chosen by
the Third Generation Partnership Project (3GPP) as the signaling protocol to set up and
control real-time multimedia sessions. In MMUSE, a mobile host (MH) is assumed to be
equipped with multiple network interfaces; each of them is assigned a separate IP address
when connected to different access networks (ANs). The MH uses the SIP protocol to set up
multimedia sessions. The architecture of the scheme is depicted in Figure 2. The session
border controller (SBC) is a device that is typically located at the border of an IP network, and
manages all the sessions for that network. A new entity, called the mobility management server
(MMS) resides within the SBC. The MMS cooperates with another entity – mobility
management client (MMC) that resides in each MH. Both the SIP user agents (UAs) on the MH
and on the corresponding host (CH) remain unaware of all the handoff procedures, which are
handled by the MMC and the MMS. On the MH, the UA sees only the MMC as its outbound
proxy and forwards the normal SIP signaling and media flows to it. MMC relays the packets
to the MMS/SBC. From there on, the packets follow the path determined by the usual SIP
routing procedure. Every time the MH moves across two ANs, a location update SIP
message is sent to the MMS. This is done over the new network so that the procedure can be
completed even if the old network is suddenly not available. If the MMS receives a call
addressed to one of its served MHs, it forwards the call to the correct interface. When the
MH changes its AN while it is engaged in a call, the procedure is almost identical. However,
in this case, the MMC sends to the MMS an SIP message that contains the additional
information required to identify the call to be shifted to new interface. To minimize the
handoff duration, the real-time transport protocol (RTP) flow coming from the MH during the
handoff is duplicated using the MMC. When the MMC starts the handoff procedures, it
sends the handover request to the MMS and at the same time, it starts duplicating the RTP
packets over both interfaces. As soon as the MMS receives the handover message, the
packets coming from the new interface are already available. The MMS performs the
switching and sends the reply back to the MMC. When the MMC receives the reply
message, it stops duplicating the packets.
Fig.2. Architecture of MMUSE [Source: (Salsano et al., 2008)]
Multi-Layer Mobility Management using Hybrid SIP and Mobile IP: In (Politis et al.,
2003), two mobility management architectures based on SIP and Mobile IP are presented.
The two approaches provide mobility in two different layers: application and network
layers respectively. The scheme is therefore called multi-layer mobility management
scheme. The SIP-based protocol uses SIP in combination with IP encapsulation mechanisms
on CHs to support mobility for all types of traffic from/to the MH. The second approach
performs separation of traffic and employs SIP in combination with network address
translation (NAT) mechanisms to support mobility for real-time traffic over UDP. The
mobility for non-real-time traffic (mainly TCP-based applications) is supported by Mobile
IP. In the SIP-based approach, if the MH moves during a session, the SIP UA sends a SIP re-
INVITE request message to each of its CHs. If a CH runs a TCP session, IP encapsulation is
used to forward packets to MH. However, if a CH runs a UDP session, the packets are sent
directly to the MH’s new address. The MH completes the handoff by sending a SIP
REGSITER message to the SIP server. For the hybrid SIP/Mobile IP scheme, the inter-
domain mobility is based on the synergy of SIP with Mobile IP. Traffic from/to an MH is
separated on the domain edge routers. SIP signaling is used to support inter-domain
mobility for real-time (RTP over UDP) traffic, while Mobile IP supports non-real-time traffic.
Network Inter-Working Agent-Based Mobility Management: In (Akyildiz et al., 2005) an
architecture has been proposed for next-generation all-IP wireless systems. Different
wireless networks are integrated through an entity called the network inter-working agent
(NIA). In Figure 3, an NIA integrates one WLAN, one cellular network, and one satellite
network. NIA also handles authentication, billing, and mobility management issues during
inter-system (inter-domain) roaming. Two types of movement of an MH are considered:
movement between different subnets of one domain (intra-domain mobility) and movement
between different access networks belonging to different domains (inter-domain mobility).
For inter-domain mobility, a novel cross-layer mobility management protocol is proposed,
which makes an early detection of the possibility of an inter-domain handoff and allows
authentication, authorization and registration of the MH in the new domain before the
actual handoff. These interoperability operations are executed by the NIA.
Fig. 3. NIA-Based Mobility Management Architecture [Source: (Akyildiz et al., 2005)]

3.2 Micro-mobility protocols

Over the past several years a number of IP micro-mobility protocols have been proposed,
designed and implemented that complement the base Mobile IP (Campbell & Gomez, 2001)
by providing fast, seamless and local handoff control. IP micro-mobility protocols are
designed for environments where MHs changes their point of attachment to the network so
frequently that the base Mobile IP mechanism introduces significant network overhead in
terms of increased delay, packet loss and signaling. For example, many real-time wireless
applications, e.g. VOIP, would experience noticeable degradation of service with frequent
handoff. Establishment of new tunnels can introduce additional delays in the handoff
process, causing packet loss and delayed delivery of data to applications. This delay is
inherent in the round-trip incurred by the Mobile IP as the registration request is sent to the
HA and the response sent back to the FA. Route optimization (Perkins & Johnson, 2001) can
improve service quality but it cannot eliminate poor performance when an MH moves while
communicating with a distant CH. Micro-mobility protocols aim to handle local movement
(e.g., within a domain) of MHs without interaction with the Mobile IP-enabled Internet. This
reduces delay and packet loss during handoff and eliminates registration between MHs and
possibly distant HAs when MHs remain inside their local coverage areas. Eliminating
registration in this manner also reduces the signaling load experienced by the network.
The micro-mobility management schemes can be broadly divided into two groups: (i)
tunnel-based schemes and (ii) routing-based schemes. In tunnel-based approaches, the
location database is maintained in a distributed form by a set of FAs in the access network.
Each FA reads the incoming packet’s original destination address and searches its visitor list
for a corresponding entry. If an entry exists, it is the address of next lower level FA. The
sequence of visitor list entries corresponding to a particular MH constitutes the MH’s
location information and determines the route taken by downlink packets. Mobile IP
regional registration (MIP-RR) (Fogelstroem et al., 2006), hierarchical Mobile IP (HMIP)
(Soliman et al., 2008), and intra-domain mobility management protocol (IDMP) (Misra et al.,
2002) are tunnel-based micro-mobility protocol.
Routing-based approaches forward packets to an MH’s point of attachment using mobile-
specific routes. These schemes introduce implicit (snooping data) or explicit signaling to
update mobile-specific routes. In the case of Cellular IP, MHs attached to an access network
use the IP address of the gateway as their Mobile IP CoA. The gateway decapsulates packets
and forwards them to a BS. Inside the access network, MHs are identified by their home
address and data packets are routed using mobile-specific routing without tunneling.
Cellular IP (CIP) (Campbell et al., 2000) and handoff-aware wireless access Internet
infrastructure (HAWAII) (Ramjee et al., 2002) are routing-based micro-mobility protocols.
Mobile IP Regional Registration: In Mobile IP, an MN registers with its HA each time it
changes its CoA. If the distance between the visited network and the home network of the
MN is large, the signaling delay for these registrations may be long. MIP-RR (Fogelstroem et
al., 2006) attempts to minimize the number of signaling messages to the home network and
reduce the signaling delay by performing registrations locally in a regional network. This
reduces the load on the home network, and speeds up the process of handover. The scheme
introduces a new network node called the gateway foreign agent (GFA). The address of the
GFA is advertised by the FAs in a visited domain. When an MN first arrives at this visited
domain, it performs a home registration - that is, a registration with its HA. At this time, the
MN registers the address of the GFA as its CoA. When the MN moves between different
FAs within the same visited domain, it only needs to make a regional registration to the
GFA. When the MN moves from one regional network to another, it performs a home
registration with its HA. The packets for the MN are first intercepted by its HA, which
tunnels them to the registered GFA. The GFA checks its visitor list and forwards the packets
to the corresponding FA of the MN. The FA further relays the packets to the MN. The use of
the GFA avoids any signaling traffic to the HA as long as the MN is within a regional
network.
Hierarchical Mobile IPv6: The basic idea of hierarchical Mobile IP (Soliman et al., 2008)
(HMIP) is the same as that of regional registration scheme. HMIP introduces a new Mobile
IP node called the mobility anchor point (MAP). An MN is assigned two CoAs - regional CoA
(RCoA) and on-link CoA (LCoA). The MN obtains the RCoA from the visited networks.
RCoA is an address on the MAP’s subnet. The LCoA is the CoA that is based on the prefix
advertised by the access router (AR). The AR is the default router of the MN and receives all
outbound traffic from it. When an MN enters a new network, it receives router
advertisement that contains the available MAPs and their distances from the MN. The MN
selects a MAP, gets the RCoA in the MAP’s domain and the LCoA from the AR. The MN
sends a binding update to the MAP. The MAP records the binding and inserts it in its
binding cache (foreign registration). The MAP sends the binding update message also to the
MN’s HA and to the CNs (home registration). When MN is outside its home network, the
incoming data to MN goes through MAP hierarchy. Messages from CN or HA are received
by the MAP, which tunnels them to LCoA. As the MN roams locally, it gets a new LCoA
from its new AR. The RCoA remains unchanged as long as the MN is within the same
network.
Fig. 4. The Architecture of IDMP [Source: (Akyildiz et al., 2004)]

Intra-Domain Mobility Management Protocol: Intra-domain mobility management

protocol (IDMP) (Misra et al., 2002) is a two-level, hierarchical, multi-CoA, intra-domain
mobility management protocol. The first level of the hierarchy consists of different mobility
domains. The second level consists of IP subnets within each domain. This hierarchical
approach localizes the scope of intra-domain location update messages and thereby reduces
both the global signaling load and update latency. The two-level hierarchical architecture
defined by IDMP is shown in Figure 4. IDMP consists of two types of entities: (i) mobility
agent (MA) and (ii) subnet agent (SA). The MA provides a domain-wide stable access point
for an MN. An SA handles the mobility of MNs within a subnet. Similar to HMIP, each MN
can get two CoAs - global CoA (GCoA) and local CoA (LCoA). The GCoA specifies the
domain to which the MN is currently attached. The LCoA identifies the MN’s present
subnet. The packets destined to an MN are first received by the HA. The HA tunnels the
packets to the MA using the MN’s GCoA. The MA first decapsulates the packets, determines
the current LCoA of the MN using its internal table, and tunnels them to the LCoA. The
encapsulated packets are received by the SA. Finally, the SA decapsulates the packets and
forwards them to the MN. When the MN moves from one subnet to another inside the same
domain, it is assigned a new LCoA. The MN registers the address of the new LCoA with its
MA. Till the registration of the new LCoA is complete, the MA forwards all packets for the
MN to the old LCoA. This results in packet drops. A fast handoff procedure has been
proposed to avoid this packet loss (Misra et al., 2002). It eliminates intra-domain update
delay by anticipating the handover in connectivity between the networks and the MNs. The
anticipation of MN’s movement is based on a link layer trigger which initiates a network
layer handoff before the link layer handoff completes. Once the MN senses a handoff, it
sends a request to the MA to multicast the packets to its SAs. The MA multicasts incoming
packets to each neighboring SAs. Each SA buffers the packets in order to prevent any loss of
packets in transit during the handoff. After the MN finishes registration, the new SA
transfers all buffered packets to the MN.
Cellular IP: Cellular IP (Campbell et al., 2000) is a mobility management protocol that
provides access to a Mobile IP-enabled Internet for fast moving MHs. The architecture of
Cellular IP is shown in Figure 5. It consists of three major components: (i) cellular IP node or
the base station (BS), (ii) cellular IP gateway (GW), and (iii) cellular IP mobile host (MH). A
Cellular IP network consists of interconnected BSs. The BSs route IP packets inside the
cellular network and communicate with MHs via wireless interface. The GW is a cellular IP
node that is connected to a regular IP network by at least one of its interfaces. The BSs
periodically emit beacon signals. MHs use these beacon signals to locate the nearest BSs. All
IP packets transmitted by an MH are routed from the BS to the GW by hop-by-hop shortest
path routing, regardless of the destination address. The BSs maintain route cache. Packets
transmitted by the MH create and update entries in BS’s cache. An entry maps the MH’s IP
address to the neighbor from which the packet arrived to the host. The chain of cached
mappings referring to an MH constitutes a reverse path for downlink packets for the MH.
To prevent timing out of these mappings, an MH periodically transmits control packets.
MHs that are not actively transmitting or receiving data themselves may still remain
reachable by maintaining paging caches. MHs listen to the beacons transmitted by BSs and
initiate handoff based on signal strength. To perform a handoff, an MH tunes its radio to the
new BS and sends a route update packet. This creates routing cache mappings on route to the
new BS. Handoff latency is the time that elapses between the handoff and the arrival of the
first packet through the new route. The mappings associated with the old BS are cleared
after the expiry of a timer. Before the timeout, both the old and new downlink routes remain
valid and packets are delivered through both the BSs. This feature used in Cellular IP semi-
soft handoff algorithms improves handoff performance.
Fig. 5. Architecture of Cellular IP [Source: (Akyildiz et al., 2004)]
Handoff Aware Wireless Access Internet Infrastructure: Handoff-Aware Wireless Access

Internet Infrastructure (HAWAII) (Ramjee et al., 2002) is a domain-based approach for
supporting mobility. The network architecture of HAWAII is shown in Figure 6. Mobility
management within a domain is handled by a gateway called a domain root router (DRR).
Each MH is assumed to have an IP address and a home domain. While moving in its home
domain, the MN retains its IP address. The packets destined to the MH reach the DRR based
on the subnet address of the domain and are then forwarded to the MH. The paths to MH
are established dynamically. When the MH is in a foreign domain, packets for the MH are
intercepted by its HA. The HA tunnels the packets to the DRR of the MH. The DRR routes
the packets to the MH using the host-based routing entries. If the MH moves across different
subnets in the same domain, the route from the DRR to the BS serving the MN is modified,
while the other paths remain unchanged. This causes a reduction in signaling message and
handoff latency during intra-domain handoff. In traditional Mobile IP, the MH is directly
attached either to the HA (i.e. the home domain router) or the FA (i.e. the foreign domain
router). Thus, every handoff causes a change in the IP address for the MH, resulting in lack
of scalability. HAWAII also supports IP paging. It uses IP multicasting to page idle MHs
when packets destined to an MH arrive at the domain root router and no recent routing
information is available.
Fig. 6. Architecture of HAWAII Protocol [Source: (Akyildiz et al., 2004)]
Summary: Various network layer micro-mobility management schemes have been

compared based on their features (Chiussi et al., 2002; Ramjee et al., 1999; Campbell et al.,
2002). Each protocol uses the concept of domain root router. In all the protocols, signaling
traffic is largely localized in a domain so as to reduce the global signaling traffic overhead.
Routing-based schemes utilize the robustness of IP forwarding mechanism. Mobile-specific
address lookup tables are maintained by all the mobility agents within a domain. In tunnel-
based schemes, registration of the mobile nodes and encapsulation of the IP packets are
performed in a local or hierarchical manner. Routing-based schemes avoid tunneling
overhead, but suffer from the high cost of propagating host-specific routes in all routers
within the domain. Moreover, the root node in routing schemes is a potential single point of
failure (Chiussi et al., 2002). Tunnel-based schemes are modular and scalable. However, they
introduce more cost and delays (Campbell et al., 2002).
4. Link Layer Mobility Management Mechanisms

Link layer mobility management mechanisms deal with issues related to inter-system
roaming between heterogeneous access networks with different radio technologies and
network management protocols. Two important considerations for designing inter-system
roaming standards are: (i) the protocols for air interface and (ii) the mobile application part
(MAP). In situations where a mobile node enters one wireless access network from another
that support the same air interface protocols and MAP, the services are seamlessly migrated.
However, when the MAPs are different for the two networks, additional network entities
need to be placed and signaling traffic are to be transmitted for inter-working. Since each
network has its own mobility management protocols, the new inter-working entities should
not replace existing systems. Rather, the entities should coexist and inter-work.
4.1 Location management protocols

For next-generation heterogeneous wireless networks, the inter-working and inter-operating
function is suggested to accommodate roaming between dissimilar networks (Pandya et al.,
1997). For existing practical systems, several solutions are proposed for some specific pairs
of inter-working systems. In these schemes, the inter-operating function is implemented in
either some additional inter-working unit with the help of dual-mode handsets (Phillips &
Namee, 1998), or a dual-mode home location register (HLR) (Garg & Wilkes, 1996) to take
care of the transformation of signaling formats, authentication, and retrieval of user profiles.
Recent research efforts attempt to design general location management mechanisms for the
integration and inter-working of heterogeneous networks. The research activities can be
grouped into two categories: location management for adjacent dissimilar systems with
partially overlapping coverage at the boundaries (Akyildiz & Wang, 2002; Wang &
Akyildiz, 2001; ETSI, 2002) and location management in multi-tier systems where service
areas of heterogeneous networks are fully overlapped (Lin & Chlamtac, 1996). All these
solutions propose additional entities that take care of inter-working issues.
Location Management for Adjacent Networks: Researchers have addressed the issues of
location management in two adjacent networks with overlapping areas (Akyildiz & Wang,
2002; Wang & Akyildiz, 2001; ETSI, 2002). Some of the protocols are discussed briefly.
Gateway Location Register Protocol: To enable inter-system roaming, a new level has been
introduced in the hierarchy of location management entities for UMTS/ IMT-2000 networks.
The new level consists of a gateway location register (GLR) (ETSI, 2002). The GLR is a gateway
that enables inter-working between two networks by suitably converting signaling and data
formats. It is located between the visitor location register (VLR) and the serving GPRS support
node (SGSN) and the home location register (HLR). When a subscriber roams, the GLR plays
the role of the HLR toward the VLR and SGSN in a visited public land mobile network
(VPLMN), and the role of the VLR and SGSN to the HLR in a home public land mobile network
(HPLMN). The GLR protocol assists the operators in lowering costs and optimizing roaming
traffic. However, the protocol is not designed for ongoing call connection during inter-
system roaming (Wang & Akyildiz, 2001). The incoming calls are routed to the home
network even when the MN is roaming. This makes the protocol inefficient.
Boundary Location Register Protocol: In (Akyildiz & Wang, 2002), a location management
mechanism has been proposed for heterogeneous network environment. It involves a
mechanism for inter-system location updates and paging. Inter-system location update is
implemented by using the concept of a boundary location area (BLA) existing at the boundary
between two systems - X and Y in Figure 7. The BLA is controlled by a boundary interworking
unit (BIU), which is connected to the mobile switching centers (MSCs) in both the systems. The
BIU queries the user’s service information, converts the message formats, checks the
compatibility of the air interfaces and performs authentication of mobile users. When an
MN is inside its BLA, it sends a location registration request to the new system. A distance-
based location update mechanism reports MN’s location when its distance from the
boundary is less than a pre-defined threshold. An entity called a boundary location register
(BLR) is used for inter-system paging. The BLR maintains in its cache the location
information of the MN and its roaming information when it crosses an intersystem
boundary. During the inter-system paging process, only one system (X or Y) is searched.
The associated MAP protocol is designed for mobile nodes with ongoing connections during
inter-system roaming (Wang & Akyildiz, 2001). Instead of performing location registration
after a mobile node arrives at the new system, the BLR protocol enables the node to update
its location and user information actively before it enters the new system. In this way, the
incoming calls to the MN during its inter-system roaming are delivered to the node.
Fig. 7. The Boundary Location Register Protocol [Source: (Akyildiz et al. 2004)]
Location Management in Heterogeneous Networks: An MN is reachable via multiple

networks when their service areas are fully overlapped. Since heterogeneous networks use
different signaling formats, authentication procedures, and registration messages, it is
difficult to merge heterogeneous HLRs into a single HLR. A multi-tier HLR (MHLR) is
proposed in (Lin & Chalmtac, 1996), where a tier manager is connected to all the HLRs. Two
types of location registration are possible: (i) single registration (SR) and (ii) multiple
registrations (MR). Under SR scheme, an MN associates with the lowest tier of the MHLR,
and receives services at low cost and high bandwidth. Under MR method, the MN registers
on multiple tiers simultaneously. The individual tiers perform their own roaming
management. The tier manager keeps track of the currently visited high-tier and low-tier
VLRs of the MN. It has been found that MR scheme involves less signaling overhead (Lin &
Chlamtac, 1996). However, since the current tier of the MN is not known to the MHLR, it
incurs a high loss when a wrong tier is selected.
Summary: To summarize, all link layer-based mobility management schemes require
additional inter-working entities for enabling information exchange between different
systems. These inter-working entities are different depending on the systems, e.g., the
GLR/BLR for inter-system location management, the MHLR for a multi-tier PCS system,
and the gateways in the integrated UMTS/WLAN system. The interworking entities
perform the following functions: (i) format translation of the signaling messages and data
packets and address translation between networks, (ii) retrieval of user profile from the
home network, (iii) acting as a gateway for signal transmission and route setup, (iv)
recording of mobility-related information during inter-system roaming, (v) negotiating QoS
when an MN enters a new network, and (vi) performing authentication during inter-system
movement. Different approaches for mobility management at the link layer address the
following issues: (i) the location where the inter-working entities are put, (ii) the degree of
coupling (loose or tight) of the entities, (iii) the timing of location registration and handoff
initiation, (iv) the way location and handoff management is performed.
5. Handoff Management Protocols

Handoff or handover is a process by which an MN moves from one point of network
attachment to another. Handovers can be classified as either homogeneous or
heterogeneous. A heterogeneous handover occurs when an MN either moves between
networks with different access technologies, or between different domains. As the diversity
of available networks increases, it is important that mobility technologies become agnostic
to link layer technologies, and can operate in an optimized and secure fashion without
incurring unreasonable delay and complexity (Dutta et al., 2008). Supporting handovers
across heterogeneous access networks, such as IEEE 802.11 (Wi-Fi), global system for mobile
communications (GSM), code-division multiple access (CDMA), and worldwide
interoperability for microwave access (WiMAX) is a challenge, as each has different quality
of service (QoS), security, and bandwidth characteristics. Similarly, movement between
different administrative domains poses a challenge since MNs need to perform access
authentication and authorization in the new domain. Thus, it is desirable to devise a
mobility optimization technique that can reduce these delays and is not tightly coupled to a
specific mobility protocol. In this section, we describe different types of handovers and
investigate the components that contribute to a handover delay. Some inter-technology and
media-independent handover frameworks are then described.
5.1 Taxonomy of handoff mechanisms

Different types of handovers may be classified based on three parameters as follows:
(i) subnets, (ii) administrative domains, and (iii) access technologies (Dutta et al., 2008).
Inter-technology: this type of handover is possible with an MN that is equipped with multiple
interfaces supporting different technologies. An inter-technology handover occurs when the
two points of attachment use different access technologies. During the handoff, the MN may
move out of the range of one network (e.g., Wi-Fi) into that of a different one (e.g., CDMA).
This is also known as vertical handover.
Intra-technology: this type of handoff occurs when an MN moves between points of
attachments supporting the same access technology, such as between two Wi-Fi access
points. An intra-technology handover may happen due to intra-subnet or inter-subnet
movement and thus may involve the layer 3 trigger.
Inter-domain: when the points of attachment of an MN belong to different domains, this type
of handoff takes place. A domain is defined as a set of network resources managed by a
single administrative entity that authenticates and authorizes access for the MNs. An
administrative entity may be a service provider or an enterprise. An inter-domain handover

possibly involves an inter-subnet handover also.
Intra-domain: handovers of this type occurs when the movement of an MN is confined
within an administrative domain. Intra-domain movement may also involve intra-subnet,
inter-subnet, intra-technology, and/or inter-technology handovers as well.
Inter-subnet: an inter-subnet handover occurs when the two points of attachment belong to
different subnets. The MN acquires a new IP address and possibly undergoes a new security
procedure. A handover of this type may occur along with either an inter- or an intra-domain
handover and also with either an inter- or an intra-technology handover.
Intra-subnet: an intra-subnet handover occurs when the two points of attachment belong to
the same subnet. This is typically a link layer handover between two access points in a
WLAN networks, or between different cell sectors in cellular networks. It is administered by
the radio network and requires no additional authentication and security procedures.
5.2 Delays in handoff

All the layers in the communication protocol stack contribute to the delay in a handoff.
Link layer delay: depending on the access technology, an MN may go through several steps
with each step adding its contribution to the overall delay before a new link is established.
For example, a Wi-Fi link goes through the process of scanning, authentication, and
association before being attached to a new access point. For intra-subnet handovers, where
network layer configurations are necessary, link layer contributes the maximum to the
overall delay.
Network layer delay: after completion of the link layer procedures, it may be necessary to
initiate a network layer transition. A network layer transition may involve steps such as:
acquiring a new IP address, detecting a duplicate address, address resolution protocol
(ARP) update, and subnet-level authentication.
Application layer delay: the delay of this type is due to reestablishment and modification of
the application layer properties such as IP address while using session initiation protocol (SIP).
The authentication and authorization procedure such as extensible authentication protocol
(EAP) includes several round-trip messages between the MN and the authentication
authorization and accounting (AAA) server causing delay in handoff.
5.3 Research work on handoff mechanisms

This section presents some of the existing handoff mechanisms proposed in the literature.
In (Hasswa et al., 2005), a vertical handoff decision function is proposed for roaming across
heterogeneous wireless networks. An optimization scheme for vertical hand off has been
proposed in (Zhu & McNair, 2004). In (Park et al., 2003), a seamless vertical handoff scheme
is proposed between a WLAN and a CDMA 2000-based cellular network. A vertical handoff
scheme between a UMTS and a WLAN network is proposed in (Zhang et al., 2003). A
connection manager detects the changes in wireless networks and makes the handoff decision.
When the MN moves from the UMTS to the WLAN network, the objective of the handoff is
to have better QoS because of the higher bandwidth of WLAN. However, in case of handoff
from WLAN to UMTS, the handoff is initiated just before the connection to WLAN breaks.
In (Efthymiou et al., 1998), a protocol for inter-segment handover (ISHO) is proposed in an
integrated space/terrestrial UMTS environment. A backward mobile-assisted handover
incorporating signalling diversity is chosen as the most appropriate handover scheme.

Based on the generic radio-access network (GRAN) concept and by using a satellite-UMTS
network architecture and functional model, the derivation of an ISHO protocol is presented.
In (McNair et al., 2000), a handoff technique is introduced that supports mobility between
networks with different handover protocols. Three types of handoffs are presented: (i)
network-controlled handoff (NCHO), (ii) mobile-assisted handoff (MAHO), and (iii) mobile-
controlled handoff (MCHO). Under NCHO or MAHO, the network generates a new
connection, finds new resources for the handoff and performs any additional routing
operations. For MCHO, the MN finds the new resources and the networks approves.
In (Stemm & Katz, 1998), a vertical handoff scheme is designed for wireless overlay
networks, where heterogeneous networks in a hierarchical structure have fully overlapping
service areas. The BSs send out periodic beacons similar to Mobile IP FA advertisements.
The MN listens to these packets and decides which BS would forward packets, which BS
should buffer packets for a handoff, and which BS should belong to the multicast group.
In (Buddhikot et al., 2003), the issues of integration of WLAN and 3G networks have been
addressed to offer seamless connectivity. Two approaches have been identified: (i) a tightly-
coupled approach and (ii) a loosely-coupled approach. In the tightly-coupled approach, the
gateway of 802.11 network appears to the upstream 3G core as either a packet control function
(PCF), in case of a CDMA2000 core network, or as a serving GPRS service node (SGSN), in
case of a UMTS network. The 802.11 gateway hides the details of the 802.11 network to the
3G core, and implements all the protocols required in a 3G access network. In the loosely-
coupled scheme, the same 802.11 gateway is used. However, the gateway connects to the
Internet and does not have any direct link to the 3G network elements such as packet data
service nodes (PDSNs), gateway GPRS service nodes (GGSNs) or 3G core network switches. In
this case, the data paths in 802.11 and 3G networks are different. The high speed 802.11
traffic is never injected into the 3G network but the end user still achieves seamless access.
5.4 Cross-layer handoff mechanisms

The cross-layer protocols for mobility management are mainly applied for handoff. Most of
these mechanisms use link layer information to make an efficient network layer handoff.
The utilization of link layer information reduces the delay in movement detection of the MN
so that the overall handoff delay is minimized.
In (Yokota et al., 2002), a low-latency handoff algorithm for a WLAN has been proposed that
uses access points and a dedicated medium access control (MAC) bridge. A seamless handoff
architecture for Mobile IP, called S-MIP is presented in (Hsieh et al., 2003) that combines a
location tracking scheme with the HMIP handoff. A vertical handoff mechanism between
IEEE 802.11 (WLAN) and IEEE 802.16e (Mobile WiMAX) networks in a wireless mesh
backbone is proposed in (Zhang, 2008). In (Dutta et al., 2008), a media-independent pre-
authentication scheme has been proposed. These four handoff schemes are discussed below.
Link Layer-Assisted Fast Handoff over WLAN: In the Mobile IP protocol, the MN
movement can be detected from advertisements of the FAs that differ from the previously
received advertisement. The new CoA is registered with the HA. However, data packets are
not forwarded to the new FA before the registration is complete. This interruption may
degrade the QoS especially in real-time applications. To tackle this issue, a handoff
mechanism is proposed in which APs in a WLAN and a dedicated MAC bridge are jointly
used to eliminate packet loss (Yokota et al., 2002). The authors have noted that the delay in
Mobile IP handoff is contributed by two elements: (i) the delay in movement detection of the
MN, and (ii) delay due to signaling for registration. The proposed mechanism reduces the
movement detection delay. It has two parts: (i) handoff for the forward direction (i.e.
mobile-terminated data) and (ii) handoff for the reverse direction (i.e. mobile-originated
data). The APs in the WLAN have the capability to notify the MAC address of an MN that
moves into their coverage areas. The MAC bridge is configured in a way that it sends only
those MAC frames whose destination addresses are registered in the filtering database (DB).
Fig. 8. Handoff Scenario in Forward Direction [Source: (Yokota et al., 2002)]
The handoff in the forward direction happens as follows. In Figure 8, the MN establishes an
association with an access point- AP1, and registers the CoA with HA. The packets destined
to the MN are encapsulated by the HA and tunneled to FA1- the FA of the MN. FA1
decapsulates the packets and sends them directly to the MN. When the signal strength of the
channel of communication between AP1 and the MN falls below a threshold, MN attempts
to find a new AP. The MN establishes association with a new AP- AP2. AP2 places the MAC
address of the MN in a MAC address registration request message and broadcasts it on the local
segment. The MAC bridge receives the address registration request. It then makes an entry
of the MAC address contained in the message and the port on which the message was
received into the filtering DB. When the MAC bridge receives a MAC frame on a port, it
refers to the filtering DB to see if the destination MAC address is registered. If the address is
registered, the MAC bridge sends it out to the corresponding port. Packets from FA1 are
thus bridged from port A to the port B of the MAC bridge, and delivered to the Network 2,
to which the MN is now connected. The MN detects its movement as it receives new agent
advertisements from FA2 and registers the new CoA with the HA. When the registration is
complete, packets destined for the MN are tunneled to FA2 and delivered to the MN. Since
no packets are bridged from that time onward, the entry for the MN in the filtering DB must
be removed upon expiration of its aging time. Thus, the MN receives packets even before
Mobile IP registration is over.
If the MAC bridge relays only those frames whose source MAC addresses are registered in
the filtering DB to the network to which MN was previously attached, then it can reduce
transmission interruption in the reverse direction as well. However, the transmission
interruption in the reverse direction is possible if the MAC bridge has only two ports. The
MAC bridge with two ports checks the source MAC address of an incoming frame from one
port with the filtering DB, and transfers it to the other port. However, if the MAC bridge has
more than two ports, the direction in which the frame should be transferred will depend on
the speed of the MN and how fast the Mobile IP registration process completes. By taking
into account that the next hop of a frame sent by the MN is always the default router of the
network where the MN has been registered, the authors have proposed a fast handoff
method in the reverse direction by registering the MAC address of the default router in the
filtering DB. The algorithm exploits the Mobile IP agent advertisement message which are
periodically broadcasted by the FAs and received by the MN. The scheme has been
evaluated in an actual network environment to measure the time required for forward and
reverse handoffs on UDP and TCP traffic. The latency due to Mobile IP handoff has been
found to be equal to that of a link layer handoff (Yokota et al., 2002).
Seamless Handoff Architecture for Mobile IP: Seamless Handoff Architecture for Mobile
IP (S-MIP) is an architecture which minimizes the handoff latency in a large indoor
environment (Hsieh et al., 2003). The architecture of S-MIP is depicted in Figure 9. It is an
extension of the HMIP architecture with an additional entity called decision engine (DE). The
DE is identical to MAP in HMIP, and makes the handoff decision for its network domain.
The MAP separates the mobility type into micro-mobility and macro-mobility. The new
access router (nAR) and the old access router (oAR) retain the same functionality and meaning
as in HMIP. Through periodic feedback information from the ARs, the DE maintains a
global view of the connection state of any MN in its network domain. DE also tracks the
movement patterns of all MNs in its domain using the signal strength information received
from the link layer and the IDs of the ARs.
Fig. 9. Architecture of the S-MIP Scheme [Source: (Hsieh et al., 2003)]

In HMIP and fast handoff mechanisms, the packet loss occurs either within the MAP and
the ARs (segment packet loss), or between the last ARs and the MN (edge packet loss).
While edge packet losses occur due to the mobility of an MN and transmission errors, the
segment-packet loss is due to the non-deterministic nature of handoffs and the resulting
switching of the data stream at the MAP after the receipt of the MAP binding update. The
design of S-MIP minimizes the edge packet and segment packet losses. Edge packet loss is
minimized by keeping the anchor point for the forwarding mechanism as close to the MN as
possible. Hence it is located at the AR that bridges the wireless network and the wired
network. Segment packet loss is minimized by using a newly developed synchronized packet
simulcast (SPS) scheme and a hybrid handoff mechanism. The SPS simulcasts packets to the
current network where the MN is attached to and to the potential access network that the
MN is asked to switch onto. The hybrid handoff strategy is MN-initiated, but network
determined. The decision as to which access network to handoff is formulated from the
movement tracking mechanism which is based on a synchronized feedback. The authors
have provided a combination of simulation results and mathematical analysis to argue that
S-MIP is capable of providing zero-packet loss handoff with latency similar to that of a link
layer delay in a WLAN environment.
A Vertical Handoff Scheme between WLAN and Mobile WiMAX Networks: In (Zhang,
2008), a vertical handoff scheme has been proposed between 802.11(WLAN) and 802.16e
(Mobile WiMAX) networks. The framework has been discussed with a wireless mesh network
(WMN) that provides high speed, scalable and ubiquitous wireless Internet services. A
wireless mesh router (WMR) is a gateway that has routing capabilities to support mesh
networking. Each WMR is assumed to have 802.11e functions, 802.16e BS functions with
point-to-multi-point mode (PMP), routing capabilities, and 802.16e subscriber station (SS)
functions with mesh mode. The MNs can connect only via mesh routers to access the
Internet using two types of links: the IEEE 802.11e and IEEE 802.16e links. The IEEE 802.16e
links between MNs and mesh routers operate in the PMP mode, while the IEEE 802.16e
links among neighboring mesh routers operate in the mesh mode. Figure 10 illustrates the
system. The links between the WMRs are 802.16e mesh links. The WMR which is connected
to the Internet with wired line is called mesh gateway (MGW). The MNs with dual network
interfaces can connect to the Internet through the WMRs by an 802.11e link or 802.16e link.
The WMRs which are connected directly or indirectly with one MGW form a domain or
subnet. The MNs connect to the WMRs using 802.11e link for high data rate and small
coverage area and 802.16e links for higher data rate and large coverage.
An MN initially sets up a connection with a WMR. The WMR forwards the IP packets from
the MN to the MGW through one or more WMRs. The MGW transmits the IP packets to the
CN in the Internet. IP packets from the CN are routed through the reverse route to the CN.
The CN may be located in the same domain as the MN. In this case, the WMRs forward the
IP packets for them. While an MN is inside the area doubly covered by the WLAN and
WiMAX, a proper vertical handoff is needed if the WLAN network is congested or if the
MN is roaming across the edge of the WLAN coverage. The author has proposed a vertical
handoff scheme for this scenario. The algorithm has four steps: (i) new network interface
scanning, (ii) new access router discovery, (iii) new network entry, and (iv) routing
information updating. After completion of these stages, the MN can transmit or receive
information data packets through the new network interface.
Fig. 10. Architecture of the Wireless Mesh Network [Source: (Zhang, 2008)]
In Figure 11, two domains are served by service providers A and B respectively. The WMRs
in the same and different domains are called intra-mesh routers and inter-mesh routers
respectively. If the CN is in the same domain as the MN, the IP packets are routed through
the intra-mesh routers only. When the MN moves to another domain, the packets from the
CN are routed via the HA. Four scenarios are considered for MN mobility.
Fig. 11. Macro-Mobility and Micro-Mobility Scenario [Source: (Zhang, 2008)]

Scenario 1: the MN is connected to the WLAN. It moves out of WLAN and connects to the
WiMAX. The movements 1a, 1b, 1c depict this situation. The WMR does not change, only
the medium access interface changes in case of 1a. The handoff occurs between intra-mesh
routers in 1c and between inter-mesh routers in case 1b.
Scenario 2: the MN is currently connected to the WiMAX. It moves into the WLAN and
either connects to the WLAN or continues with the WiMAX connection depending on the
network conditions, user preference, or application QoS requirements.
Scenario 3: the MN is located in the double-coverage area (i.e. area covered by WLAN and
WiMAX) and is currently stationary. If the WLAN is congested, the MN can switch to the
WiMAX if it can provide more bandwidth for the MN to transmit its data packets.
Scenario 4: A horizontal handoff occurs when the MN moves in 2a and 2b. In (Kim et al.,
2005), a scheme called last packet marking (LPM) has been proposed for case 2a. The
MIPSHOP (Mobility for IP: Performance, Signaling and Handoff Optimization) working
group of the Internet Engineering Task Force (IETF) has developed Mobile IPv6 fast handoff
over 802.16e networks for case 2b (Jang et al., 2008).
Media Independent Pre-Authentication for Secure Inter-Domain Handover: A media-
independent pre-authentication (MPA) scheme has been proposed in (Dutta et al., 2008). It is a
mobile-assisted, secure handover optimization scheme that works over any link layer and
with any mobility management protocol. With MPA, an MN securely obtains an IP address
and other configuration parameters for a candidate target network (CTN) - the network to
which the mobile node is being handed off. The MN is also able to send and receive IP
packets using the IP address before it attaches to the CTN. In this way, the MN completes
the binding update and use the new CoA before performing a handover at the link layer.
MPA provides four basic procedures that optimize handover for an MN. The serving
network is the network that currently serves the MN. The first procedure - pre-authentication
establishes a security association with the CTN to secure subsequent protocol signaling. The
second procedure - pre-configuration securely executes a configuration protocol to obtain an
IP address and other parameters from the CTN. The third procedure executes a tunnel
management protocol that establishes a proactive handover tunnel (PHT) between the MN and
an access router in the CTN over which binding updates as well as data packets, can travel.
Finally, the fourth procedure deletes the PHT before attaching to the CTN and reassigns the
inner address of the deleted tunnel to its physical interface after the MN attaches to the
target network. The final two procedures are collectively referred to as secure proactive
handover. Through the third procedure the MN completes higher-layer handover before
starting link layer handover. This means that the MN is able to perform all the higher-layer
configuration and authentication procedures before link layer connectivity to the CTN is
established. This can significantly reduce the handover delays.
Summary: As a macro-mobility management protocol, Mobile IP is simple, but it has several
shortcomings such as triangular routing, high-global signaling load, and high handoff
latency. Although, the route optimization mechanism eliminates triangular routing, the high
handoff latency still remains. The micro-mobility management mechanisms are not suitable
for inter-domain mobility. Most of these solutions assume one domain to be one wireless
access network or under one administrative domain. Although IDMP (Misra et al., 2002)
defines a domain based on geographic proximity where one domain consists of networks
with different access technologies in a particular geographic region, there is no procedure
specified for inter-system authentication, format transformation, and so on. In a
heterogeneous environment where users have freedom to move between different domains,
the global signaling load and corresponding handoff delay will increase significantly,
adversely affecting the network performance. The S-MIP approach (Hsieh et al., 2003)
demonstrates that along with the hierarchical architecture and procedures for fast handoff,
the link layer information used to determine the mobility pattern of the MHs can greatly
improve intra-domain handoff performance. However, the protocol cannot be extended to
support mobility between different domains, because the coverage area of one domain
might be completely covered by another domain in the hierarchical heterogeneous
environment; for example, a WLAN domain is mostly covered completely by the overlaying
2G/3G network.
6. IEEE 802.21- Media Independent Handover Services

A novel solution that ensures interoperability between several types of wireless access
network is given by the developing IEEE 802.21 standard (Eastwood et al., 2008). The work
on the standard began in 2004 and is expected to be finalized around 2010. The IEEE 802.21
is focused on handover facilitation between different wireless networks in heterogeneous
environments. The standard names this type of vertical handover as Media Independent
Handover (MIH). In MIH, the handover procedures can use the information gathered from
both the mobile terminals and the network infrastructure. At the same time, several factors
may determine the handover decision, e.g., service continuity, application class and QoS,
negotiation of QoS, security, power management, handover policy etc. IEEE 802.21
facilitates, speeds, and thereby increases the success rate of inter-technology handover
decision making and other pre-execution processes. These processes include inter-
technology candidate network discovery, target network selection, target network
preparation, and handover execution timing and initiation. IEEE 802.21 defines three
services to facilitate inter-technology handovers: (i) media independent information service
(MIIS), (ii) media independent command service (MICS), and (iii) media independent event service
(MIES). MIIS provides information about the neighboring networks, their capabilities and
available services. MICS allows effective management and control of different link interfaces
on multimodal device and enables both mobile- and network-initiated handovers. It
supports querying of target networks about the status of the rapidly changing resources.
Some MICS commands are part of the signaling between inter-radio access technology (RAT)
gateways. MIES provides events triggered by changes in the link characteristics and status.
This interface provides service primitives to the upper layers that are independent of the
access technology.
One of the most important aspects of MIH is the fact that it allows for network controlled
handovers and user controlled handovers. The advantages of the network controlled handover
lies in the lower user battery consumption since the monitoring of various network
conditions is done by the networks themselves. However, it incurs a huge signaling
overhead and a high processing load in the network elements. In user controlled handover,
the user collects necessary data and initiates the appropriate actions. The disadvantage of
this approach is the high battery power consumption.
6.1 Mobility using IEEE 802.21 in a heterogeneous IMT-advanced (4G) network

The telecommunication industry is defining a new generation of mobile wireless
technologies, called fourth generation (4G). In this regard, the International
Telecommunications Union- Radio Standardization Sector (ITU-R) has defined the concept
of IMT-Advanced that targets peak data rates of about 100 Mb/s for highly mobile access (at
speeds of up to 250 km/hr), and 1 Gb/s for low mobility (pedestrian speeds or fixed) access.
The IEEE is developing extensions to both IEEE 802.11 and 802.16 to meet IMT- Advanced
requirements. The evolving standard of IEEE 802.16m aims to achieve a data rate of 100
Mb/s in a highly mobile (25 km/hr) scenario. These data rate and mobility capabilities
make 802.16m a candidate for the high mobility portion of the IMT-Advanced standard
requirements. Another working group of IEEE 802.11n is working towards designing a very
high throughput (VHT) radio capable of data rates up to 1 Gb/s at stationary or pedestrian
speeds. Together, 802.16m and 802.11n will satisfy both the low-mobility and fully mobile
user velocity vs. data rate requirements for IMT-Advanced systems. If IEEE proposes a
combination of 802.11m and 802.11n for IMT-Advanced standard, an interworking
mechanism must be designed for tying up these two systems. In (Eastwood et al., 2008), the
authors have proposed a mobility management approach in 4G using IEEE 802.21 Media
Independent Handover (MIH) services.
7. Security in Handoff Procedures

Whenever an MN connects to a point of network access, it establishes a security context with
the service provider. During the handover process, some or all the network entities involved
in the security mechanism may change. Thus the current security context changes as well.
The MN and the network have to ensure that they still communicate with each other and
they agree upon the keys to protect their communication.
However, during handovers in networks like GSM/GPRS and UMTS no authentication is
used. This makes the handover procedures vulnerable to a hijacking attack. An attacker can
masquerade as an authentic mobile station (MS) just by sending message at the right
frequency and time slot during handover. As long as the attacker does not know the
encryption and/or integrity keys currently being used, he cannot insert valid traffic into the
channel. However, if an attacker can gain access to the key(s) (e.g. because of a missing
protection on the backbone network), he can impersonate the MS. In fact, in GSM/GPRS,
UMTS and WLAN networks, no standard protection mechanism in the backbone network
has been specified. Many GSM operators do not protect the radio link between their fixed
networks and the BSs. In UMTS, during a handover, the keys used to protect the traffic
between the MS and the previous BS are reused in communication with the next BS. While
the keys are being transmitted, they can be intercepted by an adversary, if the wireless link
is not protected.
Usually an authentication process happens before location updates and call setups. The
same mechanisms cannot however, be applied in establishing connection during a handover
process because of the stringent time constraint. In GSM, for example, the time between the
handover command and the handover complete or handover failure message is restricted to
0.5- 1.5 s. The generation of an authentication response, however, takes about 0.5 s at the MS
side. Thus an authentication overhead will cause connection disruption.
As we have seen earlier in this chapter, efficient cell prediction mechanisms can reduce the
signaling overhead between the MS and the old BS. The free time slots may be used to
forward authentication traffic between the MS, the old BS and the new BS. The MS can pre-
compute an authentication challenge and the encryption and integrity protection keys
before the actual change of channel. When the MS and the new BS establish connection, the
MS sends the pre-computed authentication response for the new BS to check. If the checking
yields positive results, a handover complete message is sent and the old BS releases its
resources. Otherwise, a handover failure happens and the MS falls back to the old channel.
8. Some Open Issues in Mobility and Handover Management

Future wireless networks will be based on all-IP framework and heterogeneous access
technologies. Design of efficient mobility management mechanisms will be playing ever
important role in providing seamless services. Following issues will play dominant roles.
QoS issues – next-generation all-IP wireless networks will have to provide guaranteed QoS
to mobile terminals. QoS provisioning in a heterogeneous wireless and mobile networks will
bring in new problems to mobility management, such as location management for efficient
access and timely service delivery, QoS negotiation during inter-system handoff, etc.
User terminals – the design of a single user terminal that is able to autonomously operate in
different heterogeneous access networks will be another important factor. This terminal will
have to exploit various surrounding information (e.g., communication with localization
systems, cross-layering with network entities etc.) in order to provide richer user services
(e.g. location/situation/context–aware multimedia services). This will also put strong
emphasis on the concept of cognitive radio and cognitive algorithms for terminal re-
configurability.
Location and handoff management in wireless overlay networks – future wireless
networks will be inherently hierarchical where access networks have different coverage
areas. Mobility management in wireless overlay networks will be a very important issue.
Mobile services – sophisticated 4G service discovery mechanisms will combine the
location/situation information and context-awareness in order to deliver users’ services in a
best possible manner. Additionally, future mobile services will require more complex
personal and session mobility management to provision personalized services through
different personalized operating environments to a single user terminal address. Whether
SIP should be the core 4G protocol, and whether the service delivering framework be the
network layer-based or application layer-based is still an open question.
Cross-Layer optimization – design of efficient cross-layer-based approaches will play a key
role is developing new mobility management schemes.
Other issues – fault-tolerance, availability of network services, enhanced security, intelligent
packet and call routing, intelligent gateway discovery and selection procedures and design
of a unified protocol stack and vertical protocol integration mechanisms are some of the
other important issues in next-generation heterogeneous networks.
9. Conclusion
In this chapter, a comprehensive discussion has been made on mobility management in
next-generation wireless networks. Issues in location registration and handoff management
have been identified and several existing mechanisms have been presented. Since global
roaming will be an increasing trend in future, attention has been paid on mechanisms which
are applicable in heterogeneous networks. Media Independent Handover Services of IEEE
802.21 standard as an enabler for handover has also been presented. Security and
authentication issues in next-generation heterogeneous networks are discussed briefly.
Finally, the chapter concludes by highlighting some open areas of research in mobility
management.
10. References
Akyildiz, I.F., McNair, J., Ho, J.S.M., Uzunalioglu, H., and Wang, W. (1999). Mobility
management in next-generation wireless systems, Proceedings of the IEEE, Vol 87,
No 8, Aug 1999, pp. 1347-1384.
Akyildiz, I.F., and Wang, W. (2002). A dynamic location management scheme for next-
generation multitier PCS systems. IEEE Transactions in Wireless Communications, Vol
1, No 1, Jan 2002, pp. 178-89.
Akyildiz, I.F., Altunbasak, Y., and Sivakumar, R. (2004a). AdaptNet: an adaptive protocol
suite for the next-generation wireless Internet. IEEE Communications Magazine, Vol
42, No 3, Mar 2004, pp. 128-136.
Akyildiz, I.F., Xie, J., and Mohanty, S. (2004b). A survey of mobility management in next-
generation all-IP-based wireless systems. IEEE Wireless Communications Magazine,
Vol 11, No 4, Aug 2004, pp. 16-28.
Akyildiz, I.F., Mohanty, S., and Xie, J. (2005). A ubiquitous mobile communication
architecture for next-generation heterogeneous wireless systems. IEEE
Communications Magazine, Vol 43, No 6, Jun 2005, pp. 29-36.
Buddhikot, M., Chandranmenon, G., Han, S., Lee, Y.W., Miller, S., and Salgarelli, L. (2003).
Integration of 802.11 and Third-Generation Wireless Data Networks, Proceedings of
IEEE INFOCOM, Vol 1, Mar-Apr 2003, pp. 503-512.
Campbell, A.T., Gomez, J., Kim, S., Valko, A.G., Chieh-Yih Wan, and Turanyi, Z.R. (2000).
Design, implementation, and evaluation of cellular IP. IEEE Personal
Communications, Vol 7, No 4, Aug 2000, pp. 42-49.
Campbell, A.T. and Gomez, J. (2001). IP micro-mobility protocols. ACM SIGMOBILE Mobile
Computing and Communications Review, Vol 4, No 4, Oct 2001, pp. 45-54.
Campbell, A.T., Gomez, J., Kim, S., Chieh-Yih Wan, Turanyi, Z.R., and Valko, A.G. (2002).
Comparison of IP micromobility protocols. IEEE Wireless Communications, Vol 9, No
1, Feb 2002, pp. 72-82.
Chiussi, F.M., Khotimsky, D.A., and Krishnan, S. (2002). Mobility management in third-
generation all-IP networks. IEEE Communications Magazine, Vol 40, No 9, Sep 2000,
pp. 124-135.
Dutta, A., Famolari, D., Das, S., Ohba, Y., Fajardo, V., Taniuchi, K., Lopez, R., and
Schulzrinne, H. (2008). Media-independent pre-authentication supporting secure
interdomain handover optimization. IEEE Wireless Communications, Vol 15, No 2,
April 2008, pp. 55-64.
Eastwood, L., Migaldi, S., Xie, Q., and Gupta, V. (2008). Mobility using IEEE 802.21 in a
heterogeneous IEEE 802.16/802.11-based IMT-advanced (4G) network. IEEE
Wireless Communications, Vol 15, No 2, Apr 2008, pp. 26-34.
Efthymiou, N., Hu, Y.F., Sheriff, R. (1998). Performance of intersegment handover protocols
in an integrated space/terrestrial-UMTS environment. IEEE Transactions on
Vehicular Technology, Vol 47, No 4, Nov 1998, pp. 1179-1199.
ETSI TS 129 120 V4.0.0 (2002). Universal Mobile Telecommunications Systems (UMTS) ;
Mobile Application Part (MAP) Specification for Gateway Location Register (GLR) ;
Stage 3, 3GPP/ETSI 2002.
Fogelstroem, E., Jonsson, A., and Perkins, C. (2006). Mobile IPv4 regional registration.
Internet draft, IETF, draft-ietf-mip4-reg-tunnel-04.txt, October 2006.
Garg, V.K., and Wilkes, J.E. (1996). Interworking and interoperability issues for North
American PCS. IEEE Communications Magazine, Vol 34, No 3, Mar 1996, pp. 94-99.
Hasswa, A., Nasser, N., and Hassanein, H. (2005). Generic vertical handoff decision function
for heterogeneous wireless networks, Proceedings of IEEE and IFIP International
Conference on Wireless and Optical Communications Networks (WOCN), March 2005,
pp. 239-243.
Haverinen, H., and Malinen, J. (2000). Mobile IP regional paging. Internet draft, IETF, draft-
haverinen-mobileip-reg-paging-00.txt, June 2000.
Hsieh, R., Zhou, Z.G., and Seneviratne, A. (2003). S-MIP: a seamless handoff architecture for
Mobile IP, Proceedings of IEEE INFOCOM, Vol 3, Mar-Apr, 2003, pp. 1774-1784.
Jang, H. J., Jee, J., Han, Y.H., Park, S.D., and Cha, J. (2008). Mobile IPv6 fast handovers over
IEEE 802.16e networks. IETF Internet draft work in progress, June 2008.
Kim, K., Kim, C., and Kim, T. (2005). A seamless handover mechanism for IEEE 802.16e
broadband access, Proceedings of International Conference on Computational Science
(ICCS), Vol 3515, May 2005, pp. 527-534.
Lin, Y-B., and Chlamtac, I. (1996). Heterogeneous personal communications services:
integration of PCs systems. IEEE Communications Magazine, Vol 34, No 9, September
1996, pp. 106-113.
McNair, J., Akyildiz, I.F., and Bender, M.D. (2000). An inter-system handoff technique for
the IMT-2000 system, Proceedings of IEEE INFOCOM, Vol 1, 2000, pp. 208-216.
Misra, A., Das, S., Dutta, A., and McAuley, A., and Das, S.K. (2002). IDMP-based fast
handoffs and paging in IP-based 4G mobile networks. IEEE Communications
Magazine, Vol 4, No 3, March 2002, pp. 138-145.
Mohanty, S. (2006). A new architecture for 3G and WLAN integration and inter-system
handover management. Wireless Networks, Vol 12, No 6, November 2006, pp. 733-
745.
Pandya, R., Grillo, D., Lycksell, E., Mieybegue, P., Okinaka, H., and Yabusaki, M. IMT-2000
standards: network aspects. IEEE Personal Communications, Vol 4, No 4, August
1997, pp. 20-29.
Park, H., Yoon, S., Kim, T., Park, J., Do, M., and Lee, J. (2003). Vertical handoff procedure
and algorithm between IEEE 802.11 WLAN and CDMA cellular network. Lecture
Notes in Computer Science (LNCS), No 2524, 2003, pp. 103-112.
Perkins, C.E., and Johnson, D.B. (2001). Route optimization in Mobile IP. Internet draft,
IETF, draft-ietf-mobileip-optim-11.txt, September 2001.
Perkins, C.E. (2008). IP mobility support for IPv4, IETF RFC 3344, draft-ietf-mip4-rfc3344bis-
07, October 2008.
Phillips J., and Namee, G.M. (1998). Personal wireless communication with DECT and PWT,
Artech House (Boston), 1998, ISBN: 0890068720.
Politis, C., Chew, K.A., and Tafazolli, R. (2003). Multilayer mobility management for all-IP
networks: pure SIP vs. hybrid SIP/mobile IP, Proceedings of IEEE Vehicular
Technology Conference (VTC), April 2003, Vol 4, pp. 2500-2504.
Ramjee, R., La Porta, T., Thuel, S., Vardhan, K., and Wang, S.Y. (1999). HAWAII: a domain-
based approach for supporting mobility in wide-area wireless networks,
Proceedings of ICNP, Oct-Nov 1999, pp. 283-292.
Salsano, S., Polidoro, A., Mingardi, C., Niccolini, S., and Veltri, L. (2008). SIP-based mobility
management in next-generation networks. IEEE Wireless Communications, Vol 15,
No 2, April 2008, pp. 92-99.
Stemm, M., and Katz, R.H. (1998). Vertical handoffs in wireless overlay networks. ACM
Mobile Networks and Applications, Vol 3, No 4, 1998, pp. 335-350.
Soliman, H., Castelluccia, C., ElMalki, K., and Bellier, L. (2008). Hierarchical Mobile IPv6
mobility management (HMIPv6). IETF draft-ietf-mipshop-4140bis-05, October 2008.
Wang, W., and Akyildiz, I.F. (2001). A new signaling protocol for intersystem roaming in
next-generation wireless systems. IEEE Journal on Selected Areas in Communications,
Vol 19, No 10, October 2001, pp. 2040-2052.
Yokota, H., Idoue, A., Hasegawa, T., and Kato, T. (2002). Link layer assisted Mobile IP fast
handoff method over wireless LAN networks, Proceedings of ACM MOBICOM,
2002, pp. 131-139.
Zhang, Q., Guo, C., Guo, Z., and Zhu, W. (2003). Efficient mobility management for vertical
handoff between WWAN and WLAN. IEEE Communications Magazine, Vol 41, No
11, November 2003, pp. 102-108.
Zhang, Y. (2008). Vertical handoff between 802.11 and 802.16 wireless access networks. Master
Thesis, Department of Electrical and Computer Engineering, University of
Waterloo, Ontario, Canada.
Zhu, F., and McNair, J. (2004). Optimizations for vertical handoff decision algorithms,
Proceedings of IEEE WCNC, Vol 1, March 2004, pp. 867-872.
GPS Total Electron Content (TEC) Prediction at Ionosphere Layer over the Equatorial Region 485
23
X
GPS Total Electron Content (TEC) Prediction at

Ionosphere Layer over the Equatorial Region
Norsuzila Ya’acob1, Mardina Abdullah1,2 and Mahamod Ismail1,2
1Department of Electrical, Electronic and Systems Eng, Universiti Kebangsaan Malaysia
2Institute of Space Science, Universiti Kebangsaan Malaysia
Malaysia
1. Introduction
Space weather is a fairly new field in science today and has very interesting effects on
humans, environment and technology in general. Scientists are now studying space weather
with a wide range of tools to try to learn more about the physical and chemical processes
taking place in the upper atmosphere and beyond. One of these tools is Global Positioning
System (GPS). GPS is currently one of the most popular global satellite positioning systems
due to global availability of signal as well as performance. GPS is a satellite-based
navigation radio system which is used to verify the position and time in space and on the
Earth. GPS nowadays allows to measure positions in real time with an accuracy of few
centimetres (Warnant et al., 2007). The advent of GPS has led to technical revolutions in
navigation as well as in fields related to surveying. The GPS system - an all-weather
satellite-based radio navigation system - can provide users on a world-wide basis with
navigation, positioning, and time information which is not possible with conventional
navigation and surveying methods.
Apart from geodesy and geophysical interest, GPS has great importance in scientific
applications. The GPS satellites that are orbiting the Earth, at altitudes of about 20,200 km,
transmit signals that propagate through the ionosphere that exists at about 60 –1500 km
above the Earth’s surface. The signals from the GPS satellites travel through the ionosphere
on their way to receivers on the Earth’s surface. The free electrons populating this region of
the atmosphere affect the propagation of the signals, changing their velocity and direction of
travel as shown at figure 1. Due to the inhomogeneity of the propagation medium in the
ionosphere, the GPS signal does not travel along a perfectly straight line (Ioannides &
Strangeways, 2000). The effects of the ionosphere can cause range-rate errors for users of the
GPS satellites who require high accuracy measurements (Bradford & Spilker, 1996).
Ionosphere is highly variable in space and time (sunspot cycle, seasonal, and diurnal), with
geographical location (polar, aurora zones, mid-latitudes and equatorial regions), and with
certain solar-related ionospheric disturbances. Ionosphere research attracts significant
attention from the GPS community because ionosphere range delay on GPS signals is a
major error source in GPS positioning and navigation. The ionosphere has practical
importance in GPS applications because it influences the transionospheric radio wave
propagation. Observed ionospheric behaviour varies over the Earth and can be generalized
into aurora, mid-latitude and equatorial regions. Ionospheric delays of 38 ~ 52 m were
observed at low-latitude region during high solar activity period at an elevation cut off
angle of 10 (Komjathy et al., 2002). The equatorial ionosphere differs significantly from
what is typically observed at mid-latitudes. The geographic bands 10-15 north and south of
the magnetic equator are referred to as the equatorial anomaly region, due to the occurrence
of Appleton anomaly (Doherty et al., 2002).
Fig. 1. Exaggerated view of GPS signal geometric paths
The parameter of ionosphere that produces most of the effects on radio signals is total
electron content (TEC). By modelling TEC parameter, the evaluation of the ionospheric error
and the correction of these ionospheric errors for differential GPS can be done. The
ionosphere causes GPS signal delays to be proportional to TEC along the path from the GPS
satellite to a receiver. TEC is defined by the integral of electron density in a 1 m2 column
along the signal transmission path. TEC is a key parameter in the mitigation of ionospheric
effects on radio system. The TEC measurements obtained from dual frequency GPS
receivers are one of the most important methods of investigating the Earth’s ionosphere. The
TEC itself is hard to accurately determine from the slant TEC because this depends on the
sunspot activity, seasonal, diurnal and spatial variations and the line of sight which includes
knowledge of the elevation and azimuth of the satellite. The highest TEC in the world occurs
in the equatorial region. In Malaysia there are few corresponding research done on the low
latitude (equatorial) ionosphere.
Ionospheric research in the equator and tropical areas has sparked interests in several
research groups. Ong and Kamarudin (2006) have conducted a research on the TEC
distribution estimation with the Bent, IRI and Klobuchar modelling by using the GPS MASS
station network. While Ho et al. (2002) reported on the typical hourly variations for quiet
ionosphere over Malaysia for 24 hours on July 14, 2000. At Universiti Kebangsaan Malaysia
(UKM) researchers have been analysing TEC since 1999. Abdullah et al. (2008) did an
analysis of TEC determination over single GPS receiver station using Precise Point
Positioning (PPP) technique. The ionosphere over Malaysia is unique because of its location
near the equator line.
In this chapter, the focus is placed on the implementation of the method to the local area
GPS reference network and the data analysis of its performance in ionospheric TEC
predictions in support of GPS positioning and navigation. It investigated ionospheric TEC
predictions using Dual frequency technique and TEC map using Bernese software (BGS)
with PPP technique. For TEC dual frequency, it assessed the errors translated from the code-
delay to the carrier-phase ionospheric observable by the so-called “Levelling Process”,
which was applied to reduce carrier-phase ambiguities from the data. The TEC data derived
from GPS pseudorange measurements have a large uncertainty because the pseudorange
has high noise level. In contrast, the noise level of carrier phase measurements is
significantly lower than the pseudorange ones. To reduce the effect of pseudorange noise on
TEC data, GPS pseudorange data can be smoothed by carrier phase measurements, for
example, by using carrier phase smoothing technique, which is also often referred to as
carrier phase levelling. Whereas, for TEC Map technique, GPS measurements from stations
at the Equatorial region were used for producing maps. The Matlab and Bernese GPS
software was used to derive TEC from GPS data.
This chapter describes the parameter of the ionosphere that produces most of the effects on
GPS signal which is the TEC. TEC is measured to estimate the impact of ionosphere to the
signal transmitted by GPS satellite to the receiver on Earth. It is a measure of the total
amount of electrons along a particular Line of sight (LOS). Ionospheric delay correction is
carried out through modelling the TEC along each satellite signal path due to high spatial
variability of the ionosphere. Prediction of communication failures and radio interference
additionally requires accurate information on TEC variations. Another technique to
calculate TEC is by using Code’s IGS TEC Map. It is based on spherical harmonic expansion
parameterizations and computed based on the BGS Software and the output is in standard
IONEX.
2. Total Electron Content (TEC) in Ionosphere

The TEC is defined as the total number of electrons integrated along the path from the
receiver to each GPS. The TEC as an indicator of ionospheric variability that derived by the
modified GPS signal through free electrons. TEC is measured in units of 1016 electrons meter
per square area, where 1016 electrons/m² = 1 TEC unit (TECU) (Abdullah, 2009).
The nominal range is 1016 to 1019 with minima and maxima occurring at midnight and mid
afternoon approximately. At night the TEC decays rather slowly due to recombination of
electrons and ions. Maximum TEC usually occurs in the early afternoon and minimum TEC
usually occurs just before sunrise. Also daily TEC variations increase as one travels from
north to south, as sunlight is more direct.
There are several methods to obtain the TEC over the reference station. In this work TEC
was obtained from dual frequency method and the IGS (International GPS Service) TEC
map.
2.1 Dual Frequency Model TEC

TEC is significant in determining scintillation and group delay of a radio wave through a
medium. Ionospheric TEC is characterized by observing carrier phase delays of received
radio signals transmitted from satellites located above the ionosphere, often using GPS
satellites. GPS satellites transmit electromagnetic waves for positioning on two frequencies
which is L1 (1575.42 MHz) and L2 (1227.60 MHz) allowing receivers equipped with dual
frequency operation to be used. This enables us to extract the ionosphere TEC along the line
of sight, from satellite to receiver.
In this work, the TEC is observed at the F layer because this region has the highest
variability of free electrons, causing the greatest effect on GPS received signal compared to
other layers. More than two-third of electron concentration are located at F2 layer. This
method is conducted by going through several processes. Figure 2 shows the flow chart of
work progress to achieve the objective of the project.
Data GPS Data collection (e.g~JUPEM)
RINEX Format Select the appropriated RINEX data File
TEC calculation Calculate TECs and TECv using Matlab software
Result
Data Analysis
Fig. 2. Flow chart of TEC processing

The process of extracting data from RINEX (Receiver Independent Exchange) file was done
by using Matlab programming language whereby the RINEX file was obtained from the
GPS receiver as shown in figure 3. The program will analyse and extract the information
needed in calculating the TEC from the observation and navigation RINEX file. The result
will show the graph of elevation angle, different phase, different delay, slant TEC (TECs)
and vertical TEC (TECv) versus time. This data of TECv were used since its value is not
depending on the location of satellite receiver compared to TECs.
Start
GPS data (RINEX)
Check RINEX file
Yes
Error
No
Assigning observation parameter order

and the data duration
Read Navigation data
Satellite elevation angle
Levelling Process
TECv obtained
End
Fig. 3. Flowchart of TEC processing

2.1.1 Determination of TEC

Dual-frequency carrier phase and code-delay GPS observations are combined to obtain
ionospheric observables related to the (TECs) along the satellite-receiver line of sight (LOS).
Pseudorange is applicable to P(Y)-codes and C/A-codes. The pseudorange equation in units
of length can be expressed as:
s
 s
 s s s
Pr  c tr  t  c r  r  c( tr   t )  I  T  mpp   (1)
s s
Where Pr is pseudorange measured at receiver; c is speed of light in vacuum, t is
transmission time of signal measured by time frame of satellite, s; t r is reception time of
s s
signal measured by the clock of receivers r;  r is signal travelling time, Pr is LOS range
s
from satellite antenna and receiver antenna,  t ,  t r is satellite and receiver clock error
due to the difference in system time; I is ionospheric induced error; T is tropospheric
induced error, mpp is multipath error and  is noise or random error.
Carrier phase is the measurement of the phase difference between the carrier signal
generated by the receiver’s internal oscillator and the carrier signal transmitted from a
satellite. The basic equation for the carrier phase measurement is:
s s
 s
 s
Lr   r  c  t r   t  I  T   N r  mpL   (2)
s s
Where Lr is phase measurement in units of length, N r is integer ambiguity between the
satellite and receiver, mpL is multipath error.
The true range or geometric range can be represented by:
 X s  xr  Y s  yr   Z s  zr 
s 2 2 2
r  (3)
Where X, Y and Z are the satellite coordinates, x, y and z are the receiver coordinates
Dual band GPS receivers were considered in the measurable linear combination (LC). Dual
frequency observations can be used to measure the ionosphere delay. This delay can then be
removed from the measurements by combining the frequencies, L1 and L2. Ionosphere delay
can then be removed from the measurements by combining the frequencies and providing
the Linear Combination (LC) solution. All observables have the dimension of length, terms
due to noise and multipath are not explicitly shown, and higher-order ionospheric terms are
ignored:
L1    I1  1 N1

2 2

L2    f1 / f2 I1  2 N 2
P1    I1
2
 2
P2    f1 / f2 I1  (4)
Where  is non dispersive delay, contains LOS, clocks and troposphere bias, I1 is dispersive
delay of first frequency and N1, N 2 is integer ambiguities on L1 and L2.
The integrated TEC from the receiver to the satellite is proportional to the accumulated
effect by the time the signal arrives at the receiver. This affects the GPS range observables: a
delay is added to the code measurements and advance to the phase measurements. To
achieve very precise positions from GPS, this ionospheric delay or advance must be taken
into account. A GPS operates on two different frequencies f1 and f2, which are derived from
the fundamental frequency of = 10.23 MHz:
f1 = 154.fo = 1575.42 MHz and
f2 = 120.fo = 1227.60 MHz (5)
A dual-frequency GPS receiver can measure the difference in ionospheric delays between
the L1 and L2 of the GPS frequencies, which are generally assumed to travel along the same
path through the ionosphere. Thus, the group delay can be obtained as:
 
1 1 
P1  P  40.3TEC    (6)
2 
 f 22 f12 
Where P1 and P2 are the group path lengths corresponding to the high GPS frequency
(f1=1575.42 MHz) and the low GPS frequency (f2 =1227.6 MHz), respectively.
The TEC can also be obtained by writing Eq. (6) as
1  f1 f 2 
TEC     P P  (7)
40.3  f1 f 2  2 1
if dual frequency receiver measurements are available;

where (P1 and P2) are the pseudoranges measured in L1 and L2, respectively.
TEC can be divided into two parts. There are:
(a) Slant TEC (TECs)
(b) Vertical TEC (TECv)
Slant TEC is a measure of the total electron content of the ionosphere along the ray path
from the satellite to the receiver, represented in figure 4. Although TECs is measured at
differing elevation angles, usually the TECv is modelled. TECv enables TEC to be mapped
across the surface of the Earth.
TEC measurements are taken from different GPS satellite observed at arbitrary elevation
angles. This causes the GPS signals to cross largely different portion of the ionosphere. To
compare the electron contents for paths with different elevation angles, the TECs must be
transformed into equivalent vertical content or TECv by dividing it by the secant of the
elevation angle at a mean ionospheric height, which usually taken to be between 350 and
450 km. Generally by referring to figure 4, the slant TEC, TECs through a given sub-
ionospheric point is obtained from Eq. (8).
2.1.2 Mapping Function

Precision monitoring of ionosphere will have profound implications in almost all areas of
GPS user communities. The ionospheric mapping function is one of the first assumptions to
consider typically when ionospheric corrections are estimated or applied from Global
Navigation Satellite System (GNSS) data. The typical assumption in many GNSS imaging
and navigation systems is to consider a fixed mapping function constant, and associated to a
2D distribution of electron content at a given effective height (typically some value between
300 and 500 km).
The line-of-sight TEC values were converted to TECv values using a simple mapping
function and were associated to an ionospheric pierce point (IPP) latitude and longitude,
assuming the ionosphere to be compressed into a thin shell at the peak ionospheric height of
350 km as illustrated in figure 4. The thin shell model was used and its height is the effective
height which is taken as the ionospheric pierce point altitude. Generally, the ionosphere can
be divided into several layers in altitude according to electron density, which reaches its
peak value at about 350 km in altitude.
Fig. 4. Ionospheric Single Model (SLM)

Source: Schaer 1996
The thin layer model currently used in GPS has deficiencies resulting from conversion of
slant TEC to effective vertical TEC. The deficiencies come from in appropriate attribution of
the thin shell height. This conversion introduces a few errors in the middle latitude where
electron density is small. But it many result in obvious error at low latitude with large
electron density and great gradient (Horvath, 2000). Usually, the ionospheric delay resulting
from observation noises, less than 1 TECU, is omitted. It is assumed that, in the two-
dimensional spherical shell model, the majority of electron density is concentrated in a thin
layer with a height of 350-450 km above the surface of the Earth.
Generally by referring to figure 4, TECs through a given sub-ionospheric point is obtained
from Eq. (8)
TECV  TEC S cos  '

 (8)
Where TECs is the value of slant TEC, χ' is the difference between 90° and zenith angle (90°-χ).
In some literature this is called the elevation-dependent single layer (or thin shell) model
mapping function, SLM where can be written as
TEC   1 1
 
F    (9)
TEC  0  cos   or sin  ' 
' 2
1  sin 
'
' Re (10)
sin   sin 
Re  hm
Where Re is the mean earth radius, hm is the height of maximum electron density, and χ and
χ’ are the zenith angles at the receiver site and at the IPP (or β’ is the elevation angle at IPP),
respectively. χ can be calculated from a known satellite position and the approximate
coordinates of the receiver location. For hm , in general the value is taken as the height
corresponding to the maximum electron density at the F2 peak. The peak altitude ranges
from 250 to 350 km at mid-latitudes and from 350 to 500 km at equatorial latitudes. Typical
value for Re and hm are set to 6371 and 450 km, respectively. The more precise mapping
function according to Schaer et al. (1996) and currently applied in the IGS Global TEC map
is the modified single layer model, M-SLM. This is defined as;
Re
'
sin    
sin a  (11)
Re  hm
where α is correction factor which is close to unity. The value is chosen to be 0.9782 when
using Re and hm as 6371 and 506.7 km, respectively and assuming a maximum zenith angle
of 80º.
F( χ ) is also known as the slant or obliquity factor in the Klobuchar model and varies from
1 to slightly above 3 at hm = 350 km. For low elevation angles slant TEC can reach until 3
times the value of TEC at zenith. However the oblique-to-zenithal thin shell conversion
including the determination of hm is still being developed further. It has also been suggested
that hm should be taken to be between 600 and 1200 km which is greater than the commonly
adopted value. If this is correct, assuming a lower value could produce an error of 15 to 30%
or more in TEC.
2.1.3 Carrier phase levelling process

GPS signals can be used to extract ionospheric parameters such as TEC. For single frequency
GPS users, models of the ionosphere such as the Klobuchar model (Klobuchar, 1987), which
is also known as the GPS broadcast model, have been constructed utilizing ionospheric
parameters given in the GPS broadcast message. It is represented by a third degree
polynomial where the coefficients of the polynomial are transmitted as part of the broadcast
message header. The TEC can also be obtained as in Eq. (7), if dual frequency receiver
measurements are available. As the TEC between the satellite and the user depends on the
satellite elevation angle, this measurement is called TECs. The TEC varies with times and
over the space, and it depends on the solar activity, user location and the PRN elevation
angle.
In practice, calculation of TEC by the above means, using pseudorange data only, can
produce a noisy result. It is desirable to also use the relative phase delay between the two
carrier frequencies in order to obtain a more precise result. Differential carrier phase gives a
precise measure of relative TEC variations but because the actual number of cycles of phase
is not known, absolute TEC cannot be found unless pseudorange is also used. Pseudorange
gives the absolute scale for TEC while differential phase increases measurement precision.
The TEC data derived from GPS pseudorange measurements have a large uncertainty
because the pseudorange has high noise level. In contrast, the noise level of carrier phase
measurements is significantly lower than the pseudorange ones. To reduce the effect of
pseudorange noise on TEC data, GPS pseudorange data can be smoothed by carrier phase
measurements. For example is by using carrier phase smoothing technique, which is also
often referred to as carrier phase levelling. Carrier phase levelling or phase smoothing is
essentially some combination of the noisy code pseudorange with the comparatively smooth
varying carrier phase. The carrier phase contains much smaller measurement error than
pseudoranges, so that ionospheric TECs can be obtained by carrier phase smoothing the
pseudoranges (Hansen et al., 2000). This was done as shown below:
Firstly, the phase observations, measured in cycles, are scaled to units of length by
multiplying with the wavelength. Because the phase measurements are ambiguous, so the
phase derived slant delay, obtained from geometry free linear combination, L4 calculated
from Eq. (12) was scaled to zero relative range error at the first epoch. This eliminates the
integer ambiguity provided there are no cycle slips.
 2 2
 
L4  L1  L2  1  f1 / f 2 I1  1N1  2 N 2  (12)
To eliminate the code multipath effect that is normally seen at both ends of the path or at
low elevation angles, the code differential delay was fitted at the higher elevation angles.
This was done by defining a shift value and was added to the relative phase to fit the code
differential delay. This results in the absolute differential delay and the remaining noise was
discarded.
Figure 5 shows the differential delays determined using the above procedure. This
smoothed differential delay (with less noise and multipath) was then translated to the
absolute TECs by multiplying with a constant (see Eq. (7)). A mapping function, SLM is
used together with Eq. (8) to convert TEC to the vertical from the slant value. For a good
description on the determination of absolute TEC from dual frequency GPS measurements
refer to Parkinson and Spilker (1996).¸
Fig. 5. Phase smoothed code differential delay
2.2 TEC Map

Bernese GPS Software (BGS) was used to map the ionosphere in this work. BGS is
commonly used by scientists for research and education, survey agencies responsible for
high-accuracy GNSS surveys, agencies responsible for maintaining arrays of permanent GPS
receiver and also commercial users with complex applications demanding high accuracy,
reliability and high productivity (Dach, 2007). TEC map has gained much attention in the
recent years because of the ionospheric effects to the GPS-based navigation application. A
range delay caused by the ionosphere during quiet and disturbed geomagnetic days can be
approximated using the measurements of TEC map.
TEC ionopheric values and maps can be delivered by the International GPS Service (IGS).
IGS has developed the global ionospheric gridded data representing the TEC over the whole
globe. Analysis centres deliver their results of TECv and DCBs in the IONosphere Exchange
(IONEX) format (Schaer et al., 1998).
In this new version of BGS, PPP was processed using BPE. BPE consists of data, user scripts
and four process control file (PCF) where one of the PCF is PPP. PCF in PPP mode has been
selected to run PPP. In this PPP.PCF, regional ionosphere model is generated and stored in
Bernese ionosphere file and in IONEX file using GPS Estimation (GPSEST) program.
GPSEST is the program that able to generate TEC maps in IONEX (Schaer et al., 1998).
GPSEST program is used to model and estimate the ionosphere. In GPSEST program,
geometry-free linear combination from the zero-difference code observations was used
because it principally contains ionospheric information. Geometry-free linear combination
of this un-differenced GPS observations is then applied in GPSEST to generate TEC map.
A MSLM was used for mapping the TEC, approximated by a spherical layer with
infinitesimal thickness assuming that all free electrons are concentrated in altitude, H, above
the spherical Earth. The altitude H of this idealized layer is set to 350 km. Based on this
model, TEC values were calculated in geographic reference system which was able to
produce the epoch-specific instantaneous regional maps of the ionosphere. Using MSLM
noted above, a vertical TEC can be obtained at IPP. It can be shown that a single GPS
receiver can probe the ionosphere in a radius of 960 km assuming 10° elevation cut-off angle
and 450 km height.
This proved that PPP technique can be used to determine TEC over single station in
Malaysia. With the new BGS version 5.0; PPP technique is now available to produce
ionosphere maps. PPP is known as a valuable tool to provide an accurate position anywhere
on Earth, also for investigating many geophysical processes at the millimetre level.
3. Analysis of TEC
The ionosphere GPS-TEC measurements were carried out using GPS receiver networks from
Department of Survey and Mapping, JUPEM. GPS data on 8 November 2005 were analyzed
for this initial analysis. The stations were at Wisma Tanah, Kuala Lumpur, (3 10’ 15.44”N;
101 43’ 03.35”E) KTPK station and Universiti Teknologi Malaysia, Johor (133’ 56.934”N,
10338’22.429”E), UTMJ station as shown in figure 6. This analysis was based on one hour
observations from 5:00 – 6:00 UT (13-14 PM (LT)) using GPS satellite PRN 23 for KTPK
station and also using GPS satellite PRN 23 for UTMJ station. The GPS data was recorded in
universal time (UT) system. The sampling time interval is 15 second and the cut-off
elevation mask is 15°. GPS data used in this project were recorded on a quiet geomagnetic
day where the geomagnetic index Kp is 1.
Fig. 6. MASS stations in Malaysia

Source: JUPEM 2009
3.1 TEC Dual- frequency using levelling process

Figure 7 to 11 show representative cases of the different situations found in the analysis. The
absolute slant TEC from the KTPK station and UTMJ station can be measured directly from
this dual frequency method. This can be calculated by using pseudorange and carrier phase
measurements from the satellites (e.g. PRN 23) used in this study. Figure 7 (a and b) shows
the elevation angle of GPS satellite PRN 23 at 5:00 – 6:00 (UT) for KTPK Station and 5:00 -
6:00 (UT) for UTMJ station. The elevation angle can be calculated from the GPS navigation
data (or ephemeris). The elevation angle for KTPK and UTMJ station can be illustrated as
below:
75
70
Elevation angle
65
60
55
50
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(a)
72
70
68
66
Elevationangle
64
62
60
58
56
54
52
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(b)
Fig. 7. (a) Elevation angle of GPS Satellite PRN 23, 5:00-6:00 (UT) (KTPK station)
(b) Elevation angle of GPS Satellite PRN 23, 5:00-6:00 (UT) (UTMJ station)
Figure 8 (a and b) clearly indicate the code TEC and phase TEC of PRN 23 for elevation from
74 to 51 and 71 to 52. The differential delay (=P2-C1) from code measurements is noisy
and influenced by multipath while the phase measurements, are ambiguous and less
effected by the multipath, were used to smooth the code differential delay. Then the vertical
TEC can be obtained. This eliminates the integer ambiguity provided there are no cycle
slips.
4.5
P2-C1
Diff. phasebeforescaledtodiff. code(P2-P1) [m]
4 L1-L2
3.5
2.5
1.5
0.5
0
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(a)
4.5
P2-C1
Diff. phasebeforescaledtodiff. code(P2-P1) [m]
4 L1-L2
3.5
2.5
1.5
0.5
0
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(b)
Fig. 8. (a) Different phase (L1-L2) before scale to different code (P2-C1) for GPS Satellite
PRN 23, 5:00-6:00 (UT) (KTPK station)
(b) Different phase (L1-L2) before scale to different code (P2-C1) for GPS Satellite
PRN 23, 5:00-6:00 (UT) (UTMJ station)
4.5
P2-C1
L1-L2
Diff. delay(P2-C1) and rel. advance(L1-L2) [m]
3.5
3
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(a)
4.5
P2-C1
L1-L2
Diff. delay(P2-C1) and rel. advance(L1-L2) [m]
3.5
2.5
2
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(b)
Fig. 9. (a) Relative range error computed from the differential carrier phase advance for
GPS Satellite PRN 23, 5:00-6:00 UT (KTPK station)
(b) Relative range error computed from the differential carrier phase advance for
GPS Satellite PRN 23, 5:00-6:00 UT (UTMJ station)
Shown in figure 9 (a and b) are the absolute ionospheric range error obtained from
differential group delay. At this stage, levelling process was applied to eliminate the code
multipath effect especially at low elevation angles. This was done by defining a shift value
and adding it to the relative phase to fit the code differential delay. For the levelling process,
it assumed the average at the elevation angle (± 60°-90°) as reference and it can be seen in
figure 7. This value was chosen because there is no multipath at the high elevation angle and
during low elevation angle multipath where it can still be seen. After the differential carrier
phase was converted to an absolute scale by fitting it to the differential group delay curve
over the desirable, low multipath portion of each pass, the differential group delay data
were simply discarded.
44
P2-C1
42 L1-L2
TEC slant scaled to P2-C1 [TECU]
40
38
36
34
32
30
28
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(a)
40
P2-C1
38 L1-L2
36
TEC slant scaled to P2-C1 [TECU]
34
32
30
28
26
24
22
20
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [decimal hr]
(b)
Fig. 10. (a) TEC slant scale to (P2- C1) TECU for PRN 23, 5:00-6:00 UT (KTPK station)
(b) TEC slant scale to (P2- C1) TECU for PRN 23, 5:00-6:00 UT (UTMJ station)
This smoothed differential delay (with less noise and multipath) was then translated to the
absolute slant TEC by multiplying it by a constant Eq. (7) as shown in figure 10 while figure
11 shows TEC vertical for GPS satellite PRN 23 for KTPK and PRN 23 for UTMJ station.
32
SLM
31.5 MSLM
31
30.5
TECv in [TECU]
30
29.5
29
28.5
28
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time of day [hr]
(a)
31.5
SLM
31 MSLM
30.5
30
TECv in [TECU]
29.5
29
28.5
28
27.5
27
26.5
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time of day [hr]
(b)
Fig. 11. (a) TEC SLM, TEC M-SLM, GPS Satellite PRN 23, 5:00-6:00 UT (KTPK station)
(b) TEC SLM, TEC M-SLM, GPS Satellite PRN 23, 5:00-6:00 UT (UTMJ station)
Figure 11 shows that SLM was used to convert the slant TEC to vertical TEC. The analysis at
an equatorial region used SLM mapping function. The peak altitude ranges from 350 to 500
km at equatorial latitudes. However, from the figure it also shows that the MSLM which is
TEC for SLM is small compared to MSLM. The vertical TEC values are precise, accurate and
without multipath, unless the multipath environment is really terrible, in which case a
small, residual amount of multipath can even be seen in the differential carrier phase.
3.2 TEC Map

TEC Map is computed with the BGS software using the PPP program and the output is in
standard IONEX format. A MSLM was used for mapping the TEC, approximated by a
spherical layer with infinitesimal thickness assuming that all free electrons are concentrated
in altitude, H, above the spherical Earth. The altitude H of this idealized layer is set to 450
km. Based on this model, TEC values were calculated in geographic reference system which
was able to produce the epoch-specific instantaneous regional maps of the ionosphere.
Using MSLM noted above, a vertical TEC can be obtained at IPP. It can be shown that a
single GPS receiver can probe the ionosphere in a radius of 960 km assuming 10° elevation
cut-off angle and 450 km height. In order to suit the geographic location to all observational
epochs, region located between 0º to 7º north of geographic latitude and 90º to 110º
longitude was selected. This map covers a 24 hour time period at intervals of 2 hours
starting from 00.00.00 hour.
Figure 12 illustrates the longitudinal vertical TEC map for KTPK station on 8 November
2005 produced with PPP technique with two-hour intervals between each map, starting
from 00:00 to 22:00 UT where local time (LT) is +8. Based on figure 12, TEC starts increasing
at 00:00 UT at about 7 TECU and gradually increased reaching a maximum level of 28 TECU
at 06:00 UT (14:00 LT) then decreased steadily until nearly 20:00 UT before sunrise and rate
of ion production is low. It showed that the maximum value of TECU usually happens near
midday while the minimum value of TECU occurs at night. The low TEC over equatorial
region is mostly due to the Kp and Dst index.
Fig. 12. TEC Map for KTPK station, 8 November 2005

A comparison was made with Global Ionosphere Maps (GIM) downloaded from Centre for
Orbit Determination in Europe (CODE) as shown in figure 13. GIM from CODE was
generated using data from about 150 GPS receivers around the globe. GIM used 5.0º and 2.5º
in longitude and latitude of special resolution with two-hour intervals. For the comparison
purpose, an area covering regional model was extracted from IGS maps. Both figures show
the same pattern of longitudinal variation of TEC starting from 00:00 UT to 22:00 UT. It is
noticeable however, that TEC from GIM is higher by about 0 to 5 TECU, as compared with
the maps generated by PPP.
Fig. 13. TEC map from IGS, 8 November 2005
This data was 12 plotted as a 24 hour contour map as a function of longitude as illustrated in
figure 12. The TEC above the reference station was extracted from interpolation of this data
at every epoch (30s interval) as illustrated in figure 14.
35
30
25
20
15
10
0
0 5 10 15 20 25
Fig. 14. TEC extracted from TEC map

The difference between the two maximum TEC is because there was no IGS station located
in Malaysia. The diurnal profile of ionospheric variation from both figure showed a good
agreement. This proved that PPP technique can be used to generate TEC map over single
receiver station. Figure 15 clearly indicates the 3-D maps based on IONEX data. The IONEX
data have been generated from the BGS software with PPP program. The effective height
obtained for 450 km (MSLM) for KTPK station with TEC is 34 TECU.
45
40 40
35
35
30
25 30
TEC
20 25
15
20
10
5 15
0 10
4
5
2
0 10 15 20
latitud 0 5
masa
Fig. 15. 3-D MSLM for KTPK station
4. Future Work
The research has also identified several possibilities of GPS methodology including accuracy
issues and further improvement on TEC that can be incorporated in future research. The
following issues and directions have been noted:
i). In this work, only two signal involve which is L1 and L2 to investigate ionospheric TEC.
Further, it would be necessary to include new improved signal, which is L5 for more
precise and accurate.
ii). Utilize data taken from other satellite navigation systems such as GLONASS, Galileo, etc
iii). More detailed work for different ionospheric condition needs to be verified and
compared for other region such as high latitude, Antartic and Artic.
Implementing this suggested further work would extend the TEC measurements at any
location with any ionospheric conditions. The outcome gives a different approach that could
be considered also for current or future GNSS augmentation systems to overcome the
ionospheric error.
5. Conclusion
In this chapter we have presented methods for processing TEC which utilize different
techniques. Two scenario of obtaining TEC were studied such as TEC dual frequency and
TEC Map. First, for TEC Dual Frequency, GPS carrier phase derived TEC provides a smooth
but relative measurement of ionospheric TEC, while code derived TEC provides a noisy but
absolute measurement. To mitigate inherent fluctuations in pseudorange due to bandwidth
limited precision, receiver noise, cycle slip, multipath etc, Levelling Process was applied to
reduce carrier-phase ambiguities from the data. As a result, the remaining noise is
discarded.
Mean while for TEC Map, analysis results showed that TEC have similar variations, where
the TEC values start to increase gradually from morning and reach its maximum at noon
and decrease around afternoon. Bernese software, the scientific GPS software packages has
already proved its ability to determine an accurate regional TEC map. The PPP technique
can be used to generate the TEC map. TEC map are needed in order to characterize the
ionospheric behaviour. The result proved that PPP technique can be performed at cm- level.
Besides, extraction of TEC information can also be done.
Considering the variability of the ionosphere in the equatorial region, it is recommended to
analyze other mapping functions to project the line-of-sight ionosphere delay into the
vertical used in the proposed approach.
6. References
Abdullah, M.; Strangeways, H.J. & Walsh, D.M.A. (2009). Improving ambiguity resolution
rate with an accurate ionospheric differential correction. Journal of Navigation,Vol.
62, No. 1, pp. 151-166, ISSN: 0373-4633.
Abdullah, M.; Bahari S.A. & Yatim, B. (2008). TEC determination over single GPS receiver
station using PPP technique, International Symposium on GPS/GNSS 2008,
November 11-14, 2008 Tokyo
Bradford, W.P. & Spilker, J.J.J. (1996). Global Positioning System: Theory and applications,
American Institute of Aeronautics and Astronautics, Vol. I and II Washington DC,
USA.
Doherty, P.H.; Dehel, T.; Klobuchar, J.A.; Delay, S.H.; Datta-Barua, S.; de Paula E.R. &
Rodrigues, F.S. (2002). Ionospheric effects on low-latitude space based
augmentation systems, Proceedings of ION GPS 2002, September 24-27, 2002,
Portland, pp. 1321-1329, Oregon.
Dach, R.; Hugentobler, U.; Fridez, P. & Meindl, M. (2007). Manual of Bernese GPS Software
Version 5.0, Astronomical Institute, University of Bern.
Hansen, A.; Blanch, J. & Walter, T. et al. (2000). Ionospheric correction analysis for WAAS
quiet and stormy. ION GPS, Salt Lake City, Utah, September 19-22, 2000, pp 634-
642, America.
Horvath. I & Essex. E.A. (2000). Using observations from the GPS and TOPEX satellites to
investigate night-time TEC enhancement at mid-latitudes in the southern
hemisphere during a low sunspot number period, Journal of Atmospheric and solar
Terissterial-Physics, Vol .62, No.5, pp. 371-391.
Ho, Y.H.; Zain, A.F.M. & Abdullah, M. (2002). Hourly variations total electron content, TEC,
for quiet ionosphere over Malaysia. Proceeding of the Annual Workshop National
Science Fellowship (NSF) 2001, Petaling Jaya, pp. 77-79, Kuala Lumpur.
Ioannides, R.T. & Strangeways, H.J. (2000). Ionosphere-induced errors in GPS range finding
using MQP modelling, ray-tracing and nelder-mead optimization. Millennium
Conference on Antennas and Propagation, AP2000, vol. II, Davos, pp. 404–408,
Switzerland.
JUPEM. (2009). MASS station. Malaysia [10 January 2009].
http://www.jupem.gov./
Klobuchar, J.A. (1987). Ionospheric time-delay algorithm for single-frequency GPS users,
IEEE Transactions on aerospace and electronic systems, Vol. 23, No. 3, pp. 325-331.
Komjathy, A. (1997). Global ionospheric total electron content mapping using the Global
Positioning System. Ph.D. dissertation. Department of Geodesy and Geomatics
Engineering Technical Report No. 188. University of New Brunswick, Fredericton,
New Brunswick, Canada. p 248.
Ong, H. P. & Kamarudin, M.N. (2006). Calculation in Estimating Total Electron Content
GPS. Universiti Teknologi Malaysia.
Parkinson, B.W. (1996). GPS error análysis, in Global Positioning System: theory and application,
Vol. 1, Edited by Parkinson & Spilker, American Institute of Aeronautics and
Astronautics, Inc., Washington D.C., pp. 469-483.
Schaer, S.; Markus, R.; Gerhard, B. & Timon, A.S. (1996). Daily Global Ionosphere Maps
based on GPS Carrier Phase Data Routinely produced by the CODE Analysis
Center, Proceeding of the IGS Analysis Center Workshop, Silver Spring, Maryland, pp.
181-192, USA.
Schaer, S.; Gurtner, W. & Feltens, J. (1998). IONEX: The IONosphere map exchange format
version 1, Proceeding of IGS Analysis Center Workshop, pp. 233-247.
Warnant, R.; Ivan, K.; Marinov, P.; Bavier, M. & Lejeune, S. (2007). Ionospheric and
geomagnetic conditions during periods of degraded GPS position accuracy: 2.RTK
events during disturbed and quiet geomagnetic conditions, Advances in Space
Research,Vol. 39, No. 5., pp. 881-888.
Performance Evaluation Methods to Study
IEEE 802.11 Broadband Wireless Networks under the PCF Mode 509
0
24

IEEE 802.11 Broadband Wireless Networks
under the PCF Mode
Vladimir Vishnevsky and Olga Semenova
ZAO Research & Development Company "Information and Networking Technologies"
Russia
1. Introduction
In the Chapter, we consider design concepts and protocols for metropolitan area wireless net-
works, realization methods for adaptive dynamic polling in these networks and investigation
of their main performance characteristics by means of stochastic polling models.
Polling mechanism is widely used in the wireless metropolitan area networks (WMANs). In
the wireless networks with PCF (Point Coordination Function), a base station polls subscriber
stations accordingly to a polling table describing the order of polling. For IEEE 802.11 Wi-Fi
networks, the polling is an option; for WiMAX networks (IEEE 802.16), it is basic. Using the
PCF in the MANs allows to avoid the problem of hidden stations, efficiently schedule an order
of station access to the wireless channel, flexibly control the radio cell operation and change
its parameters correspondingly to the current situation by adjusting only the base station.
The methods to form and keep up the polling table are not specified in the standard thus the
wireless network developers can freely decide on how to realize it. The specific polling mech-
anism and its parameters are the main factors determining the efficiency of the broadband
wireless MAN with centralized control. In the Section, we give the description of the IEEE
802.11 protocols and the main directions of their development, including the recent versions
IEEE 802.11n and IEEE 802.11 VHT. The much attention is given to development and mod-
elling of the algorithms to poll subscriber stations, the schemes of adaptive dynamic polling
(ADP). The adaptive dynamic polling is proposed to cut down the expenses of polling the
empty subscriber stations and stations that stopped working for some reason. The adaptive
dynamic polling is the prospective direction of developing the IEEE 802.11 broadband wireless
networks.
We present the models of adaptive polling system and the system with threshold polling. With
the adaptive polling order, the order of queue visit is cyclic but a server does not visit queues
that were empty at the instant of polling in the previous cycle. Under the threshold polling,
a queue is served only of its length exceeds the given threshold. Such service discipline is a
possible way to assign a priority to a queue depending on the threshold value, and it allows
server to give more attention to queues with high traffic intensity rather than spend time in
queues with low traffic.
2. Performance evaluation of the broadband wireless networks

Polling mechanism is widely used in the metropolitan area wireless transmission networks. In
the wireless networks with PCF (Point Coordination Function), a base station polls subscriber
stations accordingly to a polling table describing the order of polling. For IEEE 802.11 Wi-Fi
networks, the polling is an option; for WiMAX networks (IEEE 802.16), it is basic. In this
Section, we consider design concepts and protocols for metropolitan area wireless networks,
realization methods for adaptive dynamic polling in these networks.
2.1 Development of the broadband wireless networks: state of the art and prospects
In the recent years, the wireless transmission networks become the main direction of the
network industry development. It was provided by both the rapid Internet development
and the adoption of new progressive methods for coding, modulation and wireless data
transmission. Recently, it is obvious that broadband wireless networks are without a rival
with their efficiency of deployment, portability, price and area of potential applications.
Wireless technologies displace the wired one almost in all places where they can provide
high-quality data transmission. The tendency is evidently continuing to the future since the
wireless world is more comfortable. Nowadays, the wireless data transmission technologies
have become ingrained in everyday life of millions of people and enterprises. The modern
wireless networks allows solving variety of problems from the indoor network management
to the distributed wireless networks within a city, a region or a country. Low cost, efficiency of
deployment, wide performance capabilities to transmit data, IP telephony and video streams,
all these make the wireless technologies the most rapidly developing telecommunication area.
Rapid growth of the broadband wireless networks often called the "wireless revolution" in the
field of data transmission networks is explained by a number of their own distinctive features,
such as
– flexibility of a network topology enabling the dynamic change of the topology without
time loss when mobile users connect to the network, move or disconnect;
– high data transmission rate (up to 54 Mbit/sec);
– rapidity of designing and realization which is significant because of the strict technical
conditions to network construction;
– high unauthorized access protection level;
– high-priced laying or rent of the fiber optic or copper cable are not needed.
Recently, the wireless technologies provide effective solution of the following problems:
– mobile access to the Internet;
– organization of the wireless radiocommunication between workstations of a local area
network (organization of the wireless access to a local area network resources);
– unification of local area networks and workstations into a single data transmission
network and providing the remote access to the Internet for local area networks;
– last mile problem solution;
– interconnecting the automatic telephone systems with wireless channels of up to 54
Mbit/sec rate;
– creation of the land cellular radio modem data networks.
The mentioned features of the wireless technologies are substantially brought about by the fact
that wireless networks operating within the range 2.4-6.4 GHz are based on the technology of
broadband, or noise-type signal. The technology was initially used for the military purposes
and recently it is efficiently used for civil radio networks.
The broadband wireless technologies use two radically different methods of frequency band
utilization, they are Direct Sequence Spread Spectrum (DSSS) and Frequency Hopping Spread
Spectrum (FHSS). Both methods imply the frequency band division into n subchannels. Under
DSSS method, each data bit is coded as a sequence of n bits, and all those n bits are transmitted
simultaneously through all n subchannels, and the coding algorithm is individual for each pair
’transmitter-receiver’ so as to provide the transmission security. Under the FHSS method, a
station transmits only through one of n subchannels at each time moment periodically changing
the subchannel. Those change-overs (hops) happen simultaneously for both a transmitter and
a receiver, and their sequence is pseudo-random and is known only for ’transmitter-receiver’
pair that provide the transmission security as well.
Undoubtedly, each method has its own advantages and disadvantages. The DSSS method
allows reaching the maximal throughput and due to the n-modular redundancy it, first, pro-
vides the narrow-band interference immunity, and, second, gives the opportunity to use the
low power signal so as not to interfere with ordinary radio devises. On the other hand, the
FHSS equipment is considerably simpler and cheaper and has the broadband interference
immunity.
To work with the wireless networks, we need the special MAC (Media Access Control) proto-
cols due to the fundamental differences from the cable medium, namely the lack of complete
connection (the stations can be hidden from each other), the wireless medium is not protected
from the outside signals and its signal propagation properties are asymmetric and variable by
times. In order to provide the effective wireless medium access, the international standards,
protocols and recommendations are developed which specify the physical and MAC layers of
wireless networks: Bluetooth, ETSI Hiperlan and IEEE 802.11 for local area networks (LAN);
IEEE 802.11 using the necessary amplifiers and parabolic antennae for metropolitan area net-
works (MAN), and finally, IEEE 802.16 and callular telephony technologies modified for data
and video images transmission (GPRS, UMTS and CDMA-2000) for MANs (Vishnevsky et al.,
2009).
Among the LAN and MAN developers, the IEEE 802.11 protocol is very popular (referred to as
Radio-Ethernet as well) adopted as an international standard in 1997 and having the following
features:
• it can be used in both LANs and MANs;
• both DSSS and FHSS methods of the broadband wireless network deployment are
regulated;
• the huge number of software and hardware of the large companies (such as CISCO
Aironet, Lucent Technologies, Alvarion, etc.) in the world markets support the standard.
The IEEE 802.11 protocol determines two network development topology, they are topologies
with infrastructure and so called ad-hoc-topology. With infrastructure topology, a wireless
network has a single access point (or base station). An access point provides the synchroniza-
tion and coordination for stations within the range, transmits broadcast packets and, what is
of the great importance, can be a portal into the global network. Such topology is referred to
as the Basic Service Set (BSS). To cover the wide area, it is possible to set up several access
points working in different frequency channels and connected to the joint wired or wireless
backbone. Besides, the subscriber stations can be provided with a roaming between the access
points. Such topology is called the Extended Service Set (ESS). The ad-hoc topology called
the Independent Basic Service Set (IBSS) is the network performance scheme under which the
numerous stations are connected directly avoiding connection to the special access point. This
regime is effective when the wireless network infrastructure is not constructed (e.g., conference
hall), or can not be constructed by some reason.
The IEEE 802.11 protocol specifies the data transmission rate equal to 1 or 3 Mbit/sec, with this
the packet header and the service information can be transmitted at 1 Mbit/sec. Note that the
transmission rate did not satisfy the users even when the protocol was adopted and approved.
In order to make the wireless technology popular, cheap and, above all, to satisfy the modern
strict conditions of business applications, the developers had to set up new standards which
were the extensions of IEEE 802.11. Consider them in brief.
IEEE 802.1a. The IEEE 802.11a protocol exploits the radio frequency band of 5 GHz (5150-5250
MHz, 5250-5350 MHz and 5725-5850 MHz). In contrast to IEEE 802.11, the protocol applies
not the spectrum broadening technologies but Orthogonal Frequency Division Multiplexing
(OFDM), also referred to as multiple carrier modulation, which uses several carrier signals of
different frequencies each transmitting a number of bites. This technology allows reaching the
following data transmission rates: 6, 9, 12, 18, 24, 36, 48 and 54 Mbit/sec.
IEEE 802.11b. The IEEE 802.11b protocol involves the changes within the IEEE 802.11 physical
layer. The network operates in 2.4 GHz radio frequency band. But the other signal modulation
technology, Complementary Code Keying (CCK) allows reaching the rates 5.5 and 11 Mbit/sec
and increases the connection stability in interference and multipath signal propagation condi-
tions.
IEEE 802.11g. The IEEE 802.11g protocol as long as IEEE 802.11b operates in 2,4 GHz radio
frequency band but applies the orthogonal frequency division multiplexing (OFDM) allowing
to reach the data transmission rate equivalent to IEEE 802.11a (up to 54 Mbit/sec). Nevertheless,
the protocol enables stations to get back to rates 1, 2, 5.5 and 11 Mbit/sec, i.e to CCK modulation.
Therefore, the devices 802.11b and 802.11g are compatible within a segment of broadband
wireless network.
Recently, the standard IEEE 802.11n describing the networks with data transmission rate
100Mbit/sec on the base of antenna system technology MIMO is going to be finished. The
mobile version of the standard (IEEE 802.11p) and the addition IEEE 802.11e to provide the
guaranteed quality of service (QoS).
In 2007, the generalized standard (see IEEE Std. 802.11-2007) was approved involving all stan-
dards finished before June, 2007. They are IEEE 802.11a/b/g mentioned above and additions
IEEE802.11e/h/i/j.
The standard IEEE 802.11 is constantly being improved and developed to provide new cus-
tomer services and to increase the data transmission rate and its quality. In 2009, it is planned
to release a number of new standards being developed from 2003-2004. First of all, they are
IEEE 802.11n and IEEE 802.11s. Though those standards are being finished, many companies
have started production of devices and provide the wireless network operation based on the
draft versions of those standards.
The other standards to be approved in 2009 are:
• standard IEEE 802.11u describing communications between IEEE 802.11 networks and
outer networks;
• standard IEEE 802.11r regulating procedures of switching between subscriber stations
for delay sensible applications like IP-telephony, etc.;
• standard IEEE 802.11p for operation in dynamic environment, and for fast moving
wireless devices, in particular;
• standard IEEE 802.11v describing the wireless network control protocols;
• standard IEEE 802.11w regulating the methods to protect supervisory frames in a wire-
less network;
• standard IEEE802.11z describing the protocol for direct data exchanging between sta-
tions without using an access point.
The standard IEEE 802.11k should be also mentioned but it is not included into the generalized
standard IEEE 802.11-2007 since its final version was released at the end of 2007 only. The stan-
dard regulates the mechanisms to exchange information about radio resource, radiochannel
performance and load, noise level, etc.
The persistent growth of the transmission data volume, release of new applications, e.g. high
definition video, impose heavy demands on wireless network throughput.
In spite of high data transmission rate in third generation mobile networks based on LTE
technology and in networks IEEE 802.11n (up to 300 Mbit/sec), work on new technologies
creation in the framework of IEEE 802.11 is continued. From 2007, the standard IEEE 802.11
VNT (Very High Troughput) has been started providing a base for very high throughput local
wireless networks with nominal speed up to 500Mbit/sec within the frequency range 6 GHz.
The standard is planned to be finished in 2012.
High network throughput is obtained by the MIMO technology application with 8 spaced
antennas on both transmitting and receiving sides, by band enhancement up to 80 MHz
through multiplexing of four channels of width 20 MGz, also by using OFDMA to organize
frequency division multiple access as in IEEE 802.16. The developed standard supports
compatibility with devices working under IEEE 802.11a/b/g/n.
In Russian Federation, the new technology and both hardware and software for very high
throughput mesh-networks operating in the frequency range 60GHz (Vishnevsky & Frolov,
2009). As compared to existing mesh-networks, the proposed approach provides transmission
rate up to 1000 Mbit/sec and makes the frequency planning and operating in duplex mode
unnecessary.
In the 802.11 protocol, the fundamental mechanism to access the medium is called Distributed
Coordination Function (DCF). This is a random access scheme, based on the Carrier Sense
Multiple Access with Collision Avoidance (CSMA/CA) protocol. Retransmission of collided
packets for each station is managed according to binary exponential backoff rules (Section 2.2).
The alternative access mechanism as an option specified in IEEE 802.11 is point coordination
function (PCF) under which the coordinator station manages the centralized polling of other
stations (Section 2.3).
2.2 Medium access layer in IEEE 802.11. Distributed Coordination Function (DCF)
The IEEE 802.11 protocol is the part of IEEE 802 protocols for local area networks (LANs)and
metropolitan area networks (MANs) involving the well-known protocols IEEE 802.3 (Ethernet
LAN) and IEEE 802.5 (Token Ring LAN). The majority of IEEE 802 protocols determines the
physical and data link layers of the Open Systems Interconnection (OSI) seven-level reference
model of the International Organization for Standardization (ISO). Furthermore, the data
link layer is represented as two sub-layers, the Logical Link Control (LLC) and Medium
Access Control (MAC). Such division is conditioned by the fact that under the same LLC,
the mechanisms providing MAC can be different. The IEEE 802.11 involves the functional
description of both MAC and PHY layers. Both layers possess significant features, e.g. the
high packet loss rate due to noise and collisions, and the fact that wireless data transmission
can suffer from the unauthorized access. Logical link control is not considered in the protocol
since it is the same as in IEEE 802.2.
All questions on regulation of the wireless medium sharing by the network stations are de-
termined on MAC-layer. The necessity of such regulation rules is quite obvious. Imagine
the situation when each station of the wireless network sends data to the medium without
observing any rules. As a result of such signals interference, the destination stations can
not receive the data, and even understand the data were destined for them. Therefore, the
stringent regulating rules are essential to determine the wireless medium multiple access. The
multiple access rules can be compared to the rules of the road which regulate the road sharing
by road users.
As it is mentioned above, there are four types of IEEE 802.11 MAC layer multiple access to
wireless medium, they are Distributed Coordination Function (DCF), its extension Extended
DCF (EDCF), Point Coordination Function (PCF) and Hybrid Coordination Function (HCF).
Below, we consider these mechanisms in details.
The DCF is a method to organize peer-to-peer access to the wireless medium. The function
is based on the Carrier Sense Multiple Access/Collision Avoidance (CSMA/CA). With this
access, each station before sending a packet listens to the medium trying to detect the carrier
signal and starts transmission only when the channel is idle. But in this case the probability
that a packet collides with another one upon its transmission is high enough, when two or
more stations find out the channel is idle and start transmission at the moment when some
station is transmitting. In order to decrease the collision probability, the Collision Avoidance
(CA) mechanism is applied. The mechanism is described as follows. A station detecting the
channel idle waits for the prespecified time interval before it starts transmission. The time
interval is random given by two intervals: the Distributed InterFrame Space (DIFS) and the
random Backoff time. Consequently, each station waits for a random time before starting data
transmission that essentially decreases the collision probability since the probability that at
least two stations have the same backoff is negligible.
MAC-layer of the IEEE 802.11 protocol, its modification described in IEEE 802.11e thereof,
specifies five types of a time interval between consecutive data transmissions, namely Inter-
frame Space (IFS). The shortest one is SIFS (Short Interframe Space) used for special sequence
of data exchange, e.g. ACK transmission to acknowledge a frame successful reception. The
SIFS duration is specified by physical layer to give enough time for the system to switch from
reception to transmission or inversely. The other intervals are given in duration increasing
order: PIFS, used by stations with Point Coordination Function; DIFS, used by stations with
Distributed Coordination Function; AIFS, used in Extended DCF (EDCF), and EIFS (Extended
Interframe Space), used by stations after transmission error.
In order to provide stations with the equal access to the channel, it is necessary to determine
the appropriate algorithm to choose the backoff time which is a number of basic time intervals
called time-slots. To choose the random backoff, each station determines the Contention
Window (CW) which is the range the backoff time is chosen from. The minimal CW is 31 time
slots, and maximal one is 1023 time slots. The backoff is determined as:
Backo f f = Random(CW ) · SlotTime,

where Random(CW ) is an integer uniformly chosen in the range (0, CW − 1). At each time, the
value CW depends on the number n of the attempts failed to send a packet and is given by
CW = CW0 2n ,
where CW0 is the minimal contention window, 0 ≤ n ≤ m, and m is the maximal number of
attempts allowed to send a packet. If the m + 1-th attempt is failed the packet is discarded.
When a station tries to get an access to the channel, after the DIFS expires the backoff starts to
count down. If the channel is idle during DIFS and the backoff, the station immediately starts
transmission as the backoff counter reaches 0.
After the successful packet transmission, the CW is determined anew. If another station
transmits during the backoff time, the backoff counter is frozen until the channel becomes idle
(transmission is finished and DIFS expired). With this procedure, it is easy so see that the more
times the station freezes its backoff counter the grater the probability that the packet waiting
for transmission will not collide.
The described algorithm to get an access to the wireless medium guarantees the equal access
for all stations in the network. But under such procedure, the probability of collision (that
an arbitrary packet collides upon its transmission) is still non-zero. The collision probability
could be reduced by extension of the maximal CW. But it leads to the grater backoff value
which decreases the channel throughput. Therefore, the DCF uses the following algorithm
to minimize collisions. After each successful frame reception, the destination station sends
the ACK (ACKnowledgement) after SIFS to acknowledge the transmission was successful. If
collision happens upon transmission the sender station does not get an ACK, so it finds out
the transmission failed. The sender waits for ACK during EIFS, and if the ACK is not received
the sender increases its CW. Thus, if the CW is 31 slots for the first transmission attempt for
the second one it is 63, 127 for the third, 255 for the fourth, 511 for the fifth, and 1023 for
other attempts. It can be seen that the CW is dynamically increased with collision number
increasing, which allows reducing both the delay and the collision probability.
Note that the sender station can not receive the ACK frame indicating the transmission was
success due to collision or signal distortion. And both reasons are not distinguishable for the
sender station.
As it is mentioned above, all stations are equal to get an access to the wireless medium due to
the contention mechanism, and no station has a priority to transmit data. This restricts DCF to
provide Quality of Service (QoS). In IEEE 802.11e protocol, QoS is enabled by Enhanced DCF
(EDCF). The EDCF mechanism is similar to DCF, but the difference concerns the CW size and
the backoff counter to provide the priority access for various applications to wireless medium.
For EDCF the traffic is divided into categories (TC, Traffic Categories) which differ from each
other by priority to get an access to the transmission medium. Meanwhile, the medium access
mechanism is the same as for DCF contention based.
Before a station detecting the channel idle starts transmission, it waits for AIFS (Arbitration
InterFrame Space) and then counts down the backoff counter. The backoff counter is uniformly
chosen in the range [1, CW (TC) + 1] where CW (TC) is the contention window for the given
TC. TC priority is provided by using the different values of minimal and maximum CW and
AIFS. Thus, the minimal value of AIFS is always DIFS but it can be increased depending on
the TC.
In case of collisions which happen when two or more backoff counter drop to zero simulta-
neously and AIFS intervals are the same, the CW is increased. For EDCF, the new CW size is
determined as newCW (TC) = ((oldCW (TC) + 1)PF) − 1, where PF is the constant CW scaling
factor depending on the TC. Note that for DCF we have PF = 2.
Thus in EDCA, the TC is assigned a priority by variation of the parameters CWmin, CWmax,
AIFS, PF. Each station of the wireless network can have up to 8 TC queues for transmission.
But in this case, the TC backoff counters within the same station can drop to zero simulta-
neously which is called the virtual collision. To avoid virtual collisions, the station uses the
special queue scheduler which provides the higher priority TC with the priority access to the
transmission medium.
The considered mechanism of transmission medium multiple access control have the same
bottleneck, so-called problem of hidden stations. The situation happens when two stations
can not listen to each other directly due to natural barriers. Such stations are called hidden.
To avoid the problem, the DCF and EDCF mechanisms have the optional technique known as
Request-To-Send/Clear-To-Send (RTS/CTS).
Accordingly to RTS/CTS, before transmitting a packet, a station operating in RTS/CTS mode
"reserves" the channel by sending a special Request-To-Send short frame. An RTS frame
involves the information on the forthcoming transmission and destination station and is avail-
able to all stations in the network (except ones hidden from the sender station). It allows the
other stations to postpone transmission for the declared transmission time, and during this
time the channel is considered "virtually busy". The destination station acknowledges the
receipt of an RTS frame by sending back a Clear-To-Send (CTS) frame, after which normal
packet transmission and ACK response occurs. Since collision may occur only on the RTS
frame, and it is detected by the lack of CTS response, the RTS/CTS mechanism allows to in-
crease the system performance by reducing the duration of a collision when long messages
are transmitted.
The DCF and EDCF are simple and reliable mechanisms of multiple access to transmission
medium in IEEE 802.11 broadband wireless networks. But they have two shortcomings: the
lack of both QoS support and solution of the hidden stations problem. Thereby, the IEEE
802.11 protocol was supplemented with the alternative methods to multiple access control.
2.3 Point coordination function

The DCF mechanism described above is basic for IEEE 802.11 protocols and can be used in
both ad-hoc wireless networks and infrastructure networks (having Access Point, AP). But for
the infrastructure networks, there are more appropriate methods to control the transmission
medium multiple access, namely, Point Coordination Function (PCF) and Hybrid Coordination
Function (HCF). Note that PCF and HCF mechanisms are optional and applied for networks
with AP only.
In case of PCF, one of stations (access point) is central and called the Point Coordinator
(PC). The PC controls is imposed a responsibility to control the multiple access on the base
of the specified polling algorithm or information on station priority. Thus, the PC polls all
stations in the network listed in its polling table and organizes data transmission between
network stations. As opposed to DCF where each station decides for itself when to start
transmission, the PCF implies that only PC decides what station can get an access to the
channel. It is significant to note that this approach completely avoids the contention access
to the channel unlike DCF, makes collisions impossible and provide the priority access for
time-sensitive applications. Thus, PCF can be used to organize collision-free priority access to
the transmission medium.
The Point Coordination Function does not contradict the Distributed Coordination Function
and rather supplements it. In fact, the PCF networks can use both PCF and traditional DCF.
During network operating, the time intervals for PCF and DCF alternate.
In order to provide alternating the PCF and DCF modes, the access point which realizes PCF
has to have a priority access to transmission medium. It is possible if an access to the channel
is contention one (as for DCF) but the time interval for the access point to wait the response is
less then DIFS. In this case, when the access point tries to get an access to the channel it waits
the end of transmission as other stations do and since it has the minimal inferframe space, it
gets an access first. The interframe space for the access point is called PIFS (PCF Interframe
Space), diven that SIFS < PIFS < DIFS.
The DCF and PCF mechanisms are combined in so called superframe which is the sum of PCF
interval of the contention-free access called CFP (Contention-Free Period) and the succeeding
DCF interval of CP (Contention Period). The CP length should be enough to enable the
transmission of at least one frame by using PCF mechanism. It is necessary for the association
procedure but it is out of our topic. The superframe starts with the beacon frame, and
all stations having received it postpone their transmissions for the time determined by the
CFP. The beacons contain the information on the CFP duration and allow synchronizing the
operation of all stations.
Under PCF, the access point sends data packets (DATA) destined for stations (if available)
and asks (polls) all stations about the frames waiting for transmission by sending them the
service frame CF-POLL (invitation to transmit). The access point polls stations accordingly
to its polling list (polling table). The methods to form and keep up the polling table are not
specified in the standard thus the wireless network developers can freely decide on how to
realize it.
A station can transmit packets to the channel only when it receives the CF-POLL. Having
received the CF-POLL, the station sends the short frame containing both data (if available)
and acknowledgement of CF-POLL reception after SIFS. If there are no data to transmit, the
station answers with NULL frame containing the header only. If the access point gets the
answer frame, it waits for SIFS and polls the next station. Otherwise if there is now answer
during PIFS, the access point considers the station unaccessible and polls the next station.
In order to cut expenses, an access point can combine the CF-POLL with data transmission
(the frame DATA+CF-POLL). Similarly, the stations are allowed to combine acknowledgement
frame with data transmission (the frame DATA+CF-ACK). Under the PCF, there are four types
of a frame:
• DATA, data frame;
• CF-ACK, acknowledgement frame;
• CF-POLL, poll frame;
• DATA+CF-ACK, combined frame of data and acknowledgement;
• DATA+CF-POLL, combined frame of data and poll;
• DATA+CF-ACK+CF-POLL, combined frame of data, acknowledgement and poll;
• CF-ACK+CF-POLL, combined frame of acknowledgement and poll.
Along with the extended distributed function EDCF in the IEEE 802.11e protocol, the Hybrid
Coordination Function (HCF) is determined. Similarly as the EDCF is extension of the DCF,
the HCF extends the PCF.
Due to the fact that the PCF and HCF realize centralized non-collision priority access to
transmission medium, they completely solve the problem of hidden stations and provide QoS.
But regardless of their advantages, the PCF and HCF are harder to be realized then the DCF.
Besides, the point coordination functions imply relatively large number of service frames (CF-
POLL, etc.) that essentially increases overhead expenses of data transmission in the wireless
medium. To organize the wireless network with PCF or HCF, all stations support these regimes
and one station serves as an access point while the DCF allows organizing the ad-hoc network
without access point. The distributed coordination functions are reasonable to be used in the
simplest wireless networks without hidden stations and delay-sensitive applications.
2.4 Problem of hidden stations in wireless metropolitan area networks

In IEEE 802.11 broadband wireless network, the basic structure unit is a radio-cell having star-
shape structure: the center is a base station with omnidirectional antenna which antennae of
all subscriber stations are focused on presenting the radio bridges between wireless network
and local cable networks.
In typical conditions of the wireless metropolitan area network, the subscriber stations do
not have a radio visibility for each other (they are hidden from each other) and have to
communicate through the retransmitting base station disposed on the high altitude (tower
buildings, TV tower, etc.) and providing an access to the outer network. Thus, the broadband
wireless metropolitan area networks have the mesh structure: a mesh is a radio cell and
subscriber stations are connected with the high-speed backborn network.
Below we consider the operation of a radio cell in the broadband wireless metropolitan area
network.
A radio cell has the following features:
• all subscriber stations are hidden from each other and their antennae are focused on the
base station, i.e. data transmission between subscriber stations is possible through the
base station only;
• the distances between the base station and subscriber ones are large enough (several
kilometers) and different.
Each of the IEEE 802.11 protocols supports its own set of transmission rates. For IEEE 802.11b,
they are the channel rates 1, 2, 5.5 and 11 Mbit/sec, for IEEE 802.11a they are 6, 9, 12, 18, 24, 48
and 54 Mbit/sec. A rate is a type of modulation which modulate the carrier radio signal or its
part in the system with several carriers. The "faster" modulations are less noise-resistant, and
otherwise. For this reason, the IEEE 802.11 based devises use more noise-resistant modulations,
or lower transmission rates in cases of the signal depression and large number of losses when
transmitting frames. It is obvious that a frame of n bits is transmitted longer by rate 1 Mbit/sec
(hence it reserves a channel for a longer time) than by rate 11 Mbit/sec. In case when a
subscriber station is far off or is placed in a nasty place, the radio signal it receives from the
base station is weak, and the subscriber station automatically decrease the transmission rate.
It results in decreasing of the network throughput. In case of PCF, the situation gets worth
as it is impossible to control the subscriber station behavior. Remind, that accordingly to the
DCF and EDCF mechanisms, the subscriber station decides by itself when to transmit data.
Besides, regardless of data transmission directions within a radio cell, they come through the
base station, thus it is logical to impose the base station on the radio cell control responsibility.
It easy to see that due to the above-mentioned features, it is not reasonable to use the distributed
coordination functions in broadband wireless metropolitan area networks. Even though the
equipment supporting the centralized control is more complicated for development and pro-
duction hence it is more expensive, it much more efficiently uses two most valuable resources
of the broadband wireless network, they are frequency and throughput. The centralized
control in the MANs allows to avoid the problem of hidden stations, efficiently schedule an
order of station access to the wireless channel, flexibly control the radio cell operation and
change its parameters correspondingly to the current situation by adjusting only the base sta-
tion. The adaptive centralized control is a prospective direction of developing the IEEE 802.11
broadband wireless networks.
2.5 Adaptive dynamical polling in wireless networks

Since both the Point Coordination Function and Hybrid Coordination Function are based on
the centralized control (polling), the main attention should be focused on the method of the
centralized control realization. The specific polling mechanism and its parameters are the main
factors determining the efficiency of the broadband wireless MAN with centralized control.
The IEEE 802.11 standards allow the developers freely construct the methods of PCF and HCF
realization. The standard does not specify: the method to poll the subscriber stations, method
to construct and support the polling table, the policy to form queues of frames for transmission,
and recommendations on the relation between DCF and PCF intervals in a superframe. Further,
within the Section the main attention is given to development and modelling of the algorithms
to poll subscriber stations, the schemes of adaptive dynamic polling (ADP).
For each subscriber station, the base station forms a queue of frames. Within a queue, the
frames are ordered accordingly to the priority of traffic they belong to. Then the base station
starts the polling cycle. As the cycle starts, it polls the first station in the polling list (polling
table) sending frames from the corresponding queue to the station. Then the base station polls
it sending the CF-POLL (or QoS CF-POLL) and receives the frames to be transmitted from
the station (if available). As the base station finishes polling the subscriber one, it switches to
polling the next station accordingly to the polling table. Note that PCF leads to the overhead
expenses which can not be avoided during data exchange between the base station and the
subscriber one, they are the time to switch to subscriber stations and service frames for polling
them (CF-POLL, etc.).
The development the algorithm of adaptive dynamic polling lead to the following problems:
• choosing the method to poll the subscriber stations;
• choosing the policy to work with a subscriber station (receiving and transmission);
• development of methods to minimize the overhead expenses and;
• development of methods to optimize the system parameters.
Since the base station is substantially the central device in a radio cell of the broadband wireless
MAN, it is reasonable to provide it with the complete functionality concerning the radio cell
control, service of the queues, determination of the polling order and minimization of the
overhead expenses. For the equipment developers, it means that the base station has to carry
the majority of the network features. A subscriber station plays a passive role just answering
the service frames from the base station.
Efficiency of the broadband wireless PCF network performance considerably depends on the
method of polling mechanism realization. At the same time, the choosing of a cyclic polling
type and a queue service discipline is determined by the method of the broadband wireless
MAN (Metropolitan Area Network) application. Normally, there are two situations during the
broadband wireless MAN operating. In the first one, uplink traffic from the base station to end
stations dominates over the downlink one. Such a traffic is typical while using a broadband
wireless MAN cell as a "last mile" for Internet provider. In this case, each end station has its
own segment of LAN (Local Area Network) that gets an access to Internet via the wireless
network. The investigation of the real broadband wireless MAN (Vishnevsky, 2000) working
as the "last mile" shows that traffic from the outer network to the local segment is much higher
than the backward traffic.
In the second case, downlink traffic from an end station to the base one dominates over
the uplink one. Such a traffic is typical while using the broadband wireless network as
backbone network to transmit information from the objects "behind" the end stations to the
outer network (usually, to an information data storage center). The situation takes place when
the wireless network is used for the video monitoring systems, automatic process control
systems, telemetering, systems of collection, storage and processing of data.
Further we say a cell works in the "last mile" mode if the downlink traffic is dominate and it
works in the "data collection" mode if the uplink traffic is dominate.
The fundamental difference between two situations is that the base station (coordinator of the
wireless network cell work) knows the parameters of the queue of frames to be transmitted
to end stations in the case of dominating downlink traffic. Thus, the base station can choose
the policy to work with an end station queue before it polls the station. In the second case,
the base station knows nothing about the queues of frames to be transmitted when the uplink
traffic is dominate. Hence, it needs to poll the end station first and then to decide on the policy
to work with the queue.
When a cell works in the "data collection" mode we can neglect the data traffic from the base
station to the end ones considering just uplink traffic from the end stations to the base one.
In this case, the base station polls an end one first, that is connects to the end station and
starts transmitting. Then, the base station makes an attempt to switch to the next end station
accordingly to the polling table. Moreover, the base station does not know if the end station
will respond to the poll and if it will have the frames for transmission. In order to cut down
expenses of polling the empty end stations and stations that stopped working for some reason
we propose not to poll those stations at the next polling cycle.
Thus the rule to poll the end stations could be as follows. The base station polls a subscriber
one if it polled it in the previous cycle and the end station had frames for transmission or
it was skipped. In the case the end station did not respond to the base one in the previous
cycle or it was empty when being polled, it is skipped (not polled) by the base station in the
current cycle. Since the base station can not estimate the number of frames in the queue to
be transmitted in advance (before polling the end station) it is not reasonable to serve the
queue until it is empty. Thus, we propose the discipline to serve the end station: the base
station transmits only the frames which were present in the queue at the station polling epoch.
The adequate model to investigate the performance characteristics of the broadband wireless
network working in the "data collection" mode described above is the polling system with
adaptive polling mechanism and gated service analyzed in Section 3 where we presented the
analytical and simulation results.
When a cell works in the "last mile" mode, the data traffic from the end station to the base one
can be neglected. So we consider just downlink data traffic from the base station to the end one.
In this case, the base station sends frames from the queue to the corresponding end station.
Then, it waits for the successful data transmission acknowledgment from the subscriber station
(during PIFS) and starts sending frames from the next queue. Time to switch between queues
is random so it is impossible to say in advance if the data are transmitted successfully. If the
queue of frames is short it is obvious that data transmission costs are high. So, it is better
to serve the queue only if its length exceeds the given value called the threshold in order to
cut down the expenses. To maximize the system throughput, the queues of each end station
should be served until empty. But in the case when one or several queues are long enough,
the frame mean waiting time in the queue is large that is unacceptable for some network
applications. So, the most important system parameters are the mean queue length and the
frame mean waiting time. These parameters can be estimated by means of stochastic model
with exhaustive threshold service considered in Section 4.
3. Adaptive polling system

The polling systems are varieties of the queuing systems with multiple queues and one (or
more) server(s) common to all queues that poll the queues and serve the queued customers.
Classification of polling systems and methods to study them are presented in reviews (Levy
& Sidi, 1990; Vishnevsky & Semenova, 2006) and books (Takagi, 1986; Borst, 1996; Vishnevsky
& Semenova, 2006).
The adaptive polling mechanism assumes that the server polls not all queues in a cycle. The
queues that were empty at the previous polling epoch are skipped (not visited) by the server
and are visit in the next cycle only. Unfortunately, the exact analysis of adaptive polling
systems is cumbersome, and in this Section, we provide an approximate analysis.
We consider the polling system with N queues (having unlimited waiting space) and a single
server which is common for all queues. Each queue, say i, has a Poisson input of customers
with parameter λi . Service times at queue i are independent and identically distributed with
(r) ∞
the distribution function Bi (t) with the moments bi = 0 tr dBi (t), r ≥ 1, and Laplace-Stieltjes
∞
transform (LST) βi (s) = 0 e−st dBi (t), i = 1, N.
The server visits queues in cyclic order from queue 1 to queue N accordingly to an adaptive
scheme. A cycle is referred to as the time the server spends visiting (serving) queues 1 through
N. With adaptive mechanism, the server skips (does not visit) queues which were visited in
the previous cycle and were found empty at their polling moments. After being skipped, a
queue is always visited in the next cycle.
If queue i is polled (or visited) by the server, the switchover time is incurred with distribution
(r)
function Si (t) having the LST S̃i (t) and the moments si , r ≥ 1, r = 1, N. A polling moment
is referred to as a moment when the switchover time is finished and the server is ready to
start working at a queue. The service discipline is gated, i. e. the server serves only those
customers that presented at a queue at its polling instant. The customers arriving during the
queue service time will be served in the next cycle.
If N queues were sequentially found empty at their polling moments, the server takes a
vacation having the distribution function F(t) with LST ϕ(s), and moments ϕ(k) , k ≥ 1. After a
vacation, server starts working at the queue that he stopped polling at and went for a vacation.
The approach we use to investigate the model with several queues is based on the decom-
position of the polling system to separate queues, further analysis of a queue as a queueing
system with server’s vacations, and then, application of the obtained results to the system
with several queues. The analysis aims at deriving the first and second moments of the mean
waiting time in a separate queue and is partially based on the results obtained in (Sumita,
1988) for the queueing system with server’s vacations which are discussed in Section 3.1.
3.1 Analysis of a single queue

First, consider a single queue with server’s vacations. Within this subsection, we omit the
lower index denoting the queue number. We assume that the vacation time depends on the
fact whether the queue was empty at the moment the server finishes the previous vacation
or it was not. If not, the vacation time has distribution function H (t) with LST h(s) and the
moments h(k) , k ≥ 1. If the queue is empty when server finishes a vacation, the next vacation
period is distributed with the function H̃ (t) having the LST h̃(s), and the moments h̃(k) , k ≥ 1.
To investigate the queue, consider the Markov chain describing the queue state at the end of
vacations. Let tk be the kth epoch when vacation is finished, k ≥ 1, and itk be the number of
customers at the queue at the epoch tk . It is easy to see that the process itk , k ≥ 1, presents the
discrete-time Markov chain, and its one-step transition probabilities
pi, j = P{itk+1 = j|itk = i}, i, j ≥ 0,
have the form

j

(i)
p0, j = ỹ j , pi, j = al y j−l , i > 0, j ≥ 0,
l=0
where
∞ ∞ ∞
(i) (λt)l −λt (∗i) (λt)l −λt (λt)l −λt
al = e dB (t), yl = e dH (t), ỹl = e dH̃ (t), l ≥ 0,
l! l! l!
0 0 0
and B(∗i) (t) is the i-fold convolution of the distribution function B(t).
It can be shown that under the condition ρ = λb(1) < 1 fulfilled, the stationary state probabili-
ties exist
q( j) = lim P{itk = j}, j ≥ 0.
k→∞
These probabilities satisfy the following set of the balance equations, j ≥ 0,
∞ ∞ ∞ ∞
( j) (0) (λt) j −λt (λt)l −λt (∗i) (λt) j−l −λt
q =q e dH̃ (t) + e dB (t) e dH (t). (1)
j! l! ( j − l) !
0 i=1 0 0
By multiplying the equations (1) by the corresponding powers of z and summing them up,

it readily follows that the probability generating function (PGF) Q(z) = ∞ ( j) j
j=0 q z of the
stationary probabilities satisfies the functional equation
Q(z) = (Q(β(λ − λz)) − q(0) )h(λ − λz) + q(0) h̃(λ − λz), |z| ≤ 1. (2)
The equations of type (2) have been solved in (Sumita, 1988) on the base of the method to solve
the functional equations like (2) described in (Kuczma, 1968) and the corresponding result is
included into (Takagi, 1991), pp. 223-225.
Below, we briefly describe the solution of (2). Consider the sequence of functions η j (z), j ≥ 0,
which are defined recursively,
η0 (z) = z, η j+1 (z) = β(λ − λη j (z)), j ≥ 0. (3)

It was proven in (Sumita, 1988) that if ρ = λb(1) < 1, the sequence of functions η j (z), j ≥ 0,
converges uniformly to 1 as j → ∞ for all z, 0 ≤ z ≤ 1. By substitution η j (z) instead of z in (2),
we get
Q(η j (z)) = Q(η j+1 (z))h(λ − λη j (z)) + q(0) (h̃(λ − λη j (z)) − h(λ − λη j (z))).
Continuing the substitution for j = 0, 1, . . . , n − 1, we have

n−1
n−1
j−1

Q(z) = Q(ηn (z)) h(λ − λη j (z)) + q(0) (h̃(λ − λη j (z)) − h(λ − λη j (z))) h(λ − ληk (z)).
j=0 j=0 k =0
(4)
b

Here we assume that ck = 1 if b < a.
k =a
Now, let n → ∞ in (4). Using the normalization condition Q(1) = 1, we obtain
∞
∞
j−1

Q(z) = h(λ − λη j (z)) + q(0) (h̃(λ − λη j (z)) − h(λ − λη j (z))) h(λ − ληk (z)). (5)
j=0 j=0 k =0
Setting z = 0 at the latter equation, we get the relation for the probability q(0) :
 j−1
−1
∞
 ∞

 
q(0) = h(λ − λz( j) ) 1 − (h̃(λ − λz( j) ) − h(λ − λz( j) )) h(λ − λz(k) ) , (6)
 
j=0 j=0 k =0
where the quantities z( j) = η j (0), j ≥ 0, are given by the recursion
z(0) = 0, z( j+1) = β(λ − λz( j) ), j ≥ 0. (7)
Finally, the equations (1), (5)–(7) give the probabilities q( j) , j ≥ 0.

To obtain the performance characteristics of the system with adaptive polling scheme, we
need to derive formulas for the first three derivatives Q (1), Q (1), Q (1) of the PGF Q(z)
at z = 1, which can be found by using (5). The formula (5) involves the LSTs (h̃(λ − λη j (z))
and h(λ − λη j (z)), where the functions η j (z), j ≥ 0, are given by recursion (3). Thus, in order
to obtain the formulas for the derivatives Q (1), Q (1), Q (1) of the PGF Q(z), we have to
find the explicit expressions for the derivatives of the functions η j (z), j ≥ 0, at z = 1 :
1 − ρj
η j (1) = 1, η j (1) = ρ j , j ≥ 0, η j (1) = λ2 b(2) ρ j−1 , j ≥ 0,
1−ρ
1 − ρ2 j (1 − ρ j−1 )(1 − ρ j )
η j (1) = λ3 b(3) ρ j−1 + 3λ4 (b(2) )2 ρ j−1 , j ≥ 0.
1 − ρ2 (1 − ρ)(1 − ρ2 )
Using the formulas obtained above, we can calculate the derivatives of functions h(λ −
λη j (z)), j ≥ 0, at z = 1: h(λ − λη j (z))|z=1 = 1,
1 − ρj
(h(λ − λη j (z))) |z=1 = λh(1) ρ j , (h(λ − λη j (z))) |z=1 = λ2 h(2) ρ2 j + λ3 h(1) b(2) ρ j−1 ,
1−ρ
1 − ρj
(h(λ − λη j (z))) |z=1 = λ3 h(3) ρ3 j + 3λ4 h(2) b(2) ρ j−1 +
1−ρ
j−1 )(1 − ρ j )
1 − ρ2 j 4 (2) 2 (1 − ρ
+λh(1) ρ j−1 λ3 b(3) + 3λ ( b ) .
1 − ρ2 (1 − ρ)(1 − ρ2 )
Finally, after cumbersome simplifications, the formulas for the derivatives Q (1), Q (1),
Q (1) of the PGF Q(z) at z = 1 are obtained as
λh(1)
λ(h̃(1) − h(1) )
Q (1) = + q(0) , (8)
1−ρ 1−ρ
2 (2)
λ2 h ( 2 ) λ3 h(1) b(2) + (λh(1) )2 2ρ (0) λ (h̃ − h(2) )
Q (1) = 2
+ 2
+ q + (9)
1−ρ (1 − ρ)(1 − ρ ) 1 − ρ2

λ3 b(2) + 2ρλ2 h(1) )
+(h̃(1) − h(1) ) ,
(1 − ρ)(1 − ρ2 )
λ3 h(3) λ4 h(1) b(3) 3ρλ4 h(2) b(2) + 3ρ(1 + 2ρ)λ3 h(1) h(2)
Q (1) = + + + (10)
1 − ρ3 (1 − ρ)(1 − ρ3 ) (1 − ρ2 )(1 − ρ3 )
3 (3)
3ρλ5 h(1) (b(2) )2 + 3λ4 (1 + 2ρ2 )(h(1) )2 b(2) 6ρ3 (λh(1) )3 (0) λ (h̃ − h(3) )
+ + + q +
(1 − ρ)(1 − ρ2 )(1 − ρ3 ) (1 − ρ)(1 − ρ2 )(1 − ρ3 ) 1 − ρ3
λ4 (h̃(1) − h(1) )b(3) 3ρλ4 (h̃(2) − h(2) )b(2)

+ 3
+ +
(1 − ρ)(1 − ρ ) (1 − ρ2 )(1 − ρ3 )
3ρ2 λ3 (h̃(2) − h(2) )h(1) 3ρ(1 + ρ)λ3 (h̃(1) − h(1) )h(2)
+ + +
(1 − ρ2 )(1 − ρ3 ) (1 − ρ2 )(1 − ρ3 )

3ρλ5 (h̃(1) − h(1) )(b(2) )2 + 3λ4 (1 + 2ρ2 )((h̃(1) − h(1) )h(1) b(2) 6ρ3 λ3 (h(1) )2 (h̃(1) − h(1) )
+ + ,
(1 − ρ)(1 − ρ2 )(1 − ρ3 ) (1 − ρ)(1 − ρ2 )(1 − ρ3 )
Let ζ be the duration of the queue service (between two successive server’s vacations). Using
the relations obtained above, we can get the moments ψ̂(r) , r = 1, 2, 3 of ζ. The number
of customers served between two successive server’s vacations is a random variable, say τ,

which has the first three moments L1 = Q (1), L2 = Q (1) + Q (1), L3 = Q (1) + 3L2 − 2L1 ,
and the service time of the kth customer in a service period is a random variable, say ξk , which
τ

has the moments b(r) , r = 1, 2, 3. It is easy to see that ζ = ξk . The random variables ξk are
k =1
mutually independent and independent of τ. Thus, ζ equals to the sum of a random number
of the random variables. Using the technique of the conditional expectations, it can be shown
that the moments ψ̂(r) , r = 1, 2, 3, of the random variable ζ are
ψ̂(1) = b(1) L1 , ψ̂(2) = b(2) L1 + (b(1) )2 (L2 − L1 ), (11)

(3) (3) (1) (2) (1) 3
ψ̂ = L1 b + 3b b ( L2 − L 1 ) + ( b ) (L3 − 3L2 + 2L1 ). (12)
Denote by ψ(r) , r = 1, 2, 3, the conditional moments of the distribution of ζ, duration of the

queue service between two successive vacations, given that the queue is not empty. It is easy
to see that the conditional moments ψ(r) , r = 1, 2, 3, are expressed in the terms of the moments
ψ̂(r) , r = 1, 2, 3, as follows:
ψ̂(r)
ψ(r) = , r = 1, 2, 3.
1 − q(0)
Thus, the formulas (5)–(7) give the stationary distribution of q(i) , i ≥ 0, the number of cus-
tomers in the queue at the epoch the server finishes a vacation, and the formulas (8)–(10)
express the first three moments of that distribution. ∞
Let W (x), x ≥ 0, be the distribution of the waiting time in the queue, and w(s) = 0 e−sx dW (w)
be its LST. It is easy to see that
w (s ) = w(0) ( s ) + w(1) ( s ) + w(2) ( s ),
where w(0) (s) is the LST of the waiting time of an arbitrary customer arrived to the system
when the server was busy, w(1) (s) is the LST of the waiting time of an arbitrary customer
arrived to the system when the server was on an ordinary vacation (with distribution H (t)),
w(2) (s) is the LST of the waiting time of an arbitrary customer arrived to the system when the
server was on a special vacation (with distribution H̃ (t)).
Using the probabilistic sense of LST, one can verify the following formulas are valid
Q(β(λ(1 − β(s)))) − Q(β(s))

w(0) (s) = τ−1 h(s) ,
s − λ(1 − β(s))
h(λ(1 − β(s))) − h(s) (2)

w(1) (s) = τ−1 (Q(β(λ(1 − β(s)))) − q(0) ) , w (s) =
s − λ(1 − β(s))
h̃(λ(1 − β(s))) − h̃(s)
= τ−1 q(0) .
s − λ(1 − β(s))
These formulas lead to the following result.
Theorem 1. The LST w(s) of the waiting time is given by
τ−1
w(s) = (Q(β(s))(1 − h(s)) + q(0) (h(s) − h̃(s))).
s − λ(1 − β(s))
By differentiating w(s) at s = 0, we obtain the formula for the mean waiting time
v(2) λb(2) + 2ρh(1)

W (1) = + (13)
2v(1) 2(1 − ρ)
and for second moment W (2) of the waiting time

(2)
(2) v(3) λb(3) + 3λb(2) h(1) + 3ρh(2) (1) λb 2ρ2 h(1)
W = + +W + (14)
3v(1) 3(1 − ρ) 1−ρ 1 − ρ2
with v(k) = (1 − q(0) )h(k) + q(0) h̃(k) , k = 1, 2, 3.

3.2 Approximate analysis of the system with N queues

As it follows from (6)–(14), the key role in calculation of the performance characteristics of
the system considered is played by the form of the functions h(s) and h̃(s), the LSTs of the
distributions of vacation period after visiting an empty or non-empty queue, respectively. In
order to apply the results obtained for the system with single queue to the polling system
with N queues, we have to estimate those LSTs. It is easy to see that they depend on the time
the server spends visiting other queues, and are unknown. Below, we elaborate a procedure
of successive approximations to calculate the first and second moments of waiting time in
each queue on the base of the analysis given in Section 3.1. Note that the unknown LSTs h(s)
and h̃(s) of vacation duration are improved at each stage of the procedure. And below, we
elaborate such a heuristic procedure.
( j) (0) (r) (r)
In the procedure, we use the quantities zi , j ≥ 0, qi , ψi , r = 1, 2, 3, and Wi , r = 1, 2,
i = 1, N, given by formulas (6)–(14) where all LSTs and the corresponding moments are given
the subscript i, the number of the queue.
Since a queue being polled and found empty will not be visited at the next cycle and will be
polled again two cycles later, it seems reasonable to assume that the time when the server is
away from the queue is equal to the double intervisit time as if the queue was visited each
cycle (after being polled and found non-empty). Besides, we need to take into account the
fact that, if all queues are found empty, they are all visited in the cycle following the server’s
vacation. Thus, we assume that h̃i (w) = χ̃i (w)S̃i (w), where χ̃i (w) = (χi (w))2 + q̄i (ϕ(w +
λi (1 − βi (w)))ri (w) − χi (w)), the functions χi (s) are given by formula (17),
N
N

(0) (0) (0)
q̄i = qj , ri (w) = (q j + (1 − q j )ψ j (w))S̃ j (w).
j=1, ji j=1, ji
It follows that the moments of the distribution functions H̃ (t) and H (t) of a vacation period in
the system with single queue considered in Section 3.1 are defined as
(1) (1) (2) (2) (2) (1) (3) (3) (3) (2) (1) (2)
h̃i = χ̃i + si , h̃i = χ̃i + si + 2χ̃i si , h̃i = χ̃i + si + 3χ̃i si + 3χ̃i si ,
(l)
where χ̃i , i = 1, 3, are calculated as
(1) (1) (1) (1) (2) (2) (1) (1) (2) (2)
χ̃i = 2χi + q̄i (ϕ̂(1) + ri − χi ), χ̃i = 2χi + 2(χi )2 + q̄i (ϕ̂(2) + 2ϕ̂(1) ri + ri − χi ),
(3) (3) (1) (2) (1) (2) (3) (3)

χ̃i = 2χi + 6χi χi + q̄i (ϕ̂(3) + 3ϕ̂(2) ri + 3ϕ̂(1) ri + ri − χi ) ,
ϕ(r) , r = 1, 3, are given by
(2)
ϕ̂(1) = ϕ(1) (1 + ρi ), ϕ̂(2) = ϕ(2) (1 + ρi )2 + ϕ(1) λi bi ,
(2) (3)
ϕ̂(3) = ϕ(3) (1 + ρi )3 + 3ϕ(2) (1 + ρi )λi bi + ϕ(1) λi bi ,
(l)
and the moments ri are defined as
N

(1) (0) (1) (0) (1)
ri = (q j s j + (1 − q j )a j ),
j=1, ji
N

(2) (0) (2) (0) (2)
ri = (q j s j + (1 − q j )a j +
j=1, ji
N

(0) (1) (0) (1) (0) (1) (0) (1)
+(q j s j + (1 − q j )a j ) (qk sk + (1 − qk )ak )),
k=1, ki,k j
N
N

(3) (0) (3) (0) (3) (0) (2) (0) (2) (0) (1)
ri = (q j s j + (1 − q j )a j + 2(q j s j + (1 − q j )a j ) (qk sk +
j=1, ji k=1, ki,k j
N

(0) (1) (0) (1) (0) (1) (0) (2) (0) (2)
+(1 − qk )ak ) + (q j s j + (1 − q j )a j ) (qk sk + (1 − qk )ak +
k=1, ki,k j
N

(0) (1) (0) (1) (0) (1) (0) (1)
+(qk sk + (1 − qk )ak ) (qm sm + (1 − qm )am ))).
m=1, mi,m j,mk
Further, we discuss the problem of determining the LST h(s).

On the first step of the procedure, assume that each queue is visited (polled) in each cycle and
the server serves only one customer in a queue per a visit. From this assumption it follows
that the initial form of the LST hi (s) of the vacation (intervisit time for queue i) could be
σi
hi (s) =
s + σi
with
N

(1) (1)
(σi )−1 = (b j + s j ), i = 1, N.
j=1, ji
Then, the moments of the vacation period (intervisit time for queue i) are given by equations
(1) (2) (3)
hi = σ−1
i , hi = 2σ−2
i , hi = 6σ−3
i .
(r) ( j)
Using LSTs hi (s), i = 1, N, and the moments hi , r = 1, 3, we calculate the values of zi , j ≥ 0,
(0) (r) (r)
qi , ψi , r = 1, 2, 3, Wi , r = 1, 2, i = 1, N, by the formulas (6)–(14) for each queue i, i = 1, N.
Note that we can use the Lyapunov inequality to check the correctness of numerical calculation
(r)
of the moments. For instance, the values ψi , r = 1, 2, 3, have to satisfy the inequalities
(2) (1) (3) (1) (3) (2) 3

ψi ≥ (ψi )2 , ψi ≥ (ψi )3 , ψi ≥ (ψi ) 2 .
(r)
Having analyzed the moments ψi , r = 1, 2, 3, of queue i service period, we choose the form
of the LST ψi (s) as follows.
(2)
ψi 1
• If the value cψ = (1) is approximately equal to 1, we set ψi (s) = (1) , i.e. the
( ψi ) 2 1+sψi
vacation (intervisit time) is exponentially distributed.
• If cψ ≈ 1k , where k is some positive integer, we set

 k
 
 1 
ψi (s) =   ,
 ψi
(1)

1+s k
i.e. the vacation has Erlang-k distribution.

• If cψ > 1, we suppose that a vacation has hyper-exponential distribution, i.e.
(1) (2)
µi µi
ψi (s) = pi + ( 1 − pi ) . (15)
(1) (2)
µi +s µi +s
(1) (2) (r)

The parameters pi , µi , µi are calculated through values ψi , r = 1, 2, 3, in the
following way. First, calculate the values
(l)
ψi v3 − v1 v2 v1 v3 − (v2 )2
vl = , l = 1, 2, 3, f1 = , f2 = ,
l! v2 − (v1 )2 v2 − (v1 )2
(1) (2)
(1) 2 (2) 2 µi (µi v1 − 1)
µi = , µi = , pi = .
(2) (1)
f1 + f12 − 4 f2 f1 − f12 − 4 f2 µi − µi
If the inequalities v2 > v21 , v3 > v1 v2 , v1 v3 > v22 and f12 > 4 f2 , are fulfilled then we define
(1) (2)
the function ψi (s) from (15). Otherwise, the values µi , µi and pi are chosen to satisfy
the relations

(1) (1) (2) (1)
2(ψi )2 (2)
2ψ i
( 1 − p i ) − 2(1 − pi )pi (ψi − 2(ψi )2 )
pi , 0 ≤ pi ≤ 1, µi = ,
(2) (1) (2)
ψi 2(ψi )2 − ψi pi
(2)
(1) (2) (1) p i µi
ψi µi − 1 + pi > 0, µi = .
(1) (2)
ψi µ i − ( 1 − p i )

6pi 6(1−pi ) (3)
Note that the value pi should be chosen to minimize (1) 3 + (2) 3 − ψi for the better
( µi ) ( µi )
approximation (Kazimirsky, 2002).
• If cψ < 1 but cψ does not equal to 1k approximately, the vacation period distribution can
be approximated by the phase distribution.
• If cψ ≈ 0, we use the following form:
(1)
s
ψi (s) = e−ψi .
Now, suppose the LSTs ψi (s), i = 1, N, are known. Note that queues may have different forms
of the LST of intervisit times (server’s vacations).
Then, we make a simplifying assumption that the probability that queue i is found empty
at an arbitrary polling epoch is independent of the state of the other queues. Note that this
assumption may not be valid at all (e.g. if the other queues are empty, queue i gets more time
for its customers to be served, hence the probability that the queue is empty is greater then
in the case when all other queues are not empty). But this assumption makes it possible to
investigate the model with adaptive polling analytically. Thus, we can rewrite the form of the
LST hi (s) of vacation duration for queue i (its intervisit time) can be rewritten as
hi (w) = χi (w)S̃i (w), (16)
with
N

(0) (0)
χi ( w ) = (q j + (1 − q j )ψ j (w)S̃ j (w)). (17)
j=1, ji
The derivation of (16)–(17) is based on the probabilistic interpretation of LST as the probability
that a catastrophe from a Poisson input with parameter w does not occur during the period
considered. A catastrophe does not occur during a vacation time for the queueing system with
vacations corresponding to queue i if it does not occur during switchover time for this queue
(with probability S̃i (w)) and during the switchover and service periods for the rest of queues
in the current cycle (with probability χi (w)). The first term in (17) implies that a catastrophe
does not occur during the switchover time to queue j and the following service period with
probability 1, if the queue is not visited by the server, and with probability ψ j (w)S̃ j (w) if it
was.
Formulas (16) and (17) result in the relations for the moments of the intervisit time for queue i:
(1) (1) (1) (2) (2) (2) (1) (1) (3) (3) (3) (2) (1) (1) (2)
hi = χi + si , hi = χi + si + 2χi si , hi = χi + si + 3χi si + 3χi si ,
(18)
where
N
N

(1) (0) (1) (2) (0) (2)
χi = (1 − q j )a j , χi = (1 − q j )a j + (19)
j=1, ji j=1, ji
N
N

(0) (1) (0) (1)
+ (1 − q j )a j (1 − qk )ak ,
j=1, ji k=1, ki,k j
N
N
N

(3) (0) (3) (0) (2) (0) (1)
χi = (1 − q j )a j + 2 (1 − q j )a j (1 − qk )ak +
j=1, ji j=1, ji k=1, ki,k j
N
N
N

(0) (1) (0) (1) (0) (1)
+ (1 − q j )a j (1 − qk )ak (1 − qm )am , (20)
j=1, ji k=1, ki,kj m=1, mi,mj,mk
where, for m = 1, N,
(1) (1) (1) (2) (2) (2) (1) (1) (3) (3) (3) (2) (1) (1) (2)
am = sm + ψm , am = sm + ψm + 2sm ψm , am = sm + ψm + 3sm ψm + 3sm ψm .
Then using formulas (16)–(20) for the LSTs and the moments of the vacation duration (of
(0) (r) (r)
intervisit time for a queue), we recalculate the values qi , ψi , r = 1, 2, 3, Wi , r = 1, 2,
i = 1, N, by formulas (6)–(14) for all i, i = 1, N.
(0) (r)
The iterative procedure described above should be repeated until the values qi , ψi , r =
(r)
1, 2, 3, Wi , r = 1, 2, i = 1, N, calculated at the succeeding steps coincide with the necessary
(r)
accuracy. Thus, we get the moments, Wi , r = 1, 2, i = 1, N, of the waiting time in the queues.
3.3 Numerical results

Here we give numerical examples to illustrate how the algorithm works for polling systems
with various numbers of queues and traffic intensity comparing to the simulation results
obtained using the general-purpose simulation system GPSS World (Schriber, 1974). The
object of modeling was represented by a regional broadband wireless network consisting of
several devices operating with one base station. The rates of packet arrivals to the devices and
the rate of processing them are different. The devices are polled cyclically. Packet servicing
is gated, that is, only those packets are transmitted which were at the queue at the polling
moment. The input flows are assumed to be of the Poisson nature, and the times of packet
servicing and polling initialization are assumed to be exponentially distributed.
We assume for simulation that the system is in the stationary mode when at duplication of
the number of the packets passing through the system none of the comparison parameters
changes more than by 0.5%. In the experiments, more than three million of packets passed
through the system.
Average packet service time, switchover time, etc. in the numerical examples are taken from
the real IEEE 802.11a broadband wireless network under PCF mode with realistic packet sizes
and load levels. Thus, the obtained results correspond to such networks and mean waiting
times satisfy the real systems requirements.
Case N = 2. First, consider the symmetric system of two queues with the mean service times
(1) (1) (1) (1)
b1 = b2 = 0.311, the mean switchover times s1 = s2 = 0.091 and the mean time of
server’s vacation ϕ(1) = 0.005. The mean waiting time calculated by the Algorithm (column
"A"). Simulation results (column "S") and relative error of comparison (column "∆") are shown
in Table 1. The first column describes the customers input intensities and the corresponding
traffic intensities. Two last lines of the Table present the results obtained for different mean
server’s vacation times ϕ(1) given that λ1 = λ2 = 0.5.
A S ∆, %
λ1 = λ2 = 0.321, ρ = 0.2 0.289 0.268 7.8
λ1 = λ2 = 0.5, ρ = 0.311 0.392 0.358 9.5
λ1 = λ2 = 0.803, ρ = 0.5 0.659 0.601 9.7
λ1 = λ2 = 1.28, ρ = 0.8 1.73 1.93 10.4
ϕ(1) = 0.05 0.392 0.358 9.5
ϕ(1) = 0.1 0.417 0.384 8.6
Table 1. System with two queues
(1)
Case N = 3. Now, consider the case of three queues with symmetric service bi = 0.044,
(1) (1)
si = 0.1, i = 1, 3, ϕ(1) = 0.1. The mean waiting times Wi , i = 1, 3 obtained by using the
algorithm and simulation are presented in Table 2 for various customer input intensities. The
last two lines contain results for fully symmetric system (all λi , i = 1, 3 are the same).
Case N = 5. And, finally, consider the case of five queues with λ1 = 1, λ2 = 2, λ3 = 0.5,
(1) (1)
λ4 = 6, λ5 = 0.5, bi = 0.05, si = 0.05, i = 1, 5, ϕ(1) = 0.05. We vary the input intensity by
multiplying all λi by α which takes values 0.285, 0.714, 1, and 1.143. Thus, the traffic intensity,
ρ, varies from 0.2 to 0.8. The results are given in Table 3. The last four lines present results for
the fully symmetric system with λi = 2 multiplied by the same values of α.
A S ∆, % A S ∆, %
λ1 = 2.5 0.342 0.365 6.3 λ1 = 4.375 0.658 0.698 5.7
λ2 = 6 0.335 0.361 7.2 λ2 = 10.5 0.781 0.834 3.0
λ3 = 0.5 0.410 0.440 6.8 λ3 = 0.875 0.778 0.805 3.4
Symmetric system
λi = 3, i = 1, 3 0.387 0.382 1.3 λi = 5.25, i = 1, 3 0.702 0.771 8.9
Table 2. System with three queues
A S ∆, % A S ∆, %
(1) (1)
W1 0.203 0.220 7.7 W1 0.714 0.672 6.2
(1) (1)
α = 0.285 W2 0.199 0.215 7.4 α=1 W2 0.661 0.618 7.0
(1) (1)
ρ = 0.2 W3 0.205 0.222 7.7 ρ = 0.7 W3 0.776 0.738 5.1
(1) (1)
W4 0.197 0.203 3.0 W4 0.705 0.679 3.8
(1) (1)
W5 0.205 0.224 8.5 W5 0.776 0.745 4.2
(1) (1)
W1 1.036 0.974 6.4 W1 0.374 0.398 6.0
(1) (1)
α = 0.714 W2 0.353 0.370 4.6 α = 1.143 W2 0.967 0.934 3.5
(1) (1)
ρ = 0.5 W3 0.393 0.419 6.2 ρ = 0.8 W3 1.153 1.080 6.8
(1) (1)
W4 0.340 0.355 4.2 W4 1.152 1.120 2.9
(1) (1)
W5 0.393 0.422 6.9 W5 1.153 1.090 5.8
Symmetric system
(1) (1)
ρ = 0.2 Wi 0.207 0.216 4.2 ρ = 0.7 Wi 0.759 0.686 10.6
(1) (1)
ρ = 0.5 Wi 0.455 0.391 16.4 ρ = 0.8 Wi 1.003 1.040 3.6
Table 3. System with five queues
(1) (1) (1) (1)
The results for non-symmetric service (b1 = 0.07, b2 = 0.015, b3 = 0.1, b4 = 0.025,
(1)
b5 = 0.4) are given in Table 4.
Below, we discuss the results in brief. For the symmetric systems with N = 2 and N = 3,
the relative error of comparison grows as the traffic intensity ρ grows. But the situation is
different in case N = 5: the approximate and simulation results coincide up to 5% for low and
high traffic intensity, but for ρ = 0.5 and ρ = 0.7, the error becomes unacceptable (grater than
10%). For the system with five queues, the dependence of the relative error on the total traffic
intensity, or the traffic intensity to a queue, is not well-understood. In case of non-symmetric
system with five queues (Tables 3 and 4), the results can not be well explained. In case of
non-symmetric input of customers (Table 3), the coincidence gets better as ρ increases, but
when we make the service non-symmetric (Table 4), it holds only for queues with relatively
high traffic intensities, namely, queues 4 and 5, as long as the results for the rest of queues
behave the different way (the relative error grows as the values of ρi , i = 1, 3, grow). Note that
in Table 3, the results for queues 3 and 5 have to be the same as the queues are identical but
the simulation results differ up to 1%, which is the simulation error.
A S ∆, % A S ∆, %
(1) (1)
W1 0.251 0.250 0.4 W1 0.570 0.506 12.7
(1) (1)
α = 0.4, W2 0.248 0.244 1.6 α = 1, W2 0.535 0.475 12.7
(1) (1)
ρ = 0.2 W3 0.251 0.254 1.2 ρ = 0.5 W3 0.592 0.548 7.9
(1) (1)
W4 0.244 0.228 7.0 W4 0.516 0.455 13.4
(1) (1)
W5 0.223 0.254 12.2 W5 0.538 0.559 3.8
(1) (1)
W1 0.318 0.314 1.3 W1 1.016 0.902 12.6
(1) (1)
α = 0.6, W2 0.311 0.302 3.0 α = 1.4, W2 0.901 0.831 8.4
(1) (1)
ρ = 0.3 W3 0.322 0.325 0.9 ρ = 0.7 W3 1.095 0.994 10.2
(1) (1)
W4 0.305 0.281 8.5 W4 0.938 0.896 4.7
(1) (1)
W5 0.281 0.325 13.5 W5 1.082 1.080 0.2
Table 4. System with five queues and non-symmetric service
To complete the numerical analysis, we compare two polling schemes in the system working
in "data collection" mode described in Section 2.5. The comparison is based on the radio cell
model with parameters N = 4, input intensities λ1 = λ2 = 1500, λ3 = λ4 = λ take values
(1) (1) (1) (1)
from 1 to 200 with step 10, mean service times are b1 = b2 = b4 = 1/4500, b3 = 1/3000
(1) (1) (1) (1)
and the mean switch-over times are s1 = s2 = s3 = s4 = 1/1500. The obtained results
are shown on Fig. 1. The figure shows that adaptive dynamical polling can make significant
profit when some stations have light traffic, since the waiting times are the same for classical
cyclic polling system and one with adaptive polling in case of heavy traffic.
Fig. 1. The dependence of the mean waiting time on λ under the adaptive cyclic polling
4. System with exhaustive threshold service

In this Section, we present the polling system modelling a radio-cell working in the "last mile"
mode as described in Section 2.5. We consider the polling system described in Section 3, but
we assume that the service at a queue is exhaustive (the queue is served until empty).
The server visit a queue only if the queue length exceeds the given threshold (ki for queue i), ki ≥
0, i = 1, N. Before starting service at the queue i, the server needs a random time exponentially
distributed with the parameter si to prepare for work. This time can be considered the time
to switch to queue i which is incurred only if the queue is served. The queue also has a finite
waiting space hi (hi ≥ ki ).
The service time in queue i is exponentially distributed with parameter µi , i = 1, N. The queue
is served until it becomes empty, then the server moves to the nearest queue which has the
sufficient number of customers (having reached its threshold). If the numbers of customers in
all queues are insufficient to get service (less than ki for queue i), the server stops polling until
the number of customers in any queue reaches the threshold.
4.1 Stationary distribution of the system states and the performance characteristics
A system state at an arbitrary time t in a steady-state is presented by a random process
ξ(t) = (m(t), i(t), n(t)), t ≥ 0,
where m(t) = 0, if at time t the server is idle, m(t) = 1, if at time t the server switches to a
queue, m(t) = 2, if at time t the server serves the customers, t ≥ 0; i(t) is the number of queue
which server is attended at time t, i(t) = 0, if the server is idle; n(t) is a vector
n(t) = (n1 (t), n2 (t), . . . , nN (t)),
n j (t) is the number of customers at queue j at time t, j = 1, N.
The stochastic process ξ(t), t ≥ 0, is Markovian. Introduce the stationary state probabilities of
the process ξ(t), t ≥ 0 for r = (r1 , r2 , . . . , rN ), i = 1, N,
a(r) = lim P{m(t) = 0, i(t) = 0, n(t) = r}, 0 ≤ rm < km , m = 1, N,

t→∞
pi (r) = lim P{m(t) = 1, i(t) = i, n(t) = r}, rm = 0, hm , m = 1, N, m i, ri = ki , hi ,

t→∞
qi (r) = lim P{m(t) = 2, i(t) = i, n(t) = r}, 0, hm , m = 1, N, m i, ri 1, hi .

t→∞
The balance equations for the stationary probabilities are
N
N

λa(r) = λ j a(r − e j )I{r j >0} + µ j q j (r + e j )I{r j =0} , r1 < k1 , . . . , rN < kN , (21)
j=1 j=1
 N  N
 
 λm I {r <h } + µi
 q (r) =
 i λ j qi (r − e j )I{r j >δi j } + µi qi (r + ei )I{ri <hi } + si pi (r)I{ri ≥ki } , (22)
 m m 
m=1 j=1
 N  N
 
 λ I + s  p (r) = λ j pi (r − e j )I{r j >ki δi j } + µi−1 qi−1 (r + ei−1 )I{ri−1 =0} + (23)
 m {rm <hm } i  i
m=1 j=1
+λi a(r1 , . . . , ri−1 , ki − 1, ri+1 , . . . , rN )I{r1 <k1 ,...,ri−1 <ki−1 ,ri =ki ,ri+1 <ki+1 ,...,rN <kN } +
i−2

+ µ j q j (r + e j )I{r j =0,r j+1 <k j+1 ,...,ri−1 <ki−1 } +
j=1
N

+ µ j q j (r + e j )I{r j =0,r j+1 <k j+1 ,...,rN <kN ,r1 <k1 ,...,ri−1 <ki−1 } ,
j=i+1
0 ≤ rm ≤ hm , m = 1, N, m i, ki ≤ ri ≤ hi , i = 1, N.
By replacing one of the equations of system (21)–(23) by normalization condition
N
N

a(r) + qi ( r ) + pi (r) = 1,
r∈Λ i=1 r∈Πi i=1 r∈χi
with Λ = {(r1 , . . . , rN ) : rm < km , m = 1, N}, Πi = {(r1 , . . . , rN ) : 0 < ri ≤ hi , 0 ≤ rm ≤ hm , m =

1, N, m i}, χi = {(r1 , . . . , rN ) : ki ≤ ri ≤ hi , 0 ≤ rm ≤ hm , m = 1, N, m i}, i = 1, N, we get a
N N
system of equations for N i=1 (2hi − ki + 1) j=1, ji (h j + 1) + j=1 k j unknowns.
Having calculated the stationary state distribution, we can readily derive the following per-
formance characteristics:
1. Mean length of queue j at time when queue i is served (excluding a customer being

served): Lij = (r j − δi j )qi (r), i, j = 1, N;
r∈Πi

2. Mean length of queue j at time when the server switches to queue i: Sij = r j pi (r),
r∈χi
i, j = 1, N;

3. Mean length of queue j at time when the server is idle: U j = r∈Λ r j a(r), j = 1, N;

4. Mean length of queue j at arbitrary time: L j = N i i
i=1 (L j + S j ) + U j , j = 1, N;

5. The mean fraction of time the server is idle: ā = r∈Λ a(r).
Since the waiting space in the system is limited, some arriving customers can be lost. The
j
probability Plost that an arbitrary customer arriving to queue j is lost equals to the probability
that an arbitrary time all h j waiting places are occupied,
N
N

j
Plost = qi (r)I{r j =h j } + pi (r)I{r j =h j } , j = 1, N.
i=1 ij r∈Πi i=1 r∈χi
The mean waiting time in queue j can be obtained by Little’s law W j = Li /λi .
Consider the model of the broadband wireless network radio cell working in the "last mile"
mode with adaptive polling mechanism. The model parameters are the following: N = 4,
λ1 = λ2 = 1500, λ3 = λ4 = λ take the values 30, 60, 100. The service intensities µ1 = µ3 =
µ4 = 4500, µ2 = 3000, the intensities of switching between queues s1 = s2 = s3 = s4 = 1500.
The Fig. 2 shows the dependence of the frame mean waiting time on the threshold value ki = k
(i = 1, 4) assumed to be the same for all the queues. We see that the optimal value k varies
from curve to curve. The less the queues are loaded the grater the k minimizing the mean
waiting time. If the system has queues with low traffic it is reasonable to increase their service
thresholds. As a result, the frame mean waiting time in queues with low traffic increases
but it decreases for queues with high traffic so it can reduce the weighted sum of the mean
waiting times in the system. The optimal threshold choosing can be a troublesome problem in
practise since the threshold value depends on the relation between numbers of stations with
light traffic and heavy traffic, station parameters, etc.
Fig. 2. The dependence of the mean waiting time on the threshold
5. References
Borst, S.C. (1996). Polling Systems, Stichting Mathematisch Centrum, ISBN 90-6196-467-9,
Amsterdam.
IEEE Std 802.11-2007. Part 11: Wireless LAN Medium Access Control (MAC) and Physical
Layer (PHY) Specifications, June 2007.
Kazimirsky, A. (2002). On approximation of arrival flows with MAPs of order two, Proceedings
of IST’2002, pp. 35-40, Minsk, November 2002. (in Russian)
Kuczma, M. (1968). Functional Equations in a Single Variable, Polish Scientific Publishers, War-
saw.
Levy, H. & Sidi, M. (1990). Polling systems: applications, modeling and optimization. IEEE
Transactions on Communications, 38, 1750-1760. ISSN 0090-6778.
Schriber, T.J. (1974). Simulation Using GPSS, John Wiley & Sons.
Sumita, S. (1988). Performance analysis of interprocessor communications in an electronic
switching system with distributed control. Performance Evaluation, 9, 83-91, ISSN
0166-5316.
Takagi, H. (1986). Analysis of Polling Systems, MIT Press, Cambridge, ISBN 978-0262200578.
Takagi, H. (1991). Queueing Analysis: A Foundation of Performance Evaluation, Vol. 1, North-
Holland, ISBN 978-0444817709.
Vishnevsky, V.M. (2000). Wireless networks for broadband access to the Internet. Elektrosvyaz’,
10, 9-13, ISSN 0013-5771. (in Russian)
Vishnevsky V.M. & Frolov, S.A. (2009). Application for an invention No. 2007136908 of
03.02.2009 "Method of development of the very high throughput broadband wirelss
mesh networks".
Vishnevskii, V.M. & Semenova, O.V. (2006). Mathematical methods to study the polling
systems. Automation and Remote Control, 67, 173-220, ISSN 0005-1179.
Vishnevskii, V.M. & Semenova, O.V. (2007). Polling Systems: Theory and Applications for Broad-
band Wireless Networks, Moscow, Technosphera, ISBN 978-5-94836-166-6. (in Russian)
Vishnevsky, V.M.; Portnoy, S.L. & Shakhnovich, I.V. (2009). Encyclopaedia WiMAX. Way to 4G,
Moscow, Technosphera. (in Russian)
Next Generation Optical Access Networks: from TDM to WDM 537
25
X
Next Generation Optical Access Networks:

from TDM to WDM
Ll. Gutierrez1, P. Garfias1, M. De Andrade1,
C. Cervelló-Pastor1 and S. Sallent2
Technical University of Catalonia (UPC)1 and i2cat Foundation2
Spain
1. Introduction
Network infrastructure plays a key role in the success of added services and in user
satisfaction. It is widely accepted that Passive Optical Networks (PON) are the most
promising, cost-effective, and high-performance access network solutions. Access networks
are also commonly referred to as either ‘the last mile’ by the operators, or ‘the first mile of
the network’ in IEEE terminology. The term ’mile’ is often related to the path portion that is
used to reach the user from a network node, however access networks go as far as 20 km
depending on the technology used.
An access network comprises connections between different subscribers and a Central
Office (CO), which is attached to the metro or core network. The wired technology deployed
varies significantly from one country to another, i.e. Digital Subscriber line (xDSL), based on
copper wires; Hybrid Fiber-coax (HFC), and optical fiber.
The trends for Next Generation Access Networks (NGA) based on PON are: Wavelength
Division Multiplexing (WDM), 10 Gb/s or more, and longer reach/higher splits. NGA must
be able to cope with challenges, such as delivering diverse broadband services and
facilitating the integration of different technologies. Moreover, NGA should provide higher
bandwidths or further reduce the cost of existing delivering services, and serve as backhaul
of wireless access networks (WiFi, WiMAX). The last one constitutes a relevant issue for the
convergence between wireless and wired technologies.
1.1 Optical access networks - Fiber-to-the –x

Single-mode fiber’s properties, such as low loss and extremely wide inherent bandwidth,
make it the ideal candidate to meet the capacity challenges for today’s and for the
foreseeable future. This kind of fiber is often used in core and metropolitan networks, and
nowadays their penetration in the access domain is increasing as well (Koonen, 2006).
The Passive Optical Network (PON) is the most interesting solution, basically because there
is no active equipment installed in the field – a very satisfying feature for incumbent
operators, and also because the equipment and the feeder fibers are shared amongst users.
The application of PON technology to provide broadband connectivity to subscribers in the

access network is called fiber–to-the-x (FTTx). Depending on how deep the fiber penetrates
into the first mile, FTTx can be classified as: FTTB (Business/Building), FTTC/Cab
(Curb/Cabinet), FTTN (Neighborhood/Node), FTTP (Premises), and FTTH (Home). There
is no final agreement in the terminology to be used all over the world, but the most common
terms in the market are: FTTN/B and FTTH. The FTTN/B describes a PON where the fiber
arrives directly from the CO to the building (or business premises) and the signal is
converted and carried to the user dependencies by using copper or coaxial cables. Figure 1
depicts a FTTN/B, where the optical signal is converted to electrical signal in the very last
mile of the network, and then is carried by means of a vDSL network up to the subscriber’s
premises.
NT3
ONU 1
NT2
λd: downstream
NUCLE OX PLUS
ENC
S C N LAN RDSI 1 RDSI 2 DTE1 DTE2 DTE3 DTE4 DTE5 DTE 6
ONU i NT1
Splitter
Metro
OLT
core
λu: upstream ONU j

Metropolitan/ CORE VDSL
Network network
10 – 40 Km ONU 64 1,5 Km
Fig. 1. FTTN/B network
FTTH refers to the reach of the fiber wire till the subscriber premises. It consists of an
Optical Line Termination (OLT) at the service provider’s CO, and a number of terminals
near the end-user’s device called Optical Network Unit (ONU) or Optical Network Terminal
(ONT). PON can use single or multiple fibers for upstream and downstream traffic, with or
without WDM, being a single fiber-single channel tree the most common topology
[Figure 1].
The downstream channel is a broadcast channel, and the upstream channel is often shared
amongst users’ devices – their access must be arbitrated in order to avoid collisions amongst
stations. Each ONU transmits a burst of data that cannot be interfered by any other burst
sent from another ONU. For that reason, users’ bursts are separated by a configurable
guard-time. Such single full-duplex channel-single fiber solutions are commonly referred to
as TDM-PON.
The worldwide market reflected that, by the end of 2008, PON networks, mainly FTTB/N
and FTTH, were deployed on a large scale in Asia (78.5%) - APAC countries - and North
America (13.5%). The market in Europe was far behind that of the APAC countries or
America, scoring about 8%. However, there are many initiatives that aim to deploy PON
networks all around and it is expected that such deployment will even grow faster in a short
period of time in Europe. (Lakic & Hajduczenia, 2007).
2. Legacy TDM-PON
Two main FTTH-PON standards are currently deployed around the world: Ethernet-based
PON (EPON), approved in 2004 (IEEE 802.3ah, 2004); and Gigabit-capable PON (GPON),
born in 2003 (ITU-T G.984.[1-4], 2003). In both solutions, the equipment installed in the field
is fully passive, covering distances of up to 20 km, using a point-to-multipoint topology,
providing wider bandwidth to the end user, and allowing video broadcasting (digital
and/or analogue). The split-ratio (users per fiber) is variable – commonly, an average of 32
to 64 users.
In fact, the basic physical features for both standards are very similar – what makes them
more different is the MAC protocol and the data encapsulation scheme. While EPON carries
bursts of pure Ethernet frames, GPON encapsulates data using Generic Encapsulation
Method (GEM). Fair and efficient bandwidth allocation to users is a key issue that is out of
the scope of the standards.
2.1 The EPON standard: IEEE 802.3ah - Ethernet in the First Mile
The EPON is a point-to-multipoint network, whose topology is basically a single fiber-single
channel tree. The bit rate is symmetric, up to 1 Gb/s – therefore, the maximum bandwidth
allocated to each ONU is typically around 70 Mb/s, though it depends on the number of
active ONUs and the users’ traffic profiles. The EPON standard focuses on defining the
physical and the MAC layer as shown in Figure 2. In (Kramer, 2005) the author explains
such standard deeply.
Upper Upper
layers layers
OSI Model
LLC LLC
Application OAM (optional) OAM (optional)
Presentation MPMC MPMC

Multipoint MAC Control Multipoint MAC Control
Session MAC MAC

Media Access Control Media Access Control
Reconciliation Reconciliation
Transport
Network OLT GMII ONU (s) GMII
Data Link PCS PCS

FEC FEC
Physical (optional) (optional)
PMA PMA
PMD PMD
MDI MDI
F. O.
Fig. 2. architecture model of an EPON
Ethernet frames cannot be segmented before their transmission in an EPON. Therefore, the
transmission of a frame will be deferred to the next cycle in case it does not fit in the current
assigned timeslot. The guard-time between bursts and the segmentation-less feature has a
considerable impact on the network performance.
EPONs are composed by two types of active terminals: the OLT and the ONU. The signal is
split by a passive optical splitter at the far end of the network. It is also possible to cascade
the splitters, but it would lead to additional power losses. In such cases, it is important to
consider that the optical power budget (power loss across the network path) should not be
overtaken. In general, it can be said that the more passive devices taken in the network
trunk, the less split ratio and/or the less distance covered.
2.1.1 The MAC layer

The MAC Control Layer is divided in two sublayers: the MPCP (Multi-Point MAC Control)
and the Reconciliation sublayer as depicted in Figure 2.
The MPCP is devoted to control the access to the upstream channel amongst the subscribers,
perform a discovery process, and allocate bandwidth to each ONU. MPCP is a request-
permit protocol. Two main control messages are defined:
 GATE: sent by the OLT to the particular ONU, indicating the time and the
bandwidth allocated to it
 REPORT: sent by the ONU to the OLT, indicating the queues occupation in order to
request bandwidth for the next cycle
In each cycle, every ONU is polled, so that a REPORT message must be sent to the OLT even
if its queues are empty and no bandwidth is requested. Moreover, the OLT periodically
sends control messages to discover new ONUs aiming to join the network – the so-called
‘autodiscovery’ mode.
The algorithm to allocate the bandwidth to each ONU is an important issue to consider.
However, attention must be paid to the fact that such algorithm is not defined by the
standard – it is instead left open to the implementer. Section 5 is devoted to a brief survey
on such algorithms, i.e. the scheduling framework and the allocation computation itself. The
efficiency of such mechanism impacts directly the QoS perceived by the user.
2.2 The GPON standard. Gigabit in the access network

GPON was developed by the Full Service Access Network (FSAN) group. It is somehow
based on the former ATM access networks (APON, BPON), but GPON’s data encapsulation
(GEM) is more generic, and accepts different network protocols, such as ATM, Ethernet and
IP.
The GPON is a point-to-multipoint network as well, with two types of active terminals: the
OLT and the ONT/ONU. The user’s equipment is called ONT or ONU if no users are
directly connected to the device. Finally, the network path itself is known as ODN (Optical
Device Network) and it is usually integrated by fiber and a passive optical splitter.
2.2.1 The MAC layer

The basic transmission unit is called T-CONT (Transmission Container). The bandwidth is
guaranteed by allocating timeslots to the ONU in order to transport the T-CONT of each
communication. The bandwidth allocation algorithm is also of the request-permit type and
it is performed at the OLT. The ONU requests bandwidth each cycle, and the OLT allocates
the guaranteed transmission window in the cycle to each active ONU. Two operations
modes are possible:
 SR (Status Reporting)-DBA, where the ONU requests bandwidth to the OLT;
 and NSR (Non Status Reporting)-DBA, where the OLT monitors the incoming
traffic flows but no information is sent to the OLT from the ONUs.
2.2.2 Service provisioning

The GPON standard specifies the services supported more accurately. They match very
closely those defined in ATM networks:
 Asymmetric: Digital broadcast services, VOD, file download, etc.
 Symmetric: e-mail, file exchange, distance learning, telemedicine, etc.
 Synchronous: POTS, ISDN and circuit emulation (E1, T1). Such service is typically of
the narrow band but more strict and time bounded.
2.3 Comparison amongst both standards

A summary of both standards is shown in Table 1. There is a common concern that the ATM-
oriented technology – BPON, GPON – performs very well when the traffic is of the real time
and emulation service, i.e. T1/E1 type; while Ethernet-oriented networks perform better when
the traffic is mostly composed by pure data applications, i.e. Internet. Nonetheless, it is not so
simple to make a definitive statement about the performance, mainly because the data
collected depends on many parameters, and more importantly, on the implementation.
Reference (Hajduczenia et al., 2007) concludes that in comparable system set-ups, GPON
performs slightly better than EPON, but it is not yet a definitive statement.
Item ITU G.984 IEEE 802.3 ah

Full services
MAC Service Ethernet data
(Ethernet, TDM, POTS)
Layer
Frame GEM frame Ethernet frame
Distance 10 / 20 km (logical: 60 km) 10 / 20 km
Split ratio 64 (logical: 128) 64 max.
Upstream 155 Mb/s, 622 Mb/s, 1.25
1.25 Gb/s
(bit rate) Gb/s
Downstream
PHY 1.25 Gb/s, 2.5 Gb/s 1.25 Gb/s
(bit rate)
Layer
Bandwidth Same as above (NRZ coding) 1Gbit/s (8B10B coding)
Opt. Loss 15 / 20 / 25 dB 15 / 20 dB
Down : 1480-1500 nm Down : 1480-1500 nm
Wave-length
Up : 1260-1360 nm Up : 1260-1360 nm
Efficiency 95% 89%
Table 1. Comparison between legacy GPON and EPON
From the table above, it is important to remark that the power budget limits drastically the
range of the network and the split ratio, which is typically 64 or less in both standards.
2.4 Summary
EPON is a natural extension of the LAN systems – it bridges the gap between the LAN and
Ethernet based MAN/WAN structures. GPON, on the other hand, uses a novel
encapsulation mechanism, GEM. It is important to emphasize that the GPON has the ability
to fragment and reassemble frame fragments, including Ethernet frames. The upstream
format shown by both standards is depicted in Figure 3(a) and (b).
User 1 User 1
1 1 1 1
1
ONU 1 ONU 1
OLT
User 2
User 2
2 2
Splitter
2
OLT ONU 2
ONU 2
1 2 3
1 1 2 3 3 3
User 3
GEM Frame n User 3
Frame n 1
ONU
Ethernet 3 Frame n Frame n+1 ONU 3
ONU 1 ONU 3 Ethernet Ethernet 3 3 3
ONU 1
3 3 3
ONU 1
(a) (b)
Fig. 3. Upstream channel transmission (a) GPON. (b) EPON
Optical devices determine the resulting system price. EPON hardware parameters are very
relaxed, and therefore are more cost effective. On the other hand, GPON systems, due to
their more strict hardware requirements, are more expensive.
Moreover, EPON networks are widely deployed in the APAC regions – by the end of March
2007 there were approximately 8 million subscriber ports and 16 million CO port capacity
deployed, while GPON networks was mainly on the trial phase (Hajduczenia et al., 2007). In
contrast, other reports predict that GPON will overtake EPON as the pre-dominant
technology in a short or mid term. Some equipment manufacturers have announced the
introduction of a new family of GPON integrated access device (IAD) semiconductors that
are expected to offer high levels of integrations and better throughput performance.
The growth rate of PON deployments are of the order of 3 to 4 million subscriber ports
every 6 months, mainly in the APAC countries (Japan, Korea, China, etc.); GPON is
expected to grow first in Europe.
3. Next Generation Access networks (NGA)

Technological advances, especially in optical transmission devices, boost the upgrading of
current TDM-PON to the NGA. Benefits expected are (Heron et al., 2009):
 High Capacity, up to several gigabits per seconds
 Increased reach, allowing to include more homes and/or to reduce the number of
COs
 Wireless and wireline integration, with wireless nodes deployed deeper in the
network
 Operation improvement, broadband services or evolved existent services

 Moreover, NGA are more cost-effective than the current ones, specially because
CAPEX and OPEX are significantly reduced.
The 10G-EPON standard (IEEE 802.3av) was approved on September 2009, whereas the
schedule for the 10G GPON is to be approved in 2010.
3.1 Next Generation 10G EPON. IEEE 802.3av

The next-generation TDM-EPON is upgraded up to 10 Gb/s by the IEEE 802.3av standard.
The standard pursues the objectives listed below:
 To support subscriber access networks using point-to-multipoint topologies.
 To standardise two different single mode (SM) fiber data rate channels: symmetric -
10 Gb/s both down and up; and asymmetric - 10 Gb/s in the downstream channel
and 1 Gb/s in the upstream channel.
 To have a BER better than 10-12 at the PHY service interface.
 To define up to 3 optical power budgets that support split ratios of 1:16 and 1:32,
and distances of at least 20 km.
 To maintain a complete backward compatibility with legacy standards.
The goal is to upgrade the channel capacity for both upstream and downstream channels
gracefully, while maintaining the logical layer intact, taking advantage of the already
existing MPCP and DBA agent specifications. Moreover, 10G-EPON keeps on utilizing the
analog video delivery systems before such delivery shifts gradually to an IP-based
distribution system.
3.1.1 The 10G EPON. Physical layer

In this section, the main enhancements of the PHY layer are briefly explained. Table 2
illustrates the acronyms used to designate the new EPON power budgets:
Power Budget Power Range

Downstream Upstream Split -ratio
class insertion loss (Km)
PR10 20 dB (low) 10 Gb/s 10 Gb/s 1:16 10
1:16 (1) 20 (1)
PR20 24 dB (medium) 10 Gb/s 10 Gb/s
1:32 (2) 10 (2)
PR30 29 dB (high) 10 Gb/s 10 Gb/s 1:32 20
PRX 10 20 dB (low) 10 Gb/s 1 Gb/s 1:16 10
1:16 (1) 20 (1)
PRX20 24 dB (medium) 10 Gb/s 1 Gb/s
1:32 (2) 10 (2)
PRX30 29 dB (high) 10 Gb/s 1 Gb/s 1:32 20
Table 2. 10G EPON acronyms and Power Budget classes
Notice that the physical-medium-dependent (PMD) and Power-Budget-Class (PBC) naming

nomenclature for 10 Gb/s EPONs, is different than those for legacy EPON. Table 3
illustrates the differences between legacy TDM- EPON and 10G EPON in detail.
3.1.2 Wavelength allocation

The downstream 1 Gb/s and 10 Gb/s data streams will be WDM overlaid thus creating
indeed two independent P2MP domains.
 The 1 Gb/s downstream link remains centered at 1490 ± 10 nm
 The 10 Gb/s downstream link uses 1575-1580 nm wavelength band.
1G EPON 10G EPON

channel coding 8B10B 64B66B
- 10 Gb/s 10 Gb/s symmetric
Data rate (DS / US) 1 Gb/s 1 Gb/s symmetric
- 10 Gb/s 1 Gb/s asymmetric
Upstream (λ) 1260 – 1360 nm 1260 – 1280 nm
Downstream (λ) 1480 – 1500 1575 – 1580
# of PBC 2 3
Split ratio 1:16 1:16 / 1:32
FEC RSS (255,239) (optional) RS (255,223) (mandatory)
Table 3. Differences between 1G and 10G EPON
For the symmetric line-rate scenario, WDM channel multiplexing in the upstream channel is
discouraged due to increased cost for the ONU. The 10-Gb/s upstream channel will use the
optical window centered at 1310 nm, which is the currently the allocated window for 1 Gb/s
upstream channel.
The OLT should operate in a dual rate mode; therefore, an overlay dual stack structure will
need to be implemented from the PMD up to the Reconciliation sublayer. An OLT supporting
both downstream channels may multiplex the outputs of two transmitters using DWDM
coupler, while the optical filters at an ONU are tuned to receive one downstream wavelength.
3.1.3 The 10G EPON MAC layer. The MPCP protocol

The MPCP protocol is essentially the same in both 1G and 10G EPON networks. It should
also be considered that, for the coexistence of various line rates, the DBA in the OLT will be
responsible for scheduling not one but two mutually cross-dependent EPON systems,
sharing a single upstream channel but expecting only minor changes to the MPCP protocol.
As in legacy TDM–PONs, the DBA is out of the scope of the IEEE802.3av standard, and thus
left vendor-dependent.
3.2 Next Generation 10G GPON (NG-GPON)

The FSAN Group has been very active in upgrading the legacy GPON. The basic
requirement of NG-GPON is to offer higher capacities than GPON while maximizing the re-
use of protocols, components and infrastructure. It is mandatory to maintain the
compatibility and the coexistence with the legacy GPON systems already deployed. Such
upgrade will focus mainly on the physical layer - upgrading the rate to 10 Gb/s - but also on
the optimization of Ethernet service delivery.
3.2.1 Wavelength allocation

To reach such goal the ITU community first developed the standard G.984.5 (ITU-T G.984.5,
2007) to reserve wavelengths for the next-generation applications. Summarizing, the G.984.5
entails:
 Wavelength ranges to be reserved for future use. Service signals are overlaid via
WDM on an operating GPON system. Three optional enhancement bands (options
1, 2 and 3) are specified as depicted in Figure 4.
 Blocking filters to be supported at GPON ONT/ONU to ensure that next-generation
ONUs could be installed on currently deployed GPON, side by side with legacy
GPON ONT/ONU.
 GPON upstream wavelength reduction options, to free spectrum in the O band for
future services.
 And of course, allowing operators to gradually migrate from a working GPON
ONT/ONU to a NG-GPON ONT/ONU without disrupting existing customers.
EPON
US DS 1G DS 10G (L-band) 10G EPON
Video
US DS 1G EPON
US US DS 10G GPON
1320 1450 1560
Option 1-2 Option 3
CWDM
US
(G984.5)
1300 1400 1550 1625
1330 1450
US Option 1-1 Option 2 DFB-option
1415 1530 (G984.5)
1290 1560
1360
GPON 1550
EB (EB video)
GPON – EB
1565 (FP-option)
US
EB (EB digital)
1480
1360 1539
US Reserved DS BPON
1480
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
1271 1291 1311 1331 1351 1371 1391 1411 1431 1451 1471 1491 1511 1531 1551 1571 1591 1611 1631 1651 1671
O-band E-band S-band C-band L-band U-band λ/nm
Fig. 4. Joint wavelength allocation proposals of 10G EPON and 10G GPON
3.2.2 10G GPON evolution

Two evolution stages will compose GPON evolution: in the first step, NG-GPON1 (expected
in year 2010) will be fully compatible with legacy GPON. Its two expected goals are:
 A GPON supporting 10 Gb/s downstream, and 2.5 Gb/s upstream (referred to as
XGPON1).
 A GPON supporting symmetric service, 10 Gb/s downstream, and 10 Gb/s
upstream (referred to as XGPON2).
 A WDM option to overlay multiple GPONs and/or point-to-point overlays with
different wavelengths (i.e. WDM) on the same fiber infrastructure.
NG-GPON1 stage will be compliant with G984.5, therefore GPON legacy ONTs/ONUs could
be replaced one by one by new 10G GPON ONTs/ONUs or even added to an existent ODN.
Some voices propose a WDM-PON as an alternative solution to avoid the technical hurdles
of the NG-GPON. Among others, WDM-PON has clear advantages because it combines the
traditional PON low infrastructure cost, with the use of “colorless” optics, thus reducing
management complexity.
Finally, the second step called NG-GPON2 is expected to come later (by 2015). It aims to
develop a new GPON (probably disruptive) that does not require coexistence with Legacy
service. Other tehnologies may be considered, like higher line rate, DWDM, OFDM, and
others (Kani et al., 2009).
3.3. 10G EPON/GPON convergence

Finally, the xPON industry supports the development of a framework that will allow 10G
PON standard networks to converge. Such goal is extremely interesting in order to reduce
costs, thus increasing the 10G PON deployment worldwide.
The convergence may be achieved in two ways:
 Firstly, by aligning the 10G optics of an ITU NG-PON system with the optical layer
specification of the emerging IEEE 10G EPON. In fact, optical layer specifications
now being used by IEEE are more aligned with ITU optical budget rules than ever.
 Secondly, it will be also possible to develop an open extensible interface to the MAC
control channel – for instance, an enhanced interface might be added to the 802.3av
systems that allows the convergence with the ITU standards.
4. New Trends: the WDM-PON and Long-reach PON

The new Internet media services, such as symmetrical HD real time applications,
videoconferencing, and broadcast, among others, followed by the new type of
communications, like point-to-point or multipoint-to-point, increase drastically the end-user
bandwidth demand. This bandwidth has an exponential growth that should cope with
deployments of next generation optical access networks. The evolution of legacy PON
technologies (GPON/EPON) should provide huge network resources and cost-effective
solutions to fulfill the new user applications and network provider demands.
Requirements of Long-Reach NGA are listed below:
 Extend the geographical reach to a minimum of 100 Km
 Increase the split ratio up to 128 or more, reducing the cost per subscriber
 Increase the downlink and uplink throughput (evolution to 10 Gb/s)
 Be transparent or compatible as much as possible with current legacy PONs
 Reduce the CAPEX and OPEX (operation, deployment, and maintenance).
Finally, it is also desirable for NGAs to be deployed in a reasonable temporal horizon, 2010
to 2015, using available technologies (lasers, APD, amplifiers, new modulation schemes).
Several proposals are being launched by the standardization bodies (mainly ITU and IEEE),
as well as research projects and position research papers; however, few approaches cover
the previous requirements in their totality. The Long Reach Optical Access Network
(LROAN) objective is to increase the reach of Optical Access Network under a common
umbrella. It was proposed as an extension of the PLANET ACTS project, developed in the
late 90s with a split ratio of 2048 users over 100 Km at 2.5 Gb/s downstream rate, and 311
Mb/s upstream rate (Van de Voorde et al., 2000).
4.1 LR-PON based on the legacy GPON

The recent standard G984.6 (ITU-T G.984.6, 2008) defines a new enhancement of the legacy
GPON by extending the optical budget, and thus allowing the deployment of a longer reach
and a higher split ratio. The increase of the reach in GPON is achieved using mid-span
optical amplifier extenders, transponders or single side extenders.
The amendment allows a mid-span reach extension by adding an active device to the
passive ODN (Regenerator or Optical Amplifier) in the fiber link, between the OLT and
ONT, achieving a longer range of up to 60 km [see Figure 5].
UNI
ONT
R/S 10G (G.984.6)
UNI
extension
ONT S’/R’ R’/S’ S/R SNI
R/S
ODN Mid-span OTL OLT

UNI extender
ONT R/S
Optical Optical
Distribution Trunk line
UNI Network
R/S
ONT
Fig. 5. GPON extension architecture (ITU-T G.984.6, 2008)
The remote OLT is connected with the mid-span extender through an Optical Trunk Line
(OTL). In the same way, the ONUs are connected to mid-span using an optical distribution
network (ODN) through the R/S and S’/R’ interfaces. Reach extended (GPON-RE) adds a
long point-to-point trunk line (OTL) to a point-to-multipoint GPON ODN. GPON-RE avoids
metro network equipment, installing the OLT in the remote central office. Such proposal
reduces the number of central offices, and at the same time it applies the same split
independently of their topology and geography, simplifying the OAM operation.
Other approaches should maintain compatibility with existing OLTs to the maximum possible
extent. The extension of the reach of the PON or alternatively the split ratio of the PON is
penalized by adding an active device, thus converting the existent PON into an active PON.
The reach extender must be compatible with existing GPON 2.4/1.2 Gb/s rated ONT
equipment and ODN (power budget class B+). In the past, however, the extender
supporting a more capable ODN had already been defined as class C+ in the amendment
G.984.2.
The mid-span can be built with SOA amplifiers operating in the O band (1310 nm) and the S
band (1490nm). New devices such as wavelength filters and optical filters will be used to
extend the reach of the OTL up to 50 km and to increase the loss budget up to 54 dB. Optical
amplifiers, however, are expensive and not transparent – therefore, a 3R regenerator is
needed. A 3R-RE is composed by OLT and ONT transceivers: it re-amplifies, re-shapes and
re-times the signal. Nevertheless, such solution is not compatible with the legacy GPON.
4.2 Long Reach WDM PON

WDM-PON is the application of wavelength-division multiplexing that uses individual
wavelength for each PON network. ONUs have light sources at different tuned wavelengths
coexisting in the same fiber, increasing the total network bandwidth and the number of
users served in the optical access network. Related to communications mode, the WDM-
PON may use point-to-point communications, point-to-multipoint (like EPON/GPON trees
by each wavelength), or hybrid solutions. In the point-to-point, no dynamic bandwidth
allocation mechanisms are needed. The point-to-multipoint uses a WDM/TDM, achieving
high resource utilization efficiency.
At the same time, the ONU/ONT in a WDM-PONs are classified as: colorless or colored. In
the former one, the optical user terminal is wavelength-seeded from the remote OLT located
in the central office, using the same wavelength path for downstream and upstream
channels. In this case, the upstream optical flow is modulated using FSK, inverse return-to-
zero (RZ), or intensity modulation (IM) (Shea & Mitchell, 2009), (Martinez et al., 2008). Such
device is the preferred one because network management is gracefully reduced.
The splitter is replaced by a wavelength selective filter implemented with an arrayed
waveguide grating (AWG) when the ONU is color-sensitive and the communication mode
is of the point-to-multipoint type.
Long reach WDM-PON is feasible using low-loss AWG – in this case, the link budget is 28
dB, the splitting ratio is up to 64, and the reach increases to up to 80 km. The reference
(Mukherjee, 2006) explains largely the current WDM technology.
4.3 Long Reach DWDM-CDM PON

DWDM-CDM PON combines the code and wavelength-division multiplexing, achieving
ultra-long range due to coding gain, high bandwidth and bidirectional transmission on the
same wavelength and with a single fiber.
Typical figures are 16 λ DWDM-CDM PON, with a power budget of 42 dB at 100 km, with
32 orthogonal codes (32 users).
Some of this long reach schemes provide high split ratio, long reach -greater than 60 km-
and high symmetrical and asymmetrical bandwidth (10/2.5 Gbps) (Iwamura et al., 2007).
4.4 Current research in WDM / Long-reach networks: a survey

A proposal for a long-range architecture was implemented by the ACTS (Advanced
Communications and Technologies and Services) project, called Photonic Local Access
NETwork (PLANET) (Van de Voorde et al., 2000). The splitting factor was 2048 with a span
of 100 km in the PLANET project. The span comprehends a maximum feeder length of 90
km and a drop section of 10 km. The transport system supported on this SuperPON
architecture was based on an asynchronous transfer mode (ATM) cell. A bit rate of 2.5 Gb/s
was distributed to the optical network units (ONU) by time-division multiplexing (TDM) in
the downstream direction, whereas a time-division multiple access (TDMA) protocol was
used to share the 311 Mb/s upstream bit rate. In order to compensate the fiber and splitting
ratio losses, some optical amplifiers were housed in optical repeater units (ORUs) located at
the feeder section and between the feeder and the drop sections.
Another related proposal on LR-PON is the SuperPON architecture based on GPON by
British Telecom. This is a GPON over 135 km consistent with the standards of ITU-T. The
channels are 2488 Gb/s downstream and 1244 Gb/s upstream, using ∼1490 nm and 1552,924
nm wavelengths. Advanced 40-λ dense-wavelength-division-multiplexing (DWDM)
equipment is used to extend the physical reach and to provide fiber gain. Each wavelength
can support a split of 64 (1 × 8 followed by 1 × 8), so that a fully populated system could
support 2560 ONUs. The combined loss of the splitters and the last 10 km of the fiber is 23 dB.
The Hybrid DWDM-PON (Shea & Mitchell, 2009) by University College presents the
extension in the reach toward 100 km and an upstream bit rate of 10 Gb/s. It can potentially
support 17 TDM PONs operating at different wavelengths – each with up to 256 customers,
giving an aggregate number of 4352 customers in total. It uses 100-GHz channel spacing,
and divides the C-band into two, with one half (1529–1541 nm) carrying downstream
channels and the other (1547.2–1560.1 nm) carrying upstream channels.
Further implementations using DWDM in the backhaul to increase the fiber efficiency are
demonstrated in the EU project PIEMAN (Shea & Mitchell, 2007). In this architecture, the
network has a 100-km reach with a 32 wavelength DWDM backhaul. Each 10 Gb/s
wavelength is uniquely allocated to a PON with a 512-way split, enabling the network to
support (32 × 512) up to 16,384 users with an average bandwidth of ~20 Mb/s. By using
dynamic bandwidth allocation and 10 Gb/s-components in the ONU, it is possible for each
user to burst at 10 Gb/s. [see table 4]
Project Reach (Km) #λ DS/UP (Gb/s) # ONTs

PLANET 100 1 2.5/0.311 2.048
Super-PON 135 40 2.5/1.25 2.560
Hybrid PON 100 17 10/10 17*256=4.352
PIEMAN 100 32 10/10 32*512=16.384
SUCCESS 25 4*16 1.25/1.25 4*16
SARDANA 100 >1 10/10 1000
Table 4. Long Reach-PON projects
One of the most promising recent WDM-PON network, is the so-called SUCCESS network
(Kazovsky et al., 2007). The SUCCESS-HPON architecture is based on a topology consisting
of a collector ring and several distribution stars connecting a central office (CO) and several
optical networking units (ONUs). It uses Coarse WDM (CWDM) and dense WDM (DWDM)
technologies. Each ONU has its own dedicated wavelength for both upstream and
downstream transmissions on a DWDM grid to communicate with the OLT. The
communication is a half-duplex communication – the tunable transmitters at the OLT are
used for both downstream and upstream modulated frames by ONU’s. Furthermore, a
scheduling algorithm has been developed to keep track of the status of all shared resources
and arrange them properly in both time and wavelength domains, including the control for
both tunable transmitters and tunable receivers. The research is also focused on the
evaluation performance of two scheduling algorithms: 1) batching earliest departure first
(BEDF); and 2) sequential scheduling with schedule time framing (S3F).
Lastly, there are also important investments in optical technologies in Europe. The current
FP7 framework of the CE launched in 2007 funds several projects related to optical
technologies. Amongst others, is interesting to consider the project called Single-fiber
Advanced Ring Dense Access Network Architecture (SARDANA) – its goals are quite
ambitious: up to 1024 users per PON, 10 Gb/s data rate, remote passive amplification and
wavelength-agnostic customer equipment. Finally, it also aims to score well in traffic

balancing, as well as being highly scalable and allowing cascading (Prat, 2008).
5. Dynamic Bandwidth Allocation (DBA) and QoS provisioning in EPON
One of the main challenges of a TDM-PON is to schedule the transmissions and the
bandwidth allocation in the upstream shared channel efficiently. The issue of developing
appropriate scheduling algorithms was an important topic of research in the past. Goals of
the scheduler are: be efficient; support the QoS of each traffic flow requirements; allocate
fair bandwidth to users reducing delay and jitter; and finally, it must be computationally
simple enough.
To guarantee the efficiency and scalability of EPON in terms of resource management,
numerous contributions have been presented. There are two main strategies: the fixed
bandwidth allocation (FBA), and the dynamic bandwidth allocation (DBA). The first one
allocates the same transmission slots to every ONU in every service cycle. It is a simple
scheme but it does not perform optimally. On the contrary, the dynamic policy allocates the
transmission in the upstream channel based on each ONU’s requested bandwidth,
consequently the dynamic scheme provides a more fair, efficient and flexible bandwidth
allocation.
5.1 DBA algorithms legacy EPON network

The DBA of the EPON should acomplish QoS requirements to deliver different services,
such as multimedia traffic. DBA algorithms proposed so far for EPONs deal with different
criteria and can be categorized as shown in Figure 6.
Bandwidth
Allocation
DBA Grant Scheduling

computation Sizing Framework
Centrilized Distributed Fixed Dynamic Online Offline
Prediction Non-
oriented Prediction
Linear Constant
Elastic Gated Limited
Constant Credit
Fig. 6. Bandwidth Allocation Criteria
DBA algorithms may be either centralized or distributed. Besides, the dynamic grant sizing
–bandwidth allocation- might be based on a prediction oriented or non-prediction oriented
approaches.
Some examples of algorithms prediction-oriented are: (Hee-Jung Byun et al., 2003),

(Yuanqiu Luo & Ansari, 2005), (Yongqing Zhu et al., 2006) and (Lannoo et al., 2007). Finally,
authors in reference (Kramer et al., 2002b) proposed different DBAs explained in the IPACT
section below.
On the other hand, the scheduling framework determines the way scheduling decisions are
made. There are two main frameworks to consider: online and offline scheduling. With
online scheduling, the OLT makes scheduling decisions “on-the-fly” based on individual
requests and without global knowledge of the network. On the other hand, offline
scheduling requires a full knowledge of the network status, thus its scheduling decisions are
computed after having received the requests from all of the ONUs. (McGarry et al., 2006).
In many settings, the online scheme performs better than the offline scheduling, but with
less control of channel transmission times.
5.2 Centralised vs. distributed scheduling

The OLT computes the bandwidth allocation in the centralised scheduling, which is the
most common approach. On the other hand, the distributed approach contemplates the
participation of both OLT and ONUs. Bandwidth allocation is calculated by ONU though it
is also authorized by OLT.
In what follows, we present a description of a centralized DBA, the Interleaved Polling with
Adaptive Cycle Time (IPACT), and a distributed DBA, Dynamic Distributed Scheduler for
EPON (DDSPON), and a performance comparison among them.
5.2.1 IPACT
The IPACT is one of the early works that became very popular in the literature (Kramer et
al., 2002b). The cycle period adjusts to the bandwidth requirements of the ONUs, and the
definition of a maximum transmission window does not allow ONUs with high traffic level
to monopolize the bandwidth resource.
ONU Bytes RTT

ONU Bytes RTT
ONU Bytes RTT
1 550 200
1 6000 200
1 550 200
2 3200 170 2 3200 170
2 5700 170
3 1800 120 3 1800 120
3 1800 120
Tx 6000 3200 1800
t
OLT
Rx 6000 550 3200 5700
Tx 6000 550
ONU 1
6000
Rx
Tx 3200 5700
ONU 2
Rx 3200
Tx 1800 4400
ONU 3
1800
Rx
Fig. 7. Interleaving Polling mechanism in IPACT (Kramer et al., 2002b).
What is more interesting in this proposal is that IPACT uses an interleaved polling
approach, in which the next ONU is polled before the transmission of the previous one is
finished in order to utilize the efficiently the channel, as depicted in Figure 7.
The IPACT grant sizing is performed using five different alternatives: fixed, limited, gated,
constant credit, linear credit and elastic. Summarizing, the prediction-oriented DBAs are:
constant credit, linear credit and elastic. The credit approach - constant or linear - grants the
ONU’s requested bandwidth plus an extra amount of bandwidth; while the elastic approach
basically limits the maximum cycle time. The rest of the options – fixed, limited and gated-
are non prediction-oriented DBAs. The limited approach allocates no more than a
predefined amount of bandwidth to an ONU; and the gated one grants the requested
bandwidth without any limitation. Finally, the authors demonstrate that the limited
discipline is more efficient than the gated one and it has been the most preferred one to
compare with in the literature.
5.2.2. DDSPON
The DDSPON (De Andrade et al., 2007) is a DBA developed by some of the authors of this
chapter. This DBA requires an MPCP extension, because some extra information must be
supplied and therefore carried in control messages – mainly the weight vector (). This data
vector allows ONUs to compute its transmission window size. Such parameter represents a
proportional weight set up according to each ONU’s guaranteed bandwidth agreement. The
ONU computes the required bandwidth (Ri) and its current weight i , and then reports
such value to the OLT in a report message. The interleaving polling mechanism is applied
here as well as IPACT does [see Figure 8].
Cycle n-1 Cycle n

Cycle n+1
Tx R,Φ1 R,Φ2 R,Φ3 R,Φ1
OLT t
Rx G,Φ G,Φ G,Φ
Tx R,Φ1 R,Φ1
ONU 1
Rx
Tx R,Φ2
ONU 2
Rx
Tx R,Φ3
ONU 3
Rx
Fig. 8. DDSPON Polling Mechanism.
The average transmission window size of ONUi is computed by the equation below:
Φi
Wi = WMAX ; [N : of ONUs] (1)
∑
N
j =1
Φj
where WMAX is the maximum transmission window size (bits) that corresponds to the
maximum cycle time (TMAX).
WMAX = TMAX * Upstream rate (2)
The DDSPON process is as follows:
 The OLT receives a report messages from ONUi along the cycle n containing the
window size (Ri(n)) computed by itself as in equation 5, and the weight (Φi)
ONU1 ONU2 ONU3 …… ONUN

Φ1(n) Φ2(n) Φ3(n) …… ΦN(n)
 The OLT sends the Gate message to the ONU in the next cycle (n+1) including the
weight vector Φ (with weights of all other ONUs) and the time to start the ONU’s
transmission. Then, the ONUi transmits the data according to the value of Ri(n)
previously computed, and calculates the new values of Ri(n+1) and Φi(n+1).
Φ iconf
Wi (n + 1) = WMAX (3)
Φ iconf + ∑ j =1; j ≠ i Φ j (n)
N
And Ri(n+1) is computed
Ri (n + 1) = min (Wi ( n + 1), Qi ); [Qi : queue size] (4)
Finally, the new weight for next cycle n+1 is calculated based on the request value for cycle
n+1:
Φ i (n + 1) =
[
Ri (n + 1) Φ iconf + ∑ j =1; j ≠ i Φ j (n)
N
] (5)
WMAX
Notice that each ONU schedules the size of its transmission window dynamically. DDSPON
is executed in an online framework because the scheduling process is executed without the
need of waiting the reports from the rest of the ONUs. Moreover, by getting the weight
vector, each ONU is able to get an ‘idea’ of the rest of the ONUs’ loads, which is
characteristic in offline DBAs.
5.2.3. Performance evaluation

This section illustrates the performance evaluation of DBA algorithms. First we define the
power ratio as in equation (Kleinrock, 1975):
φmax φmax : throughput

P= ; (6)
D D : delay
The power ratio evaluates the DBA’s efficiency. Therefore the performance is evaluated by
computing: packet delay, average queue size and throughput for different setting scenarios
from low to high offered loads.
The comparison between DDSPON versus the IPACT was conducted by event-driven
simulations using the OPNET Modeler simulator. These simulations considered an ideal
channel and identical network parameters. To be more accurate, the distance parameter was
modified through the different scenarios of simulations from long (20 km) to short distances
(5 km), and the traffic model considered was self-similar in order to obtain more realistic
results.
The setting scenario consists of 16 ONUs with the same nominal weight (1/16), 1 Gb/s line
rate, 8 µseg. guard interval, large buffer size to avoid packet drops, and finally single traffic
class per ONU.
The main results obtained are presented below. Firstly, Table 5 below presents the average
queue size and packet delay for different offered loads and in the long distance scenario.
Average queue size (bytes) Average packet delay (ms)

Offered
IPACT DDSPON IPACT DDSPON
load
scenario 1 scenario 1 scenario 1 scenario 1
0,05 942 694 0,286 0,29025
0,2 1809 1890 0,373 0,41108
0,4 3292 3497 0,567 0,62216
0,6 7078 6766 1,050 1,00087
0,8 157538 59307 19,095 7,09164
1 425367 148205 41,337 14,3108
Table 5. Queue and delay size (average) for 20 Km
Figure 9a presents the average packet delay, and Figure 9b presents the average queue size
for different distances in a high loaded scenario (H=0.8). Notice that results provided have
important implications, for instance the IPACT requires larger buffer size to avoid the
packets loss than DDSPON.
Offered Load 0,8 Offered Load 0,8
1,6E+05
0,02
0,015 1,2E+05
Bytes
sec
0,01 8,1E+04
0,005 4,1E+04
0
6,4E+02
18<d<20 10<d<20 10<d<11 4<d<5
18<d<20 10<d<20 10<d<11 4<d<5
ONU-OLT Distance (km)
IPACT DDSPON
IPACT DDSPON
(a) (b)
Fig. 9. Average comparison between IPACT and DDSPON. (a) packet delay (b) queue size.
It is interesting to consider in the figures above that DDSPON presents an increase in both,
the average queue size and average packet delay, when the setting scenario is
heterogeneous, i.e. ONUs are far distributed between 10 and 20 km, this variation is roughly
about 21% higher (for loads of 0,8) than the results obtained in the homogeneous scenario.
Figure 10 clearly shows that IPACT’s average packet delay is worse than the DDSPON’s
one. The percentage of improvement of DDSPON over IPACT goes from 30.6% to 65.4% for
the different distance ranges.
70
60
50
40
%
30
20
10
0
18<d<20 10<d<20 10<d<11 4<d<5
offered load 0,8 offered load 1
Fig. 10. Average packet delay percentage difference: DDSPON over IPACT.
To conclude, the DDSPON remains stable versus IPACT with the variation of distances, and
the most remarkable is that DDSPON presents significant improvements versus the IPACT
in all simulations performed, being more relevant in highly loaded scenarios.
5.3 QoS provisioning in an EPON

The bandwidth allocated to each ONU is guaranteed by the DBA (inter-ONU scheduling),
but the QoS is guaranteed internally by the ONU, the so-called intra-ONU scheduling
process. The EPON follows the IEEE QoS policy defined in the standards IEEE 802.1P/Q.
There are up to eight sub-queues in each ONU depending on the traffic type; the intra-ONU
scheduling is of the strict type, hence, queues are served in order of priority. Such procedure
does not perform optimally in light load networks, so many algorithms, such as (Hsueh et
al., 2003), (Kramer et al., 2002a), (Jing Xie et al., 2004), (Kramer et al., 2004) or (Yuanqiu Luo
& Ansari, 2005) among others, propose to maximise intra-ONU scheduling.
According to the QoS policy, (McGarry et al., 2004) classifies DBA algorithms into two
categories: algorithms with statistical multiplexing (no QoS guaranteed), and algorithms
with quality of service. The latter is further separated into algorithms with absolute QoS
assurances, i.e. those that follow the integrated services paradigm; and algorithms with
relative QoS assurances, which provide different QoS levels according to the traffic classes,
i.e. differentiated services.
DBA algorithms introducing the support of differentiated services use strict priority
scheduling. Some of the examples developed in the past can be listed: (i) in (Choi & Huh,
2002), where bandwidth is allocated according to traffic priority requests; (ii) in (Kramer et
al., 2002a), which combines IPACT and strict priority queuing in order to support QoS; (iii)
(Assi et al., 2003) introduced an approach that consists of distributing the fairly excessive
bandwidth amongst the highly loaded ONUs. It takes also into consideration different
traffic classes, so that requested bandwidth consists of high, medium and low priority; and
(iv)in (Jing Xie et al., 2004), the authors propose the division of the frame, but in this case the
frame is divided into multiple subframes according to the different traffic classes, in order to
reduce the delay of high priority and medium priority classes; the size is variable depending
on the request, and through the definition of weights for each class, it is possible to avoid
bandwidth monopolization.
The aforementioned algorithms are the main contributions regarding DBA with
differentiated QoS support; more references can be found in (Zheng & Mouftah, 2009).
Furthermore, QoS contributions in DBAs are summarized in (McGarry et al., 2008).
5.4 Enhanced DBA algorithm for WDM-EPON networks

To provide higher bandwidth in PONs, a WDM technique can be performed incorporating
multiple wavelengths in either, the upstream or downstream direction so a WDM-PON has
many advantages such as increasing network capacity, in terms of bandwidth or user
scalability.
The new challenge for WDM-EPON is to allocate bandwidth to ONUs in both time and
wavelength domains, maximizing the whole network efficiency. Therefore, DBA algorithms
initially designated for EPON require modifications to exploit the multichannel architecture.
The bandwidth management problem can be split into two sub-problems: grant sizing
(bandwidth allocation) and grant scheduling (wavelength selection). Such algorithms
hereinafter are known as Dynamic-Wavelength and Bandwidth Allocation (DWBA).
Grant sizing is not analyzed anymore because any of the aforementioned DBAs may be
used. Instead, two main approaches cope with grant scheduling by: improving a former
DBAs, e.g. SIPACT (Clarke et al., 2006); or developing new mechanisms, for instance,
applying a well-known scheduling theory (Pinedo, 2002). The backward compatibility of the
MPCP is mandatory, but some extensions to the MPCP must be considered to deal with the
wavelength discovery and scheduling.
The reference (McGarry et al., 2008) introduces the concept of just-in-time (JIT) which is a
hybrid scheduling framework between offline and online. The OLT schedules the grant
based on the report messages accumulated since the last channel became available. The
ONUs that have not been allocated to a wavelength yet, are scheduled together across all
wavelengths as soon as a wavelength becomes available. The online JIT scheduling
framework gives the OLT more opportunity to make better scheduling decisions.
The simplest grant scheduling policy is to assign the next available supported channel
(NASC) to the ONU which means that the OLT must know which upstream channel will
first turn idle according to its polling table; such policy is not optimal in all cases and it does
not consider ONUs that support different wavelengths.
The approach of selecting the wavelength by using the scheduling theory seems much better
policy (McGarry et al., 2008), e.g. Shortest Path First (SPT), Longest Path First (LPT), and
Least Flexible Job First (LFJ) amongst others (Pinedo, 2002). LFJ first schedules transmissions
to the ONUs that support the fewest number of wavelength channels at the earliest available
supported channel. The LFJ policy is optimal because it minimizes under certain conditions
the length of the schedule.
Wavelength
Allocation
Scheduling Scheduling
Framework Wavelength
Policy
Online Hybrid NASC LFJ

Offline SPT ...
Online / Offline
Fig. 11. Wavelength allocation scheduling
Figure 11 classifies the scheduling framework and the scheduling wavelength policies
explained above.
5.4.1 Survey of DWBA

As an evolution of IPACT in WDM, some variants of IPACT are addressed in different
proposals, e.g. in (Kae Hsiang Kwong et al., 2004). The authors propose an algorithm so-
called WDM IPACT with a single polling table (WDM IPACT-ST). The grant scheduling is
of the NASC type and the grant sizing is performed according the IPACT (with fixed,
limited or gated approach). This approach requires new devices at both ends of the fiber
links to support simultaneous transmissions over multiple wavelengths.
In (Clarke et al., 2006), the authors developed a DBA called Simultaneous and Interleaving
Polling with Adaptive Cycle Time (SIPACT). SIPACT allows different architectures to poll
ONUs, either intra-wavelength (on the same wavelength) or inter-wavelength (amongst
different wavelengths), simultaneously but depending on the set of wavelengths supported
by each individual ONU.
The authors in (Dhaini et al., 2007) presented several DWBA variants also based on former
EPON DBAs algorithms and compared their performance. The three different approaches
compared depend on the weight of the individual ONU, the two more interesting ones
consider “on the fly” (online) mechanisms. Simulations performed showed that such
approach presents a better throughput and delay performance.
Finally, the recent proposal (McGarry et al., 2008) addressed the queuing delay and channel
utilization through the scheduling theory, which is concerned to scheduling a set of jobs
with specific processing times to be executed on a set of machines. In this case, the ONUs
represent the jobs, the grant size is represented by the processing time, and the channels are
represented by machines.
6. Conclusions
This chapter discoursed on optical fiber based access networks. While the current backbone
networks support high capacities; the last mile for the access network remains a bottleneck.
New enhancements of the legacy standards are coming soon allowing the upgrade the
upstream/downstream channel line rate to 10 Gb/s. Such new standards are intended to be
compatible with existent Legacy TDM-PONs.
But the more important step towards NGA is the implementation of the WDM technology.
There are huge research efforts worldwide in developing metro and access networks based
on WDM, using either DWDM or CWDM. The research is focused on improving the optical
devices as well as in developing new architectures to handle the multi-wavelength channel
efficiently. Such solutions are not only based on the type of network -such as WDM-PON or
10G PON- but also on hybrid technologies such those presented in section 4.
Moreover, incumbent operators are interested in the so-called Long Reach-PON (LR-PON)
that will help the growing process of PON deployment. LR-PON is a very cost-effective
solution because the CAPEX and the OPEX of the network are lower mainly due to the fact
that the number of equipment interfaces, network elements, and nodes are reduced, and
moreover, the network management complexity is also simplified.
The allocation of bandwidth to users in the upstream shared channel of the network is
addressed by appropriate DBAs. Among them we present IPACT and DDSPON, which are
representatives of centralized and distributed DBAs, respectively. The results provided
demonstrate that DDSPON is more efficient than IPACT. New DBAs for WDM-PON
networks are a key issue to manage and fairly distribute the resources. Proposals presented
in section 5, especially those based in the scheduling theory are very promising, but further
research should still be carried out.
Furthermore, the new goals are directed to support scalable networks and to help the
coexistence between legacy and next-generation PONs.
7. References
Assi, C.M.; Yinghua Ye; Sudhir Dixit & Ali, M.A. (2003). Dynamic bandwidth allocation for
quality-of-service over Ethernet PONs. IEEE Journal on Selected Areas in
Communications, Vol.21, No.9, pp. 1467-1477.
Choi, S. & Huh, J. (2002). Dynamic Bandwidth Allocation Algorithm for Multimedia
Services over Ethernet PONs. ETRI Journal, Vol.24, No.6, pp. 465-468.
Clarke, F.; Sarkar, S. & Mukherjee, B. (2006). Simultaneous and interleaved polling: an
upstream protocol for WDM-PON. Optical Fiber Communication Conference, 2006 and
the 2006 National Fiber Optic Engineers Conference. OFC 2006, pp. 3.
De Andrade, M.; Gutierrez, L. & Sallent, S. (2007). DDSPON: A Distributed Dynamic
Scheduling for EPON. IEEE International Conference on Signal Processing and
Communications, 2007. ICSPC 2007, pp. 840-843.
Dhaini, A.R.; Assi, C.M.; Maier, M. & Shami, A. (2007). Dynamic Wavelength and
Bandwidth Allocation in Hybrid TDM/WDM EPON Networks. Journal of Lightwave
Technology, Vol.25, No.1, pp. 277-286.
Hajduczenia, M.; da Silva, H.J.A. & Monteiro, P.P. (2007). 10G EPON Development Process.
ICTON '07. 9th International Conference on Transparent Optical Networks, Vol.1, pp.
276-282.
Hee-Jung Byun; Ji-Myung Nho & Jong-Tae Lim (2003). Dynamic bandwidth allocation
algorithm in Ethernet passive optical networks. Electronics Letters, Vol.39, No.13,
pp. 1001-1002.
Heron, R.; Storry C. & Patel S (2009). Future Directions in Fiber Access Networks.
Proceedings NOC, Valladolid (Spain).
Hsueh, ; An, ; Kim, & Kazovsky, (2003). A new media access control protocol with quality of
service and fairness guarantee in ethernet-based passive optical networks.
Proceedings of the 7th joint conference on information sciences, pp. 1392.
IEEE 802.3ah (2004). IEEE Standard for Information Technology. "Ethernet in the First Mile".
ITU-T G.984.[1-4] (2003). ITU - T Recommendations Gigabit-capable PON (GPON).
ITU-T G.984.5 (2007). ITU - T Recommendations Gigabit-capable PON (GPON).
ITU-T G.984.6 (2008). ITU - T Recommendations Gigabit-capable PON (GPON).
Iwamura, H.; Gupta, C.; Kashima, M.; Tamai, H.; Watanabe, R.; Ushikubo, T. & Kamijoh, T.
(2007). 42dB Loss Budget Hybrid DWDM-CDM-PON without Optical Amplifier.
Conference on Optical Fiber Communication and the National Fiber Optic Engineers
Conference, 2007. OFC/NFOEC 2007, pp. 1-3.
Jing Xie; Shengming Jiang & Yuming Jiang (2004). A class-based dynamic bandwidth
allocation scheme for EPONs. The Ninth International Conference on Communications
Systems, 2004. ICCS 2004, pp. 356-360.
Kae Hsiang Kwong; Harle, D. & Andonovic, I. (2004). Dynamic bandwidth allocation
algorithm for differentiated services over WDM EPONs. The Ninth International
Conference on Communications Systems, 2004. ICCS 2004, pp. 116-120.
Kani, J.; Bourgard, F.; Cui, A.; Rafel, A.; Campbell, M.; Davey R. & Rodrigues, S. (2009).
Next-Generation PON - Part I: Technology Roadmap and General Requirements.
IEEE Communications Magazine, Vol.47, No.11, pp. 43-49.
Kazovsky, L.G.; Wei-Tao Shaw; Gutierrez, D.; Ning Cheng & Shing-Wa Wong (2007). Next-
Generation Optical Access Networks. Journal of Lightwave Technology, Vol.25, No.11,
pp. 3428-3442.
Kleinrock, L. (1975). Queueing systems. John Wiley & Sons, 0471491101; New York.
Koonen, T. (2006). Fiber to the Home/Fiber to the Premises: What, Where, and When?
Proceedings of the IEEE, Vol.94, No.5, pp. 911-934.
Kramer, G.; Banerjee, A.; Singhal, N.K.; Mukherjee, B.; Sudhir Dixit & Yinghua Ye (2004).
Fair queueing with service envelopes (FQSE): a cousin-fair hierarchical scheduler
for subscriber access networks. IEEE Journal on Selected Areas in Communications,
Vol.22, No.8, pp. 1497-1513.
Kramer, G.; Mukherjee, B.; Dixit, S.; Ye, Y. & Hirth, R. (2002a). Supporting differentiated
classes of service in Ethernet passive optical networks. Journal of Optical Networking,
Vol.1, No.8, pp. 280-298.
Kramer, G.; Mukherjee, B. & Pesavento, G. (2002b). Interleaved Polling with Adaptive Cycle
Time (IPACT): A Dynamic Bandwidth Distribution Scheme in an Optical Access
Network. Photonic Network Communications, Vol.4, No.1, pp. 89-107.
Kramer, G. (2005). Ethernet passive optical networks. McGraw-Hill, 0071445625, New York.
Lakic, B. & Hajduczenia, M. (2007). On optimized Passive Optical Network (PON)
deployment. Second International Conference on Access Networks & Workshops, 2007.
AccessNets '07, pp. 1-8.
Lannoo, B.; Verslegers, L.; Colle, D.; Pickavet, M.; Demeester, P. & Gagnaire, M. (2007).
Thorough analysis of the IPACT dynamic bandwidth allocation algorithm for
EPONs. Fourth International Conference on Broadband Communications, Networks and
Systems, pp. 486-494.
Martinez, J.J.; Gregorio, J.I.G.; Lucia, A.L.; Velasco, A.V.; Aguado, J.C. & Binue, M.A.L. (2008).
Novel WDM-PON Architecture Based on a Spectrally Efficient IM-FSK Scheme Using
DMLs and RSOAs. Journal of Lightwave Technology, Vol.26, No.3, pp. 350-356.
McGarry, M.P.; Reisslein, M. & Maier, M. (2008). Ethernet passive optical network
architectures and dynamic bandwidth allocation algorithms. Communications
Surveys & Tutorials, IEEE, Vol.10, No.3, pp. 46-60.
McGarry, M.P.; Maier, M. & Reisslein, M. (2004). Ethernet PONs: a survey of dynamic
bandwidth allocation (DBA) algorithms. IEEE Communications Magazine, Vol.42,
No.8, pp. 8-15.
McGarry, M.P.; Reisslein, M.; Colbourn, C.J.; Maier, M.; Aurzada, F. & Scheutzow, M. (2008).
Just-in-Time Scheduling for Multichannel EPONs. Journal of Lightwave Technology,
Vol.26, No.10, pp. 1204-1216.
McGarry, M.P.; Reisslein, M. & Maier, M. (2006). WDM Ethernet passive optical networks.
Mukherjee, B. (2006). Optical WDM networks. Springer, 0387290559, New York.
Pinedo, M. (2002). Scheduling: theory, algorithms, and systems. Prentice-Hall, 0130281387,
Upper Saddle River.
Prat, J. (2008). Next-Generation FTTH passive optical networks: research towards unlimited
bandwidth access. Springer, 9781402084690, London.
Shea, D.P. & Mitchell, J.E. (2009). Architecture to integrate multiple PONs with long reach
DWDM backhaul. IEEE Journal on Selected Areas in Communications, Vol.27, No.2,
pp. 126-133.
Shea, D.P. & Mitchell, J.E. (2007). Long-Reach Optical Access Technologies. IEEE Network,
Vol.21, No.5, pp. 5-11.
Van de Voorde, I.; Martin, C.M.; Vandewege, I. & Oiu, X.Z. (2000). The superPON
demonstrator: an exploration of possible evolution paths for optical access
networks. IEEE Communications Magazine, Vol.38, No.2, pp. 74-82.
Yongqing Zhu; Maode Ma & Tee Hiang Cheng (2006). IPACT with Grant Estimation for
EPON. 10th IEEE Singapore International Conference on Communication systems, 2006.
ICCS 2006, pp. 1-5.
Yuanqiu Luo & Ansari, N. (2005). Bandwidth allocation for multiservice access on EPONs.
Zheng, J. & Mouftah, H.T. (2009). A survey of dynamic bandwidth allocation algorithms for
Ethernet Passive Optical Networks. Optical Switching and Networking, Vol. In Press,
Corrected Proof.
Building energy efficiency design for telecommunication base stations in Guangzhou 561
26
X
Building energy efficiency design for

telecommunication base stations in Guangzhou
Yi Chen, Yufeng Zhang and Qinglin Meng
State Key Laboratory of Subtropical Building Science, Department of Architecture, South
China University of Technology
China
1. Introduction
Telecommunication base stations (TBSs), which are the basis of the telecommunications
network, consume more energy than other public buildings due to their high inner heat
density and special operating schedule. The number of mobile phone users in China exceeds
500 million, and the telecommunications network has become the largest in the world,
consisting of more than 100,000 TBSs. The annual energy consumption of the network is 20
billion kWh, one third of which is used by TBSs. Guangzhou, as one of the fastest
developing cities in China, has more than one thousand existing TBSs and the total annual
electricity bills are in the tens of millions of RMB, 25% of which is the cost of air
conditioning. With the rapid growth of telecommunications, energy conservation for TBSs is
gaining greater attention in China.
Previous studies on energy conservation in TBSs mainly focused on improvement of
efficient air-conditioning systems (Nakao et al., 1988; Maeda et al., 2005; Choi et al., 2007),
indoor airflow optimization (Hayama & Nakao, 1989; Dan & Matti, 2000), use of renewable
energy (Makhkamdjanov, 2006), and other energy saving and monitoring techniques
(Schmidt & Shaukatullah, 2003). There are very limited studies on building energy efficiency
design of TBS. Building energy efficiency design, which is known as passive cooling
technology, is very popular in the traditional buildings and well inherited and applied in
the modern buildings in Southern China. Compared with the active one, such as air-
conditioning, passive cooling technology has notable advantages on utilizing natural
cooling capacities with no or few energy consumption. Many studies report building energy
efficiency designs of walls, roof, glazing, shading and natural ventilation for residential
buildings (Feng, 2004), institutional buildings (Athanassios et al., 2007) and high rise
apartments (Cheung et al., 2005), however, no such studied were performed on TBS
buildings. Nakao et al. (Nakao et al., 1988) studied a thermal control wall for TBS, which can
lose heat by using a two-phase loop-type thermo-siphon system integrated inside the wall.
The heat transmission coefficient of the wall was found to be one to ten times of the
ordinary wall and the annual energy saving was estimated to be 20%. The study proposes a
new energy efficiency design of wall for TBS, however, no other designs, such as shading or
natural ventilation were mentioned.
This chapter reports the study on building energy efficiency design for TBSs in Guangzhou.
Through field investigation of a typical TBS in Guangzhou, the basic information of TBS was
achieved, the key factors influencing energy consumption of TBS were determined and
several building energy efficiency designs were proposed. The effect on the annual cooling
load for each energy efficiency design was analyzed and a combined effect was achieved for
several combinations of the designs by building simulation. The building energy efficiency
design strategy for TBS in Guangzhou was concluded. It was found that ventilation design
is the primary choice for energy efficiency TBS building design. A new ventilation cooling
technology (VCT) was proposed. Based on the field investigation, the application feasibility
of VCT was studied systematically and the optimization of airflow organization for VCT
was analyzed.
2. Field investigation
2.1. Climate conditions in Guangzhou
Guangzhou is located at latitude 23°08’N and longitude 113°19’E. Summer is hot and humid
and winter is warm. The average air dry-bulb temperature is 28.4 °C in July and 13.3 °C in
January. The mean daily temperature variation is 7.5 °C. The relative humidity is about 83%
in summer and 70% in winter. The average annual rainfall is about 1705 mm of which 80%
falls between April and September.The long, hot and humid summer in Guangzhou creates
a huge demand for energy for cooling.
2.2. Field investigation methods

A typical TBS in Guangzhou was chosen for investigation. The TBS is located on the top of a
library, with dimensions of 4.66 m long × 4.66 m wide × 2.8 m high. The plan view of the
typical TBS is shown in Figure 1. The structure and conductive thermal resistance of the
building envelopes of the typical TBS are shown in Table 1. Two air conditioners are
installed for cooling with a temperature set point of 25 °C. The telecommunication
equipment and the air conditioners work continuously, 24 h per day and 365 days per year.
There is no window on the walls, mainly for the safety of the telecommunication equipment.
Fig. 1. Plan view of a typical TBS.

Building Conductive thermal

Structure
envelope resistance (m2K/W)
External cement/sand plaster (20mm)

Walls Lime-sand brick (180mm) 0.210
Internal gypsum plaster (20mm)
Polyurethane foam plastics (50mm)

Roof 1.376
External cement/sand plaster (20mm)

Floor Reinforced concrete (100mm) 0.104
Table 1. Structure and conductive thermal resistance of the building envelopes of the typical
TBS.
The thermal characteristics of the building envelopes, the heat dissipation of the
telecommunication equipments and the performance of the air conditioners were
investigated through physical measurements. The measurements were carried out during
the winter in Guangzhou from 18:00 h on January 8 to 13:00 h on January 10 (44 h in total).
Measurements taken included the interior and exterior surface temperatures of the walls,
roof and floor (Figure 2), the exterior surface temperature of the equipments, the velocity
and temperature of the air exhausted from the equipments (Figure 3), the ambient room air
temperature, and the return and supply air velocities and temperatures of the air
conditioners. The temperature was measured using thermocouples with an accuracy of
±0.5 °C and recorded by a data logger at 5 min intervals. Air velocity was measured by a
heated-sphere anemometer with a range of 0.1-30 m/s and an accuracy of ±5%. The energy
consumption of the TBS was measured using a power meter.
Fig. 2. The measuring positions on the building envelopes.

Outlet
Inlet 2
Inlet 1
Fig. 3. Telecommunication equipment measurements.
2.3. Field investigation results
2.3.1. Inner heat source

Heat loss of the telecommunication equipments is the main inner heat source for the TBS,
which can be transported by convective and radiative ways. The convective heat loss was
calculated by using the airflow rate, ambient room air temperature (i.e. inlet air
temperature) and outlet air temperature. The radiative heat loss was calculated by using the
exterior surface temperature of the equipments and the interior surface temperatures of the
envelopes. It was found that the total heat loss of the equipments was maintained constant
at a level of 4.35 kW, regardless of communication work loads or outdoor weather
conditions. The density of the inner heat source was about 200 W/m2, which is much larger
than the normal public buildings.
2.3.2. Thermal performance of building envelopes

Figure 4 shows the measured results of the south wall. Outdoor air temperature obtained
from the local meteorological observatory in Guangzhou is included in Figure 4 as well. The
interior surface temperature, mainly influenced by the indoor air temperature, was
comparatively stable (around 25 °C). The exterior surface temperature, affected by the
outdoor air temperature and solar radiation, fluctuated greatly with time from 12.5 to 37.5
°C. The mean outdoor air temperature was 12.8 °C and the highest was only 18.3 °C during
the measuring period.
40
35
exterior surface
30 tempera ture
Temperature(℃)
25 outdoor air
20 tempera ture
15 interior surfa ce
tempera ture
10
5
0
Time
1- 9 0: 00
1- 9 3: 00
1- 9 6: 00
1- 9 9: 00
1- 8 15: 00
1- 8 18: 00
1- 8 21: 00
1- 9 12: 00
1- 9 15: 00
1- 9 18: 00
1- 9 21: 00
1- 10 0: 00
1- 10 3: 00
1- 10 6: 00
1- 10 9: 00
1- 10 12: 00
1- 10 15: 00
1- 10 18: 00
Fig. 4. Temperatures of interior and exterior surface of the southern wall and outdoor air
temperature change with time.
It was decided to use the steady state heat transfer equation to make an initial observation
on the variation of the heat transfer amount and direction, rather than the dynamic
equation, which is relative complicated. The steady state heat transfer equation is:
Q  ( e   i) / R  S (1)
is the total heat transfer (W),  e is the exterior surface temperature (°C),  i is
where
Q
the interior surface temperature (°C), R is the conductive thermal resistance (m2 K/W) and S
is the area (m2) of the building envelopes. Based on the thermal characteristics (see Table 1)
and the measured temperatures, heat transfer of each building envelope was calculated and
shown in Figure 5.
1000
800
600 east wall
400
Heat transfer(W)
south wall
200
west wall
0
-200 north wall
-400 roof
-600 floor
-800
-1000
1-8 15:00
1-8 18:00
1-8 21:00
1-9 12:00
1-9 15:00
1-9 18:00
1-9 21:00
1-10 0:00
1-10 3:00
1-10 6:00
1-10 9:00
1-10 12:00
1-10 15:00
1-10 18:00
1-9 0:00
1-9 3:00
1-9 6:00
1-9 9:00
Time
Fig. 5. Heat transfer change with time for each building envelope.
The variation of heat transfer with time was slight for the roof and floor, and large for the
walls, especially for the south wall. Guangzhou is located in the northern hemisphere and
the sun lies in the southern sky during the measurement, so the south wall gained more heat
from solar radiation than the roof and the other walls, resulting in the largest heat transfer
change with time. The heat gain (heat transfer from outdoor to indoor) was highest at noon
for the south wall and roof and at 15:00 for the west wall. The heat loss (heat transfer from
indoor to outdoor) was highest at 7:00 for all the envelopes.
The heat transfers of all the building envelopes were summed up and the change of the total
heat transfer with time is show in Figure 6. In the period of the investigation, the time for
the building envelopes to gain heat is only 2 h and to lose heat in the rest of 42 h. Moreover,
the maximum of hourly heat loss (3444 W) is far more than the one for heat gain (114 W).
The building envelopes of the TBS works in most of time together with the air conditioners
to discharge the inner heat generated by the telecommunication equipments and gains heat
in only very short time due to the strong solar radiation.
6.0 30
Outdoor air temperature(℃)

envelopes and cooling load of air
5.0
Total heat transfer of building
26
4.0
22
3.0
18
conditioners(W)
2.0
1.0 14
0.0 10
-1.0
6
-2.0
-3.0 2
-4.0 -2
1-8 15:00
1-8 18:00
1-8 21:00
1-9 12:00
1-9 15:00
1-9 18:00
1-9 21:00
1-10 0:00
1-10 3:00
1-10 6:00
1-10 9:00
1-10 12:00
1-10 15:00
1-10 18:00
1-9 0:00
1-9 3:00
1-9 6:00
1-9 9:00
Time
Total heat transfer of building envelopes

Cooling load of air conditioners
Outdoor air temperature
Fig. 6. Total heat transfer of building envelopes, cooling load of air conditioners and outdoor
air temperature.
The cooling load of the air conditioners was calculated by using the measured results of the
return and supply air temperature and airflow rate. The change tendencies with time are
very similar for the cooling load of the air conditioners, the total heat transfer of the building
envelopes and outdoor air temperature (see Figure 6). The lower the outdoor air
temperature, the more the heat discharged by the building envelopes and the lower the
cooling load of the air conditioners. Enhancement of the heat transfer of the building
envelopes could save energy by decreasing the cooling load and shortening runtime of the
air conditioners.
Figure 7 shows the measured results for the single-phase electrical power of the TBS.
Electrical power was maintained at a low level from 5:10 pm to 5:45 pm on Jan 8. That is
because the door was open during this period, and the heat was dissipated through the
open door by natural ventilation. This shows that internal heat can be removed directly by
ventilation, and thus the energy consumption for the air conditioning can be significantly
reduced.
2.5
Electricity power (kW)

2.0
1.5
1.0
0.5
0.0
17:10 17:25 17:40
17:55 18:10 18:25 18:40 18:55
Time
Fig. 7. Single-phase electrical power variation of the TBS with time.
The basic information of the TBS, including the thermal characteristics of the building
envelopes, the inner heat density, the indoor air temperature set point, the operating
schedule of equipments and air conditioners were achieved and the thermal performance of
the building envelopes during the measuring period was studied by the field investigation.
Based on this information, building simulation was applied to study the thermal
performance of the TBS for a whole year and estimate the energy saving potentials for each
building energy efficiency design.
3. Building simulation
3.1 Simulation methods
A dynamic thermal simulation program named DeST was applied to estimate the annual
cooling load of the typical TBS. DeST was originally developed in 1989 by Tsinghua
University in China based on the IISABRE simulation environment (Hong, et al. 1997) and
validated by comparison with both well-known international thermal simulation programs
(Bloomfield, 1994) and experimental results (Zhang, et al. 2004). DeST has become a reliable
simulation program and is widely applied in building design and for national standards in
many countries (Yan, et al. 2004).
Figure 8 shows the model of the typical TBS in DeST created according to the real plan of the
TBS. The parameters for the simulations were set according to the basic information
achieved by the field investigation. The structure and thermal characteristics of the building
envelopes were set as in Table 1. The telecommunication equipments run all day with a total
heat of 4.35 kW. The heat density of the lighting and the people were ignored due to the
limited period of occupancy. The air conditioners operated continuously with a temperature
set point of 25 °C and a relative humidity set range between 5% and 85%. The ventilation
rate between the outdoor air and the TBS room was 0.5 h-1. The absorption coefficient of
solar radiation of the exterior surfaces of the walls and roof was 0.7. Outdoor climate data
was set according to typical Guangzhou annual meteorological data.
Fig. 8. Building simulation model for the typical TBS.
3.2 Simulation results and discussion

Simulation on the performance of the investigated TBS shows that the accumulation of
annual cooling load is 36891 kWh. It can be seen that 97% of the inner heat source becomes
the cooling load of the air conditioners and only 3% is discharged by the building envelopes
in the investigated TBS. Well building energy efficiency design, which enhances the heat
dissipation by the building envelopes, can save energy consumption of cooling. Heat
transfer coefficient and solar absorptance are the key factors determining heat transfer of
building envelope. In addition, shading and ventilation, which are widely used in the hot
and humid region of China, are considered as the potential energy efficiency designs for
TBS as well.
3.2.1. Temperature set point adjustment

Thermal performance of building energy efficiency design is affected by many factors, such
as inner heat density, operating schedule and environmental requirements. The information
of these factors were obtained by the field investigation and directly inputted as the known
conditions into building simulation, except for the indoor temperature set point. The
temperature set point for the air conditioners in a typical TBS is 25 °C, which takes human
comfort requirements into consideration. In practice, the time for which people are present
in the TBS is so short that the comfort requirements can be ignored, and the indoor
temperature set point of the TBS can be raised to 30 °C according to the environmental
requirements of the telecommunication equipment (Standardization Institute of Posts and
Telecommunications, 1995). The performance of the TBS with the adjustment of temperature
set point was simulated and the results show that the annual cooling load is decreased by
21% compared with the one without adjustment. Meanwhile, the role of building envelopes
becomes more important and their heat dissipation accounts for 23% of inner heat source.
The following analysis on the building energy efficiency design is based on the model with
the temperature set point adjustment.
3.2.2. Design of heat transfer coefficient

The heat transfer coefficients of the walls and roof can be varied from 0.5 to 6W/m2 K when
choosing different local materials and structures in Guangzhou. The simulation results show
that (see Figure 9) the relationship between the annual cooling load and the change of heat
transfer coefficient is linearly and an increase in heat transfer coefficient is beneficial to the
reduction of cooling load. The concept of better insulation saving more energy is not
applicable to the special kind of buildings such as TBS in Guangzhou. The annual cooling
load was found to be less sensitive to the change of heat transfer coefficient of the roof due
to the strong solar radiation absorption of the roof during a whole year. Considering the
practical available materials and structures in Guangzhou and safety of the communication
equipments, the walls and roof were determined as 60mm reinforced concrete walls with
plaster and 50mm reinforced concrete roof with plaster. The percentage saving in annual
cooling load achieved by the designs is 11.8% and 3.1% separately.
1800
Annual cooling load per square
1600
1400
meter(kWh/m2 )
1200
1000
800 Walls
600
Roof
400
200
0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
Heat transfer coefficients(W/m2 K)
Fig. 9. Effect of heat transfer coefficients of walls and roof on annual cooling load.
3.2.3. Design of solar absorptance

The solar absorptance of the outside surface of the external walls and roof was changed
from 0.7 to 0.1 to represent different external finishes. It was found that the annual cooling
load has a linear relationship to the solar absorptance of the external surfaces, and the lower
the solar absorptance, the higher the saving that can be achieved (see Figure 10). A reduction
in solar absorptance from0.7 to 0.1 can achieve a 7.5% and 1% saving in annual cooling load
for the walls and roof separately. The design of shading on the walls or roof can be included
into the design of solar absorptance by transforming the shading coefficient into the
resulting solar absorptance.
1600
1400
1200
meter(kWh/m2 )
1000
Wa lls
800
Roof
600
400
200
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Solar absorptance of the external surfaces
Fig. 10. Effect of solar absorptance of exterior surface of walls and roof on annual cooling load.
3.2.4. Ventilation design

The frequency distribution of outdoor air dry-bulb temperature in a typical year in
Guangzhou is shown in Figure 11. The percentage of the time when outdoor air temperature
is lower than the indoor temperature set point of TBS (30 °C) is more than 90%, which
indicates a great potential of energy saving can be achieved by ventilation. Air exchange rate
has a significant impact on cooling load. The ventilation rate between the outdoor air and
the TBS room when applying VCT was changed from 0.5 h-1 to 300 h-1 and the annual
cooling load was simulated, and is shown in Figure 12. The ventilation rate has a significant
effect on the annual cooling load. When the ventilation rate increases from 0.5 h-1 to 50 h-1,
the annual cooling load per square meter decreases from 1439 kWh/m2 to 514 kWh/m2 and
the rate of decline is 64%. As the ventilation rate continues to increase, the annual cooling
load does not reduce significantly, which indicates that a reasonable ventilation rate for VCT
is 50 h-1. The increase of air exchange rate can be achieved in practice by design of
mechanical ventilation.
3000
Frequency(h)
2500
2000
1500
1000
500
0
0～5 5～10 10～15 15～20 20～25 25～30 30～35 ＞35
Outdoor air dry-bulb temperature (℃)
Fig. 11. Frequency distribution of outdoor air dry-bulb temperature in Guangzhou.
1600
1400
1200
meter/(kWh/m2 )
1000
800
600
400
200
0
0 50 100 150 200 250 300
Ventilation rate between the outdoor air and the TBS
room(h-1 )
Fig. 12. Effect of ventilation rate on annual cooling load.
The internal heat in the TBS remains constant while the outdoor air temperature changes are
comparatively large, and thus the time when the outdoor cold air can be used changes along
with the outdoor air temperature. Figure 13 shows that the runtime of the air conditioners is
greatly shortened by applying VCT. The air conditioners need to run year-round in a TBS
without VCT, while they run mainly on a part-time basis from May to October in the TBS
with applied VCT. The annual runtime of the air conditioners amounts to 2509 h, which
accounts for only 29% of the total of 8760 h. Thus significant energy savings are made.
800
700
Typical TBS
Running time (h)

600
500 Typical TBS
using VCT
400
300
200
100
0
1 2 3 4 5 6 7 8 9 10 11 12
Month
Fig. 13. Monthly runtime of air conditioners in a typical TBS and in a typical TBS using VCT.
3.2.5. Combinations of building energy efficiency designs

The savings of annual cooling load from various building energy efficiency designs are
shown in Figure 14. It can be seen that ventilation design achieves the highest saving of
more than 64.2%, followed by an 13.4% saving from heat transfer coefficient design of walls.
The savings by using other energy efficient designs are less than 10%. The combination of
the energy efficiency designs are always preferred and applied in practice to achieve a better
thermal performance of a building. As the impact on energy saving may be strengthened or
weakened by the combination, several possible combinations of the building energy
efficiency designs were analyzed by using building simulation. The combined effect is
approximately equal to the sum of the effect of each design for the combination of the
designs of walls and roof (see Figure 15), which is because the impact of walls on cooling
load can be considered independent from the one of roof. The combined effect is
significantly greater than the sum for the combination of heat transfer coefficient design and
solar absorptance design, which can be explained as followings: Not only the heat loss, but
also the heat gain will increase with heat transfer coefficient design, and the latter one can be
weakened by solar absorptance design, resulting in a higher energy saving. The
combinations of ventilation design with others do not work much better than ventilation
design, which is because that the most of inner heat source will be taken away through well
ventilation and the impact of heat transfer of walls and roof becomes very slight.
70 64.2
Annual cooling load saving (%)
60
50
40
30
20 13.4
10 6.1
2.9 0.9
0
Ventilation Heat transfer Solar Heat transfer Solar
design coefficient absorptance coefficient absorptance
design of design of design of design of
walls walls roof roof
Building energy efficiency design
Fig. 14. Annual cooling load saving from various building energy efficiency design.
80
68.6 68.5
70
Annual cooling load saving (%) 60
50
40
30 27.2
20 14.0
10 6.8
0
Ventilation with Ventilation with Heat transfer Heat transfer Solor
heat transfer solor absorptance coefficient and coefficient design absorptance
coefficient and designs of walls solor absorptance of walls and roof design of walls
solor absorptance and roof designs of walls and roof
designs of walls and roof
and roof
Combinations of building energy efficiency designs

Fig. 15. Annual cooling load saving for combinations of building energy efficiency designs.
4. Study of ventilation cooling technology

It was found that ventilation design is the primary choice for energy efficiency TBS building
design. However, there are no studies on TBS ventilation design for energy conservation to
date. The envelopes of the existing TBSs are well sealed without any ventilation and the
significant cooling capacity of outdoor air in winter, spring, autumn and morning and
evening in summer is neglected, resulting in huge energy costs for air conditioning. To
make full use of the natural outdoor cooling resources, a new ventilation cooling technology
(VCT) was proposed. By applying VCT, the outdoor cold air is imported by fans to dissipate
the internal heat of telecommunication equipment directly. The fans, which consume much
less electrical power, are substituted for the air conditioners for environmental control when
the temperature and humidity of the outdoor air meet the equipment requirements. The
runtime of the air conditioners is thus greatly shortened, resulting in remarkable energy
savings.
However, there are several practical problems for the application of VCT to the TBS such as:
a control strategy for the fans and air conditioners for an environment with guaranteed
temperature and humidity to ensure reliable operation of the equipment; the temperature
and humidity of the environment itself in the TBS; chemically and mechanically active
substances and their conditions in the TBS; the condensation risk on the internal surfaces of
the building envelopes; additional costs for reconstruction and the payoff period. Therefore,
a feasibility analysis for VCT relative to the above problems should also be carried out.
4.1 Feasibility analysis
4.1.1 Control strategy

Figure 16 shows a schematic drawing of the VCT applied in the typical TBS. The fans and
filters are installed on the exterior walls. Air conditioners are also installed as auxiliary
cooling equipment to guarantee the environmental temperature and humidity requirements.

There are three situations where the fans and air conditioners are controlled by the control
system, i.e.: the fans and air conditioners are both switched off; the fans are switched off and
the air conditioners are turned on; and the fans are switched on and the air conditioners are
turned off. Figure 17 shows the control strategy for the VCT. If the indoor temperature Ti is
lower than the temperature set point of Tacs, the fans and air conditioners are both switched
off. Otherwise, if the outdoor temperature To is higher than the set point of Tfs or the indoor
relative humidity Φi is higher than the set point of Φs, the air conditioners are turned on
and the fans are switched off. However, if To is lower than Tfs and Φi is lower than Φs the
air conditioners are turned off and the fans are switched on. These cycles are repeated at
regular intervals.
Fig. 16. Schematic drawing of VCT.
Fig. 17. Control strategy for VCT.

4.1.2 Temperature and humidity environment

The simulation results indicate that the time percentage for an indoor air temperature
between 25 °C and 30 °C accounts for 60%, while temperatures between 20 °C and 25 °C
account for 20% and those between 5 °C and 20 °C account for 20% (Figure 18). The
telecommunication equipment in the typical TBS, called RBS2202, is designed for normal
operation in the climatic/mechanical conditions of class 3.1 of ETSI EN 300 019-1-3
(European Telecommunications Standards Institute, 2004). Relative humidity conditions for
environmental class 3.1 range from 5% to 85%. The indoor air temperature is much higher
than the outdoor air temperature; consequently, the air relative humidity decreases
markedly when air is transported into the room. The relative humidity of the air in the TBS
is lower than 85% over the majority of the year. The relative humidity set range of air
conditioners is from 5% to 85% in the building simulation model, i.e. the room air is
humidified when the relative humidity is less than 5% and is dehumidified when the
relative humidity exceeds 85%. The simulation results show that there is no need for the
room air to be humidified year-round and no need for it to be dehumidified for the vast
majority of the year. The dehumidification load is only 6.5 kWh annually. Under the control
of the VCT control system, the fans will be turned off and the air conditioners will be
switched on automatically to dehumidify the air when the indoor relative humidity exceeds
the limit of the range, and consequently the humidity conditions are guaranteed.
70
60
Percentage (%)
50
40
30
20
10
0
<5 5～15 15～20 20～25 25～30 >30
Temperature (℃)
Fig. 18. Time percentage of the indoor air temperature.
4.1.3 Chemically and mechanically active substance conditions

Table 2 shows the atmospheric environment in Guangzhou in recent years and the relative
conditions of the chemically and mechanically active substances used in the equipment
(European Telecommunications Standards Institute, 2004). The sulphur dioxide and
nitrogen oxide content levels meet the acceptable chemically active substance conditions for
the equipment. Considering the maximum sedimentation dust content of 8.41 mg/m2h in
recent years, filters with filtration efficiency of 85% will be able to maintain an acceptable
indoor sedimentation dust content under these conditions. Lower resistance, lower wind
pressure loss, ease of maintenance and long cycles for replacement of the filter materials
should also be considered during filter selection.
Sulphur Nitrogen Dust

dioxide oxides sedimentation
Environmental parameter Year
(mg/m3) (mg/m3) (mg/m2h)
2001 0.051 0.071 8.41
2002 0.058 0.068 7.98
Atmospheric environment in 2003 0.059 0.072 8.31

Guangzhou 2004 0.077 0.073 8.21
2005 0.053 0.068 7.37
2006 0.054 0.067 7.81
Active substances conditions

0.3 0.5 1.5
in equipment
Table 2. Atmospheric environment in Guangzhou in recent years and active substance
conditions in equipment.
4.1.4 Condensation risk on the interior surfaces of the building

Water condensation on the interior surfaces of the building envelopes may have negative
effects on the telecommunication equipment. Condensation occurs when the interior surface
temperature is lower than the air dew-point temperature. The greater the difference there is
between the interior surface temperature and the air dew-point temperature, the lower is the
possibility of condensation. The air dew-point temperatures were obtained from typical
annual meteorological data for Guangzhou and the interior surface temperature was
calculated using the building simulation model. Taking the north wall, whose interior
surface temperature is the lowest, for analysis, Figure 19 shows that the interior surface
temperature is higher than the air dew-point temperature from 1.7 °C to 24.7 °C. Therefore,
the condensation risk is quite small.
Fig. 19. Difference between the north wall interior surface temperature and the air dew-
point temperature.
4.1.5 Cost-effectiveness analysis

The ventilation rate between the outdoor air and the TBS room reaches 50 h-1 when fans
with total power of 0.22 kW are installed. Because of the high internal heat density, the time
when the fans and air conditioners are both turned off is very short. Therefore it can be
assumed that the fans run when the air conditioners are switched off (see Figure 13). Table 3
shows that the energy conservation is about 49% for application of VCT to the typical TBS.
Additional cost for reconstruction is not more than 5000 RMB, which includes the costs of
the fans, filters, ventilation control system and temperature and humidity detectors.
Additional maintenance costs for replacing the filter materials are only some hundred RMB
annually. The payoff period for the VCT is less than two years, based on a rate of 0.8 Yuan
per kilowatt-hour of electricity. The majority of the TBSs in Guangzhou can be reconstructed.
Provided that one thousand TBSs are reconstructed and VCT is applied, more than 3 million
RMB will be saved annually in electricity bills.
Typical TBS using

Typical TBS
VCT
Annual cooling load of air conditioners

31267 11182
( kWh)
Coefficient of performance of air

3.4 3.4
conditioners
Power consumption of air conditioners

9196 3289
( kWh)
Total power of fans

0 0.22
( kW)
Runtime of fans
0 6251
(h)
Power consumption of fans

0 1375
( kWh)
Total power consumptions

9196 4664
( kWh)
Annual Energy conservation

0 49
(%)
Table 3. Annual energy conservation of VCT.
4.2 Optimization of airflow organization

The installation location of the fans and openings and the layout of the equipment all
influence the temperature and air velocity distribution in the TBS, and therefore influence
the heat dissipation of the equipment. Computational fluid dynamics (CFD) simulation
provides detailed spatial distributions of the air velocity, air pressure, temperature,
contaminant concentration and turbulence by numerically solving the governing
conservation equations of fluid flows. It is a reliable tool for the evaluation of thermal
environments. A CFD code, PHOENICS, was applied to simulate the temperature and air
velocity distributions for the different cases of airflow organization in the typical TBS.
4.2.1 CFD model

Figure 20 shows the model of the typical TBS in PHOENICS created according to the real
plan and dimensions of the TBS. The parameters for the simulations were set according to
the field investigation results. The simulations were performed under steady state
conditions using a k-e turbulent model. A mesh of 0.1 m × 0.1 m × 0.1 m blocks was used.
The key boundary conditions needed for the calculation comprise:
Openings: External air temperature was set to 22 °C, which is the mean air temperature of
Guangzhou according to typical annual meteorological data. External pressure was zero
relative to the indoor atmospheric pressure.
Fans: The fan delivery was set according to the ventilation rate of 50 h-1.
Envelopes: The walls, floor and roof were assumed to be adiabatic.
Cabinet model: The cabinet model was created according to the actual cabinet. Each cabinet
has two inlets, one outlet and two heat sources. Each heat source was assigned a fixed total
heat flux of 271.9Wand the total heat flux for all heat sources in the TBS amounted to 4.35
kW. The wind speed of the outlet was set at the measured value of 1.83 m/s. The actual
cabinet and the cabinet model are shown in Figure 21.
Batterie Fan
Air Cabinets
Openin
Fig. 20. Model of the typical TBS in PHOENICS.
Outlet
Outlet
Heat
source
Inlet
Inlet
(a) (b)
Fig. 21. Actual cabinet (a) and cabinet model (b)
4.2.2 Airflow organization

There are cabinets, batteries, air conditioners, fans and openings in the typical TBS with
applied VCT. Taking the convenience of the reconstruction into consideration, the sizes of
the openings and the fans should be minimized. The fans and openings are installed on the
exterior walls. The cabinets are relocated for facilitation of the airflow. Several airflow
organization cases are proposed as follows:
Case 1: One opening and one fan with dimensions of 0.4 m × 0.4 m are installed, where the
opening is 0.2 m high and the fan is 1.8 m high.
opening is 1.8 m high and the fan is 0.2 m high.
opening and the fan are both 0.6 m high.
Case 4: Two openings and two fans with dimensions of 0.3 m × 0.3 m are installed, where
the openings are 0.2 m high and the fans are 1.8 m high.
The total deliveries of the fans in all cases mentioned above were set according to the
ventilation rate of 50 h-1. The layout of each case is shown in Figure 22.
(a) (b)
(c) (d)
Fig. 22. Layout of each case: (a) Case 1, (b) Case 2, (c) Case 3 and (d) Case 4.
4.2.3 Results and discussion

As cold air is taken into the cabinet through the inlets and dissipated from the outlet, the
average temperature of the inlets of all the cabinets for each case is used to evaluate the heat
dissipation efficiency. Figure 23 illustrates the contour of indoor temperature of each case at
z = 0.3 m. The average temperature of the inlets of each cabinet and of all the cabinets for
each case are presented in Figure 24. In Case 2, the cold air is mixed with the hot air exhaust
from the cabinets, and the heat dissipation efficiency is greatly reduced. In Case 3, the cool
air is quickly dissipated to the outdoors without sufficient heat exchange with the indoor air.
In Case 4, due to the additional openings and fans, which are uniformly distributed in the
region of the cabinets, the average temperature of the inlets of all cabinets drops by 1.1 °C
compared with Case 1. In this study, Case 4 serves as the optimal design.
(a) (b)
(c) (d)
Fig. 23. Contour of indoor temperature (°C) of each case at z=0.3m: (a) Case 1, (b) Case 2, (c)
Case 3 and (d) Case 4.
Average temperature (℃) 30

case1
25
20 case2
15 case3
10 case4
5
0
all cabinet cabinet cabinet cabinet cabinet cabinet cabinet cabinet
cabinets 1 2 3 4 5 6 7 8
Fig. 24. Average temperature of the inlets of each cabinet and of all cabinets for each case.
Based on the above comparison, the following results are generated:

1) The openings should not be set too close to the fans to avoid short circuits.
2) The openings should not be set at the height of the outlets of the cabinets, and the fans
should not be set at the height of the inlets of the cabinets to avoid mixing of cold and hot air.
3) The openings and fans when uniformly distributed in the region of the cabinets can
provide a uniform temperature field and better heat dissipation efficiency.
4) Results show that optimization of the airflow organization has a strong influence on the
heat dissipation efficiency for VCT.
5. Conclusions
TBS buildings have large numbers, high energy consumptions and great potentials on
energy conservation. Through field investigation of a typical TBS in Guangzhou, the basic
information of TBS was achieved, the key factors influencing energy consumption of TBS
were determined and several building energy efficiency designs were proposed. The effect
on annual cooling load was analyzed for each building energy efficiency design and the
combined effect were achieved for several combinations of the designs by building
simulation. The building energy efficiency design strategy for TBS in Guangzhou can be
concluded as followings: The indoor air temperature set point of TBS should be elevated to
30 °C regardless of human thermal comfort requirement. Ventilation design is the prior
choice for building energy efficiency design of TBS, and other designs are not necessarily
considered when ventilation design is applied. For the TBS which is not feasible to apply
ventilation design due to practical problems, the combination of designs of high heat
transfer coefficient and low solar absorptance walls and roof is strongly recommended.
Study of ventilation cooling technology (VCT) was proposed. The application feasibility of
VCT was analysed systematically. The results show that the temperature and humidity
requirements of the equipment can be fully met under the linked control of the fans and air
conditioners. The chemically and mechanically active substance conditions can be met by
using filters with filtration efficiency of 85%. The condensation risk on the interior surfaces
of the building envelopes is quite low. The energy conservation achieved by the use of VCT
is about 49%, and the payoff period is less than two years. The application of VCT in the
TBSs in Guangzhou is feasible, and obvious economic and social benefits will be achieved if
VCT can be broadly applied. The PHOENICS CFD code was applied to simulate the
temperature and air velocity distribution for several different airflow organization cases in
the typical TBS. The simulation results show that optimization of airflow organization has a
strong influence on the efficiency of heat dissipation for VCT.
Acknowledgments
The authors are thankful for the financial support and cooperation of Guangzhou Mobile
Communication Corporation. We especially thank the Department of Building Science of
the School of Architecture in Tsinghua University for important contributions to the testing.
6. References
Nakao, M.; Hayama, H. & Uekusa, T. (1988). An efficient cooling system for
telecommunication equipment rooms, Proceedings of the 10th International
Telecommunications Energy Conference, pp. 344-349, San Diego, Canada, October
1988
Maeda, Y.; Seshimo, Y. & Okazaki, T. (2005). Study of a cooling system for the
telecommunication base site, ASHRAE Transactions, Vol. 111, No. 2, pp. 746-755,
2005
Choi, J.; Jeon, J. & Kim, Y. (2007). Cooling performance of a hybrid refrigeration system
designed for telecommunication equipment rooms, Applied Thermal Engineering,
Vol. 27, No. 11-12, pp. 2026-2032, August 2007
Hayama, H. & Nakao, M. (1989). Air flow systems for telecommunications equipment
rooms, Proceedings of the 11th International Telecommunications Energy Conference, pp.
1-7, Florence, Italy, October 1989
Dan, N. & Matti, K. (2000). Application of CFD technique in thermal design of a
telecommunication base station, Proceedings of 9th International Flotherm User
Conference, Orlando, USA, October 2000
Makhkamdjanov, B.M. (2006). Technological model of the independent power supply with
converters of renewable energy for base station of mobile communication,
Proceedings of Internet, 2006 2nd IEEE/IFIP International Conference, Central Asia, pp.
1-4. Sept. 2006
Schmidt, R.R. & Shaukatullah, H. (2003). Computer and telecommunications equipment
room cooling: a review of literature, IEEE Transactions on components and packaging
technologies, vol. 26, No. 1, pp. 89-98, March 2003
Feng, Y. (2004). Thermal design standards for energy efficiency of residential buildings in
hot summer/cold winter zones, Energy and Buildings, Vol. 36, No. 12, pp. 1309-1312,
December 2004
Athanassios, T.; Andreas, K.A. & Panagiota, K. (2007). Simulation of facade and envelope
design options for a new institutional building, Solar Energy, Vol. 81, No. 9, pp.
1088-1103, September 2007
Cheung, C.K.; Fuller, R.J. & Luther, M.B. (2005). Energy-efficient envelope design for high-
rise apartments, Energy and Buildings, Vol. 37, No, 1, pp. 37-48, January 2005
Nakao, M.; Ohshima, K. & Jitsukawa, H. (1988). Thermal control wall for telecommunication
equipment rooms, Proceedings of the 10th International Telecommunications Energy Conference,
San Diego, Canada, pp. 280-284, October 1988
Hong, T.; Zhang, J. & Jiang, Y. (1997). IISABRE: an integrated building simulation
environment, Building and Environment, Vol. 32, No. 3, pp. 219-224. May 1997
Bloomfield, D.P. (1994). Final report of IEA Annex 21 (Calculation of Energy and
Environmental performance of buildings), CRC Publications, London, 1994.
Zhang, X.; Xie, X.; Yan, D. & Jiang, Y. (2004). Building environment design simulation
software DeST (3): validation of dynamic simulation results of building thermal
progress, HV & AC, Vol. 34, No.9, pp. 37-50, 2004 (in Chinese)
Yan, D.; Xie, X.; Song, F. & Jiang, Y. (2004). Building environment design simulation
software DeST (1): an overview of developments and information of building
simulation and DeST, HV & AC, Vol. 34, No.7, pp.48-56, 2004 (in Chinese)
Standardization Institute of Posts and Telecommunications. (1995). Environmental conditions
for telecommunication base rooms, GF 014-1995, 1995, China (in Chinese)
European Telecommunications Standards Institute. (2004). Equipment Engineering (EE);
Environmental conditions and environmental tests for telecommunications equipment;
part1-3: classification of environmental conditions; stationary use at weather protected
locations, ETSI EN 300 019-1-3, 650 Route des Lucioles, F-06921 Sophia Antipolis
Cedex, France, 2004. [Online]. Available: www.etsi.org.
Dynamic Space-Code Multiple Access (DSCMA) System:
A Double Interference Cancellation Multiple Access Scheme in Wireless Communications System 583
27
X
Dynamic Space-Code Multiple Access (DSCMA)

System: A Double Interference Cancellation
Multiple Access Scheme in Wireless
Communications System
Chee Kyun Ng1, Nor Kamariah Noordin1, Borhanuddin Mohd Ali1,
and Sudhanshu Shekhar Jamuar2
1Department of Computer and Communication Systems Engineering,
Faculty of Engineering, Universiti Putra Malaysia, Malaysia.
2Department of Electrical Engineering,
Faculty of Engineering, Universiti Malaya, Malaysia.
1. Introduction
It is well known that cellular mobile phone systems have evolved from 1G and 2G that use
frequency and time division multiple access (FDMA and TDMA) systems respectively, to
code division multiple access (CDMA) of third generation (3G) systems (Chen et al., 2006).
Furthermore, the exploitation of spatial diversity from the emergence of advance antenna
technologies such as smart antenna and space time signal processing have given rise to
induce another multiple access scheme called space division multiple access (SDMA)
systems (Fang, 2002). Among these schemes, the system capacity and spectrum efficiency
are the key factors to compare the performances of various mobile communication systems.
Since radio frequency (RF) spectrum is a limited resource, these techniques have
approached their fundamental limitations. Flexible utilization of such resources in space,
time and code has led to great improvement in system capacity. For a given bandwidth, the
system capacity for narrowband radio systems such as FDMA and TDMA is dimension or
bandwidth limited. In contrast, the system capacity of CDMA and SDMA systems is
interference limited. Any reduction in interference in CDMA and SDMA systems converts
directly and linearly into increased capacity (Yu et al., 2004), (Chen et al., 2008).
Multiple access schemes such as FDMA and TDMA increase their system capacity and
spectrum efficiency by dividing the different network planning phases more clearly into
individual parts to allow different frequencies to be used at different time moments
(Castaeda & Lara, 2008). In CDMA systems, the same frequency is used simultaneously in
adjacent cells and the interference level should be taken into account in the coverage-
planning phase (Niemela & Lempiainen, 2003). Furthermore, cell splitting and sectorisation
to form SDMA systems with use of directional antenna could also result in increase of
system capacity and spectrum efficiency over the omnidirectional antenna system (Godara,
1997). Although these approaches do significantly increase the system capacity and
spectrum efficiency, each scheme basically is attempting a more efficient use of the same
resource.
It is well known that CDMA system is characterized as being interference limited.
Independent simultaneous transmissions by mobile users at different locations in a cell give
rise to the near-far phenomenon. To combat the near-far problem, power control is used to
ensure equal signal levels are received from all mobile users at different location (Hashem &
Sousa, 1997). Therefore, power control is considered the most important system requirement
for CDMA systems to increase the system capacity on the reverse link by overcoming the
near-far problem (Cameron & Woerner, 1996), (Uthansakul, 2002). Since all the cells can
operate with the same channel in CDMA cellular network, a significant source of
interference apart from traffic in its own cell is the traffic from neighbouring cells. Thus, the
system capacity of CDMA systems is determined by the amount of co-channel interference
that it can tolerate, which is comprised of intra-cell interference and inter-cell interference
(Wu et al., 1998). If the traffic load in neighbouring cells is reduced, more traffic can be
accepted in the observed cell (Chatovich & Jabbari, 1999). However, because of power
control from observed cell base station (BS), transmitting a high power level in reverse link
may result in high interference to neighbouring cell BS (Hashem & Sousa, 1997). Therefore,
in CDMA systems, if the capacity of a single cell increases it creates higher interference to its
neighbouring cells and thus impacts their capacity.
Other approach that shows a promise for substantial capacity enhancement is the use of
spatial filtering with exploitation of smart antenna at cell site BS (Zheng et al., 1996). Hence,
the deployment of SDMA system has been recognised as one of the most promising
techniques for controlling co-channel interference in cellular systems, leading to the
required system capacity improvement (Liberti & Rappaport, 1998). The beamforming
ability of smart antenna technology has been adapted to increase the gain of the desired
signal while null interference sources resulting in the improvement of the system capacity
(Huang et al., 2001). The narrow beams from smart antenna are steered toward desired
users in order to filter out interference caused by co-channel users located in the same cell
and from adjacent cells (Galvan-Tejada & Gardiner, 2001).
However, in order to achieve an ideal SDMA system, smart antenna must carefully form its
radiation patterns to capture the desired user and to nullify sufficiently interfering users.
Therefore, the smart antenna requires high accuracy in propagation channel response
estimation (Cho et al., 2002). If there are N elements antenna array used in a smart antenna
system, it is only possible to accommodate N – 1 users in reverse link (Rapajic, 1998), (Kim et
al., 2001). Actually in the randomness of mobile users distribution, this is not always
possible to eliminate interferers by null-steering in the corresponding arrival directions.
Hence, there will be a probability of two or more mobile users located near to each other.
This means that the co-channel interferences will occur among these mobile users when
adaptive beams steering smart antenna are employed. On the other hand, the present of
sidelobes from smart antenna system will further reduce the signal to interference ratio (SIR)
performance of each mobile user. Hence, more sidelobes interferences are radiated in the
direction of the desired user main lobe pattern. These sidelobes interferences can
significantly reduce the system capacity if multiple beams are synthesized from smart
antenna to accommodate the density of mobile users in a particular area.
The wireless channel usually characterized by the path loss, shadowing and fading
(Feuerstein et al., 1994). In urban areas, multipath propagation is common, whereby the
receiver observes a number of copies of the transmitted signal, each with a different time
delay (Adachi et al., 2005). This provides a form of multipath fading. In a digital
communication system, the delay-spread of multipath propagation could also cause inter-
symbol interference (ISI) (Lien & Cherniakov, 1998). The characteristics of the spreading
sequences in CDMA system provide a crucial effect on the performance of the whole
communication systems. This signature sequences in general determine how much
interference is received at a receiver from other mobile users and influence the extraction
capability of the desired signal from noise-like spectrum (Xie & Rahardja, 2005). On the
other hand, since the reverse link of a CDMA system is usually asynchronous, in the sense
that the arrival times for each mobile user signal are different (Thompson et al., 1996), (Choi
et al., 2007). Therefore, the spreading sequences of CDMA systems are characterized with ISI
as well as multiple access interference (MAI) (Peterson et al., 1995), (Guo & Wang, 2008). In
multipath propagation environment, multiple copies of transmitted signal arrive at receiver
with different time delay will cause ISI. A MAI occurs if the orthogonality among spreading
sequences is lost (Ishida et al., 2000). The MAI is caused by asynchronous in a CDMA
system where each mobile user will observe interference from all other mobile users in the
system, since the transmitted signal will not be orthogonal in delay-spread environment
(Thompson et al., 1996). Traditional CDMA spreading sequences such as m-sequence
(Golomb, 1992), Gold codes (Gold, 1967), and Kasami codes (Kasami, 1966), exhibit non-zero
cross-correlation which results in high MAI in asynchronous reverse link transmission.
Another family of orthogonal codes is constituted by Walsh codes (Harmuth, 1970) and
orthogonal Gold codes (Popovic, 1997), do retain their orthogonality in the case of perfect
synchronization, but also exhibit non-zero cross-correlation in asynchronous transmission
(Wei et al., 2005). Recently, an attractive family of large area synchronized (LAS) CDMA
spreading sequences is introduced in (Li, 2003) has exhibited zero correlation zone (ZCZ) or
interference free window (IFW) near zero delay time offset, resulting in zero ISI and MAI
within the IFW. The LAS spreading sequence is constituted by the combination of Large
Area (LA) code (Li, 1999) and Loosely Synchronous (LS) code (Staňczak et al., 2001). More
specifically, the interference-free in CDMA system only become possible when the
maximum channel-induced delay-spread is within the designed IFW duration. However, in
the system design especially using omnidirectional antenna, not all multipath signal
components arrive within IFW time offset. Since the total duration of IFW expressed in
terms of the number of chip intervals depend on the minimum zero padding implanted
between non-zero pulses interval, thus the number of minimum zero padding must be
increased to maximum delay-spread of the channel in LAS sequence in order to
accommodate all multipath signal components. This implies that the duty ratio of LAS
spreading sequences is low when the number of minimum zero padding is increased.
Therefore, a specific drawback of LAS-CDMA is that its relatively efficient orthogonal codes
demanded in wireless systems are limited, and hence reduce its spectrum efficiency. Besides
that, the implementation of LAS sequences is very complex that additional components are
necessary.
There have been many multiple access systems for the cellular system designed to improve
its system performance. Several works have been carried out to show the improvement in
the system capacity using the joint multiple access system. A careful selection of joints
multiple access from two or more individual systems can determine the fitness of the joint
system. Interference-limited systems such as CDMA and SDMA are susceptible to time of
arrival (TOA) and angle of arrival (AOA) of individual user signals. Thus, a non-uniform
traffic can severely degrade the performance of CDMA and SDMA systems. In this chapter,
a joint multiple access of CDMA and SDMA system is proposed. The performance of this
joint multiple access system is also vulnerable to the non-uniform traffic. Although the
performance of this joint multiple access system has been previously studied in several
papers (Liberti & Rappaport, 1994), (Naquib et al., 1994), (Buracchini et al., 1996) and (Ng &
Sousa, 1998), none of them considers to evaluate the most realistic of system performance in
this joint multiple access.
In this chapter, a new approach called dynamic space-code multiple access (DSCMA)
system arising from the combination of CDMA and SDMA systems is designed, and its
system performances are then investigated. An innovative approach to eliminate the
existing interferences in DSCMA system is introduced. The spreading sequences of Large
Area Synchronous Even Ternary (LAS-ET), which exhibited an interference free window
(IFW) in their correlation, are exploited here. The spatial signature from smart antenna
narrower beam is exploited to drive all the multipath propagation signals to arrive within
the IFW in reverse link transmission. The size of IFW is adaptable with the size of smart
antenna beamwidth through dynamic space-code (DSC) algorithm. Therefore, the result of
combined dominant signature from DSCMA system will yield a perfect interference
cancellation so that the system capacity increases dramatically.
2. The Properties of Orthogonal CDMA Sequences

Traditional ways of separating multiple access signals in time or frequency such as TDMA
and FDMA are relatively simple by making sure that the signals are orthogonal and non-
interfering. However, in CDMA different mobile users occupy the same bandwidth at the
same time. They are separated from each other through the use of a set of orthogonal
sequences. Two waveforms x and y are said to be orthogonal to each other if their cross-
correlation, Rxy(0) over T period is zero in time shift  (Lee, 1998), where
1 T
R xy ( )  lim
T   2T  x(t ) y(t   )dt
T
(1)
In discrete time, the two sequences x and y are orthogonal if their cross product Rxy(0) over
T period is zero (Wang et al., 2007). The cross product of Rxy(  ) is defined as
R xy ( )   x(t ) y(t   )
  T
(2)
As an example, the following two sequences or codes, x and y are orthogonal.
x   1,1,1,1 (3)
y   1,1,1,1 (4)
Hence, their cross-correlation is zero.
Rxy(0) = (–1)(–1) + (–1)(+1) + (+1) (+1) + (+1) (–1) = 0 (5)
In order for the set of codes to be used in a multiple access scheme, an additional property is
needed. In addition to the zero cross-correlation property, each code in the set of orthogonal
codes must have an equal number of +1s and –1s (Faruque, 1996). This second property
gives that particular code the pseudorandom nature. A direct sequence CDMA (DS-CDMA)
system spread the baseband data by directly multiplying the baseband data pulses with a
peudorandom or PN sequence that is produced by a PN code generator. A single pulse or
symbol of the PN waveform is called a chip, where the chip rate is much higher than the
data bit rate (Lee, 1991).
2.1 Welch Bound in CDMA Systems

The CDMA system is a multiple access scheme in which several independent users access a
common communication channel by modulating their data symbols with preassigned
spreading sequences. The receiver observes the sum of the transmitted signals in additive
white Gaussian noise (AWGN) channel. The decoder for a given mobile user treats the sum
of the interfering signals from other mobile users as noise. The spreading sequences are
chosen to create good single user channels for the individual coding systems. In fact,
however, the channel created by the spreading sequences is susceptible to MAI (Rupf &
Massey, 1994). In 1974, Welch in (Welch, 1974) had shown that the lower bound for the
1

acceptable sidelobes of auto-correlation and cross-correlation functions are set around SF 2 ,
where SF is the spreading factor or processing gain of the system. This lower bound is called
as Welch bound (Li, 2003). Signature sequences that maximize the sum capacity in the
uplink of CDMA systems in AWGN channel are known to satisfy Welch’s bound on the
total squared correlation with equality (Heath et al., 2004).
2.2 LAS-ET Sequences

The original LAS codes proposed in (Li, 1999) are synthesized by seeding LS codes in LA
codes to improve it spectrum efficiency. An N p LA codes are synthesized in such a manner
that the N p non-zero 1 pulses from m-sequences oriented are positioned as shown in
Table 1. This arrangement forms a configuration of LA( N p , K 0 , Lc ) where K 0 is the
minimum number of zero padding in pulse interval of non-zero pulses which determine the
size of IFW delay-spread in term of chips, while having a total code length of Lc chips.
0 38 78 120 164 210 258 308 360 414 470 530 592 660 732 808 847
C1 + + + + + + + + + + + + + + + + +
C2 + + + + - - - - + - + - - + + - +
C3 + + + - - - - + - + - - + + - + +
C4 + + - - - - + - + - - + + - + + +
C5 + - - - - + - + - - + + - + + + +
C6 + - - - + - + - - + + - + + + - +
C7 + - - + - + - - + + - + + + - - +
C8 + - + - + - - + + - + + + - - - +
C9 + + - + - - + + - + + + - - - - +
C10 + - + - - + + - + + + - - - - + +
C11 + + - - + + - + + + - - - - + - +
C12 + - - + + - + + + - - - - + - + +
C13 + - + + - + + + - - - - + - + - +
C14 + + + - + + + - - - - + - + - - +
C15 + + - + + + - - - - + - + - - + +
C16 + - + + + - - - - + - + - - + + +
Table 1. The arrangement of 16 LA(16,38,847) sequences
In order to exploit the characteristics of LA sequences proposed in (Li, 1999) without

altering the size of its IFW, a modified version of the sequence such LAS-ET sequences (Ng
et al., 2009) is employed in DSCMA instead of LAS-CDMA sequences proposed in (Li, 2003)
which exhibit a small IFW. Figure 1 shows the correlation properties of the
LAS  ET 16,38,818 sequences. As can be seen in these figures, the correlation properties of
LAS  ET 16,38,818 sequences are similar to the original proposed LA16,38,847  sequences
which exhibited a large IFW around the origin. The cross-correlation value of
LAS  ET 16,38,818 sequence in zero delay spread is 4.03 x 10-17.
(a)
(b)
Fig. 1. Correlation properties of LAS  ET 16,38,818 sequence; (a) auto-correlation and (b)
cross-correlation.
3. Reverse Link Capacity of SDMA System

The conventional SDMA systems increase its capacity by spatial filtering the interferences.
The system continuously adapts its narrower beam from smart antenna system to steer each
mobile user with the main lobe while isolating interferences with nulls. Hence, SDMA is
allowed to reuse the limited radio resources (frequency, time and code) within a cell. From
Equation (1) in (Ng et al., 2008), the nulls’ AOA,  nulls of the SDMA radiation pattern occur
at
  h  
 nulls  cos 1 2  
 (6)
  N e 2 
where N e is the number of elements in smart antenna system,  is progressively phase

shift, and h is any integer but not equal to 0, n, 2n, …..
Figures 2a and 2b show the typical SDMA system for N e = 8 and 32 respectively with 90o
AOA of the desired user. For N e = 8, the nulls to accommodate interfering users are
occurred at 41.41o, 60o, 75.52o, 104.48o, 120o and 138.59o, while the nulls for N e = 32 are
occurred at 20.36o, 28.96o, 35.66o, 41.41o, 46.57o, 51.32o, 55.77o, 60o, 64.06o, 67.98o, 71.79o,
75.52o, 79.19o, 82.82o, 86.42o, 90o, 93.58o, 97.18o, 100.81o, 104.48o, 108.21o, 112.02o, 115.94o, 120o,
124.23o, 128.68o, 133.43o, 138.59o, 144.34o, 151.05o and 159.64o. These figures show that the
system capacity, K of SDMA system is direct proportional to the number of antenna
elements in smart antenna with expression below (Rapajic, 1998)
K  Ne 1 (7)
The interfering users are only allowed to be located at null AOAs, otherwise co-channel
interferences between mobile users will occur. Any additional mobile user into this system
after the limited nulls are fully occupied will also cause co-channel interference to other
mobile users.
(a)
(b)
Fig. 2. Radiation pattern of SDMA system for (a) N e = 8 and (b) N e = 32.
It has been reported that smart antenna can synthesize a high directive beam toward the
desired user while nulling the interfering users to increase capacity. However, fully nulling
the interfering users in SDMA system do not take place because there are two major
interference sources, which are side-lobes and co-channel interferences. The interfering
users will not always locate at the nulls of the desired user radiation pattern especially in
randomly distributed traffic environment as shown in Figure 3.
Fig. 3. Co-channel and side-lobes interferences to the desired user, C from randomly located
interfering users of 1, 2, 3, 4,….., 16
4. Dynamic Space Code Multiple Access (DSCMA) System

Non-uniformly distributed traffic usually degrades the performance of CDMA and SDMA
systems severely in the reverse link. The imperfect correlation properties of the traditional
CDMA spreading sequences result in ISI and MAI at non-zero delay spread. The random
positions of mobile users will cause MAI among them in the SDMA system, where positions
at nulls of the desired user radiation pattern are rarely achieved. Therefore, the non-uniform
traffic causes loss of orthogonality to distinguish each mobile user in the conventional
interference limited systems.
Here, a promising solution to deploy the BS with smart antenna system to perform the joint
multiple access of CDMA and SDMA systems is proposed. The CDMA and SDMA systems
are adapted to each other dynamically to form DSCMA system. This proposed multiple
access scheme is a novel interference cancellation scheme that employ the spreading
sequences of CDMA system into spatial signatures of SDMA system through DSC
algorithm. In DSC algorithm, the size of dedicated IFW from LAS-ET spread sequence is
adapted dynamically to the size of half power synthesized beamwidth from smart antenna
beamforming system as shown in Figure 4. In this joint multiple access scheme, each user is
assigned an LAS-ET sequence within a high directivity beam. Hence, the integration of these
two signature schemes, spatial filtering and spreading sequence, creates a dominant
signature scheme called DSC signature. Therefore, by using this dominant signature
scheme, the inherent interferences in CDMA and SDMA systems environment can be
eliminated.
Fig. 4. Performance of various beamwidths in smart antenna system over IFW region from
correlation property of LAS  ET 16,38,818 sequences.
As shown in previous section, the co-channel interference between two mobile users in
SDMA system occurs when both of them are located close to each other. For example,
assuming that the desired user, C is located at AOA of 90o to the smart antenna axis while
other mobile users are randomly located within the AOA of 0o to 180o as shown in Figure 3.
It is observed that the 11th user’s beam is located very near to C with only 0.6o separation.
This phenomenon causes co-channel interference between them while other mobile users
also contribute interferences to C through their sidelobes radiation pattern. It is possible to
mitigate these co-channel interferences by CDMA spreading sequences. However, all
traditional CDMA spreading sequences are self-interference systems when all signals from
each mobile user arrive at the BS in asynchronous manner. The auto-correlation and cross-
correlation properties of traditional CDMA sequences are not orthogonal at non-zero delay
spread,   0 . Thus, it shows that interference occur among the mobile users in
asynchronous transmission environment.
Therefore, it is necessary to prefer spreading sequences that exhibit zero correlation between
each other to drive all the asynchronous signal components to drop within the smart antenna’s
narrow beam maximum propagation delay spread. Hence, a spreading sequence that exhibits
large size of IFW is required to accommodate large beamwidth of smart antenna radiation
pattern. Considering Figure 3 again, there is group of beams with theirs AOA respectively to
accommodate 17 randomly distributed mobile users. Each beam is assigned to different mobile
user with an LAS sequence order, C1, C2, C3, C4, C5,…, Cm, where m is the maximum
number of total available sequence. These sequences are assigned to mobile users in
chronological order upon their arrival and can be reused dynamically whenever needed.
To illustrate how directive beam can improve the reverse link in a single cell of DSCMA
system, consider the case in which each mobile user has an omnidirectional antenna, and the
BS tracks each mobile user in the cell using a directive beam. Assume that the beam pattern,
G   in formed such that the pattern has a maximum gain in the AOA of the desired user.
Such a directive pattern can be formed using an N e elements smart antenna array. Assume
that K users in the single cell of DSCMA system are non-uniformly distributed throughout a
cell. On the reverse link, the power received from the desired user signal is Pr , 0 with
maximum gain of G0  0  . The received powers from K – 1 interfering users are given by Pr ,i
for i  1,2,....., K  1 . Then the average total received interference power, I seen at the desired
user AOA,  0 at the BS is given by
 K 1

I  E
  G  P
i 1
i 0 r ,i 

(8)
where Gi  0  is the ith interference gain level of smart antenna radiation pattern seen at the
AOA of the desired user. The value of Gi  0  can be obtained from Equation (1) in (Ng et
al., 2008) and is given as
1 sin N e k cos 0   i 2 

Gi ( 0 )  (9)
N e sin k cos 0   i 2 
where k is given as 0.5 for half wavelength spacing between elements to avoid the
appearance of grating lobe in the system, and parameter  i is the phase shift of the smart
antenna to steer the beam in  i direction of ith interfering user. If the perfect power control
is applied such that the received power at the BS antenna from each mobile user is the same,
then Pr ,i  Pc for each of K users, and hence the average interference power seen by the
desired user is given by
 K 1

I  Pc E 
 G  
i 1
i 0 (10)
5. Reverse Link Interferences in DSCMA System

In DSCMA system, a BS equipped with smart antenna transmits signal to each mobile user
in forward link transmission using a synthesized narrow beam and a dedicated spreading
sequence. The signal is perfectly synchronized at transmission so that it arrives at mobile
receiver in synchronism with zero delay spread. Consequently, due to the orthogonalities of
both spatial signature and spreading sequence in zero delay spread among the K users in a
cell, each mobile receiver can demodulate its own signal without interference from other
transmitted signals that share the same channel.
However, this synchronism in forward link transmission cannot be maintained in reverse
link transmission where all the signals from K users are rather arrived at BS in
asynchronized manner. Thus, the signals from the other mobile users appear as additive
interference to the desired user signal if the orthogonalities of both spatial signature and
spreading sequence among them are loss in non-zero delay-spread. The reverse link
interferences are twofold: the interference arising from K – 1 users in the same cell or can be
known as intra-cell interference, and the interference arising from mobile users in
neighbouring cells or also called as inter-cell interference. Hence, the system capacity of
DSCMA is examined by considering both intra-cell and inter-cell interference environments
in reverse link transmission.
5.1 Intra-cell Interference

Suppose that each cell has K randomly distributed mobile users. With the use of perfect
instantaneous power control, all K user signals are arriving at the BS with the same power
level S within the same cell. Therefore, the intra-cell interference, I int ra from K – 1 interfering
users is given as
I int ra  ( K  1) S (11)
In DSCMA system, the K – 1 interferences power level are not same in the AOA of the
desired user,  0 . Nevertheless, the interfering signals from K – 1 users are still received at
the same power level S from theirs respective AOA through perfect power control. Most of
these interfering signals contribute merely side-lobe interferences with Gi  0   S in  0
direction. Some of the interfering signals are also received at the same power level,
Gi  0   S when they are at the same AOA of the desired user.
The arbitrarily interferences level, Gi  0  as shown in Figure 3 with 16 interfering users can
be analogously as multiple dots along the line of radiation pattern as shown in Figure 5.
This is assuming that all radiation patterns for all mobile users are same. Hence, from (9)
and (11), the intra-cell interference in AOA of mobile user C, I int ra  0  yields to
K 1 K 1
S sin N (k cos( 0 )   i / 2)
I int ra ( 0 )  S   G ( )  N  
i 1
i 0
i 1
sin(k cos( 0 )   i / 2)
(12)
Fig. 5. The analogously random side-lobes interferences from 16 interfering users in AOA of
the desired user
5.2 Inter-cell Interference

In the multi-cell of DSCMA system, the interference analysis in reverse link becomes
complicated. This is because the mobile users are power controlled by their own cell BS. The
membership of the user is determined by the maximum pilot signal power among the cells
and not the minimum distance from a cell BS. The mobile users are connected to a BS that
offers the lowest signal attenuation rather than the closest BS (Gilhousen et al., 1991).
Because of power control, the interference level received from mobile users in neighbouring
cells depends on two factors: attenuation in the path to the desired user's cell BS, and
attenuation in the path to the mobile user's cell BS. Thus, in the fourth power law of
distance, the user’s transmitted power Pt can be expressed as (Chatovich & Jabbari, 1999)
Pt  Pr r 410( 10 )
(13)
where Pr is the received signal power at its BS,  is the log-normal Gaussian random
variable with zero mean and standard deviation,  of 8 dB, and r is the distance from the
mobile user to BS. Since only average power levels are considered, the effects of multipath
fading are ignored. To evaluate inter-cell interference, I int er  0  in DSCMA, consider an
interfering user located in mth neighbouring cell at a distance rm from its base station BS m
and r0 from the desired user base station BS 0 as shown in Figure 6.
Fig. 6. Inter-cell interference environment model.
If Pt is its transmit power, the received power S at its BS is given by
Pt Gm ( m )
S (14)
10 ( m 10 ) rm4
where Gm ( m ) is the antenna gain in the AOA of the interfering user to its cell BS m , and  m
is the Gaussian random variable representing the shadowing process in its cell. Then the
interference I received at BS 0 is given by
Pt G0 ( mi )
I (15)
10 ( 0 10 ) r04
where G0 ( mi ) is the antenna gain of BS 0 in the AOA of the ith interfering user from mth
neighbouring cell to BS 0 , and  0 is the Gaussian random variable representing the
shadowing process in the desired user cell.
Hence, from (14) and (15), the interference to signal ratio, I/S is given by
4
I  rm   G0 ( mi ) 
   
  G ( )   10
( m  0 ) 10
(16)
S  r0   m m 
where the first term is due to the attenuation caused by distance and blockage to the given
BS, while the third term is the effect of power control to compensate for the corresponding
attenuation to its BS. Since  m and  0 are independent their difference has zero mean and
variance 2 2 (Cooper & Nettleton, 1978). The second term reveals the total antenna gain
received in the AOA of the interfering user to the BS0, and its value is less than unity. In
DSCMA, this second term will only has a maximum value when an interfering user from a
neighbouring cell is located at the same AOA of the desired user,  0 . Then G0 ( mi ) in (16)
will become
6 K 1
1 sin N e (k cos( mi )   0 / 2)
G0 ( mi ) 
Ne
 m 1 i 1
sin(k cos( mi )   0 / 2)
(17)
For all values of the parameters in (16), I S is less than unity. If its value is not less than
unity then the user would switch to the other cell BS. Therefore, I int er  0  in DSCMA is
found by summing (16) for all mobile users in the first tier neighbouring cells
6 K 1 4
r   G ( ) 
I int er ( mi )  
m 1 i 1
S   mi
 r0i



  0 mi   10 ( mi  0i ) 10
 Gm ( m ) 
(18)
where Gm ( m ) =1 which the gain to theirs mobile user is 1.
6. DSCMA System Signalling

To simplify the derivation, only the baseband signal of transmitted signal is being
considered. Hence, in DSCMA system, the transmitted signal from the ith user, s i (t ) that
occupies the ith spreading sequence for i  1,2,....., K  1 can be written as
si (t )  P d i (t )ci (t ) 0t T (19)
Assuming that the desired user is user 0 and all the other K – 1 users are interfering users. The
received signal, r t  is a sum of the transmitted signals from all K users and corrupted by its
additive complex Gaussian thermal noise, n(t ) in an AWGN channel. The signal of each
mobile user arrives at a different propagation delay,  i . Thus, the received signal at the BS
equipped with smart antenna beamforming network in the AOA of user 0 can be expressed as
K 1 K 1
r  0 , t   
i 0
si t   i  Gi ( 0 )  nt   
i 0
P Gi ( 0 ) d i t   i ci t   i   nt  (20)
where Gi ( 0 ) denotes as ith radiation pattern gain of ith user in the AOA of user 0. This
equation signifies the combination of CDMA and SDMA by the terms Gi ( 0 ) . When
G0 ( 0 ) is normalized to 1, the generated Gi ( 0 ) will be less than 1 if the ith interfering user
is not located at the same AOA as user 0. For a conventional CDMA system, Gi ( 0 ) will take
the value of 1. This signal is then despread with the spreading sequence of user 0 at the
receiver. A correlation-based detector is used to obtain the appropriate decision, z 0 which
can be derived as

z 0  r ( 0 , t )c0 (t   0   e )dt
0
(21)
where  e is the sequence synchronization error, which degrades the auto-correlation

properties. If prefect synchronization is assumed as in the case of directional antenna, where
the multipath fading effect is neglected, then  0   e  0 . Thus, Equation (21) leads to

z0  r ( 0 , t )c0 (t )dt
0
 K 1

 

 P G0 ( 0 ) d 0 (t )c0 (t ) 
T 2
i 1
P Gi ( 0 ) d i (t   i )ci (t   i )c0 (t )
 dt
(22)
0
 n(t )c (t ) 
 0 
 S  I 
7. Probability of Error Evaluation in DSCMA System over AWGN Channel

A bit error rate (BER) expression for DSCMA is derived over MAI from the other K – 1 users
in an AWGN channel. The derivation is performed at the baseband level, which will
simplify the analysis. From the previous section, the first term of (22), S is the transmitted
signal of user 0, where
T
S

0
P G0 ( 0 ) d 0 (t )c02 (t )dt (23)
Considering that G0 ( 0 )  1 , c02 (t )  1 and d 0  1 , thus Equation (23) becomes
S   PT (24)
The term  in (22) is the noise component due to n(t ) in AWGN channel, which
corresponds to the despread term of n(t ) attributes to

  nt c0 t dt
0
(25)
Since n(t ) is the zero mean AWGN having a variance of  2  N 0 2 , thus  is also a zero
mean Gaussian variable and a variance of Var   , which is derived as
Var[ ]  E[ 2 ]
 T T


 0 
 E  n(t )c0 (t )dt n(u )c0 (u )du 
0  (26)
T T

  E[n(t )n(u)]c (t )c (u)dtdu
0 0
0 0
But E[n(t )n(u )] is the auto-correlation of n(t ) , where
N0
E[n(t )n(u )]   (t  u ) (27)
2
Therefore, the variance of (26) becomes
N0 T T
Var[ ] 
2 0 0 
 (t  u )c0 (t )c0 (u )dtdu
N T

 0 c02 (u )du
2 0
(28)
N 0T

2
The term I in (22) is the MAI component of the K – 1 interferers, which is given by
K 1

T
I
i 1

P Gi ( 0 ) d i (t   i )ci (t   i )c0 (t )dt
0
(29)
T
Since
 c (t   )c (t )dt
0
i i 0 is cross-correlation between sequences ci and c0 , thus
 c (t   )c (t )dt  R
0
i i 0 T
i ,0 (30)
and d i (t   i )  1 , therefore I is reduced to
K 1
I   PT 
i 1
Gi ( 0 ) Ri , 0 (31)
The signal to interference plus noise ratio (SINR) is then given as
S2
SINR  (32)
Var    I 2
Therefore, from (24), (28) and (31), the SINR for the DSCMA system can be expressed as
PT 2
SINR  K 1
N 0T
2
 PT 2  G ( )R
i 1
i 0
2
i,0
1
 N 0T K 1


2
 PT 2 
Gi ( 0 ) Ri2,0 
(33)
  i 1 

PT 2
 
 
1
 N K 1

 0 
 2 PT
i 1
Gi ( 0 ) Ri2,0 

Since the amplitude of each mobile user’s signal is P , then the energy per bit is Eb  PT .
The Equation (33) leads to
1
 1
K 1

SINR  
 2 Eb N 0
  i 1
Gi ( 0 ) R 

2
i ,0 (34)
Assuming that the combining noise and interference components have a Gaussian
distribution, then the BER is given as
  
1
 

K 1
 2 
1
BER  Q SINR  Q 
  b N 0
2 E
  i 1
Gi ( 0 ) Ri2, 0  
 
(35)
 
where Q(x) is the Gaussian Q-function.

For the interference limited of DSCMA system where thermal noise is not a factor due to
Eb E
 SF c (36)
N0 N0
where the resultant of Eb N 0 is large enough to cause the thermal noise become negligible.
Therefore, Equation (35) is reduced to
 K 1  
1
  2
BER  Q 
  i 1
Gi ( 0 ) Ri , 0  
2
 
(37)
 
which is a expression of BER performance in DSCMA over AWGN channel.

8. BER Performance and System Capacity in DSCMA System

Two different types of BS antenna, omnidirectional antenna and smart antenna are exploited
for BER performance comparison through simulation. The system simulation will evaluate
the BER performance of DSC algorithm in DSCMA by considering interference from both
intra-cell and inter-cell interferences. The BER expressions over these interferences have
been derived in the previous section. From these BER expressions the system capacity in
DSCMA can be estimated by looking at the number of mobile users that the system can
support at 0.001 BER (BER < 10-3). Since the cell is split into three sectors, the inter-cell
interference sources are only considered from two neighbouring cells for each sector. The
system chip rate is 1.2288 Mcps resulting at a data rate of 4.8 Kbps for sequence length of
about 256 chips per bit. This data rate is used to transmit the multimedia type data. The data
transmission will go through a wireless channel with fourth power of distance loss and 8 dB
shadowing. The voice activity factor is not taken into consideration in this simulation. The
system parameters for the system simulation are summarised in Table 2.
Cell radius, R Unity

Number of sectors per cell 3
Number of interfering neighbouring cells per sector 2
Type of data Multimedia
Spreading factor, SF About 256
System chip rate, W 1.2288 Mcps
Data rate, R 4.8 Kbps
Path loss exponent¸  4.0
Standard deviation of shadowing,  8 dB
Table 2. Summary of the system parameters
The radiation pattern of a smart antenna represents gains of different AOA along a 120o
azimuth span sector. It is assumed that K separate narrow beams and K different spreading
sequences can be generated from BS and directed to each of K users within a sector of interest.
Assume that a sector antenna beamwidth is 120o in the three sectors per cell configuration.
This beamwidth size attributes to a maximum excess delay,  max of 142 chips for chip rate, Rc
of 1.2288 Mcps in separation of 10 Km. Thus, the correlation property, Ri , 0 of the spreading
sequences between ith interfering user and the desired user is randomly taken within 142
chips delay spread. Nevertheless, when the smart antenna is exploited, the beamwidth
becomes narrower as a function of the number of elements, N e . Hence, the maximum excess
delay,  max of this narrower beam will be reduced. The maximum excess delay,  max for the
smart antenna system with different number of elements, N e has been shown in Table 3.
Number of Beamwidth, BW Maximum angular Maximum excess delay,  max

elements, N e (o ) spread,  max (o) ( s ) (chip)
1 120 60 115 142
4 25.5 12.75 15.1 18.5
8 12.75 6.375 7.45 9.15
16 6.375 3.19 3.7 4.56
32 3.19 1.59 1.85 2.27
64 1.59 0.8 0.93 1.14
Table 3. Performance of smart antenna for various N e
This table showed that the narrower beam of a smart antenna will reduce the TOA of a
transmitted signal. This implies that the maximum excess delay of a channel can be
reduced when more elements in a smart antenna system is exploited. To analyse the
system performance in DSCMA, all the simulations are executed by considering smart
antenna systems with N e = 4, 8, 16, 32, 64 in the channel models of AWGN, Rayleigh
fading, and fading with diversity gain. The simulations also consider spreading sequences
such as m-sequence, Gold, Walsh-Hadamard and LAS-ET for performance comparison in
DSCMA system. For the simulation in AWGN channel, Equation (37) from previous
section is used to analyse the BER performance in DSCMA system.
Fig. 7. BER performance of CDMA system for various spreading sequences

For comparison purposes, the BER of conventional CDMA and SDMA systems will be the first
to be evaluated. In the CDMA system, the omnidirectional antenna gain, Gi ( 0 ) for all mobile
users are normalized to one in the whole sector. Therefore, the BER of the CDMA system is
evaluated based on the correlation property of the spreading sequences. Figure 7 shows the
BER performance of CDMA system for various spreading sequences. The system capacity of
the CDMA system for these spreading sequences for BER < 10-3 are shown in Table 4.
Sequences Capacity, N u (BER < 10-3)

m-sequence 18
Gold 29
Walsh-Hadamard 58
LAS-ET 104
Table 4. System capacity of CDMA system for various spreading sequences
On the other hand, in the SDMA system, the correlation property, Ri2, 0 is normalized to one
because it is not a factor of improvement in the conventional SDMA system. The only parameter
that is used to evaluate BER performance of SDMA system is the antenna gain of interfering
users, Gi ( 0 ) in the direction of the desired user. These antenna gains of interfering users are
taken randomly from the radiation pattern of smart antenna within the 120o sector. Figure 8
shows the BER performance of the conventional SDMA for various antenna beamwidths. The
SDMA system capacity of BER < 10-3 for various beamwidths is given in Table 5.
Fig. 8. BER performance of SDMA system for various numbers of elements

Number of elements, N e Capacity, N u (BER < 10-3)

4 1
8 1
16 1
32 2
64 3
Table 5. System capacity of SDMA system for various numbers of elements
In the AWGN channel, it is necessary for DSCMA to perform in perfect synchronous

manner to obtain the orthogonality among the mobile users in zero delay spread. However,
perfect synchronisms rarely exist because each mobile user signal arrives at BS receiver with
different delay. Therefore, the orthogonality between spreading sequences is no longer held
in non-zero delay spread.
The BER performance of the DSCMA system in AWGN channel for various spreading
sequences, vis m-sequence, Gold, Walsh-Hadamard and LAS-ET sequences are shown in
Figures 9a - 9d respectively. In each figure, the BER performance is evaluated by exploiting
the smart antenna with different number of elements, N e = 4, 8, 16, 32 and 64 elements.
Additionally, the system capacity of the DSCMA system based on BER < 10-3 for these
spreading sequences are shown in Tables 6a - 6d respectively.
Fig. 9a. DSCMA system BER performance for m-sequence with different number of antenna
elements, N e in AWGN channel

4 48
8 25
16 173
32 374
64 420
Table 6a. DSCMA system capacity of BER < 10-3 for m-sequence with different number of
antenna elements, N e in AWGN channel
Fig. 9b. DSCMA system BER performance for Gold sequence with different number of

4 69
8 119
16 194
32 259
64 530
Table 6b. DSCMA system capacity of BER < 10-3 for Gold sequence with different number of
Fig. 9c. DSCMA system BER performance for Walsh-Hadamard sequence with different
number of antenna elements, N e in AWGN channel

4 1147
8 6.85 x 103
16 2.75 x 104
32 8 x 104
64 2.4 x 105
Table 6c. DSCMA system capacity of BER < 10-3 for Walsh-Hadamard sequence with
different number of antenna elements, N e in AWGN channel
Fig. 9d. DSCMA system BER performance for LAS-ET sequence with different number of

4 2.4 x 1033
8 5.67 x 1033
16 1.27 x 1034
32 2.04 x 1034
64 2.9 x 1034
Table 6d. DSCMA system capacity of BER < 10-3 for LAS-ET sequence with different number
of antenna elements, N e in AWGN channel
All the evaluated system capacities in DSCMA system here are based on the interference
level that the system can tolerant. Hence, these attained results are not the ideal system
capacity performance. There are some other factors need to be considered in conforming to
this issue such as the totally bandwidth available, the number of spreading sequences that
can be synthesized, and the limitation of system signal processing. Therefore, all the
attained results are only suited for the comparison purpose and are not representing the real
scenario.
In general, it shows that the DSCMA system performance is improved with spreading
sequences of m-sequence, Gold, Walsh-Hadamard and LAS-ET in ascending order as in
conventional CDMA system. This is because the mean square correlation property, E Ri2, 0  
for i  1,2,3,....., K  1 between (K – 1) interfering users and the desired user are decreased in
ascending order from these spreading sequences. Further improvement is exhibited when
the number of elements, N e in smart antenna is increased as in conventional SDMA system.
This is because the interference gain factor, Gi ( 0 ) for i  1,2,3,....., K  1 in the direction of
the desired user,  0 is reduced due to the fact that the sidelobe levels of interfering users
radiation pattern presented in  0 direction is decreased while the number of nulls is
increase when the number of elements in smart antenna is increase.
All the energies from higher number of antenna elements, N e are transformed into high
gain in the main lobe of smart antenna. It can be seen that there is considerable
improvement when the beamwidth of the smart antenna become narrower. The narrower
beam from higher number of antenna elements, N e also contributes to the interference
reduction due to the fact that the probability of users’ beam interfere to each other is low in
narrower beam compared to wider beam. Therefore, higher system capacity can be observed
in DSCMA when number of elements, N e is increased. And this improvement can achieve
as high as 2.9 x 1034 users per sector when LAS-ET sequences together with DSC algorithm
are used. On the other hand, this is interesting to find that in traditional spreading
sequences, their system performances suppose to be appeared in randomness because of
non-uniform distribution in their correlation property within the concerned delay-spread.
This is obviously occurred in m-sequences.
9. Conclusion
It can be concluded that non-uniform traffics can severely degrade the performance of
CDMA and SDMA cellular systems. The DSCMA system described in this chapter is a
double signatures system that can distinguish more users by cancelling the existing
interference in multipath environment. This multiple access system uses LAS-ET sequences
to create IFW near zero delay spread in its cross-correlation function. In order to ensure all
the signal components drop within the IFW, a narrow beam with higher directivity smart
antenna system is exploited. The size of IFW is adapted to the smart antenna half-power
beamwidth using DSC algorithm. Therefore, all the interferences induced in non-uniform
traffics can be dramatically reduced in DSCMA system and thus resulting in higher system
capacity.
10. References
Adachi, F., Garg, D., Takaoka, S. & Takeda, K. (2005). Broadband CDMA techniques, IEEE
Wireless Communications, vol. 12, no. 2, pp. 8 - 18.
Buracchini, E. et al. (1996). Performance Analysis of A Mobile System Based on Combined
SDMA/CDMA Access Technique, IEEE 4th International Symposium on Spread
Spectrum Techniques and Applications Proceedings, vol. 1, pp. 370 - 374.
Cameron, R. & Woerner, B. (1996). Performance Analysis of CDMA with Imperfect Power
Control, IEEE Transactions on Communications, vol. 44, pp. 777 - 781.
Castaeda-Camacho, J. & Lara-Rodriguez, D. (2008). Teletraffic Analysis of an Overlaid

System Using CDMA and TDMA With Cell Coverage Area Restriction, IEEE
Transactions on Vehicular Technology, vol. 57, no. 2, pp. 828 - 846.
Chatovich, A. & Jabbari, B. (1999). Effect of Non-uniform Traffic Load on Erlang Capacity of
CDMA, IEEE 49th Vehicular Technology Conference Proceedings, vol. 1, pp. 816 - 820.
Chen, H. H., Chiu, H. W. & Guizani, M. (2006). Orthogonal Complementary Codes for
Interference-Free CDMA Technologies, IEEE Wireless Communications, vol. 13, no. 1,
pp. 68 - 79.
Chen, H. H., Chiu, H. W. & Guizani, M. (2008). On Next Generation CDMA Technologies:
The REAL Approach for Perfect Orthogonal Code Generation, IEEE Transactions on
Vehicular Technology, vol. 57, no. 5, pp. 2822 - 2833.
Cho, K. et al. (2002). Novel Smart Antennas for Applying SDMA to Cellular Mobile
Communication Systems, IEEE Antennas and Propagation Society International
Symposium Proceedings, vol. 4, pp. 652 - 655.
Choi, W., Andrews, J. G. & Heath, R. W. (2007). Multiuser Antenna Partitioning for Cellular
MIMO–CDMA Systems, IEEE Transactions on Vehicular Technology, vol. 56, no. 5,
Part 1, pp. 2448 - 2456.
Cooper, G. R. & Nettleton, R. W. (1978). A Spread Spectrum Technique for High Capacity
Mobile Communications, IEEE Transactions on Vehicular Technology, vol. VT-27, pp.
264 - 275.
Fang, X. (2002). More Realistic Analysis for Blocking Probability in SDMA Systems, IEE
Proceedings Communications, vol. 149, no. 3, pp. 152 - 156.
Faruque, S. (1996). Cellular Mobile Systems Engineering, Norwood, MA, Artech House.
Feuerstein, M. et al. (1994). Path Loss, Delay Spread, and Outage Models as Functions of
Antenna Height for Microcellular System Design, IEEE Transactions on Vehicular
Technology, vol. 43, no. 3, pp. 487 - 498.
Galvan-Tejada, G. M. & Gardiner, J. G. (2001). Theoretical Model to Determine the Blocking
Probability for SDMA Systems, IEEE Transactions on Vehicular Technology, vol. 50,
no. 5, pp. 1279 - 1288.
Gilhousen, K. S. et al. (1991). On The Capacity of A Cellular CDMA System, IEEE
Transactions on Vehicular Technology, vol. 42, no. 2, pp. 303 - 312.
Godara, L. C. (1997). Applications of Antenna Arrays to Mobile Communications, Part I:
Performance Improvement, Feasibility, and System Considerations, Proceedings of
the IEEE, vol. 85, no. 7, pp. 1031 - 1060.
Gold, R. (1967). Optimal Binary Sequences for Spread Spectrum Multiplexing, IEEE
Transactions on Information Theory, vol. 13, no. 4, pp. 619 - 621.
Golomb, S. W. (1992). Shift Register Sequences, Aeqean Park Press.
Guo, D. & Wang, C. C. (2008). Multiuser Detection of Sparsely Spread CDMA. IEEE Journal
on Selected Areas in Communications, vol. 26, no. 3, pp. 421 - 431.
Harmuth H. F. (1970). Transmission of Information by Orthogonal Functions, Springer-
Verlag.
Hashem, B. & Sousa, E. (1997). On the Capacity of a Cellular DS/CDMA System under Slow
Multipath Fading and Fixed Step Power Control, IEEE 6th International Conference
on Universal Personal Communications Record, vol. 2, pp. 352 - 355.
Heath, R. W. et al. (2004). Construction of Equiangular Signatures for Synchronous CDMA

Systems, IEEE 8th International Symposium on Spread Spectrum Techniques and
Applications Proceeding, pp. 708 - 712.
Huang, C. et al. (2001). A Joint Multiple Access Scheme for Uplink Channels of CDMA
Systems, International Conferences on Info-tech and Info-net (ICII 2001) Proceedings, vol.
2, pp. 323 - 328.
Ishida, Y. et al. (2000). Wideband CDMA Base Station with Co-channel Interference
Canceller, IEEE 51st Vehicular Technology Conference (VTC 2000-Spring) Proceedings,
vol. 1, pp. 293 - 297, Tokyo.
Kasami T. (1966). Weight Distribution Formula for Some Class of Cyclic Codes, Technical
Report R-285 (AD 632574), Coordinated Science Laboratory, Univ. of Illinois,
Urbana.
Kim, S. W. et al. (2001). Performance of Smart Antennas with Adaptive Combining at
Handsets for the CDMA2000 System, International Conference on Third Generation
Wireless and Beyond Proceeding, San Francisco, pp. 882 - 887.
Lee, W. C. Y. (1991). Overview of Cellular CDMA, IEEE Transactions on Vehicular Technology,
vol. 40, no. 2, pp. 291 - 302.
Lee, W. C. Y. (1998). Mobile Communications Engineering, McGraw Hill.
Li, D. (1999). A High Spectrum Efficient Multiple Access Code, Fifth Asia-Pacific Conference
on Communications and Fourth Optoelectronics and Communications Conference
(APCC/OECC '99) Proceedings, vol. 1, pp. 598 - 605.
Li, D. (2003). The Perspectives of Large Area Synchronous CDMA Technology for the
Fourth-Generation Mobile Radio, IEEE Communications Magazine, vol. 41, no. 3, pp.
114 - 118.
Liberti, J. C. Jr. & Rappaport, T. S. (1994). Analytical Results for Capacity Improvements in
CDMA, IEEE Transactions on Vehicular Technology, vol. 43, no. 3, pp. 680 - 690.
Liberti, J. C. Jr. & Rappaport, T. S. (1998). Smart Antennas for Wireless Communications,
Prentice Hall PTR, Upper Saddle River, NJ.
Lien, S. Y. & Cherniakov, M. (1998). Analytical Approach for Multipath Delay Spread Power
Distribution, IEEE Global Telecommunications Conference (GLOBECOM 98)
Proceedings, The Bridge to Global Integration, vol. 6, pp. 3680 - 36858.
Naguib, A. F. et al. (1994). Capacity Improvement with Base-Station Antenna Arrays in
Cellular CDMA, IEEE Transactions on Vehicular Technology, vol. 43, no. 3, pp. 691 -
698.
Ng, B. & Sousa, E. S. (1998). Performance Enhancement of DS-CDMA System for Wireless
Local Loop, IEEE International Telecommunications Symposium (ITS) Proceedings,
vol.1, pp. 78 - 82.
Ng, C. K., Noordin, N. K., Khatun, S., Mohd Ali, B., Jamuar, S. S. & Ismail, M. (2008).
Directional Diversity of Smart Antenna in LAS CDMA Systems, Wireless Personal
Communications, Springer Netherlands, vol. 46, pp. 305 – 316.
Ng, C. K., Noordin, N. K., Mohd Ali, B., Jamuar, S. S. & Ismail, M. (2009). Spectrum
Efficiency Enhancement in Dynamic Space Coded Multiple Access (DSCMA)
System, Wireless Personal Communications, Springer Netherlands, 0929-6212 (Print)
1572-834X (Online), DOI 10.1007/s11277-009-9682-7.
Niemela, J. & Lempiainen, J. (2003). Impact of the Base Station Antenna Beamwidth on
Capacity in WCDMA Cellular Networks, The 57th IEEE Semiannual Vehicular
Technology Conference (VTC 2003-Spring) Proceedings, vol. 1, pp. 88 - 84.
Peterson, R. L. et al. (1995). Introduction to Spread Spectrum Communications, Prentice Hall
International Edition.
Popovic, B. M. (1997). Efficient Despreaders for Multi-code CDMA Systems, IEEE 6th
International Conference on Universal Personal Communications Record, vol. 2, pp. 516 -
520.
Rapajic, P. B. (1998). Information Capacity of The Space Division Multiple Access Mobile
Communication System, IEEE 5th International Symposium on Spread Spectrum
Techniques and Applications Proceedings, vol. 3, pp. 946 - 950.
Rupf, M. & Massey, J. L. (1994). Optimum Sequence Multisets for Synchronous Code-
Division Multiple-Access Channels, IEEE Transactions on Information Theory, vol. 40,
no. 4, pp. 1261 - 1266.
Stańczak, S. et al. (2001). Are LAS-Codes a Miracle?,” IEEE GLOBECOM ’01 Proceedings, vol.
1, (San Antonio, Texas), pp. 589 - 593, November 2001.
Thompson, J. S. et al. (1996). Smart Antenna Arrays for CDMA Systems, IEEE Personal
Communications, pp. 16 - 25.
Uthansakul, P. & Uthansakul M. (2002). Number of Elements in Smart Antenna System
Effects on Both Perfect and Imperfect Power Control for Reverse Link of CDMA
Cellular System, The International Technical Conference on Circuits/Systems, Computer
and Communications Proceedings, Thailand.
Wang, X., Ahonen, T. & Nurmi, J. (2007). Applying CDMA Technique to Network-on-Chip,
IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 15, no. 10, pp.
1091 - 1100.
Wei, H. et al. (2005). Interference-Free Broadband Single and Multicarrier DS-CDMA, IEEE
Communications Magazine, vol. 43, no. 2, pp. 68 - 73.
Welch, L. R. (1974). Lower Bounds on the Maximum Cross Correlation of Signals, IEEE
Transactions on Information Theory, vol. IT-20, pp. 397 - 399.
Wu, J. S. et al. (1998). Hot-Spot Traffic Relief with A Tilted Antenna in CDMA Cellular
Networks, IEEE Transactions on Vehicular Technology, vol. 47, no. 1, pp. 1 - 9.
Xie, S. & Rahardja, S. (2005). Performance Evaluation for Quaternary DS-SSMA
Communications with Complex Signature Sequences over Rayleigh-Fading
Channels, IEEE Transactions on Wireless Communications, vol. 4, no. 1, pp. 266 - 277.
Yu, J. et al. (2004). Reverse-Link Capacity of Power-Controlled CDMA Systems with
Beamforming, IEEE Transactions on Vehicular Technology, vol. 53, no. 5, pp. 1423 -
1433.
Zheng, Z. et al. (1996). Capacity Improvement with Base Station Antenna Arrays in Cellular
CDMA, IEEE 46th Vehicular Technology Conference, Mobile Technology for the Human
Race Proceedings, vol. 2, pp. 1303 - 1306.
Video Streaming in Evolving Networks under Fuzzy Logic Control 613
28
X
Video Streaming in Evolving Networks

under Fuzzy Logic Control
Martin Fleury, Emmanuel Jammeh, Rouzbeh Razavi,
Sandro Moiron and Mohammed Ghanbari
University of Essex
United Kingdom
1. Introduction
Internet Protocol TV (IPTV) and other video streaming services are expected to dominate
the bandwidth capacity of evolving telecommunications networks. In fact, managed, all-IP
networks are under construction with video largely in mind. In these networks, a variety of
broadband access networks will form the final link to the home across which video is
streamed from proprietary servers. Co-existing with these networks or as an extension of
them, the traditional, best-effort Internet will continue to support applications such as
video-on-demand, Peer-to-Peer (P2P) streaming, and video clip selection.
This Chapter will begin by broadly surveying research and development of video streaming
across evolving telecommunications networks under the categories of best-effort and
managed networks. In particular, the Chapter will introduce the different forms of control
that are necessary to ensure the quality of the delivered video, whether live or pre-encoded
video, when for the latter bitrate transcoding may be required. The concentration will be on
single-layer unicast distribution though simulcast, bandwidth reservation, multicast, and
other forms of delivery will be touched upon.
With the growth in computational power, rate-distortion (R-D) control has emerged as an
effective way to optimise the output encoded bitstream. In R-D control, the optimal choice
of compression rate (and hence codec output bitstream rate) relative to improvement in
video quality is sought. (The choice is generally found through the method of pre-set
Lagrangian multipliers with trial codec settings repeatedly tested to find the best result.)
Though attempts have been made to integrate R-D control and network congestion control
(Chou & Miao, 2006), often congestion control has been considered separately as best-effort
networks are prone to fluctuations in available bandwidth. In all-IP networks, though traffic
in the core of the network will be switched, the variety of access network types poses a
problem to servers that may be oblivious of the final hop technology. When broadband
wireless (IEEE 802.16 d,e, WiMAX (Fleury et al., 2009)) access links are involved error
control is especially important.
The Chapter will then specialise to consider in what ways fuzzy logic controllers (FLCs)
have been applied to rate control and congestion control. A feature of this Chapter will be
consideration given to the growing prominence of type-2 fuzzy logic in networked
multimedia control, bringing greater robustness in the face of unforeseen network

conditions. To illustrate the application of fuzzy logic control, the Chapter will include two
case studies. One of these will show how type-2 logic can improve upon type-1 logic, both of
which forms of congestion control improve upon traditional controllers respectively within
managed networks and within the Internet. The design of the controllers is illustrated for
non-specialists, showing how type-2 controllers extend type-1 FLCs. From the results of
simulations, FLCs in a managed network are shown to be superior to traditional congestion
controllers. Transcoding is presented as an effective way to apply fuzzy logic control.
With the advent of IPTV, statistical multiplexing has again become an important issue for
managed networks. Unlike traditional broadcast channels, network distribution may
involve changes in available bandwidth and streaming conditions because of the variety of
possible access types and coexisting traffic. In the second case study, an FLC is used to
integrate two video complexity measures to achieve an effective combination of video or TV
channels. The intention is dynamically to reduce the bandwidth allocation to channels that
are already of high enough quality and increase the quality of streams with potentially
greater coding complexity. Simulation results are presented to show the value of the
approach applying the state-of-the-art H.264 codec. This case study will also include a
review of other forms of statistical multiplexing.
2. Video streaming
2.1 Streaming basics
In video streaming, the compressed video bitstream is transmitted across a network to the
end user’s decoder (prior to display) without the need for storage other than in temporary
buffering. Its advantage over progressive download from a network point-of-view is that
the throughput is only that required to render the video at the user’s display. Download
risks overloading the network by too high a throughput. If download is not progressive,
then the user has to wait an intolerable time before (say) viewing a 2 hr movie. There are
also issues of commercial confidentiality if the video is stored on the user’s machine.
Downloading video does permit Variable Bitrate (VBR) to be transported. In VBR, the codec
quantization parameter (QP) is fixed leading to a constant quality. The alternative is to set a
target bit rate for Constant Bitrate (CBR) video and allow fluctuating quality but with a gain
in controllability. The main problem with VBR is that due to a strong variation in the
number of bits allocated to each of the frame types (Lakshman et al., 1998) the rate is highly
variable (Van der Auwera & Reisslein, 2009). Long video streams are also not statistically
stationary in time, which causes a problem when attempting to model video input to a
network. This variability is accentuated in the H.264/Advanced Video Codec (AVC)
(Schwarz, 2007) and it is reported (Van der Auwera et al., 2008) that the variability is
accentuated the more so in the Scalable Video (SVC) extension to H.264, with the result that
prior smoothing of VBR streams is contemplated. (The reason for increased variability is
attributable to the increased number of motion estimation modes in H.264/AVC and in
H.264/SVC, the addition of hierarchical B-frames.)
In temporal smoothing, multiple encoded frames are accumulated so that the compressed
bitstream can be packetized and sent at a desired average bitrate. This form of traffic
shaping has the disadvantage for video streaming that end-to-end latency is increased by
the number of frames accumulated. For ‘conversational’ video services, which have an
additional latency introduced by the need to encode each frame, the effect on the viewer can
be disconcerting. Ideally end-to-end latency should be no longer than 200 ms. For this
reason, in services such as teleconferencing and videophone, CBR is preferable. However,
for pre-encoded video at a significant cost in computational complexity (Salehi et al., 1998) it
is also possible through optimal smoothing to send video frames (or rather their compressed
bitstream) in advance of their decode time, provided it is known that overflow (or
underflow) at the playback buffer will not occur. In the best-effort Internet, jitter introduced
by cross-traffic congestion will disrupt these calculations but in those network cores in
which ATM or virtual ATM is still in place optimal smoothing has a role. Unfortunately, the
presence of access networks of differing types prior to the consumer’s home, or reduced
bandwidth links prior to campus and corporate networks introduces an ill-behaved section
within the end-to-end path.
Video is known as a delay-sensitive service but in fact there are varying levels of
intolerance, and a limit of 200 ms has been mentioned. However, for one-way streaming the
delay requirements are less stringent. For example, channel swapping or VCR-like control is
restricted to 500 ms intervals, because anchor or key frames at which switching can occur
are placed at these intervals within a stream. Another form of delay is start-up delay, with
Video-on-Demand (VoD) services hoping to make this imperceptible (< 20 ms), which is
perhaps possible on the Internet if the Resource ReSerVation Protocol (RSVP) (Zhang et al.,
1997) were to be widely deployed. Variation in delay (jitter) is also important in terms of
media synchronization (between audio and video) (Blakowski & Steinmetz, 1996). However,
there are also display deadlines to be met, implying that a jitter buffer should be
dimensioned to absorb any variation in delivery (assuming Internet delivery). For reference
frames (one used for predictive motion estimation), their data is still of value for decoding
future frames even if they miss their display deadline. Too large a receiver buffer will lead
to increased end-to-end latency and start-up delay, while too small a buffer may cause
overflow. This is why adaptive buffers have been contemplated in the research literature
(Kalman et al., 2002).
Video streaming is also known as a loss-tolerant service. However, this is misleading as the
loss of more than 10% of packets will generally lead to a noticeable deterioration in the
quality of the video unless: error-resilience measures have been taken; error control through
some form of acknowledgements (ACKs) is used (as in the Windows Media system);
Forward Error Correction (FEC) is in place; or error concealment can be applied. A
combination of these methods is preferable as part of an error response strategy and
unequal error protection (UEP) is possible. In UEP, protection is prioritized according to
compressed video content or the structure of the video. Acknowledgments are possible but
their impact on delay must always be judged. For example, in (Mao et al., 2003) layered
streaming was attempted across an ad hoc network in which multi-hop routing and broken
links can lead to high levels of delay. In layered streaming (Mao et al., 2003), a more
important base layer allows a basic reconstruction of the video while one or more
enhancement layers can improve the quality. However, because of the high risk of delay, in
(Mao et al., 2003) it was only possible to send one ACK at most to secure the base layer.
Though FEC schemes with linear decoder complexity (Raptor codes, a variety of rateless
erasure codes) have been developed (Shokrollahi, 2006), FEC generally leads to delay in
encoding. Because of the additional delay involved in sending acknowledgments (or
negative acknowledgments), when there is a long round-trip-time careful engineering needs
to be applied if rateless erasure coding is to be used. In rateless or Fountain coding

(MacKay, 2005), additional redundant data can always be generated, while in conventional
forms of channel coding such as Reed-Solomon, there is a threshold effect whereby if the
channel noise or packet erasures pass the level of protection originally provided then all
data are lost.
Error resilience techniques, the range of which have been expanded in the H.264 codec
(Wenger, 2003), are based on source coding. Error resilience results in lower-delay and as
such is suitable for real-time, interactive video streaming, especially video-telephony and
video conferencing. However, due to the growing importance of broadband wireless access
networks, error resilience is also needed to protect video streaming to the home. This is
because physical-layer FEC is already present and, therefore, application-layer FEC may
duplicate its role. The exception is if application-layer FEC can be designed to act as an outer
code after inner coding at the physical layer, in the manner of concatenated channel coding.
Compressed frame data is often split into a number of slices each consisting of a set of
macroblocks. In the MPEG-2 codec, slices could only be constructed from a single row of
macroblocks. Slice resynchronization markers ensure that if a slice is lost then the decoder is
still able to continue with entropic decoding. Therefore, a slice is a unit of error resilience
and it is normally assumed that one slice forms a packet, after packing into a Network
Abstraction Layer unit (NALU) in H.264. Each NALU is encapsulated in a Real Time
Protocol (RTP) packet. Consequently, for a given frame, the more slices the smaller the
packet size and the less risk of packet loss through bit errors.
In H.264/AVC, by varying the way in which the macroblocks are assigned to a slice (or
rather group of slices), Flexible Macroblock Ordering (FMO) gives a way of reconstructing a
frame even if one or more slices are lost. Within a frame up to eight slice groups are
possible. A simple FMO method is to continue a row of macroblocks to a second row, Figure
1a, but allow disjoint slice groups (Lambert et al., 2006). Regions of interest are supported,
Figure 1b. Checkerboard slice group selection, Fig, 1c allows one slice group to aid in the
reconstruction of the other slice group (if its packet is lost) by temporal (using motion vector
averaging) or spatial interpolation. Assignment of macroblocks to a slice group can be
general (type 6) but the other six types pre-define an assignment formula, thus reducing the
coding overhead from providing a full assignment map.
Data partitioning in H.264/AVC separates the compressed bitstream into: A) configuration
data and motion vectors; B) intra-coded transform coefficients; and C) inter-coded
coefficients. This data form A, B, and C partitions which are packetized as separate NALUs.
The arrangement allows a frame to be reconstructed even if the inter-coded macroblocks in
partition C. are lost, provided the motion vectors in partition A survive. Partition A is
normally strongly FEC-protected at the application layer or physical layer protection may be
provided such as the hierarchical modulation scheme in (Barmada et al., 2005) for broadcast
TV. Notice that in codecs prior to H.264, data partitioning was also applied but no
separation into NALUs occurred. The advantage of integral partitioning is that additional
resynchronization markers are available that reset entropic encoding. This mode of data
partitioning is still available in H.264 and is applied to I-frames.
slice group 0 slice group 1
slice group 0
slice group 1
slice group 0
slice group 2
slice
slice group 0
group 1
slice group 2
slice group 1
Fig. 1. Example FMO slice groups and types (after (Lambert, 2006) a) Continuing row (type
0) b) geometrical selection (type 2) c) checkerboard selection (type 1)
The insertion of intra-coded macroblocks into frames normally encoded through motion-
compensated prediction allows temporal error propagation to be arrested if matching
macroblocks in a previous frame are lost. Intra-refresh through periodic insertion of I-frames
with all macroblocks encoded through spatial reference (intra-coded) is the usual way of
catching error propagation. However, I-frames cause periodic increases in the datarate when
encoding at a variable bitrate. They are also unnecessary if channel switching points and
VCR functions are not required.
This brief review by no means exhausts the error-resilience facilities in H.264, with
redundant frames, switching frames, and flexible reference frames also considered in
(Stockhammer & Zia, 2007). We have referred to H.264/AVC anchor frames as I-frames for
consistency with previous codecs. In fact, H.264 uses Instantaneous Decoder Refresh (IDR)-
frames for the same purpose, whereas H.264 I-frames allow motion estimation reference
beyond the Group of Pictures boundary.
Error concealment (Wang & Zou, 1998) is the process of concealing errors at the decoder.
However, the form of error concealment is implementation dependent because of the
complexity of these algorithms. In fact, for reasons of speed, previous frame replacement is
often preferred. If lost frames are replaced by the last frame to arrive successfully there is a
danger of freeze frame effects. When there is rapid motion or scene cuts then partial
replacement of macroblocks from the previous frame will result in obvious blocky effects.
For error concealment in H.264/AVC (Vars & Hannuksela, 2001) the motion vectors of
correctly received slices are computed if the average motion activity is sufficient (more than
a quarter pixel). Research in (Vars & Hannuksela, 2001) gives details of which motion vector
to select to give the smoothest block transition. It is also possible to select the intra-coded
frame method of spatial interpolation, which provides smooth and consistent edges at an
increased computational cost. Experience shows a motion-vector-based method performs
best except when there is high motion activity or frequent scene changes (Kim & Kim, 2002).
2.2 Streaming systems

In networked video delivery, systems are classically divided (Chou, 2007) into streaming
and broadcast systems. In the former, video is pre-encoded before storage and access by a
server, while in the latter there is no storage before server access and multicast over a
network. A further distinction in this model is that in streaming a control path exists,
whereas the presence of many receivers in a broadcast system means that feedback would
be impossible to manage. Feedback can be used for congestion control but it can also return
VCR commands, typically through the Real Time Streaming Protocol (RTSP) (Schulzrinne et
al., 1998). Nevertheless, it is possible to stream both pre-encoded and online or live video
because, after feedback notification of congestion, the streaming rate can be changed
through bitrate transcoding (Assunção & Ghanbari, 1997) (Sun et al., 2005). One problem
that fast transcoding may face in the H.264 codec is error drift when transcoding I-frames
(Lefol et al., 2006).
Scalable video also allows rate control as a response to network conditions or target device
capability but a full discussion of the variety of multi-layer or scalable options such as Fine
Grain Scalability (Radha et al., 2001), Multiple Description Coding (Wang, 2005), signal-to-
noise ratio (SNR) scalability (Pesquet-Popescu et al., 2006) would require another chapter.
Rich though the scalable options are commercial Internet operators seem to prefer simple
schemes such as simulcast as used by RealVideo. In simulcast, multiple streams are stored
(or encoded online) at different rates and selected according to network conditions. In
H264/AVC, stream switching frames allow a smoother transition between low and higher
quality stream at lower cost in bandwidth than through switching at I-frames.
At the target device, video is first buffered in a playout buffer, decoder or client buffer (there
are various alternative names) prior to access by the decoder. This buffer will vary in size
depending on the capabilities of the device. Large buffers are not advisable for battery-
powered devices because of both active and passive energy consumption. Nevertheless
some buffering is required to absorb variation of delay (jitter) over the network.
Because of motion-compensated prediction coding it is always necessary to store packets
prior to decode, especially if VBR is in use. An additional render buffer, able to store a few
frames prior to display, is also generally present. Apart from buffer overflow in the
intermediate buffers of routers through congestion, buffer overflow at the playout buffer is
also possible. Packets arriving too late for their display or decode deadlines may also be
dropped. It is also possible, because of jitter, for buffer underflow to occur. In fact, in the
Windows Media system (Chou, 2007) the receiver monitors the buffer level to detect
network congestion. Again like RealVideo, Windows Media uses simulcast, with the
receiver signaling the server to swap to a lower rate stream when it detects congestion.
However, the Windows Media receiver or client is not only reliant on buffer monitoring,
because packet loss at the receiver is also taken into account.
3. Congestion control
In this Section, the focus is on congestion control of single stream unicast for IPTV and other
multimedia services. Because the main thrust in congestion control research is to provide an
enhanced service through VBR delivery, this Section concentrates on that whereas in Section
4 on statistical multiplexing, multi-channel delivery of CBR streams is considered. The latter
is likely to be a broadcast service.
3.1 IPTV and unicast streaming

Real-time video applications, such as IPTV, video-on-demand (VoD), and network-based
video recorder interest telecommunication companies, because of their high bitrates, though
they also risk overwhelming existing networks if it is not possible to control their flows. The
unicast variety of IPTV is very attractive because it allows streaming of individual TV
programs at a time chosen by the end user. Broadly speaking, two types of heterogeneous
delivery network exist: 1) the familiar Internet, with best-effort Internet Protocol (IP)
routing, i.e. an unmanaged IP network; and 2) All-IP networks, which retain IP packet
framing but, particularly in the network core, switch packets (across Clos switches) rather
than employ packet routers, i.e. a managed IP network. These IP networks are generally
referred to as converged networks, as they combine a traditional telephone service (through
Voice-over-IP) with data delivery (normally high speed Internet access) and TV (through
IPTV). The marketing term for such a combined service is ‘triple-play’ and if mobility is
added then this term becomes ‘quadruple-play’.
IPTV services are in active commercial development for converged telephony networks,
such as British Telecom's 21st Century Network (21CN) (Geer, 2004) or the all-IP network of
KPN in the Netherlands.. Within the 21CN, video streaming is sourced either from
proprietary servers or from an external Internet connection, with best-effort routing. Before
distribution from the server to individual users, multiple videos streams will share a
multimedia channel, an example being MPEG-2 Transport Stream which serves for
H.264/AVC pre-encoded streams. These video streams could represent different TV
channels that can be selected by the IPTV user. However, when the multimedia channel
leaves the core network it is commonly delivered across an access network such as
Asymmetric Digital Subscriber Line (ADSL) (Zheng & Liu, 2000), when different delivery
conditions apply.
On the Internet, video streams must coexist with other data traffic, while in emerging All-IP
networks multimedia traffic may predominate. In an All-IP network, as in the Internet, a
capacity restriction may still exist at the connection between the network core and the access
network, of which the technology can be cable (Vasudevan et al., 2008), broadband wireless
(IEEE, 2004), or connections to the Video Serving Office (Han et al., 2008) from which video
is typically distributed over Asymmetric Digital Subscriber Line (ADSL) connections. Note
also that Internet traffic may be directed through an All-IP network by means of the
common agency of IP framing.
In the Internet, a tight link (or more loosely a bottleneck), which commonly exists at the
network edge before a corporate or campus network (Cisco, 2000), is the link of minimum
available bandwidth on a network path. Strictly the term ‘bottleneck’ defines the bandwidth
capacity of a network path, which while the path exists is a constant, though the term may also
be loosely applied to a tight link. A tight link is a dynamic concept, as its location will vary
firstly over time according to background traffic patterns and secondly according to the
network path’s route, which is not fixed because of dynamic routing on the Internet. These two
factors can create uncertainty in any video streaming response. Available bandwidth is
restricted by coexisting cross-traffic, which is most likely carried by the Transmission Control
Protocol (TCP) and predominantly originates from web-servers or P2P file transfer (Xie et al.,
2007). Transport-layer protocols like TCP, sitting above IP, are responsible for end-to-end
negotiation of delivery between applications. On All-IP networks, coexisting traffic across a
network sub-channel or pipe is more likely to arise from other proprietary video servers and
be carried by the minimal User Datagram Protocol (UDP) as directed by congestion
controllers. A pipe is a virtual bandwidth restriction imposed by quality-of-service
requirements that must balance the requirements of other types of traffic and the capacity of
the access network. As in the Internet, All-IP congestion controllers should be end-to-end over
the network path, allowing a general solution in the sense that the nature of the access network
bottleneck may not be known in advance. In an All-IP network, statistical multiplexing of VBR
video sources within a video pipe may increase its efficiency but there is no spare capacity for
greedy acquisition of bandwidth by independently controlled video servers. We return to the
subject of statistical multiplexing within the IPTV pipe in Section 4.
Congestion control is vital to avoid undue packet loss from the fragile compressed video
stream. At the sub-frame level, because variable-length coding (VLC) prior to outputting
the bitstream introduces a dependency between each encoded symbol, there is fragility that
error resilience techniques such as decoder synchronization markers and reversible VLC
only partially address. Because successive video frames are broadly similar (except at scene
cuts and changes of camera shots), only the difference between successive frames is encoded
in order to increase coding efficiency. Consequently, at the frame-level, removing temporal
redundancy introduces a dependency on previously transmitted data that implies lost
packets from reference frames will have an impact on future frames.
Unicast video streaming, which brings increased flexibility and choice to the viewer over
multicast delivery, is achieved by determining the available bandwidth and adapting the
video rate at a live video encoder or an intermediate transcoder. Fuzzy Logic Control (FLC)
is suited to congestion control (Jammeh et al., 2007), because of the inherent looseness in the
definition of congestion and the uncertainty in the network measurements available,
together with the need for a real-time solution. Within video coding it has previously found
an application (Grant et al., 1997) in maintaining a constant video rate by varying the
encoder quantization parameter according to the output buffer state. This is a complex
control problem without an analytical solution. Fuzzy logic is gaining acceptance in the
video community, witness (Rezaei et al., 2008), but it turns out that further improvements
are possible with interval type-2 (IT2) fuzzy logic.
3.2 Fuzzy logic control for congestion

In our application, FLC of congestion is a sender-based system for unicast flows. The
receiver returns a feedback message indicating changes to the delay experienced by video
stream packets crossing the Internet. This allows the sender to compute the network
congestion level and from that the FLC estimates the response. The same controller also
should be able to cope with a range of path delays and with video streams with differing
characteristics in terms of scene complexity, motion, and scene cuts.
Traditional, type-1 FLC is not completely fuzzy, as the boundaries of its membership
functions are fixed. This implies that there may be unforeseen traffic scenarios for which the
existing membership functions do not suffice to model the uncertainties in the video stream
congestion control task. IT2 FLC can address this problem by extending a Footprint-of-
Uncertainty (FOU) on either side of an existing type-1 membership function. In IT2 fuzzy
logic, the variation is assumed to be constant across the FOU, hence the designation
`interval'. Though the possibility of type-2 fuzzy systems has been known for some time
(Zaddeh, 1975), only recently (Mendel, 2007) have algorithms become available to calculate
an IT2 output control value at video rate. The first IT2 controllers (Hagras, 2007) are now
emerging, in which conversion or retyping from fuzzy IT2 to fuzzy type-1 takes place before
output. For video streaming there are important practical advantages. Not only does such
a controller bring confidence that re-tuning will not be needed when arriving traffic displays
unanticipated or un-modeled behavior but the off-line training period required to form the
membership functions can be reduced.
We now compare type-1 FLC for congestion control of video streaming to an IT2 FLC and
compare the performance in the presence of measurement noise that is artificially injected to
test the relative robustness. The delivered video quality in terms of Peak Signal-to-Noise
Ratio (PSNR) is equivalent to the successful type-1 FLC when the measurement noise is
limited and under test results in a considerable improvement when the perturbations are
large. We go on to compare the IT2 FLC to a non-adaptive approach and to congestion
control by two well-known controllers, TCP-friendly Rate Control (TFRC) (Handley et al.,
2003) and TCP Emulation at Receivers (TEAR) (Rhee et al., 2000), one sender-based and the
other receiver based. These are tested by their ability to support multiple broadband
connections over an all-IP network. However, firstly we introduce fuzzy logic control.
3.3 Fuzzy logic control

Figure 2 is a block diagram of FLC of congestion, with two inputs, the packet delay factor,
df, and delay samples to form a trend (whether packet delay is increasing or decreasing).
The formation of these inputs is described in Section 3.4. These inputs are converted to
fuzzy form, whereby their membership of a fuzzy subset is determined by predetermined
membership functions. This conversion takes place in the fuzzifier and trend test units of
Figure 1. The fuzzy outputs are then combined in the inference engine through fuzzy logic.
Fuzzy logic is expressed as a set of rules which take the form of linguistic expressions. These
rules express experience of tuning the controller and, in the methodology, are captured in a
knowledge database. The inference engine block is the intelligence of the controller, with the
capability of emulating the human decision making process, based on fuzzy-logic, by means
of the knowledge database and embedded rules for making those decisions. Lastly, the
defuzzification block converts inferred fuzzy control decisions from the inference engine to
a crisp or precise value, which is converted to a control signal. The control signal causes the
quantization parameter of the video stream to be changed, thus adjusting the output
bitstream.
Fig. 2. FLC delay-based congestion controller

,QDIX]]\VXEVHWHDFKPHPEHULVDQRUGHUHGSD
LUZLWKWKHILUVWHOHPHQWRIWKHSDLUEHLQJD
PHPEHURIDVHW S DQGWKHVHFRQGHOHPHQWEHLQJWKHSRVVLELOLW\LQWKHLQWHUYDO>@WK
PHPEHULVLQWKHIX]]\VXEVHW7KLVVKRXOGEH
FRPSDUHGZLWKD%RROHDQVXEVHWLQZKLFKHYHU\
PHPEHURIDVHW6LVDPHPEHURIWKHVXEVHWZLWKSUREDELOLW\WDNHQIURPWKHVHW^‘LQZKLFK
DSUREDELOLW\RIUHSUHVHQWVFHUWDLQPHPE
HUVKLSDQGUHSUHVHQWVQRQPHPEHUVKLS
7KH )/& GHWHUPLQHV LQFLSLHQW FRQJHVWLRQ
IURP RQH ZD\ SDFNHW TXHXLQJ GHOD\ LQ
LQWHUPHGLDWHURXWHUEXIIHUV7KHTXHXLQJGHOD\
LVDPHDVXUHRIQHWZRUNFRQJHVWLRQDQGWKH
UDWLR RI WKH DYHUDJH TXHXLQJ GHOD\ WR WKH PD[LPXP TXHXLQJ GHOD\ LV D PHDVXUH R
ERWWOHQHFNOLQNEXIIHUIXOOQHVV)RU i
HDFKUHFHLYHGSDFNHWLQGH[HGE\
OWDi = Tr − Ts,
ZKHUHTr LV WKH UHFHLYH WLPH RI WKH FXUUHQW SDFNHW

Ts LV WKH
DQG
WLPH WKH SDFNHW ZDV VHQW
:KHQLWLVDSSURSULDWHWKHFRPSXWHGOWDiXSGDWHVWKHPLQLPXPDQGPD[LPXPRQHZD\
GHOD\V OWDV OWDmin DQGOWDmax RQ D SDFNHWE\SDFNHW EDVLV 6XEVHTXHQWO\ WKH
PD[LPXPTXHXLQJGHOD\LVIRXQGDV maxQD = OWDmax − OWDmin
QDiLVFRPSXWHGIURPWKHPHDVXUHGGHOD\DQG
7KHTXHXLQJGHOD\RYHUWKHQHWZRUNSDWK
WKHPLQLPXPGHOD\
QDi = OWDi − OWDmin
DQG DQ H[SRQHQWLDOO\ZHLJKWHG DYHUDJHHXLQJ RI WKHGHOD\

TX IRU WKH LWK UHFHLYHG SDFNHWLV
IRUPHGE\
avgQDi = (1 − α) × avgQDi−1 + α × QDi
ZKHUH  1 LV D IRUJHWWLQJ FRQVWDQW ,Q

α WHVWV
ZDV VHW WR $ GHOD\ IDFWRUDf LV
FRPSXWHGIURPWKHDYHUDJHTXHXLQJGHOD\DQGWKHPD[LPXPTXHXLQJGHOD\
Df = avgQDi /maxQD
ZKHUHDf UDQJHV EHWZHHQ >@ ZLWK LQGLFDWLQJ

LQFLSLHQW
QR FRQJHVWLRQ LQGLFDWLQJ IXOO
EORZQ FRQJHVWLRQ ZLWK VKDGHV RI LQ FRQJHVWLRQ EHWZHHQ Df
FLSLHQW DQG
LV DQ HDUO\
QRWLILFDWLRQRIFRQJHVWLRQDQGLV
WKHILUVWLQSXWWRWKH)/&
$WUHQGDQDO\VLVPHWKRGLVXVHG
WRGHWHUPLQHWKHJHQHUDOWUHQG
RIWKHDYHUDJHGHOD\,QHDFK
PHDVXUHPHQWHSRFKDQXPEHU k RITXHXHGHOD\VDPSOHVDUHJURXSHGLQWR
τJURXSVZKHUH
  k :HXVHWKHSDLUZLVHFRPSDULVRQWHVW3&7WRGHWHUPLQHWKHRYHUDOOWUHQGRIWKH
TXHXHLQJGHOD\DVVKRZQLQ
 I (M i
 M i 1 )
TPCT  i 2
 1
ZKHUHMiLVWKHPHGLDQRIJURXS
iDQGI(X)LVLI XKROGVDQGRWKHUZLVH7KHYDOXHRI
TPCT
LVVHQWEDFNWRWKHVHQGHUZKHUHDIX]]LILHU
GHWHUPLQHVZKHWKHUWKHOHYHOZDVLQFUHDVLQJR
QRWDFFRUGLQJWRDPHPEHUVKLSIXQFWLRQ
IT2 input membership functions for Df and trend are constructed, Figure 3, as an extension
KDWWKH of the type-1 FLC through an FOU at the boundaries of the formerly crisp (fixed)
membership functions. Assuming the usual singleton input of Df (or TPCT ), an interval set
K requires just an upper and lower value to be resolved to form the resulting FOU in the
corresponding output set. For example, Figure 4 shows two IT2 membership functions for
input sets A and B, each with an FOU. Singleton input X is a member of each with different
degrees of membership. Strictly, an infinite number of membership functions (not all
RI necessarily triangular) can exist within the FOUs of sets A and B, but IT2 sets allow the
upper and outer firing levels to be taken, as shown in Figure 4. The minimum operator
(min) acts as a t-norm on the upper and lower firing levels to produce a firing interval.
The firing interval serves to bind the FOU in the output triangular membership function
shown to the right in Figure 4. The lower trapezium outlines the FOU, which itself consists
of an inner trapezoidal region that is fixed in extent. The minimum operator, also used by us
as a t-norm, has the advantage that its implementation cost is less than a product t-norm. (A
t-norm or triangular norm is a generalization of the intersection operation in classical logic.)
Once the FOU firing interval is established, Center-of-Sets type reduction was applied by
means of the Karnik-Mendel algorithm, which is summarized in (Mendel, 2007). Type
reduction involves mapping the IT2 output set to a type-1 set. In practice, defuzzification of
this type-1 output fuzzy set simply consists of averaging maximum and minimum values.
The result of defuzzification is a crisp value that determines the change in the video rate.
(a)
RU (b)
Fig. 3. IT2 FLCC (a) Delay factor (Df) (b) Trend membership functions.
The Development of Crosstalk-Free
Scheduling Algorithms for Routing in Optical Multistage Interconnection Networks 643
29
Traditional float charges: are they suited to stationary antimony-free lead acid batteries? 665
30
negative electrode (IEffect0-) and of the positive electrode (IEffect0+):
“Current Balance” is the result theoretically expected when applying a current to the
battery. So it is the subtraction of the applied low current and the self-discharge current.
Current Balance: (7)
Calculated rate
Applied Calculated
Effective current of side reactions
Executed service Current Current Balance
I Effect (A/Ah) I Side-reac
during 356 days I Applied I Balance (A/Ah)
(A/Ah)
(A/Ah)
NAM PAM NAM PAM NAM PAM
Initial state
- - - - - - -
Self-discharge
0 -9 -30 -9 -30 -9 -30
Low current 25 -1.4 12 16 -5 -26.4 -13
idem 50 8 22 41 20 -42 -28
idem 100 23 42.6 91 70 -77 -57.4
idem 200 24 50 191 170 -176 -150
Table 4. Equivalent currents obtained from the variations of sulfate level and the duration of
the experiments, calculated current balances and side reaction rates.
Side reactions are always present inside the lead acid cell at open circuit as well as under
low currents. At open circuit, the side reaction rate is the self-discharge current. Under low
currents, side reaction rates can be calculated by subtracting the effective currents to the
applied current.
Side reaction rate under low currents: (8)
In figure 9, Effective Currents (EC) are compared to Current Balances (CB) and Side
Reaction Rates (SRR) for negative and positive plates.
At open-circuit in figure 9, variations of sulfate content in mole percent are +3.9% for the
NAM and +12.9% for the PAM (Nguyen et al., 2008). One can calculate from these sulfate
content variations the self-discharge currents during 356 days of the experiment: -9 A/Ah
and -30 A/Ah for the negative and positive electrod es respectively. The self-discharge rate
of the positive electrode is more than three times faster than the self-discharge rate of the
negative electrode.
have an efficiency that is higher than 100%. The low current effect is then not only a
compensation of the self-discharge process, but a modification and/or a rate reduction of
the secondary reactions involved in this process. Indeed, one observes a minimum of the
side reaction rate at 25 μA/Ah (cf. figure 9).
Secondary reactions at the positive electrode are the oxygen evolution and the positive grid
corrosion.
Oxygen evolution takes place at open-circuit, charge and discharge potentials. This reaction
is known to increase its rate with positive polarization, so it cannot be the reason of the
reduction of the self-discharge process observed in this case.
On the contrary, corrosion can slow down under positive polarizations. This phenomenon is
well known in the field of corrosion of metals as anodic protection, or more generally as
corrosion passivation.
In the case of lead acid batteries, it is also well known that corrosion reactions are different
at open-circuit and during charging. More exactly, corrosion reactions of the positive grid
consist of two steps:
Pb  O 2   PbO  2e  Reaction 11
PbO  O 2   PbO 2  2e  Reaction 12
In which, O2- ions are brought by migration from the active material across oxidation layers.
Only the first step, Reaction 11, operates at open-circuit while these both steps occur during
charging.
Indeed, at open-circuit, divalent lead (Pb2+) is the stable state for lead. Only the first step of
the corrosion reaction occurs but not the second step (Reaction 12, in which Pb2+ would
oxidize into Pb4+), because Pb4+ is not the thermodynamically stable state for lead. As at
open-circuit the two electrons of Reaction 11 cannot be evacuated by the external circuit,
they are used by reduction of Pb4+ into Pb2+ in the discharge reaction of the positive active
material:
PbO 2  H 2SO 4  2H 3O   2e   PbSO 4  4H 2 O Reaction 8
The combination of Reaction 11 and Reaction 8 is then:
PbO 2  Pb  H 2SO 4  PbSO 4  PbO  H 2 O Reaction 6
When PbO is not protected with a dense PbO2 layer, it reacts chemically with sulfuric acid to
form PbSO4 as follows:
PbO  H 2SO 4  PbSO 4  H 2 O Reaction 10
The overall corrosion reaction at open-circuit is then:

PbO 2  Pb  2H 2SO 4  2PbSO 4  2H 2 O Reaction 9
When this reaction occurs, usually in prolonged open-circuit conditions, positive grids are
no more protected. Indeed, the formation of lead sulfate in the corrosion layers leads to
mechanical stress, causing the formation of cracks and the destruction of the protective
layer.
During charging (under low currents), Reaction 12 takes place. Under positive polarization,
Pb4+ (in PbO2) is indeed the stable species of lead. PbO2 produced in Reaction 12 is formed in
the outer part of the corrosion layer. As it is dense, issued from the dense inner layer, and
stable in sulfuric acid solution, it takes the role of protecting the PbO inner layer and the
positive grid from the electrolyte (Ruetschi, 2004, Berndt, 1997, Garche, 1995).
Under positive polarizations, grid corrosion, which is part of the self-discharge process of
the positive electrode, is then slowed down by the formation of a protective layer of dense
PbO2. Therefore, for the positive electrode, this can be the reason why the effective current is
sensibly higher than the calculated current balance, as long as corrosion is important part of
the positive self-discharge process. Benefit of 25 μA/Ah low current is then triple:
(i) Compensation of the self-discharge:

For intermittent charge, in the prospect of replacing open-circuit periods by low current
periods, it is obvious that periodical charges would no longer be necessary or at least would
be required less frequently.
(ii) Slowing down of the corrosion rate:

As mentioned previously, several authors indicate a minimum corrosion zone. This
minimum is generally situated at positive polarizations in the [30, 80] mV range. Indeed, at
25 μA/Ah, the strong effect observed can be attributed to an important reduction of the
corrosion rate. But in this case the positive polarization is in the order of 2 mV much lower
than the preceding published values. It must be noticed that our results, obtained from long
duration tests (over 19 months) at room temperature, differ from these published results,
generally obtained at accelerated conditions.
(iii) Reduction of water consumption:

It concerns water involved in corrosion reactions. In these reactions, the oxygen is taken
from the positive active material. In turn, the active material takes oxygen from water of the
electrolyte. The oxygen, finally locked in the corrosion product layers, cannot be recombined
to reform water. So, the decrease of water consumption due to corrosion is important for the
battery life span.
5.2 Capacity measurement

Table 5 gives the capacities of VRLA batteries before (Cref) and after (Cfin) 6 months at open-
circuit, at low currents of 29 and 105 μA/Ah, and at 2.27V float charge. These experiments
were done at room temperature (18 ± 3°C). The battery at open-circuit for 6 months lost (264
- 327)/327 = -19.3 % of its capacity compared to its reference capacity; the capacity of the
battery subjected to 105 μA/Ah increased by (329 - 313)/329 = +5.1%.
Figure 10 (a) shows the capacity variations given for each battery in percent of their
reference capacities. These percentages are also the state of charge variations induced by the
applied currents during the 6 months of experiments. The capacity variations are converted
to the equivalent currents or as named above the effective currents. These effective currents
are presented as a function of the applied current in figure 10 (b). The currents balances and
the side reaction rates are calculated as the same way in the section 5.1. At open-circuit, the
capacity reduction of 63 Ah after 6 months allows calculating the average self-discharge
current, which is 36 μA/Ah.
Reference Applied Current during

Final capacity Capacity Variation
Battery capacity 6 months at room temperature
Cfin (Ah) (Cfin - Cref)/ Cref (%)
Cref (Ah) 18 ± 3°C (μA/Ah)
B115 340.0 ~ 414 (float charge at 2.27V) 379.0 11.8
B116 313.0 105 329.0 5.1
B117 318.0 29 308.0 -3.1
B118 327.0 0 264.0 -19.3
Table 5. Capacities of the VRLA batteries, which were subjected to open circuit, to 29 and
105 μA/Ah low-currents and to 2.27 V float charge during 6 months at room temperature
(18 ± 3°C) as well as their reference capacities before this service.
Capacity Variation ΔC = (Cfin - Cref)/Cref %
CB
EC
Self-discharge current
SRR
(a) (b)
Fig. 10. VRLA Batteries were tested during 6 months at room temperature (18 ± 3°C).
(a) Relative capacity variations (state of charge variations) vs. applied currents.
(b) Equivalent currents calculated from the capacity variations vs. applied currents.
The capacity variation has a strong increase for the lowest applied current (29 μA/Ah). It
appears that this low current, whose value is 80 % of the average self-discharge current
(36 μA/Ah), compensates 84 % of the self-discharge. Such a high efficiency (superior to
100%) suggests that these VRLA batteries could be positive limited at this mid-life state.
Further experiments confirmed that these batteries were indeed positive limited.
Increases of the battery state of charge at the low current of 105 μA/Ah and at float current
of 414 μA/Ah indicate that the charge procedure we used does not lead to a real full charge.
Charge is then completed slowly at currents beyond 50 μA/Ah. The reference capacity is
supposed to be the maximum capacity the battery is able to discharge after a full charge. But
what is a full charge? We all know that absolute full charge of a lead acid battery does not
exist. It is an asymptotic state. What we generally call a full charge is in fact the result of a
compromise between the search of a high state of charge and the time we can spend on this
operation. In practice it is considered that the end of charge is reached, when there is no
charging current or voltage evolution during at least 2 hours. Our “full charge” procedure
for VRLA batteries was a classical IUi charge. According to the results above, we must
consider that our charge procedures were not sufficient to reach states of charge as high as
after a few months of float charge or even of low current charge.
According to experiment results, we supposed that when combined with low currents in the
order of 25 to 50 μA/Ah, a refresh charge every 6 months or every single year is sufficient to
maintain antimony-free batteries in a good state of charge. Higher currents, as traditional
float currents (10 times higher), are not necessary and not suited, as they would increase
corrosion and water loss. We consider using a refresh charge only when the battery has been
subjected to a discharge service.
6. New management system for standby antimony-free battery

Valve regulated lead acid batteries - VRLA batteries - have been developed and used for
about 30 years in standby applications. They have shown several advantages compared to
flooded lead acid batteries: spill-proof, reduced weight, free from excessive gas evolution or
acid spillage, reduced maintenance and reduced cost. However, limitations have been also
observed concerning system reliability and battery service life.
Several reliability prediction methods have been used such as complete discharge test (Piller
et al., 2001), open-circuit voltage measurement (Bullock et al., 1997), conductance testing
(Kniveton, 1995), internal resistance and impedance measurements (Hariprakash et al., 2004,
Huet, 1998, Karden et al., 2000, Rodrigues et al., 2000, Shukla et al., 1998, Piller et al., 2001).
Among these, the complete discharge test is well known as the most reliable one but it
requires service interruption. This has driven EDF R&D to develop the “Stationary
Multibat” system, which consists of a new design of the electrochemical storage and an
adapted electronic battery management system (Desanti and Schweitz, 2006). This system,
combining redundancy and automated periodical capacity measurements, increases
reliability and allows a real time monitoring of the battery state of health. Redundancy not
only ensures the continuity of service in the case of a cell failure, but also enables complete
discharges to be periodically performed without service interruption.
Standby VRLA batteries maintained under a constant float voltage to compensate self-
discharge encounter the problem of short service life, e.g. about 3-4 years compared to the
so-called 20-year design and to the 20-year lifetime of conventional lead acid batteries in
similar conditions (Misra, 2007). Indeed, VRLA batteries under float charges are
permanently overcharged and different failure modes have been observed, such as
corrosion of positive grid alloys, electrolyte dry-out and thermal runaway (Berndt, 1997,
Feder, 2001, Ruetschi, 2004, Wagner, 1995, Dai et al., 2006).
A new management system for standby VRLA batteries has been developed at EDF using
the Stationary Multibat system (Desanti and Schweitz, 2006) to ensure system reliability and
the Low-current method (Nguyen et al., 2008) to improve the battery life span.
Figure 11 (a) describes a simplified schematic diagram of the Stationary Multibat system.
Three battery-pack strings in parallel instead of one pack are used in order to improve the
system reliability. In case of a cell failure such as an open circuit, the system loses only one
battery-pack string, i.e. one-third of the total capacity. This configuration also allows each
string to be discharged completely across the test resistance to evaluate its real state of
health. Each battery string is periodically (e.g. every 6 months) discharged and charged.
Fig. (b) describes how to integrate and to operate Low-current method of maintaining the
charge on the Stationary Multibat system:
 A resistance branch is added to reduce the traditional float current with a factor of
5 to 10. In standby state, Programmable Logic Controllers (PLC) K1 and K2 are
open; the battery is maintained at charge with a low current via the resistance R.
 On backup demand, a voltage drop appears on the DC bus, the battery
immediately provides electricity to the DC load via the diode D. Then PLC K1
closes; the battery directly supplies power to DC load.
 During the periodical discharge test (e.g. every 6 months) to evaluate the battery state
of heath, PLC K2 closes and K1 stays open. The battery-pack string is discharged
across the test resistance. The low current, still provided to the battery pack string, is
a parasitic but negligible effect (< 0.1%). Next, the charge is operated via K1. The
charge current is controlled by a Pulse Wave Modulation (PWM). When the charge
finishes, K1 re-opens, the battery is re-maintained the charge with a low current via R.
All those operations are done sequentially and automatically so that no intervention is
required from the maintenance.
Battery Management System
Test Resistance
Charger
Discharge test
DC loads
=
DC bus DC bus
String in test K1 R K2
D
Battery
System
(a) (b)
Fig. 11. (a) Simplified schematic diagram of the Stationary Multibat system. (b) Zoom on the
power electronic components of one of the three battery-pack strings using Low-current
method.
This system is expected to fulfill three targets:

 Increasing the system reliability by a redundancy in the design of the battery
system and by an assessment method to control battery state of health.
 Improving battery life spans by the Low-current method, leading to reduced
corrosion and water loss.
 Decreasing maintenance costs.
7. Conclusion
Antimony-free lead-acid batteries were tested for several months at open-circuit, float
charge and intermediate rates of charge – called “Low-current” in this chapter. The
following conclusions can be drawn from this study:
 In antimony-free lead acid batteries, the self-discharge rate of the positive electrode
is higher than that of the negative electrode.
 Low-currents beyond 25 μA/Ah appear to be able to maintain the state of charge
for both positive and negative plates.
 The effect of 25 μA/Ah low current on the positive active material is much higher
than a simple compensation of the self-discharge. This can be explained by a
passivation phenomenon of the positive grid corrosion. In other words, it
constitutes an anodic protection of the positive grid.
 The use of low-current periods in place of open-circuit periods in the intermittent
charging method should increase the life span of VRLA batteries, as it both lowers
the water consumption and the corrosion rate of positive grids.
Combining the Low-current method, battery redundancy and automated periodical capacity
measurements, VRLA batteries should provide long life spans and high reliability. Several
management systems of this kind are being experimentally used at EDF.
Acknowledgment
The authors gratefully acknowledge financial support from the Association Nationale de la
Recherche Technique (ANRT - France).
8. References
ALZIEU, J. & DILLENSEGER, G. (2006) Method for maintaining the charge of a lead storage
battery. France, Electricite de France.
BERNDT, D. (1997) Maintenance-Free batteries, Taunton, Somerset, New York, Chichester,
Toronto, Brisbane, Singapore, Research studies press LTD, John Wiley & Sons INC.
BERNDT, D. & TEUTSCH, U. (1996) Float charging of VRLA batteries: A balancing act
between secondary reactions. Journal of Electrochemical Society, 143.
BODE, H. (1977) Lead Acid batteries, New York, London, Sydney, Toronto, John Wiley &
Sons.
BRECHT, W. B. (1998) Strategies for overcoming the adverse effects of imbalances in the
second order reactions in valve regulated lead acid cells. Telecommunications Energy
Conference, 1998. INTELEC. Twentieth International.
BRECHT, W. B., FEDER, D. O., MCANDREWS, J. M. & WILLIAMSON, A. J. (1989) The
effect of positive polarization on grid growth, cell performance and life. II.
Telecommunications Energy Conference, 1989. INTELEC '89. Conference Proceedings.,
Eleventh International.
BRECHT, W. B., FEDER, D. O., MCANDREWS, J. M. & WILLIAMSON, A. J. (1990) Low
float technology battery. United States, C & D Charter Power Systems, Inc.
(Plymouth Meeting, PA).
BRISSAUD, C., REUMONT, G., SMAHA, J. P. & FOCT, J. (1997) Structural and
morphological study of damage in lead/acid batteries during cycling and floating
tests. Journal of Power Sources, 64, 117-122.
BULLOCK, K. R., WEEKS, M. C., BOSE, C. S. C. & MURUGESAMOORTHI, K. A. (1997) A
predictive model of the reliabilities and the distributions of the acid concentrations,
open-circuit voltages and capacities of valve-regulated lead/acid batteries during
storage. Journal of Power Sources, 64, 139-145.
COOPER, A. & MOSELEY, P. T. (2003) Progress in overcoming the failure modes peculiar to
VRLA batteries. Journal of Power Sources, 113, 200-208.
CULPIN, B. (2004) Thermal runaway in valve-regulated lead-acid cells and the effect of
separator structure. Journal of Power Sources, 133, 79-86.
DAI, C., YI, T., WANG, D. & HU, X. (2006) Effects of lead-foam grids on performance of
VRLA battery. Journal of Power Sources, 158, 885-890.
DESANTI, J. D. & SCHWEITZ, G. (2006) Decreasing Owning Costs of MV/LV Substations
Backup Batteries. Telecommunications Energy Conference, 2006. INTELEC '06. 28th
Annual International.
DILLENSEGER, G. (2004) Caractérisation de nouveaux modes de maintien en charge pour
batteries stationnaires de secours. Sciences et Techniques du Languedoc. Montpellier,
Université de Montpellier II.
ELGH, R. (1994) Tests on valve regulated lead acid batteries at different environmental
temperatures and float voltages. Telecommunications Energy Conference, 1994.
INTELEC '94., 16th International.
FEDER, D. O. (2001) 1950-2001: more than one-half century of learning how to live with each
new generation of telecommunication standby batteries. Telecommunications Energy
Conference, 2001. INTELEC 2001. Twenty-Third International.
FEDER, D. O. & CAROSELLA, G. (1994) The never ending pursuit of float voltage
uniformity in stationary reserve battery plants. Telecommunications Energy
Conference, 1994. INTELEC '94., 16th International.
GARCHE, J. (1995) Corrosion of lead and lead alloys: influence of the active mass and of the
polarization conditions. Journal of Power Sources, 53, 85-92.
GIESS, H. (1997) Investigation of thermal phenomena in VRLA/AGM stationary lead/acid
batteries with a thermal video imaging system. Journal of Power Sources, 67, 49-59.
GIESS, H. (2001) The operation of VRLA monoblocs with an on/off float charge regime.
Telecommunications Energy Conference, 2001. INTELEC 2001. Twenty-Third
International.
GUN, J. P., FIORINA, J. N., FRAISSE, M. & MABBOUX, H. (1997) Increasing UPS battery life
main failure modes, charging and monitoring solutions. Telecommunications Energy
Conference, 1997. INTELEC 97., 19th International.
HARIPRAKASH, B., MARTHA, S. K., JAIKUMAR, A. & SHUKLA, A. K. (2004) On-line
monitoring of lead-acid batteries by galvanostatic non-destructive technique.
Journal of Power Sources, 137, 128-133.
HAWKINS, J. M., MOORE, L. E. E. & BARLING, L. O. (1995) Aspects of the float and
temperature behaviour of lead-acid batteries in telecommunications applications.
Telecommunications Energy Conference, 1995. INTELEC '95., 17th International.
HU, J., GUO, Y. & ZHOU, X. (2006) Thermal runaway of valve-regulated lead-acid batteries.
Journal of Applied Electrochemistry, 36, 1083-1089.
HUET, F. (1998) A review of impedance measurements for determination of the state-of-

charge or state-of-health of secondary batteries. Journal of Power Sources, 70, 59-69.
KARDEN, E., BULLER, S. & DE DONCKER, R. W. (2000) A method for measurement and
interpretation of impedance spectra for industrial batteries. Journal of Power Sources,
85, 72-78.
KITA, A., MATSUI, T., KASAI, Y. & KISHIMOTO, K. (1993) Uniterruptible power system.
United States, Yuasa Battery Company Limited (Takatsuki, JP).
KNIVETON, M. W. (1995) Reducing the cost of maintaining valve-regulated lead/acid
batteries in telecommunications applications. Journal of Power Sources, 53, 149-152.
KOSAI, M., YASUKAWA, S., OSUMI, S. & TSUBOTA, M. (1997) Effect of antimony on
premature capacity loss of lead/acid batteries. Journal of Power Sources, 67, 43-48.
LANDER, J. J. (1956) Further studies on the anodic corrosion of lead in H2SO4 solutions.
Journal of the Electrochemical Society, 103, 1-8.
MARTINEZ, T. G. & NOVAK, A. F. S. (1990) Increased float voltage and the effects of
negative self-discharge on flooded lead calcium telecommunications cells.
MISRA, S. S. (2007) Advances in VRLA battery technology for telecommunications. Journal
of Power Sources, 168, 40-48.
MISRA, S. S., NOVESKE, T. M. & WILLIAMSON, A. J. (1994) Maintenance and reliability of
standby battery systems: flooded vs. valve regulated lead acid battery.
MISRA, S. S. & WILLIAMSON, A. J. (1995) Impact of grid corrosion in valve regulated lead-
acid battery on standby float service. Telecommunications Energy Conference, 1995.
INTELEC '95., 17th International.
MUNERET, X., COUX, M. & LENAIN, P. (2000) Analysis of the partial charge reactions
within a standby VRLA battery leading to an understanding of intermittent
charging techniques. Telecommunications Energy Conference, 2000. INTELEC. Twenty-
second International.
NAKAMURA, K., SHIOMI, M., TAKAHASHI, K. & TSUBOTA, M. (1996) Failure modes of
valve-regulated lead/acid batteries. Journal of Power Sources, 59, 153-157.
NGUYEN, T. M. P., DILLENSEGER, G., GLAIZE, C. & ALZIEU, J. (2008) Between floating
and intermittent floating: Low-current self-discharge under compensation.
Telecommunications Energy Conference, 2008. INTELEC 2008. IEEE 30th International.
PAVLOV, D. (1993) Premature capacity loss (PCL) of the positive lead/acid battery plate: a
new concept to describe the phenomenon. Journal of Power Sources, 42, 345-363.
PAVLOV, D. (1997) Energy balance of the closed oxygen cycle and processes causing
thermal runaway in valve-regulated lead/acid batteries. Journal of Power Sources, 64,
131-137.
PETERS, K. (1996) Review of factors that affect the deep cycling performance of valve-
regulated lead/acid batteries. Journal of Power Sources, 59, 9-13.
PILLER, S., PERRIN, M. & JOSSEN, A. (2001) Methods for state-of-charge determination and
their applications. Journal of Power Sources, 96, 113-120.
REID, D. P. & GLASA, I. (1987) A new concept: intermittent charging of lead acid batteries
in telecommunication systems. Telecommunications Energy Conference, 1987.
INTELEC '87.
ROCCA, E. & STEINMETZ, J. (2003) Passivation phenomenon of low antimony alloys in

deep discharge conditions of lead-acid batteries. Journal of Electroanalytical
Chemistry, 543, 153-160.
RODRIGUES, S., MUNICHANDRAIAH, N. & SHUKLA, A. K. (2000) A review of state-of-
charge indication of batteries by means of a.c. impedance measurements. Journal of
Power Sources, 87, 12-20.
ROSSINOT, E., LEFROU, C., DALARD, F. & CUN, J. P. (2001) Batteries in standby
applications: comparison of alternate mode versus floating. Journal of Power Sources,
101, 27-34.
ROSSINOT, E., LEFROU, C. & GUN, J. P. (2003) A study of the scattering of valve-regulated
lead acid battery characteristics. Journal of Power Sources, 114, 160-169.
RUETSCHI, P. (2003) Silver-silver sulfate reference electrodes for use in lead-acid batteries.
Journal of Power Sources, 116, 53-60.
RUETSCHI, P. (2004) Aging mechanisms and service life of lead-acid batteries. Journal of
Power Sources, 127, 33-44.
RUETSCHI, P. & ANGSTADT, R. T. (1964) Anodic oxidation of lead at constant potential.
Journal of the Electrochemical Society, 111, 1323-1330.
SHIOMI, M., OKADA, Y., TSUBOI, Y., OSUMI, S. & TSUBOTA, M. (2003) Study of PCL
mechanism: Influence of grid/PAM state on PCL. Journal of Power Sources, 113, 271-
276.
SHUKLA, A. K., GANESH KUMAR, V., MUNICHANDRAIAH, N. & SRINATH, T. S. (1998)
A method to monitor valve-regulated lead acid cells. Journal of Power Sources, 74,
234-239.
STEVENSON, P. (1996) Durée de vie des accumulateurs étanches. REE, N°1, 57-60.
WAGNER, R. (1995) Failure modes of valve-regulated lead/acid batteries in different
applications. Journal of Power Sources, 53, 153-162.
WILLIHNGANZ, E. (1968) Accelerated testing of stationary batteries. Electrochemical
Technology, 115, 338-341.
Neighbor Discovery: Security Challenges in Wireless Ad hoc and Sensor Networks 693
31
X
Neighbor Discovery: Security Challenges

in Wireless Ad hoc and Sensor Networks
Mohammad Sayad Haghighi and Kamal Mohamedpour
K.N. Toosi University of Technology
Iran
1. Introduction
Wireless ad hoc and sensor networks are infrastructureless systems with self-configuration
capabilities. Neighbor discovery protocols are fundamental requirements in the construction
of self-organizing networks. Each computer (node) must discover who its neighbors are in
order to be able to coordinate with them for any later communication. This goal is usually
accomplished through broadcasting methods in the initial phases of network deployment.
Along with the development of neighbor discovery protocols, security threats also
introduced and some researchers discovered new forms of attacks for neighbor discovery
scenarios. Authors in this field have given different definitions on what a neighbor is and to
what extent an adversary is equipped. We will first clarify these differences by making some
categorizations and definitions before proceeding to the introduction of attacks and
solutions.
After the introduction of initial concepts, we classify the attacks into two general groups
and explain the solutions for each of the groups numerating the pros and cons of them. We
will review the current external attacks solutions for the neighbor verification problem first
and start with the early simple methods which tried to defeat the relaying attacks (like the
wormhole one) using distance estimation methods. This family of protocols relied on time
stamps which needed tight clock synchronization among the nodes and was quite
impractical in distributed networks specially the sensor ones. After that, we introduce the
descendants of that family of protocols which resolved the clock synchronization problem
by using challenge-response-like methods.
Then, the recent efforts on formal description of the time-based and time- and location-
based neighbor discovery protocols are explained. These researches led to the conclusion
that time-based protocols can only secure the neighbor discovery under some strict
conditions. Time- and location-based protocols are generally more secure than the time-
based ones alone.
Next, we will argue why all of these protocols are vulnerable to the internal attacks and
introduce other methods for defeating internal adversaries. Describing the mostly
cryptographic solutions in this domain, we outline their pros and cons. As we will see,
almost all of these protocols are either unable to resist the invasion of an internal adversary
equipped with both powerful transmitter and sensitive receiver, or need an initial setup
phase in which the network is assumed to be secure. Adding the lack of mobility support to
the specifications of this family of solutions, we move on to present our solution for a special
type of internal attack which statistically tries to block the sensitive broadcasting internal
adversary in mobile dense networks. This approach works based on the inherent
characteristics of the dense networks and only the medium access control protocol
parameters are changed. Therefore, it imposes a very low cost to the system to create higher
levels of robustness.
At the end a conclusion is made which suggests combinational methods to be employed in
the neighbor verification protocols to achieve an acceptable level of security against both the
internal and external attackers.
1.1 Fundamental Concepts and Definitions

In wireless ad hoc networks, computers (nodes) are usually stand-alone entities working in
cooperation with each other in order to fulfill a desired task (Rubinstein et al., 2006). In such
networks, nodes are located centimeters to hundred meters away from each other but can
communicate through their wireless transceivers.
Sensor networks can be thought to be a subgroup of ad hoc networks with some specific
characteristics. They are usually deployed (densely) in an area to sense or monitor quantities
of desired form (Akyildiz et al., 2002). For example in a battlefield, enemy movements can
be detected and localized by a distributed sensor network. Unlike the ad hoc networks, the
messages created by sensor nodes are always destined to be received by a single target
named “Sink”. The sensors cooperate to deliver the messages to the sink in a multi-hop
manner.
The unattended nature of sensor networks and the high number of nodes creates some
constraints for the designer. Each node must survive days or even years with a single source
of power. Therefore, the protocols must not be too power consuming and should minimize
both the amount of processing and the number of transmissions. The processing power and
the amount of RAM1 and ROM2 of a sensor node are also quite limited. So, generally,
designing a secure protocol is much harder in the sensor networks than in the ad hoc ones.
It is obvious that before a distributed cooperative network begins to work, nodes need to
know their neighbors to form local structures. Any long-distance communication must be
made in a multi-hop manner. The transmission radii of the nodes are limited and hence,
each node should pass the messages to one/some of its neighbors in order to participate in
the delivery of messages.
Neighbor discovery means determining whether a wireless device (node) is directly
reachable without the assistance of any other device according to the predesigned rules of
the network or not. Neighborhood can be unidirectional or bidirectional depending on
whether only one side is able to deliver its messages to the other side or both sides are
capable of doing so.
A neighbor-discovery attacker tries to deceivingly convince the nodes to believe that they
are neighbors of a specific set of nodes (possibly including the adversary herself), when they
are actually not. There are various types of attacks on neighbor discovery scenarios. The
effectiveness level of an attack depends on whether the adversary is a part of the network or
1Random Access Memory

2Read-Only Memory
not. In the next sub-section, we make some definitions on adversaries’ types and their level
of capability.
1.2 Types and Capabilities of Adversaries

To describe the current solutions for the neighbor verification problem in the rest of this
chapter better, we categorize them into two groups based on their resistance to internal
(intrusive) or external (non-intrusive) attacks.
In external attacks the adversary is not able to compromise the nodes and hence, does not
have access to the private information like the cryptographic keys and communication codes
stored in the memory of the nodes whereas in the internal ones has (Khelladi et al., 2005).
Usually, an external attacker is only able to overhear (eavesdrop), relay (replay) or block
(jam) the packets. On the other side, an internal attacker is capable of masquerading himself
as a legal node and thus can imitate all the behaviors of a healthy node. Having the private
cryptographic keys, she can even generate fake (but authenticated) messages to obtain a
higher number of neighbors to what a traditional healthy node does. It is rather obvious that
the second type of adversary is much more powerful than the first one.
1.3 Effective Attacks on Neighbor Discovery

In this part, we briefly introduce the currently known neighbor-discovery-related attacks in
ad hoc and sensor networks. There are a few general attacks which have effects on neighbor
discovery, and a few others which specifically address the neighbor-discovery-related
issues.
One of the oldest external passive attacks is eavesdropping. Regardless of the protocol
architecture, an adversary is always able to overhear wireless communications. There is
little chance for the designer to block eavesdropping. However keyed cryptographic
operators (like the encryption ones) are quite useful in keeping the external adversaries from
extracting sensitive information out of the transmitted signal (Zhu et al., 2006)(Du et al.,
2005)(Du et al., 2006). Neighbor discovery protocols are no exception. The protocol designer
must seal the places where the information might leak during wireless transmissions.
The active invasions that target the availability of network services are called Denial of
Service (DoS) attacks. DoS attacks can be planned to work on any layer of the network
protocol stack depending on how much weak that layer is. Jamming can be well categorized
into the physical layer DoS attacks group. There are only a few non-perfect classic solutions
like spread spectrum communication for this attack (Pickholtz et al., 1982). Other types of
DoS attack also exist among which some try to excessively overload a badly designed
protocol run on a resource-limited machine (Djenouri et al., 2005). So, one can easily
conclude that in sensor networks, designing a DoS-resilient neighbor discovery protocol is
more complicated than in ad hoc networks. Ignoring the heuristic solutions, the classic
countermeasures for protocol-related DoS attacks are easy-to-compute checksums and
ciphers that reject massive fake messages.
Relaying and replaying are two other simple but powerful attacks (Papadimitratos et al.,
2008). In the replay attack, an adversary uses an old packet which was previously generated
by a healthy node in order to deceive another healthy node in the future. To overcome this
problem researchers have suggested using timestamps (in clock-synchronized networks)
and nonces (Du et al., 2006)(Shokri et al., 2008).
Relaying attacks are harder to detect. The adversary relays the healthy nodes’ packets
instantaneously (either at the physical layer or in a store-and-forward manner) in another
part of the network. Wormhole attack is a well-known representative of this family. In the
wormhole (tunneling) attack, two (or more) adversarial nodes try to transfer information
through a dedicated channel between themselves and then use it in another part of the
network (Hu et al., 2006). This attack can be implemented both by internal and external
adversarial nodes. In the external form, they simply relay the packets either in a store-and-
forward manner or instantaneously at the physical layer. In the internal form, they also have
the opportunity to alter the packet contents intelligently before forwarding.
Every neighbor discovery protocol is composed of a series of packet transmissions. In a
weak protocol, these attacks can be launched to relay neighbor discovery packets to other
areas of the network, in order to convince distant nodes to believe that they are true
neighbors. Figure 1 shows two healthy nodes A and B, and two adversarial nodes C and D.
A and B cannot see each other directly since their transmission radii (shown with circles
around them) are small compared to their distance. However, C and D can relay the
neighbor-discovery-related packets to create a virtual link.
Fig. 1. The wormhole attack: Two adversarial nodes C and D, relay the packets between
healthy nodes A and B, to deceivingly convince them to believe that they are in the vicinity
of each other
Hello flooding is a broadcast-type attack that was originally designed for sensor networks
(Karlof et al., 2003). However, its concept can be adopted in the ad hoc networks too. Most
of the neighbor discovery protocols (as well as the routing protocols) use a broadcasted
packet called “Hello” or “Beacon” to announce a node’s presence to its neighbors. Nodes
receiving this message assume that the sending node is one of their neighbors. An internal
attacker can launch a hello flooding attack by simply broadcasting the hello message with
very high power. This way she tries to convince a lot of nodes that she is one of their
neighbors and the victims add her to their table of neighbors. If the adversary is well
equipped, she might even have a low noise sensitive receiver which enables her to receive
distant weak signals and thus turn this attack into a bidirectional one. Figure 2. shows the
adversarial and healthy transmission ranges.
This attack was initially designed for the internal attackers. However a simple repeater-like
external adversary can also launch a similar attack through boosting a healthy node’s
transmission power and acting as a “man in the middle”. Fortunately, as we will see, there
are more solutions for the external threats than the internal ones.
Fig. 2. A broadcast-type attack (hello flooding) diagram in a network with regular

transmission range of r. The adversary (A) transmits messages with high power pretending
to be a neighbor of B whereas she is actually R meters away (R>r)
Broadcast attacks can also be implemented in routing protocols in the route discovery phase
(e.g. Routing Request (RREQ) message broadcasting) to shorten the adversary’s path to the
destination and thus putting her into most of the routes.
2. Neighbor Discovery External Attacks Countermeasures

If the maximum transmission range of a healthy node is r, then a secure neighbor discovery
protocol must reject any claiming neighbor farther than this distance. The most challenging
threat in designing externally resistant neighbor discovery protocols is the wormhole (or
relaying) attack and the easiest way to block this attack is estimating the distance between
the nodes.
The very early methods of neighbor verification relied on the round trip propagation time
(RTT) measurement to estimate the approximate distance between two nodes and compare
it with a maximum allowable value. Brands et al. were pioneers in using this method with
their one-bit exchange proposal (Brands et al., 1993). The distance bounding method they
proposed was a cryptographic protocol that put an upper-bound on the distance between
two users (nodes for example) and did not let the protocol be manipulated. This scheme was
able to resist man-in-the-middle like attacks, however, Singelee et al. later claimed that
Brands’ scheme is unable to stop what they called “terrorist fraud attack” which is an
internal attack in our terminology (Singelee et al., 2005). After the introduction of distance
bounding concept, similar protocols with names like “Echo protocol“ (Sastry et al., 2003)
appeared which were all based on the same basic idea of round trip time measurement.
In (Hu et al., 2003), to prevent wormhole attacks in ad hoc networks, a method called
“packet leashes“ was introduced and an idea similar to the signal trip delay measurement
was repeated. The authors proposed two types of methods which could resist the (external)
wormhole attacks: geographical leashes and temporal leashes. Geographical leashes fall into
the category of time-and-location based protocols which we will introduce in the next parts.
However, it does not actually use the timing data directly. It assumes that the sender sends
its location (with the maximum relative error equal to �) along with the timestamp showing
the transmission time (with a maximum relative error of ∆). If the nodes move with a
maximum speed of �, then this method gives an upper bound on the distance which is
found by � � �|�� |� � �� .
The authors also proposed a time-based approach which they called “temporal leashes”. In
this approach the transmission time is (authentically) written in the packet and the receiver
can then decide on the distance from the duration of signal flight. To authenticate the
timestamp (and location in the previous part), digital signatures and the hash-chain-based
(tree) authentication methods have been used. However the original idea comes from
TESLA authentication method (Perrig et al., 2000).
Regardless of the protocol types, for both parties to be able to measure the distance, each
must have the chance to challenge the other. So, at least three messages must be exchanged.
For unidirectional neighbor discovery, exchanging two messages suffices (Sayad et al.,2008).
Korkmaz addressed the RTT measurement, focusing on the difference of signal propagation
speed in the healthy nodes and adversaries’ channels (Korkmaz, 2005). In this method, a
combination of power and delay-related criteria has been used. The simplified diagram of
Korkmaz protocol is depicted in Figure 3. Here, “M“ is an authenticated request message
and “ACK“ is its reply containing �� , �� and �� which is the processing delay in B.
Fig. 3. Korkmaz’s neighbor discovery protocol
If the worst case signal propagation speed between two nodes with distance R is s, and the
maximum speed of signal propagation in the adversaries‘ media is vadv (vadv>s), then the
��
upper and lower bound estimates of the RTT are found as [ , �. Any claiming neighbor
��
with measured RTT lower than is accepted while anyone with a RTT larger than � is
��
rejected. Not all the measurements which fall in this interval are accepted. If the actual
speed of signal propagation in the network is x, then Korkmaz proposes using a hard
decision making threshold for the RTT samples falling in the abovementioned interval. The
following formula normalizes the measured RTT.
��
��
�� (1)
�
� ��
The hard decision making threshold is the acceptable level of confidence i.e. for �� we
assume that the two nodes are neighbors and for �� we assume they are not. This
threshold is equivalent to an effective distance �� . Obviously,
this distance can be either smaller or larger than R. So this type of decision making is always
prone to either rejecting correct neighbors or accepting some incorrect ones which opens a
vulnerability window for the system. To mitigate this flaw, Korkmaz proposed to combine
the RTT-measurement-based method with a power-measurement-based one. Generally, the
relationship between the distance, transmitted and received signal strengths is:
� ��
� � �� (2)
�
where k and n are constants and are determined by the characteristics of channel and
communicating devices. Since the ACK message contains the measured power fields, A can
easily compute the distance (d) using equation (2). However, since k and n fluctuate in the
real world scenarios it is proposed to verify the following equilibrium instead:
௉೟೘ ௉೟
ൌ ௉ೌ (3)
௉ೝ೘ ೝೌ
Korkmaz claims that if the nodes are not actual neighbors i.e. they communicate through
relaying adversaries, then the values of ܲ௥೘ and ܲ௥ೌ will be altered depending on the
transmission power used during relaying and the distance between the adversaries and
legitimate nodes, and hence, the above equilibrium does not hold. However, it can be easily
verified that if the relaying adversaries collaborate with each other, this countermeasure is
easily neutralized. For example in Figure 1, if C tells D to send the signal received from A, to
B with the same power she received (and vice versa), then eq. (2) is satisfied. Also it should
be emphasized that this method only addresses probabilistic security facing external attacks.
Obviously B, as an internal attacker can deceive A if she lies about the processing time.
As an alternative approach, a few authors suggested searching for graph abnormalities to
detect the relaying attackers either in a distributed or centralized manner. Maheshwari et al.
proposed a distributed geometrical algorithm to search for graph abnormalities in wireless
networks (Maheshwari et al., 2007). The core idea of their work was to find impossible cases
which do not occur in healthy graphs. If two nodes are claimed not to be neighbors then the
number of their independent neighbors (those neighbors of the two nodes who cannot see
each other) is quite limited. For example consider Figure 4 in which two healthy nodes a and
b are at their farthest possible range of neighborhood. In this case, the nodes which are
considered to be neighbors of both of them can only be found in the intersection of the two
coverage areas (the hatched area). Mashewari et al. have proved that no more than two
independent neighbors can be found in the hatched area in this case. So if a third common
neighbor comes in, then it must be covered by one of the previous common neighbors. The
number of independent common neighbors decreases even more when the two nodes
further move away from each other. So if the adversaries launch a wormhole attack like the
one in Figure 5, then two independent nodes a and b which are far from each other, will
have three common independent neighbors e, f and g and according to the argument we
made there can be no more than two common independent neighbors in this case and hence,
a fraud has been detected.
Fig. 4. Mashewari et al.‘s diagram to demonstrate the maximum number of independent

common neighbors of two nearly independent nodes
Fig. 5. A sample wormhole attack which is detected by Mashewari et al.‘s forbidden

structure search
However this was a simple example and the authors have extended this one-hop graph-
based idea to a k-hop one to obtain a higher wormhole attack detection probability.
In a combined work of distance measurement and graph abnormality detection, Shokri et al.
proposed what they called a practical secure neighbor verification protocol for sensor
networks (Shokri et al., 2008). The core idea of their measurement part contribution relies on
having two different transceivers (one Radio Frequency (RF) and one Ultra Sonic (US))
which is a bit similar to the echo protocol approach (Sastry et al., 2003). Since the sensors
have simple processors of microsecond clock pricision, the RF transceiver is used for clock
synchronization-like purposes (or simply measuring the difference of time). Distance
measurement with RF signals requires a nanosecond clock precision. So the RTT
measurement (and consequently the distance measurement) is done more accurately using
the slow-propagating US signals with the precision at hand.
This method is actually composed of three smaller sub-protocols. Assuming that all pairs of
the nodes already have a symmetric shared key (like �� between A and B), each initiating
node A, sends a request message to (arbitrary) node B over the RF channel. B replies to this
request with another message. Then A broadcasts a message through its ultrasonic
transmitter to reveal its nonce whose hash was previously sent at the request transmission
phase. This is similar to the delayed authentication methods (like TESLA method (Perrig et
al., 2000)) which bind the authenticity of the next step to the previous one. At last an ACK is
sent to each of the candidates (like B). The whole message exchange of sub-protocol 1 is
depicted in Figure 6. Notice that �� . � denotes an encryption with key K, and �� and ��
are two nonces. �� . � means that the whole message is protected with a message
authentication code under key K. Node B‘s side times are measured with its own clock
which is not necessarily synchronized with A. At the end of this phase, B can compute its
distance to A using �� . ��
′ ′
� �� where � is the speed of slowly
propagating ultrasonic signal.
After this stage, the nodes exchange their local table of distances as the second step of the
protocol so that each obtains a local view of the network topology. At the third stage, each
node starts to run a series of tests on the information obtained in the previous stages. First of
all, a node eliminates those links with distances outside the acceptable range. Second, the
symmetry of the links is verified meaning that an arbitrary node A checks the ��
equaility for any candidate B. Third, for any claiming neighbor, the testing node tries to find
two other candidates in the table and then, knowing the distances between them, it assumes
hypothetical (but wisely selected) positions for these three candidates so that the triangular
inequalities hold for each possible set of three nodes (including itself). At last, it verifies
whether the four nodes form a convex quadrilateral3 or not. The authors have shown that
the above criteria, prevent wormhole attacks with two colluding adversaries and also resist
well facing wormhole attacks with more than two adversaries.
Fig. 6. The sub-protocol 1 (distance measurement) of Shokri et al.’s method
Although Shokri’s protocol has novel contributions and combined many ideas like delayed
authentication and geometrical tests, it suffers from some design flaws. In an
infrastructureless wireless network, neighbor discovery is the first protocol which is run i.e.
there is no information about the existence of the nearby nodes prior to starting the protocol.
So there are no agreed mutual cryptographic keys and hence A cannot send authenticated
messages to a node like B which is not even known to be in its vicinity. Sensor nodes are
extremely resource-limited devices and in a network with thousands of sensors, it is
practically impossible to store a large number of symmetric keys in a node’s memory and
this has been a challenging security problem thus far (Chan et al., 2003)(Du et al., 2005)(Zhu
et al., 2003 & 2006). Using LEAP (Zhu et al., 2003) or LEAP+-like (Zhu et al., 2006) key
distribution methods selected by the authors, is problematic since these protocols
themselves have neighbor discovery protocols in their initial setup phase and static
networks do not need multiple-time independent neighbor discoveries. Besides, LEAP+
itself is capable of blocking wormhole attacks after the key establishment phase. With a
rather high number of message transmissions (which includes the hidden stage of table
exchanges) and a high number of link verification computations, this solution is not very
energy conserving. Also the use of two transceivers (RF and US) is in contradiction to the
basic assumption in sensor networks design which is cost effectiveness. It should be
mentioned that this protocol is unidirectional in our terminology and is designed to prevent
external wormhole attacks in static networks, thus has limited functionality facing internal
attackers and also in highly mobile networks. However, compared to the older centralized
approaches for graph abnormality detection (Rasmussen et al., 2007), both of the
abovementioned distributed approaches are valuable.
3A convex quadrilateral with four vertexes A, B, C and D is characterized by

൫‫ ܤܣ‬ൈ ሬሬሬሬሬԦ
ሬሬሬሬሬԦ ሬሬሬሬሬԦ ൈ ‫ܦܥ‬
‫ ܥܤ‬൯൫‫ܥܤ‬ ሬሬሬሬሬԦ൯൫‫ܦܥ‬
ሬሬሬሬሬԦ ൈ ‫ܣܦ‬
ሬሬሬሬሬԦ൯൫‫ܣܦ‬
ሬሬሬሬሬԦ ൈ ሬሬሬሬሬԦ
‫ ܤܣ‬൯ ൐ Ͳ
Poturalski et al., in a series of evolutionary papers and technical reports, tried to classify
neighbor discovery protocols into two generic time-based and time-and-location-based
classes and provide systematic rules to formally verify the security of these protocols against
external attacks (Poturalski et al., 2007) (Poturalski et al., 2008). In their initial technical
report, they formally derived a so-called impossibility result for the time-based neighbor
discovery protocols stating that it is impossible for a neighbor discovery protocol which
solely relies on signal propagation time measurements to provide seamless security
(Poturalski et al., 2007). Figure 7 demonstrates the “impossibility result“ informally. If the
maximum allowable distance of two healthy neighbors is r, then the maximum signal travel
time would be r/s where s is the speed of signal propagation in the channel. If two nodes
cannot reach each other directly, either due to a large distance (Figure 7c) or a barrier
(Figure 7b) then one or more relaying adversaries can deliver the messages and as long as
the time of signal flight is less than r/s, this relay will not be detected. However, if the
imposed delay of a relay is more than r/s (i.e. ∆�� r/s) then time-based protocols can
become secure. It is also shown that designing secure time-and-location based protocols is
possible since the receiving nodes have the sender’s location (in an authentic manner carried
by the message) and compute the valid distance themselves and then compare it with the
one obtained from time measurements to detect probable attacks.
(a)
(b)
(c)
Fig. 7. There is no possibility for B (or A) to always distinguish between these three paths in
a propagation time measuring protocol as long as the propagation delay does not exceed r/s.
vadv is the signal propagation speed in the adversaries‘ channel which is higher than s.
Later, each of the two classes of protocols proposed by Poturalski et al., was divided into
two groups; Beacon (B) and Challenge/Response (CR) protocols (Poturalski et al., 2008). In
B-protocols, a node broadcasts some information without having the response of the other
side. The receivers are supposed to add the sender to their list of neighbors after some
processing and hence, this method is unidirectional. On the other side, CR-protocols
support the minimum three-phase message exchange requirement for a bidirectional
neighbor discovery. However the samples the authors have mentioned for the CR protocols
are not really bidirectional (although they could be as in their original framework). In
Poturalski’s examples, CR protocols create the chance of challenge for one side only, and all
the other nodes must run the same protocol themselves to complete the neighbor discovery
task.
The authors claimed a secure neighbor discovery protocol is characterized with two
properties: “correctness” i.e. if the protocol declares two nodes neighbors at some time, they
must indeed be neighbors at that time, and “availability” meaning that if two nodes remain
neighbors for some time (TP) then the protocol must detect this neighborship.
They defined seven types of events to describe their rules and protocols with. Table 1
summarizes six of them which we will deal with, along with their description.
Receive(A;t;m) A receives the first bit of message m at t

Bcast(A;t;m) A broadcasts message m at t
Fresh(A;t;n) Nonce n is freshly generated by A at t
Neighbor(A;t;B;C;t‘) At t, A declares B has been a neighbor of C at t‘(unidirectional)
NDstart(A;t) A starts a neighbor discovery with all the nodes at t
NDstart(A;t;B) A starts a neighbor discovery with B at t
Table 1. Some of the events symbols and their description in Poturalski et al.‘s literature.
Based on these definitions, they formally defined security requirements for the four possible
groups of neighbor discovery protocols and presented a sample pseudo-code for each group
satisfying those set of requirements. To unify the demonstrations, we have converted the
codes to message diagrams. Figure 8 and Figure 9 show Poturalski’s sample protocols for
beacon time-based (B/T) and beacon time and location-based (B/TL) neighbor discovery
protocols respectively. In Figure 10 and Figure 11, challenge-response versions of these
protocols are depicted. Notice that some of the messages in these figures are actually
intended to be received by a single node. However to comply with the broadcast-type
message transmission symbol shown in Table 1 (and in the pseudo codes), they are drawn
with broadcast-like arrows.
In all of these figures, len{.} stands for an operator giving the length of a message
transmission (in seconds for example), r is a node regular transmission range and s is the
signal propagation speed in the network communication channel.
Fig. 8. A beacon and time-based protocol pseudo code (B/T)
Fig. 9. A beacon and time-and-location-based protocol pseudo code (B/TL)

Fig. 10. A challenge-response time-based protocol pseudo code (CR/T)
Fig. 11. A challenge-response time-and-location-based protocol pseudo code (CR/TL)
As we described in the previous parts, if there are no challenges and responses the
adversaries can easily relay the broadcasted beacon packets to other parts of the network.
Also notice that the protocols times are written from a third-party‘s point of view. In
practice, B and A have time synchronization problems too, which even widens the
vulnerability window more in beacon-based protocols.
In all of these sample protocols, nodes need to verify the authentication codes attached to
the end of messages which in turn makes the protocols dependent on another key
distribution protocol (involving either storage of all the keys in the memory of the nodes
that makes it only suitable for ad hoc networks or employment of a key agreement protocol
which inherently needs a neighbor discovery protocol itself).
Regardless of these practical issues, each of these simplified representatives of neighbor
discovery protocols has been proven to satisfy Poturalski’s mathematical security
requirements under some specific conditions. Beacon time-based (B/T) protocols are secure
if the adversary’s relaying delay is greater than or equal to ‫ݎ‬Ǥ ‫ି ݏ‬ଵ . This is a rather obvious
constraint since with less than this bound, the adversary can relay some messages without
adding too much delay and keep the overall propagation and relaying delays below the
acceptable threshold ‫ݎ‬Ǥ ‫ି ݏ‬ଵ (refer to Figure 7). It can also be verified that the availability
property is also satisfied with TPB/T � ��A , t, MACA �t�� i.e. if two nodes remain
neighbors for this amount of time (and the B/T neighbor discovery is started) then, this
neighborship will be definitely detected by the protocol. Similarly, for a TL protocol to be
secure, the designer must take a communication media whose signal propagation speed is
close to the maximum possible speed in an adversary’s channel (or simply the light speed
(s=c=vadv)). If less than this speed is used (i.e. vadv>s), even though A has both B‘s and its own
locations, a pair of adversaries can simply launch the wormhole attack keeping the overall
RTT equal to 2×distance(A,B)/s which is sufficient to satisfy the protocol criterion. The
security conditions of other protocols can be found similarly. Table 2 summarizes the
conditions under which each of the four groups of neighbor discovery protocols is secure.
Protocol
Security Constraints Applied
Type
1. ∆��
B/T
2. TPB/T =len{IDA,t,MACA(t)}+ ��
1. ∆��
CR/T
2. TPCR/T =len{IDB,n}+len{MACB(n)}+ ��
B/TL 1. ∆��
2. � � ��
CR/TL 3. TPCR/TL =∞ (depending on distance)
Table 2. Conditions for each class of neighbor discovery protocols to support Poturalski’s
formal description of security requirements
We investigated some of the main external-attack-related researches and the positive and
negative aspects of them in this section. In the next section we focus on the internal attacks'
countermeasures which are different from the previous methods in nature and mostly are
cryptographic solutions.
3. Neighbor Discovery Internal Attacks Countermeasures

There are a few internal attacks affecting the performance of neighbor discovery protocols.
Needless to say some of the external attacks have the internal form too. For example, in a
wormhole attack, two internal relaying adversaries can even alternate the packet contents in
order to deceive healthy entities. Unfortunately compared to the external attacks described
in the previous section, the number of these attacks is not limited at all. This is because these
attacks are more related to the protocol-specific features than the nature of neighbor
discovery. As there are many variants for each neighbor discovery protocol family, the
internal attacks are also numerous. However, there are a few common attacks that can
conceptually cover some of these threats, and, among them, broadcast attacks are more
outstanding.
Almost all the algorithms which use a broadcasted data are susceptible to being invaded by
the broadcast-type attackers. In the sensor networks hello flooding case, a loud
advertisement can convince many surrounding nodes that the adversary is one of their
neighbors. This attack can be launched in the routing algorithms too e.g. the adversary can
rebroadcast the received RREQ packet with high power to be a part of the best routes to the
destination with a high probability. Notice that unlike the previous part, in internal attacks,
the adversary is a legitimate node i.e. a node is captured and compromised by an adversary
and all of its cryptographic keys and private information are known to the adversaries.
Broadcast attacks had been previously addressed with different names somewhat but Karlof
and Wagner were the first who introduced this attack specially in sensor networks (Karlof et
al., 2003). Consider a simple form of neighbor discovery in which a well-equipped internal
(legitimate) adversary tries to broadcast a packet called hello (beacon) and every node that
hears this packet adds the sender to its neighbors list. Since one side of the protocol which
sends the authenticated messages is adversary herself, it is impossible to rely on either time
or location information given by the other party. So generally to defeat an internal attacker
every node must rely on its own data. The same argument holds true for the CR protocols. If
the adversary gives false (but authenticated) location information to the other party, then
colluding with other relaying adversaries she can easily bypass the normal security
checkpoints.
As a solution for the broadcast-type attacks, Karlof et al. proposed to verify bidirectionality
of the links (Karlof et al., 2003). They assumed that the adversary has a high-power
transmitter but an ordinary receiver and thus is unable to capture distant weak signals. If
the receiver (B) sends valuable information in reply to the broadcasted message (hello), on
which the adversary (A) must rely for future communications, then she will be defeated
since she cannot hear B. The initial raw idea of verifying bidirectionality of the links was not
developed much by the authors, however, it was mentioned that this countermeasure is
useless if the adversary has both a high-power transmitter and a sensitive receiver.
In sensor networks, some authors then tried to limit the nodes communication ranges
through cryptographic methods (Du et al., 2005)(Lin et al., 2005)(Zhu et al., 2006). The early
methods, tried to pre-load a large number of pairwise keys in each node’s memory (key pre-
distribution). But for sensor networks, this was an impractical solution since the amount of
memory each sensor has is quite limited. Probabilistic key pre-distribution schemes were
proposed to solve this problem. In these methods, after the deployment, every node tries to
find some common keys with its neighbors through a so-called mutual key discovery
protocol (Eschenauer et al., 2002). However these solutions had problems too. The common
key discovery was itself another protocol which was needed to be secured. Besides
probabilistic approaches do not always guarantee providing a common key. So another
approach was adopted which was letting the nodes themselves establish the keys after
deployment (Lin et al., 2005)(Zhu et al., 2006). This implies the use of a negotiation protocol
between the nodes when they are deployed. It is rather obvious that to protect these
negotiations against external attackers the messages must be encrypted with some key and
since the mutual keys are not known at this phase yet, usually a pre-loaded global key is
used.
A general assumption made by these methods is that compromising a node takes time but to
stop an attacker who can capture and compromise the nodes, the whole negotiation should
not take more than Tmin seconds. To prevent further attacks, nodes themselves delete the
global key from their memory after Tmin seconds.
Secure Cell Relay (SCR) is one of the distributed key establishment/routing protocols which
resists broadcast attacks (Lin et al. 2005)(Du et al., 2006). It uses a three-way handshake
protocol to avoid the unidirectional link problem. There are two versions of this protocol but
the most recent one in which the location information has also been used is depicted in
Figure 12. In this figure, ‫ ܭ‬is the global shared key, EK the encrypting operator, and N0 is a
nonce. ‫ܭ‬஻௕ is defined as the B’s broadcast key and ‫ܭ‬஺஻ is the private key between A and B
used for later communications. At the end of this process, B adds A to its neighbors list and
stores ‫ܭ‬஺஻ and ‫ܭ‬஻௕ in a table for the future. After completion of the protocol for every node,
all the nodes delete the global key form their memory.
Fig. 12. Secure Cell Relay (SCR) neighbor discovery protocol message diagram
But SCR neighbor discovery protocol is weak in many aspects. The use of time stamps forces
the designer to somehow maintain a synchronized clock which is problematic in distributed
networks. Besides with a good design, in a three-phase message exchange protocol, both
nodes can add each other to their neighbors list because both of the parties had the chance to
challenge the other one. However, SCR messages need to be modified to provide this
feature. Also notice that after erasing the global key ‫ܭ‬, there is no chance for a node to
update its neighbors list. This property makes the protocol unsuitable for mobile scenarios.
LEAP (Zhu et al., 2003) and then LEAP+ (Zhu et al., 2006) are key distribution protocols for
sensor networks which also block the broadcasting adversary. The main goal of LEAP+ was
to create four sets of keys for each node; one set of pairwise keys for inter-neighbor
communications, one key for local broadcasting, one globally-shared network key and one
key to communicate with the sink. As the local broadcast key is made from the local
pairwise keys and dealing with the attack the two others are not needed, we only focus on
the construction of pairwise keys.
Here again the assumption of an adversary-free immune network after the initial
deployment is necessary. So we assume that for an interval which is at least Tmin seconds
there is no internal attacker present in the network. To prevent external attacks during this
period, there is a globally shared key pre-loaded into memory of the nodes.
If fk is a one-way keyed pseudo-random function (like encryption operators) with key k,
then the pairwise key construction in LEAP+ can be summarized as shown in Figure 13.
Fig. 13. The simplified form of key establishment in LEAP+ protocol at the initial
deployment phase
“A” finds the pairwise key by computing ‫ܭ‬஺஻ ൌ ݂௄ಳ ሺ‫ܦܫ‬஺ ሻ where ‫ ܭ‬஻ is found by applying the
global key to the one-way function with B’s ID as the input argument (‫ ܭ‬஻ ൌ ݂௄ ሺ‫ܦܫ‬஻ ሻ). Once
this process is done for every node, the global key is erased from the memory along with
‫ ܭ‬஻ . After the pairwise key establishment phase, all the communications are done in an
encrypted manner with the previously generated keys. Since the adversary is supposed to
capture a node not sooner than Tmin seconds, she is unable to send meaningful messages to
nodes farther than the communication range of a normal node. So, broadcast attacks are
thwarted this way. Also, the adversary cannot make new pairwise keys anymore since the
required global key is missing in the memory of the captured node.
Although this protocol performs well in terms of resource consumption and complexity, it
has drawbacks too. A closer look at the protocol reveals that B’s presence in this challenge-
response-like protocol is not as strong as it should be. The only role of B is announcing its
presence by sending back its ID. It does not even generate any random number to maintain
the freshness of the key. Although it has been assumed that the network is devoid of any
internal adversaries for at least Tmin seconds (since for example it takes at least Tmin seconds
to intrude into a tamper-resistant device memory), this is not a valid assumption for the
external ones. If external adversaries are present before the nodes deployment, they are able
to launch relay-based attacks which simply means, in the above protocol, B could be a
relayed distant node. After Tmin seconds, the adversary intrudes into A’s memory and at the
end, she has a large-range communication capability for which she had planned before. It is
also obvious that due to the erasure of the global key after Tmin seconds, nodes are unable to
restart the neighbor discovery protocol in the future and thus, this protocol does not support
mobility. Any mobility-supporting security framework must allow periodical updates of the
neighbors list. This implicitly involves participation of either time or a random number
(nonce) in the protocol to block replay attacks. The authors have virtually limited the range
one node can communicate in, through cryptographic methods. With the erasure of global
key (and the static network the authors assumed) the probable future neighbor discoveries
are limited to the detection of lost connections due to power depletions or failures.
In (Sayad et al., 2008), as an alternative solution, especially in sensor networks in which
broadcast attacks are more devastating, we proposed a probabilistic robust design
framework for the internal broadcast attacks which has a very low complexity and can even
be combined with any other secure neighbor discovery protocol. To define a probabilistic
robustness, we shall first differentiate three general attack profiles.
Fig. 14. (A) An optimal secure neighbor discovery protocol broadcast attack profile (B) An
unprotected neighbor discovery protocol broadcast attack profile (C) A broadcast attack-
resilient neighbor discovery profile
As shown in Figure 14., facing broadcast-type attacks, the behavior of neighbor discovery
protocols can be categorized into three different profiles. When the transmission power of a
node or sensor increases, the number of neighbors is also increased (conforming with the
different transmission ranges nodes have based on their available power resources), but if
the power of a node goes beyond a threshold corresponding to the maximum allowable
transmission range, then these three family of protocols behave differently. Regarding the
broadcast attacks, type A profile is the desired optimal secure neighbor discovery protocol
i.e. if an adversary tries to increase her power in order to obtain more neighbors, she will not
succeed. Complex cryptographic solutions may be put in this group.
Type B profile shows the performance of an unprotected neighbor discovery protocol. The
adversary’s payoff increases as its power goes up. In type C which is actually the focus of
our work, the maximum payoff (possible number of neighbors) of the adversary is not as
low as an optimal secure solution but definitely bounded. In this approach, the designer
tries to simplify the protocol in expense of losing some (but not all) of the security. This is a
general framework however and the amount of penalty paid for simplicity depends on the
designer’s approach. It is worth mentioning that the initial parts of all these three profiles
are similar, because as long as an adversary’s behavior falls in the class of healthy nodes’
behavior, she receives the same amount of payoff as a normal node does. So, there is no
instant solution which can decrease the payoff of an internal adversary to less than that of a
healthy node.
We propose that one can wisely manipulate Medium Access Control (MAC) protocol
parameters to achieve a type-C resistivity against broadcast attacks in sensor networks
without consuming too much energy. In a compromised network, when an adversary that is
equipped with both powerful transmitter and sensitive receiver, broadcasts a hello-like
request or beacon (hello flooding), a lot of nodes receive it almost simultaneously. In a two
(or more) way handshake protocol, these nodes will start to compete to grab the channel
and send the reply messages in order to announce their presence. Healthy nodes have small
transmission and reception ranges. Therefore, roughly speaking, those nodes located farther
than the carrier sense range of each other will try to send the reply messages back almost
simultaneously. The main idea is to tune the channel access and transmission parameters so
that the responses of these distant nodes collide with each other at the adversary’s receiver
due to the high density in arrival. If these reply messages contain valuable information
which is vital for future communications, then as the adversary is unable to decode the
reply messages correctly, she is obliged to reduce her power. This is somehow similar to the
well-known hidden node effect in wireless ad hoc networks. Figure 15. shows the attacker
and sample distant nodes colonies along with their corresponding transmission ranges.
Fig. 15. Transmission ranges of normal and adversary nodes in Sayad et al.’s framework.
Adversary’s high transmission power attracts many healthy nodes. However, far healthy
nodes do not see each other and hence start to transmit reply packets simultaneously
As in the neighbor discovery phase no infrastructure is already formed, a random channel

accessing method is preferred. So one can expect that this approach would be a probabilistic
one at the end. Although the design framework introduced is quite general, we give an
instance protocol in which the notion of important information delivery in the reply packets
transmission phase is contrived. The sample protocol in Figure 16 tries to establish a mutual
key between the two nodes using Diffie-Hellman key agreement protocol (Diffie & Hellman,
1976) while following the framework rules.
It should be noticed that neighbor discovery message collisions must be minimized in a
healthy scenario i.e. since the only difference between healthy and adversary nodes is their
transmission ranges, high-colluding protocol design for the adversary might also affect the
normal behavior of the network and this is to be avoided. In our example, this leads to a
trade-off problem.
To make a demonstration of the above proposal’s efficiency, for channel accessing, IEEE
802.15.4 WPAN standard was chosen (IEEE 802.15.4, 2006). However, some modifications
were made to the protocol to adapt its neighbor discovery to the classic three-phase type,
and the minimum and maximum back-off exponents were considered as the parameters to
be tuned. With a node distribution density of 0.01 (1 node in 100m2), the attack profile for
different values of the minimum back-off exponent is shown in Figure 17.
As it can be clearly seen, this method behaves like a type-C attack-resistant protocol. With
the maximum back-off exponent set to eight and the identical reply packets which were
seven back-off periods long, the minimum back-off exponent was swept. With this
configuration, the adversary’s payoff increases as the minimum back-off exponent goes up.
Fig. 16. An example of Sayad et al.‘s framework. Reply message includes a part of the
common key which will be used for future communications
Fig. 17. The broadcast attack profiles on Sayad et al.‘s sample protocol. The adversary
receives a negative feedback from the network in response to her greediness in obtaining
more neighbors
Notice that we also have to keep the healthy neighbor discovery scenario as intact as
possible i.e. any profile closer to the intersection of the mean number of a healthy node’s
neighbors line and the normal transmission range line is preferred. With a high minimum
back-off exponent, the profile approaches the ideal point for healthy scenarios, however, the
maximum (but still restricted) payoff of the adversary is also increased. Now it is up to the
network designer to set how much security (with respect to the broadcast attacks) should be
sacrificed for sake of gaining more efficiency.
The abovementioned design framework is a concept which can be freely combined with the
other neighbor discovery protocols and since it has a very low cost, is quite applicable in
resource-limited sensor networks. Unlike the cryptographic methods (and many other
described protocols), this framework supports (high) mobility of the nodes. Besides, in this
approach, the limiting assumption about the initial adversary-free period of the network
operation at the time of nodes deployment has been removed.
4. Conclusion
In this chapter, we introduced many of the current threats and solutions for the neighbor
discovery problem in wireless ad hoc and sensor networks. Separating the attacks into
internal and external ones, depending on how much the protocols are resistant to each of
these attacks, we categorized them into two groups as well.
Many of the current solutions focus on the external relay-based attacks while little effort has
been made to mitigate the internal attackers’ damage. Most of the current solutions for the
internal attacks are cryptographic which are either resource consuming or do not support
mobility.
Numerating all the solutions, we proposed a probabilistic countermeasure for the internal
broadcast attacks. It is obvious that combinational attacks require combinational solutions.
Our design framework for the internal broadcast attacks can be combined with nearly all the
other solutions to provide higher levels of security facing both internal and external attacks.
Acknowledgement
This work has been financially supported by Iran Telecommunication Research Center
(ITRC). The authors also wish to thank Prof. Vijay Varadharajan form Macquarie University
and Alireza Mohammadi-nodooshan from K.N.Toosi University of Technology for their
scientific support.
5. References
Akyildiz, F.; Ian, W.S., Sankarasubramaniam, Y. and Cayirci, E. (2002). A survey on sensor
networks, IEEE Communication Magazine, vol. 40, Aug. 2002., pp. 102-114
Brands, S.; Chaum, D. (1993). Distance-bounding protocols, Proceedings of Eurocrypt ’93:
Workshop on the theory and application of cryptographic techniques on Advances in
cryptology, pp. 344-360, May 1993, Springer
Chan, H. ; Perrig, A. & Song, D. (2003). Random key predistribution schemes for sensor networks,
Proceedings of IEEE Security and Privacy Symposium, pp. 197-213, 2003
Diffie, W.; Hellman, M.E. (1976). New directions in cryptography, IEEE Transactions on
Information Theory, vol. 22, No.6, Nov. 1976, pp. 644-654
Djenouri, D.; Khelladi, L., Jamel, D. and Badache, N. (2005). A survey of security issues in
wireless ad hoc and sensor networks, IEEE Communication Surveys, vol. 7, 4th
quarter 2005, pp. 2-28
Du, W.; Deng, J., Han, W.S., Varshney, P.K., Katz, J., Khalili, A. (2005). A pairwise key
predistribution scheme for wireless sensor networks. ACM Transactions on Information
and System Security (TISSEC), 2005 , pp. 228-258, 2005, ACM Press
Du, X. ; Xiao, Y., Chen, H.H. (2006). Secure cell relay routing protocol for sensor
networks: Research Articles, Vol. 6, No. 3, Wireless Communications and Mobile
Computing, May 2006, pp. 375-391
Eschenauer, L. ; Gligor, V. (2002). A Key Management Scheme for Distributed Sensor

Networks, Proceedings of ACM Conference on Computer and Communications Security,
pp. 41-47, USA, Apr. 2002, Washington DC
Hu, Y.C; Perrig, A., Johnson, D.B. (2006). Wormhole attacks in wireless networks, IEEE
Journal on Selected Areas in Communications, Vol. 24, No. 2, Feb. 2006, pp. 370-380
IEEE 802.15.4 Standard (2006). Wireless Medium Access Control (MAC) and Physical Layer
(PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs),
IEEE, Sep. 2006
Karlof, C. ; Wagner, D. (2003). Secure Routing in Sensor Networks: Attacks and
Countermeasures, Elsevier Ad hoc Networks, Vol. 1, No. 2-3, Sep. 2003, pp. 293-315
Korkmaz, T. (2005). Verifying Physical Presence of Neighbors against Replay-based Attacks
in Wireless Networks, Intenational Journal of Information Technoilogy, Vol. 11, No. 2,
2005., pp. 9-20
Lin, F. ; Du, X. (2005). Secure cell relay routing protocol for sensor networks, Proceedings of
IEEE Computing and Communications Conference, pp. 477-482, Apr. 2005
Maheshwari, R.; Gao, J. & Das, S.R. (2007). Detecting wormhole attacks in wireless networks
using connectivity information, Proceedings of IEEE International Conference on
Computer Communications, pp. 107-115, May 2007
Papadimitratos, P.; Poturalski, M., Schaller, P., Lafourcade, P., Basin, D., Capkun, S.,
Hubaux, J.P., (2008). Secure neighborhood discovery: a fundamental element for
mobile ad hoc networking, IEEE Communication Magazine, vol. 46, No. 2, Aug. 2002.,
pp. 102-114
Perrig, A.; Canetti, R., Song, D., Tygar, D. (2000). Efficient Authentication and Signing of
Multicast Streams over Lossy Channels, Proceedings of IEEE Symposium on Security
and Privacy, May 2000
Pickholtz, R.L.; Schilling, D.L. & Milstein, B. (1982). Theory of spread spectrum
communications: a tutorial, IEEE Transactions on Communications, Vol. 20, No. 5,
May 1982, pp. 855-884
Poturalski, M.; Papadimitratos, P., Hubaux, J.P. (2008). Towards provable secure neighbor
discovery in wireless networks, LCA Report (EPFL University), Oct. 2008
Poturalski, M.; Papadimitratos, P., Hubaux, J.P., (2007). Secure Neighbor Discovery in
Wireless Networks: Formal Investigation of Possibility, LCA Report (EPFL
University), 2007
Rasmussen, K.B.; Capkun, S. (2007). Implications of radio fingerprinting on the security of
sensor networks, Proceedings of International Conference on Security and Privacy for
Emerging Areas in Communication Networks, pp. 331-340, Sep. 2007
Rubinstein, M.G.; Moraes, I.M., Campista, M.E.M., Costa, L.H.M.K., Duarte O.C.M.B. (2006).
A survery on wireless ad hoc networks, In: Mobile and Wireless Communication
Networks (from International Federation for Information Processing (IFIP)), Guy Pujolle
(Ed.), Vol. 211, pp. 1-33, Springer, Boston
Sastry, N.; Shankar, U. & Wagner, D. (2003). Secure verification of location claims,
Proceedings of ACM workshop on Wireless Security, pp. 1-10, 2003, ACM Press
Sayad Haghighi, M. ; Mohamedpour, K. (2008). Securing wireless sensor networks against
broadcast attacks, Proceedings of International Symposium on Telecommunications, pp.
49-54, Aug. 2008
Shokri, R.; Poturalski, M., Ravot, G., Papadimitratos, P., Hubaux, J.P. (2008). A low-cost
secure neighbor verification protocol for wireless sensor networks, LCA Report
(EPFL University), Oct. 2008
Singelee, D.; Preneel, B. (2005), Location verification using secure distance bounding
protocols, Proceedings of IEEE International Conference on Mobile Ad hoc and Sensor
Systems Conference, pp. 840-846, Nov. 2005
Zhu, S.; Setia, S. & Jajodia, S. (2003). LEAP: efficient security mechanims for large-scale
distributed sensor networks, Proceedings of ACM conference on Computer and
Communications Security, pp. 62-72, May 2003
Zhu, S.; Setia, S. & Jajodia, S. (2006). LEAP+: Efficient security mechanisms for large-scale
distributed sensor networks, ACM Transactions on Sensor Networks, Vol. 2, No. 4,
Nov. 2006, pp. 500-528
New Trends in Network Anomaly Detection 715
32
X
New Trends in Network Anomaly Detection

Yasser Yasami1 and Saadat Pourmozaffari
Amirkabir University of Technology
Iran
1. Introduction
Computer networks are complex interacting systems composed of individual entities such
as various devices, workstations and servers. Nowadays, Internet Protocol (IP) is used as a
dominant layer 3 protocol. The evolving nature of IP networks makes it difficult to fully
understand the dynamics of the systems and networks. To obtain a basic understanding of
the performance and behavior of these complex networks, large amount of information need
to be collected and processed. Often, network performance information is not directly
available, and the information obtained must be synthesized to obtain an understanding of
the ensemble behavior.
Traditional signature-based intrusion detection techniques use patterns of well-known
attacks to match and identify known intrusions. The main drawback of theses techniques is
inability to detect the newly invented attacks. To obtain sufficient information about
complex network traffic and compensate for the weaknesses of traditional Intrusion Detection
Systems (IDS), Anomaly Detection Algorithms (ADA) are used [G.Maselli & L.Deri, 2003; K.
Hwang et al., 2004; A. Lazarevic et al., 2003]. Theses algorithms can be employed as a useful
mechanism to analyze network anomalies and detect misbehaviors issued by users, or even
unknown signature viruses and worms.
There are two main approaches to study or characterize the ensemble behavior of the
network [M. Thottan & C. Ji, 2003]: the first is inference of the overall network behavior and
the second is to analyze behavior of the individual entities or nodes. The approaches used to
address the anomaly detection problem depend on the nature of the data that is available for
the analysis. Network data can be obtained at multiple levels of granularity such as
network-level or end-user-level. The methods presented in this chpater are host-based
ADA's and are categorized in the latter approach.
In this chapter, we present some ADA's developed based on some classification methods.
The goal of this chapter is to classify each user's behavior as anomalous or normal actions in
an unsupervised fashion. Four different algorithms are disccusd and compared based on
some defined measures.
The experiments are performed on a real evaluation network test bed. Instances are
captured in eight consecutive weeks, three weeks of training and five weeks of testing. Some
1Yasser Yasami is currently an instructor of computer science & engineering at Payam Nour
University.
ARP anomaly criteria are defined. These criteria are applied to the three weeks training
instances for generating normal ARP traffic.
Performance evaluation of the approaches is conducted by five performance measures:
Sensitivity, Specificity, Negative Likelihood Ratio, Positive Predictive Value, and Negative
Predictive Value. Finally some comparisons are performed based on the defined measures.
2. Background and Related Works

Network anomaly detection is a vibrant research area. ARP anomaly detection in particular
has been of great interest. Some methods for anomaly detection are based on switch
characteristics, such as performance and backplane [D. Ármannsson et al., 2005]. In such
methods switch characteristics must be known. Our knowledge is limited to theoretical
backplane speed mentioned in datasheets. But, because switch processing power, especially
when forwarding and flooding small packets, does not equal to that of theory and
performance of switches in high load, small packet traffic degrade dramatically, so using
such algorithm, encounters functional limitations.
In other researches [D. Whyte et al., 2005], feature-based approaches for host-based analysis
of ARP anomaly detection have been suggested. To achieve more accuracy on the results,
more inputs factors to these algorithms are needed to be defined. Furthermore, the proposed
factors have correlation with each other. None of these works include any suggestion about
correlation between the factors, which affect on their precision.
The proposed algorithm in [Shekhar R. Gaddam et al., 2007] is a supervised ADA. We are
not provided with a set of anomalous and normal labeled training instances, mostly. So,
supervised algorithms such as the one proposed in [Shekhar R. Gaddam et al., 2007] are
confronted with limitations in practical applications. Furthermore, the majority of the works
proposed in [N. Ye et al., 2004; D. Mutz et al., 2006; S. Kumar & E.H. Spafford, 1994; C.
Kruegel & G. Vigna, 2003] evaluate the performance of anomaly detection methods on the
measurements drawn from one application domain, thereby addressing the problem of
anomaly detection on limited data instances collected from a single application domain.
There are some other approaches that apply machine learning techniques like symbolic
dynamics [A. Ray, 2004], multivariate analysis [N. Ye, et al., 2002], neural-networks [Z.
Zhang et al., 2001], self-organizing maps [S.T. Sarasamma et al., 2005], fuzzy classifiers [J.
Gomez & D.D. Gupta, 2001] and others [H.S. Javitz & A. Valdes, 1991; I. Levin, 2000; D.Y.
Yeung & C. Chow, 2002; R. Agarwal & M.V. Joshi, 2000; G. Qu et al., 2005]. Almost all of
these anomaly detection approaches apply single machine learning techniques while recent
advances in machine learning show that selection, fusion and cascading [A. Verikas et al.,
1999; J. Kittler et al., 1998; L.I. Kuncheva, 2002] of multiple machine learning approaches
have a better performance yield over individual approaches.
3. Network Anomalies
Network anomalies typically refer to circumstances when network operations deviate from
normal network behavior. The anomalies can arise due to various causes such as
malfunctioning network devices, bad configuration in network services and operating
systems, network overload, malicious denial of service attacks, ill advised applications
installed by users, high level users’ effort to discover network and gather information about
it and its devices, and network intrusions that disrupt the normal delivery of network
services. These anomalous events will disrupt the normal behavior of some measurable
network data. The definition of normal network behavior for measured network data is
dependent on several network specific factors such as dynamics of the network being
studied in terms of traffic volume, the type of network data available, and types of
applications running on the network. Accurate modeling of normal network behavior is still
an active field of research, especially the online modeling of network traffic.
Some of intrusions and malicious usages don’t have significant effects on network traffic
(i.e. ARP Spoofing). So such misbehavior is not addressed in this chapter. Other types of
attacks are based on broadcasting large number of packets with abnormal behavior, as in the
case of DoS attacks. Abnormality is generally different from large number of packets,
although large number of packets introduces abnormality to network traffic, too. High
percentage of packets, degrade network performance. There are other types of attacks which
apply broadcast traffic for detecting live hosts in network. Network anomalies can be caused
by some unintentional and curious motivations, too. To detect these anomalies an algorithm
is introduced in this chapter. The main objective of the ADA's is detection of zero-day
worms and viruses broadcasting ARP requests to find vulnerable hosts. Besides, the
approach will be very effective in preventing unwanted traffic, too.
4. Anomaly Detection by Stochastic Learning Automata

In this section the proposed method based on Stochastic Learning Automata (SLA) is
described. A learning algorithm that constructs host-based learning models of normal ARP
behavior from attack-free network ARP traffic is presented. Behavior that deviates from the
learned normal model signals possible novel attacks.
4.1 Formal Description of SLA

An automaton is a machine or control mechanism designed to automatically follow a
predetermined sequence of operations. The stochastic emphasizes the adaptive nature of the
automaton. This adaptation is the result of learning process.
Formally, the automaton can be represented by quintuple {,,,F(•,•), H(•,•)} [K. S.
Narendra & M. A. L. Thathachar, 1989], where :
 is a set of internal states. At any instant n, the state (n) is an element of the finite
set which is as follow:
= {i |i, 1 ≤ i ≤ s} (1)
 is a set of actions (or outputs of the automaton). The output or action of an
automaton at the instant n, denoted by (n), is an element of the finite set .
Description of is as below:
= {i |i, 1≤ i ≤ r} (2)
 is a set of responses (or inputs from the environment). The input from the
environment (n) is an element of the set which could be either a finite set or an
infinite set, such as an interval on the real line:
{i | i, 1≤ i ≤ m} or {(a,b)} (3)
 F(•,•): is a function that maps the current state and input into the next
state. F can be deterministic or stochastic:
(n1) F[(n),(n)] (4)
 H(•,•): is a function that maps the current state and input into the
current output. If the current output depends on only the current state, the
automaton is referred to as state-output automaton. In this case, the function
H(•,•) is replaced by an output function G(•): , which can be either
deterministic or stochastic:
(n) G[(n)] (5)
The automaton applied for our application is of type of the later case.
4.2 General Reinforcement Scheme

In order to describe the reinforcement scheme, p(n) is defined as a vector of action
probabilities :
Pi(n) = P( (n) = i ) , 1 < i < r (6)
Updating action probabilities can be represented as follow:
P(n+1) = T[p(n), (n), (n)] (7)
where T is a mapping. This formula says the next action probability p(n+1) is updated based
on the current probability p(n), the input from the environment and the resulting action.
The general scheme for updating action probabilities for an r-action automaton in an
environment with  is as follow:
if (n) = i , when  = 0:
pj(n+1) = pj(n) – gj(p(n)), j, j ≠ i (8.a.1)

r
pi (n  1)  pi (n)   g ( p(n))
k 1
k (8.a.2)
k i
and when  = 1 :
Pj(n+1) = pj(n) + hj(p(n)), j , j ≠ i (8.b.1)

r
pi (n  1)  pi ( n)   h ( p(n))
k 1
k (8.b.2)
k i
Where gk and hk, k = 1, 2, …, r are continuous, nonnegative functions with the following
assumptions :
0 < gk(p(n)) < pk(n) (9.a)

r
0 p
k 1
k (n)  hk ( p ( n))  1, 1  i  r , 0  p k  1
(9.b)
k i
4.3 Why Using SLA

Nowadays in Intrusion Detection researches, efforts are mainly focused on misuse detection
direction, since it is strait forward and easy to implement. But it has some inherent
disadvantages. It is difficult to gather required information on known attack (content of TCP
packets must be checked while maybe not enough). The most severe disadvantage is that it
possibly can not detect attempts to new and unforeseen vulnerabilities.
These disadvantages make Anomaly Detection approaches a vibrant research area. Here we
will make some effort to do host-based anomaly detection.
In order to model normal user’s behavior, we believe that a good model should be able to
give a reasonable explanation of the real system. Here SLA is used to satisfy this condition.
Our reasons are as follow:
1. A computer user of a system should have some kind of routine behavior, especially
for long-term computer users. This is what anomaly detection is based on.
2. Each computer user should be in some kind of state, when using computer. This state
corresponds to what he currently mainly wants to do. For example, at one time, the
user wants to browse web sites for shopping, at another time, he wants to make
programming, etc. In each state, the user will mainly do some correspondent actions
which are different with other states. So from statistic aspect, the distribution of every
kind of connections or commands in each state will be different from other states.
3. Transition from one state to another can be treated roughly as state transition process
of Finite State Machine. For the Steady State Duration time, we treat it as Gaussian
(Normal) distribution, since human doing a task is not without remembering, so
exponential distribution can not be used. On state transition decision, because human
usually make decision on which task he will do next based on the previous several
tasks he has done, so we treat the transition probability with conditional transition.
So from above three aspects, we believe SLA can be used for modeling of computer user’s
behavior in an understandable and accurate way.
4.4 Modeling Normal Behavior of ARP Traffic with SLA

For each node of network one automaton is learned from attack-free network ARP traffic.
In this approach, there is one state corresponding to each node in the network. So that, the
set of internal states for each node learning automaton is defined as follow:
= {IPi | 0 ≤ i ≤ s} (10)
Where IPi is the IP address of node i and s is the number of existing nodes in the network.
The set of actions (or outputs of the automaton) is a set of triples as follow:
= {(IPi, i, i2) | i, 1 ≤ i ≤ r} (11)
Where IPi is the state identity and i and i2 are the Average and Variance of steady state
duration, respectively which are defined as below:
ni
t
j 1
ij
(12)
i 
ni
  E [t ]  E [t i ] 2
i
2
i
2
(13)
Where tij is the elapsed time after jth ARP request with destination IP address corresponding
to i until next ARP request with source IP address i issues. ni is the number of occurrence of
the i (i.e., the number of ARP requests issued with destination IP address corresponding to
state i).
E[ti] and E[ti2] in the second expression are expected values (mean) of the random variables
ti (steady state duration of state i) and ti2, respectively.
There is one action (output) for each state of the automaton, so we have (r = s).
The environment (network in our model) interacts with this automaton by introducing ARP
requests to it. , the set of responses (or inputs from environment) is defined as follow:
{reqi | 0 ≤ i ≤ m} (14)
Where reqi is ARP request with destination IP address i. As stated earlier, the normal model
is learned for each node x in the network. Therefore, the source IP address of all of the
members of set  is same as the node whose normal model is under learning. It is obvious
that m in this definition is equal to the number of nodes. We have (m = s).
Each ARP request causes a transition from one state (the state corresponding to destination
IP address of the previous ARP request) to another state (the state corresponding to the ARP
request destination IP address). The formal description of transition function (F) is as stated
below:
IPn+1 = F (IPn, reqn+1) (15)
The transition function F of the automaton is deterministic and the result of this function is
uniquely specified for each state. For each special state X and reqY issued from the
environment, the automaton changes its state from X to Y and this is deterministic.
The current output of the model is dependant on only current state, so the automaton is
state-output. The formal description of output function is as below:
G(IPn) = (IPn, n, n2) (16)

This function is stochastic and nondeterministic, because the output set  is updated
whenever the environment interacts with the automaton (whenever an ARP request issues).
The elements of the set G are denoted by gij. The value of this element represents the
probability that the action performed by the automaton is (IPj, j, j2) given the automaton is
in state IPi :
Gij = P(n) = (IPj, j, j2) | (n) = IPi], 1 < i, j < s (17)
In short the automaton takes an input from the environment and produces an action based
on this.
The automaton is a variable-structure one. Although, transition function F is deterministic
and does not change over time, but the output function is stochastic and its value changes
over time.
4.5 The Reinforcement Scheme of the Proposed Model

The reinforcement scheme is the basis of learning process for learning automata. The formal
definition of reinforcement scheme can be obtained by substitution of functions in Equation
(8) as follow:
If (n) = IPn and the environment issues reqn then award function will be as:
a
g k ( p(n))  (18)
n 1
If (n) = IPn and the environment issues reqm, m  n then penalty function will be as:
b
hk ( p(n))  (19)
n 1
Therefore, the formal definition of the reinforcement scheme is given as:

if (n) = IPn , when environment response is reqn:
a
p j ( n  1)  p j ( n )  , j, j ≠ i (20.a.1)
n 1
(r  1).a
pi (n  1)  pi (n)  (20.a.2)
n 1
and when environment response is reqm, m  n:
b
p j (n  1)  p j (n)  , j, j ≠ i (20.b.1)
n 1
(r  1)  b
pi (n  1)  pi (n)  (20.b.2)
n 1
4.6 Anomaly Detection

The purposed algorithm constructs a learning model of normal ARP traffic for each existing
host in the network. The model resulted from this learning process is named as normal model.
Discussion of producing normal ARP traffic will come, latter. Online network traffic is
compared by the normal model in a process referred to as “matching process”. Any deviation
from the normal model is an indication of anomaly which is quantified as Anomaly Score
(AS) parameter. The algorithm makes decisions on normality or abnormality of each node
by means of the calculated AS parameter in matching process.
An important parameter in anomaly detection is an accurate threshold value. An indication
of normality is used for this purpose, referred to as Normal Score (NS). This parameter is
described latter.
- Anomaly Score (AS)

We apply AS as an anomaly indicator of a node ARP traffic and obtain it from weighted
summation of Partial Anomaly Scores (PAS'es) as the following equation denotes:
N
K nj A nj
AS  K s As  n2
Pij
(21)
Where N in the above equation is the number of previous environment responses (ARP
requests issued in matching process) affecting on the AS value and decision making on
normality or abnormality of the corresponding node, j is the state in the learning model
which the node will be in after nth ARP request and Anj is the PAS, explained latter.
AS is the PAS corresponding to the state which the node will be inside after the first ARP
request ((N-1)th previous ARP request) and can be stated formally as follow:
AS = Ai1, 1≤i≤r (22)
K nj is coefficient of the participating term in weighted summation of AS and is dependent

on state j in learning model, such that occurrence of a state with low probability has more
effect on AS than occurrence of a state with high probability. So the purposed value for this
parameter is simply the inverted state probability:
K nj  Pj (n) 1 ( 23)
This justification can be described for transition probability residing in denominator. Pij(n) is
conditional probability of transition from state i to state j caused by nth ARP request
(environment response req nj ), given the sequence of observed transitions in matching
process. It can be formally described as follow:
Pij ( n)  P (Tij T I I T I I ...T I ) (24)

12 2 3 n  2 I n 1
Where, Tij is transition from state i to state j, TI I TI I ...TI I is sequence of transitions in

1 2 2 3 n2 n1
matching process, Ix indexes correspond to states where the node will be in after nth ARP
request in matching process (after reqIn ), as I1 and In-1 indexes correspond to states S and i,
n
respectively. This conditional probability is calculated in learning process as follow:
P(TI1I 2 TI 2 I 3 ...TI n  2i Tij )

Pij (n)  (25)
P (TI1I 2 TI 2 I 3 ...TI n  2i )
- Partial Anomaly Score (PAS)

Defined as deviation from average steady state duration in a state:
 ( nj  t nj ) 2
 , if t nj   nj
A nj   ( j )
n 2
(26)

0, if t nj   nj
Where t nj is the time interval between nth and (n+1)th ARP requests in matching process,
such that the node will be in state j after nth ARP request.
Steady state duration values, greater than average value (jn) for each state, indicate normal
behaviors, so its effect does not participate in AS.
There are some states in learning model which environment doesn't interact with them
(there is not any reqi for such states labeled with i). For such states we have:
i = (IPi, i, i2) = (IPi, 0, 0) (27.a)

Pi(n) = 0 (27.b)
A similar problem exists about sequences of transitions with these conditions. Minimum
state probability among probabilities of all of the existent states is used for probability of
such states and minimum probability among sequences of transitions corresponding to the
node for the probability of such sequences of transitions, as formally stated below:
Pi(n) = MIN {Pk(n) | 1 ≤ k ≤ r} (28.a)

Pij(n) = MIN {Pik(n) | 1 ≤ k ≤ r} (28.b)
Also, we considered statistical parameters of each state i satisfying conditions stated in

equation (27) (parameters i, and i2), as follow:
 in = MAX {  kn | 1 ≤ k ≤ r } (29.a)
 = MIN {  | 1 ≤ k ≤ r }
i
n n
k (29.b)
It means taking the worst case value for these parameters.
- Normal Score (NS)

NS, as hinted above, is an indicator of normality degree and is a function of partial NS'es
(PNS). We define PNSi as normality score at ith time interval in learning process. NS is
calculated as the following equation states:
NS = MAX {PNSi} (30)
For calculating PNSi, we use the same method as used for calculation of AS, but in this case
for normal ARP traffic. We get this normal traffic from purified traffic in different time
intervals and calculate PNS for each interval. It is obvious that to obtain a right value for NS
from PNS's, these time intervals should be of the same length.
- Threshold calculations
An estimation of ARP traffic normality is required for a right threshold values. The NS value
gives this estimation to the hand. It is an indication of maximum value of AS without being
detected as abnormal. ASi values calculated in matching process, satisfying the inequality
NSi ≤ Thi < ASi for threshold value of Thi of node i, are detected as abnormal. Making
decisions on Th values affect on False Negative and False Positive and accuracy of algorithm.
There are various ways to this problem. One most simple and feasible way is to get
threshold k times of NS. For example, the chosen value for k is 1.2 in [Kai Hwang, Hua Liu
& Ying Chen, 2004].
5. Anomaly Detection by K-Means and ID3 Decision Trees

5.1 Anomaly Detection by K-Means Algorithm
The K-Means algorithm [R. Duda et al., 2000] groups N data points into k disjoint clusters,
where k is a predefined parameter such that k < N. The steps in the K-Means clustering-
based anomaly detection method are as follows:
1. Select k random instances from the training data subset as the centroids of the
clusters C1, C2,…,Ck.
2. For each training instance X:
a. Compute the Euqlidean distance:
D(Ci, X), i = 1...k.
Find cluster Cq that is closest to X.

b. Assign X to Cq. Update the centroid of Cq. (The centroid of a cluster is the
arithmetic mean of the instances in the cluster.)
3. Repeat Step 2 until the centroids of clusters C1, C2,…, Ck stabilize in terms of mean-
squared-error criterion. Finally, the algorithm aims at minimizing an objective
function (here, squared error function). The objective function J:
k N
2
J   X i( j )  c j (31)
j 1 i 1
where the term:

2
X i( j )  c j
is a chosen distance measure between a data point X i( j ) and the cluster centre cj, is
an indicator of the distance of the N data points from their respective cluster
centers.
5.2 Anomaly Detection by ID3 Decision Trees

The ID3 decision tree learning algorithm [T. Mitchell, 1997] computes the Information Gain G
on each attribute A, as:
Sv
G( S , A)  Entropy( S )  
vValues( A)
S
Entropy( Sv ) (32)
where S is the total input space and Sv is the subset of S for which attribute A has a value v.
The Entropy(S) over c classes is given by:
c
Entropy (S)    p log
i 1
i 2 (pi ) (33)
where pi represents the probability of class “i”. The probability of class i is calculated as
follow:
Ni
pi  c
(34)
N
j 1
j
where, Nx is the number of training instances in class x.

The attribute with the highest information gain, say B, is chosen as the root node of the tree.
Next, a new decision tree is recursively constructed over each value of B using the training
subspace S-{SB}. A leaf-node or a decision-node is formed when all the instances within the
available training subspace are from the same class. The algorithm constructs the ID3
decision tree with the normal purified traffic. Anomaly detection is performed using this
tree by traversing the tree with features of test instance. If the traverse reaches a leaf node,
the test instance will be detected as normal; else it will be detected as abnormal.
5.3 Anomaly Detection by Combined K-Means Clustering & ID3 Decision Trees
We are provided with a normal (purified) training data set Xi where each instance
represents an n-dimensional vector. The approach has two phases: training and testing.
During training phase, K-Means-based anomaly detection method is first applied to
partition the training space into k disjoint clusters C1, C2,…, Ck. Then ID3 decision tree is
trained with instances in each K-Means cluster. The K-Means clustering method ensures
that each training instance is associated with only one cluster. However, if there are any
subgroups or overlaps within a cluster, the ID3 decision tree trained on that cluster refines
the decision boundaries by partitioning the instances with a set of if-then rules over the
feature space.
The combined application of the two algorithms overcomes some limitations of each
algorithm when applied individually. For example, selection of a right value for parameter k
in the K-Means clustering algorithm can affect on the overall accuracy of the algorithm.
Considerably little values of k, compared to inherent number of natural subgroupings
within the training data will lead to overlapping subgroups within clusters. This problem is
compensated by ID3 decision tree constructed in each cluster.
The testing phase of the algorithm includes two phases: the candidate selection phase and
candidate combination phase. In the first phase, AS from K-Means clustering and decisions
from ID3 decision tree are extracted. In the second phase, the final AS is gotten from
combined results of K-Means clustering and ID3 decision tree. The algorithms applied for
candidate selection and candidate combination is explained below.
5.4 The Candidate Selection Phase

Steps of the candidate selection algorithm are as follow:
1. For each test instance Zi, 1 ≤ i ≤ n:
a. Compute Euclidean distance:
D(Zi, rj) , j = 1 < j < n,
and find f clusters closest to Zi.

b. Compute K-Means AS and extract decisions of ID3 decision trees for f
nearest candidate clusters.
2. Return Anomaly Score Matrix for Zi.
The algorithm gets test instances Zi, 1 ≤ i ≤ n, and the parameter f as inputs and gives matrix
AS[f×2] for each test instance Zi. f is a user defined parameter. If DT1, DT2,…, DTk be the ID3
decision tress on clusters C1, C2,…, Ck formed by applying the K-Means method on the
training instances, and r1, r2,…, rk be the centroids of clusters C1, C2,…, Ck, respectively, then
given a test instance Zi, the candidate selection procedure extracts AS'es from f candidate
clusters G1, G2,…,Gk. The selected f candidate clusters are f clusters in C1,C2,…,Ck that are
nearest to Zi in terms of Euclidean distance between Zi and the cluster centroids.
Let l1, l2,…, lf be the centroids of candidate clusters G1, G2,…, Gf and the Euclidean distances
between the test vector Zi and the f candidate clusters is as follow:
D(Zi, lj) = dj, 1 ≤ j ≤ f (35)
the AS for each of the f candidate clusters is calculated by K-Means clustering as follow:
dj
AS j  P(C j )  [1  ], 1  i  n, 1  j  f
k
(36)
 D( Z , r )
a 1
i l
P(Cj) in the above equation is probability of cluster Cj and is calculated as the following
equation indicates:
NCj
P (C j )  k
(37)
N
i 1
Ci
where the nominator in the above equation is the number of training instances of cluster Cj
and the denominator is the total number of training instances.
Term:
dj
1 k
 D( Z , r )
a 1
i l
in equation (36) is referred to as Scaling Factor (SF). It scales P(Cj) by weighing it against the
ratio of the Euclidean distance between the cluster j and Zi, and the sum of Euclidean
distances between Zi and the clusters C1, C2,…, Ck. The SF penalizes the probability of cluster
Cj with its distance from the test vector Zi, such that little distance Dj yields a high ASj and
voice versa. The decision from the ID3 decision trees associated with the f candidate clusters
are either "0" or "1" representing normal or abnormal test instances, respectively. This output
is depended on that the decision trees can be traversed reaching to a leaf node.
For each test instance Zi, the candidate selection phase outputs a matrix AS[f×2] with AS
calculated by K-Means clustering algorithm and decision extracted from ID3 decision tree.
The final AS resulted from combined application of the two algorithms is extracted by
Candidate Combination. A detailed explanation of this algorithm follows.
5.5 The Candidate Combination Algorithm

Candidate Combination algorithm take as input the output of Candidate Selection
algorithm, the AS matrix including the ASj, 1 < j < f, values of the K-means clustering
algorithm and the decisions of ID3 decision trees over f candidate clusters. The algorithm
then order the f candidate clusters G1, G2,…, Gk in AS matrix such that the distances
d1,d2,…,df between Z and the candidate clusters G1, G2,…, Gf , respectively, satisfy
d1<d2<…<df . The first ASj value of the K-means clustering algorithm in the ordered AS
matrix satisfying each of the below conditions, is selected as the final AS value of the
combined K-Means clustering and ID3 decision tree algorithm. The conditions are:
ASj ≤ 0.5 & DTj = 0

ASj > 0.5 & DTj = 1
Where DTj = "0" or DTj = "1", means ID3 decision tree of cluster j classifies the test instance
as normal or abnormal, respectively. Finally, for each test instance Zi, an AS value in
continuous closed interval [0,1] is yielded from the combined application of the two
algorithms. The Threshold Rule is used for classification of the test instance Zi as an
anomalous or normal instance. The threshold rule for classifying a test instance Zi that
belongs to cluster Cr is as follow:
Assign Z1 if AS >,

Otherwise Z0.
where  is a predefined threshold.

6. Evaluation Test Bed Network and Data Set

Our test bed network in this research work was a typical network with about 900 active
devices. With exception of a few servers, all of hosts run different Microsoft Windows
platforms like windows 98, 2K, XP and 2003 Server. The network is connected to Internet via
a 2Mbps link and about 200 stations and 5 servers are connected to internet concurrently.
Internet access was enabled for majority of hosts.
The captured data set contains approximately, eight weeks, three weeks of training and two
weeks of test data. The main resources of abnormalities in our evaluation test bed network
are malicious softwares, malfunctioning network devices, ill-advised applications, scanning
tools and high level user’s efforts for network discovery. To capture network traffic we used
a computer connected to the core switch of the network. Capturing traffic and some
statistical parameters from it, is performed in real-time interaction with our prototype, by
setting the sniffing machine's NIC in sniffing mode. A sniffing tool in VC++ powered by
WinPcap application programming interface has been developed for traffic capturing. The
implementations have been performed by Matlab V.7.5.0.342.
Some anomaly criteria are defined and applied to the captured ARP traffic to generate
normal training instances. These anomaly criteria are as follow:
 ARP rate: this criterion is defined as the overall number of ARP requests divided by the
length of time over which these were observed.
 Burstiness: if we define the maximum instantaneous ARP request rate for a device to be
the inverse of the shortest observed inter-request time between two consecutive
requests from that devise, burstiness can be defined as the ratio of maximum request
rate to the ARP rate. The burstiness characteristics of ARP traffic for our evaluation test
bed network are illustrated in figure (1). This diagram shows that most devices in
normal operation do not send ARP request broadcasts in bursts.
900
800
700
600
Device ID
500
400
300
200
100
0
0 500 1000 1500 2000 2500 3000 3500 4000
Burstiness
Fig. 1. Burstiness characteristics of ARP traffic for our test network.

 Sequential scans: sequential scan is defined as ARP requests with sequential destination
IP addresses. ARP requests in normal conditions have not sequential destination IP
addresses.
 Dark space: is defined as ARP requests with destination IP addresses not included in
address space of network.
 Repetitive Requests: this criterion is defined as ARP Requests within time intervals
smaller than expiration time of corresponding entries in ARP tables. ARP tables
maintained by each host or network device are updated when an ARP request is issued.
This caching mechanism prevents repetitive ARP requests.
7. Evaluation & experimental results

Our evaluation is based on the following criteria:
 Sensitivity: probability that a test result will be positive when there is anomaly (True
Positive or TP).
 Specificity: probability that a test result will be negative when the there is not anomaly
(True Negative or TN).
 Negative likelihood ratio: ratio between the probability of a negative test result given the
presence of the anomaly and the probability of a negative test result given the absence of
the anomaly, i.e. Negative likelihood ratio = False negative rate / True negative rate =
(1-Sensitivity) / Specificity.
 Positive predictive value: probability that the anomaly is present when the test is positive.
 Negative predictive value: probability that the anomaly is not present when the test is
negative.
ARP traffic has been applied to detect network abnormalities in the approach, as stated
before. In K-Means clustering, ID3 Decision trees, and the combinatorial approach of
these two algorithms we are provided with training and test data set Xi , 1 ≤ i ≤ N,
where Xi represents a 9-dimentional vector as follow:
Xi=(Si1, Di1, Ti1, Si2, Di2, Ti2, Si3, Di3, Ti3) (38)
7
Day1
Day2
6
Day3
Day4
5
Sequence Length
1
2 3 4 5 6
Average Sequence Iteration
Fig. 2. Iteration of sequences with different length.
Six'es and Dix'es in the data instance Xi are source and destination IP addresses of ARP
requests. We have used three successive ARP request characteristics in each data instance Xi.
Tix is the discretely quantized time interval between two successive ARP requests. The
main reason for using characteristics of multiple ARP requests in each data instance Xi, is
that user activities include some sequential activities which are depended on the state the
user is in and what he mainly wants to do in that state. So, individual ARP requests can not
be applied for detecting abnormalities. Figure (2) presents average iteration of ARP requests
sequences with different length within four different days. As it is obvious from this figure,
there is an egregious difference between iteration of sequences of length 3, 4 such that
iteration of sequences of length 4 is as close to 1 as desired. So, we have used three
successive ARP requests in each data instance Xi.
The experimental results within five successive weeks are represented in table (1). The
experiments are based on five evaluation measures, described above. The Sensitivity,
Specificity, Negative Likelihood Ratio, Positive Predictive Value, and Negative Predictive
Value characteristics of K-Means clustering, ID3 Decision tree, SLA-based and the
combinatorial K-Means+ID3 is illustrated in figures (3) to (7). These figures show that:
Sensitivity Characteristics
ID3+K-Means Clustering SLA K-Means Clustering ID3
0.98
0.96
0.94
0.92
0.9
Sensitivity
0.88
0.86
0.84
0.82
0.8
0.78
1 2 3 4 5
Weeks
Fig. 3. Comparison of Sensitivity characteristics.
Specificity Characteristics
ID3+K-Means Clustering SLA K-Means ID3
1
0.99
0.98
0.97
Specificity
0.96
0.95
0.94
0.93
0.92
0.91
1 2 3 4 5
Weeks
Fig. 4. Comparison of specificity characteristics.

Negative Likelihood Ratio Characteristics
0.18
0.16
Negative Likelihood Ratio 0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
1 2 3 4 5
Weeks
Fig. 5. Comparison of negative likelihood ratio characteristics.
Positive Predictive Value Characteristics
0.95
Positive Predictive Value
0.9
0.85
0.8
0.75
0.7
1 2 3 4 5
Weeks
Fig. 6. Comparison of positive predictive value characteristics.
Negative Predictive Value Characteristics
1
0.99
Negative Predictive Value
0.98
0.97
0.96
0.95
0.94
0.93
0.92
0.91
1 2 3 4 5
Weeks
Fig. 7. Comparison of negative predictive value characteristics.

Week 1 2 3 4 5
Total Number of Test Instances 679100 856200 987000 598400 832700
The Number of Abnormality Instances 4500 10900 3400 5900 4200
Sensitivity 0.938326 0.958148 0.933155 0.961538 0.957303
Specificity 0.999704 0.99942 0.999929 0.999747 0.999903
Negative Likelihood Ratio 0.061692 0.041876 0.06685 0.038471 0.042701
Positive Predictive Value 0.955157 0.956444 0.980337 0.976563 0.981567
Negative Predictive Value 0.999585 0.999444 0.999746 0.999578 0.999771
Table 1. Evaluation results of the K-means+ID3 in five weeks.
1) K-Means+ID3 method has better performance than the other methods in terms of
all defined measures.
2) The performance of the SLA-based approach is in-between the combinatorial
approach and each of individual K-Means and ID3.
3) Individual K-Means has better performance than individual ID3.
Malicious softwares by issuing a large number of ARP packets in little time intervals had
large effects on traffic abnormalities. They assigned high percentage of triggered alarms to
themselves. Bad-configured applications were the main origins of abnormality after
malicious softwares. Curious users' activities, malfunctioning or bad configured network
devices, and etc. affect on network traffic abnormalities but had a lower portion in it.
8. Conclusion
This chapter presented some anomaly detection approaches for classification of anomalous
and normal activities in computer network ARP traffic. The proposed approaches use some
well-known machine learning methods: the SLA, K-Means clustering and the ID3 decision
tree learning approaches. As described, in SLA-based approach a learning algorithm has
been used for modeling of normal ARP traffic behavior. Making decisions on abnormal
behavior of each device in the network is based on comparison of online behavior of each
host by its normal model. In the combinatorial approach based on K-means and ID3
decision trees, the K-Means method was first applied to partition the training instances into
k disjoint clusters. The ID3 decision tree built on each cluster learns the subgroups within
the cluster and partitions the decision space into finer classification regions; thereby
improving the overall classification performance. This combinatorial method was compared
with the individual K-Means and ID3 methods and the other proposed approaches based on
SLA in terms of the overall classification performance defined over five different
performance measures. Results on real evaluation test bed network data sets show that: the
proposed method outperforms the individual K-Means and the ID3 compared to the other
approaches. The performance of SLA is in-between the proposed combinatorial K-
Means+ID3 and individual K-Means and ID3, in terms of all the five performance measures
over the real network ARP traffic data set.
Further research should be carried out to evaluate the performance of the proposed
approaches with other combinatorial approaches which can be developed by different
clustering approaches.
Acknowledgements
This work was partially supported by the Payam Nour University and Iran Information
Technology Research Center.
9. References
A. Lazarevic, A. Ozgur, L. Ertoz, J. Srivastava, and V. Kumar, “A Comparative Study of
Anomaly Detection Schemes in Network Intrusion Detection”, Proceedings of SIAM
Int’l Conference Data Mining, May 2003.
A. Ray, “Symbolic Dynamic Analysis of Complex Systems for Anomaly Detection”, Signal
Processing, vol. 84, no. 7, pp. 1115- 1130, 2004.
A. Verikas, A. Lipnickas, K. Malmqvist, M. Bacauskiene, and A. Gelzinis, “Soft Combination
of Neural Classifiers: A Comparative Study”, Pattern Recognition Letters, vol. 20, pp.
429-444, 1999.
C. Kruegel and G. Vigna, “Anomaly Detection of Web-Based Attacks”, Proceedings of ACM
Conference Computer and Communication Security, Oct. 2003.
D. Ármannsson, G. Hjálmtýsson, P. D. Smith, L.Mathy; “Controlling the Effects of
Anomalous ARP Behaviour on Ethernet Networks”, Proceedings of the 2005 ACM
conference on Emerging network experiment and technology, 2005, pp. 50–60.
D. Mutz, F. Valeur, G. Vigna, and C. Kruegel, “Anomalous System Call Detection,” ACM
Trans. Information and System Security, vol. 9, no. 1, pp. 61-93, Feb. 2006.
D. Whyte, E. Kranakis, P. Van Oorschot, “ARP-Based Detection of Scanning Worms within
an Enterprise Network”, Proceedings of Annual Computer Security Applications
Conference (ACSAC 2005), Tucson, AZ, Dec. 5-9, 2005.
D. Y. Yeung and C. Chow, “Parzen-Window Network Intrusion Detectors”, Proceedings of
16th Int’l Conf. Pattern Recognition, vol. 4, pp. 385- 388, Aug. 2002.
G. Maselli, L.Deri, “Design and Implementation of an Anomaly Detection System: an
Empirical Approach”, Proceedings of Terena TNC 2003, May 2003, Zagreb, Croatia.
G. Qu, S. Hariri, and M. Yousif, “A New Dependency and Correlation Analysis for
Features”, IEEE Trans. Knowledge and Data Eng., vol. 17, no. 9, pp. 1199-1207, Sept.
2005.
H. S. Javitz and A. Valdes, “The SRI IDES Statistical Anomaly Detector”, Proceedings of IEEE
Symp. Security and Privacy, pp. 316-326, May 1991.
I. Levin, “KDD-99 Classifier Learning Contest: LLSoft”s Results Overview”, SIGKDD
Explorations, vol. 1, pp. 67-75, Jan. 2000.
J. Gomez and D.D. Gupta, “Evolving Fuzzy Classifiers for Intrusion Detection”, Proceedings
of 2002 IEEE Workshop Information Assurance, June 2001.
J. Kittler, M. Hatef, R.P.W. Duin, and J. Matas, “On Combining Classifiers”, IEEE Trans.
Pattern Analysis and Machine Intelligence, vol. 20, no. 3, pp. 226-239, Mar. 1998.
K. Hwang, H. Liu, and Y. Chen, “Protecting Network-Centric Systems with Joint Anomaly
and Intrusion Detection over Internet Episodes”, IEEE IPDPS- 2005, Oct.8, 2004.
K. S. Narendra, M. A. L. Thathachar, Learning Automata: an introduction, Prentice-Hall,
1989.
Kai Hwang, Hua Liu, Ying Chen , “Cooperative Anomaly and Intrusion Detection for Alert
Correlation in Networked Computing Systems”, IEEE Trans. Dependable and Secure
Computing, November 24, 2004.
L. I. Kuncheva, “Switching between Selection and Fusion in Combining Classifiers: An

Experiment”, IEEE Trans. Systems, Man, and Cybernetics, vol. 32, no. 2, pp. 146-156,
Apr. 2002.
M. Thottan and C. Ji, “Anomaly Detection in IP Networks”, IEEE Trans. Signal Processing,
vol. 51, no. 8, pp. 2191-2204, 2003.
N. Ye, S. M. Emran, Q. Chen, and S. Vilbert, “Multivariate Statistical Analysis of Audit Trails
for Host-Based Intrusion Detection”, IEEE Trans. Computers, vol. 51, no. 7, pp. 810-
820, 2002.
N. Ye, Y. Zhang, and C.M. Borror, “Robustness of the Markov-Chain Model for Cyber-
Attack Detection”, IEEE Trans. Reliability, vol. 53, no. 1, pp. 116-123, 2004.
R. Agarwal and M.V. Joshi, “PNrule: A New Framework for Learning Classifier Models in
Data Mining (A Case-Study in Network Intrusion Detection)”, Technical Report
DSTO-GD-0286, Dept. of Computer Science, Univ. of Minnesota, 2000.
R. Duda, P. Hart, and D. Stork, Pattern Classification, second ed. Wiley Publishers, Oct. 2000.
S. Kumar and E.H. Spafford, “A Pattern Matching Model for Misuse Intrusion Detection”,
Proceedings of 17th Nat’l Computer Security Conf., pp. 11-21, Oct. 1994.
S. T. Sarasamma, Q. A. Zhu, and J. Huff, “Hierarchical Kohonen Net for Anomaly Detection
in Network Security”, IEEE Trans. Systems, Man, and Cybernetics-Part B, vol. 35, no.
2, Apr. 2005.
Shekhar R. Gaddam, Vir V. Phoha, Kiran S. Balagani, “K-Means+ID3: A Novel Method for
Supervised Anomaly Detection by Cascading K-Means Clustering and ID3
Decision Tree Learnimg Methods”, IEEE Transactions on Knowledge and Data
Engineering, Vol. 19, No. 3, March 2007.
T. Mitchell, Machine Learning. McGraw-Hill, 1997.
Z. Zhang, J. Li, C.N. Manikopoulos, J. Jorgenson, and J. Ucles, “HIDE: A Hierarchical
Network Intrusion Detection System Using Statistical Preprocessing and Neural
Network Classification”, Proceedings of 2001 IEEE Workshop Information Assurance,
pp. 85-90, June 2001.
An Efficient Energy Aware Routing Protocol for Real Time Traffics in Wireless Sensor Networks 735
33
X
An Efficient Energy Aware Routing Protocol for

Real Time Traffics in Wireless Sensor Networks
Amir Hossein Mohajerzadeh1 and Mohammad Hossein Yaghmaee2,
1Department of computer engineering, Ferdowsi university of Mashhad,
Mashhad, Iran, [email protected]
2Department of computer engineering, Ferdowsi university of Mashhad,
Mashhad, Iran, [email protected]
1. Introduction
Wireless Sensor Networks (WSN) have recently been extensively deployed and researched.
They are composed of a high number of small and simple nodes where most of them have to
function as a router in an ad hoc manner. Because of limited energy sources in sensor
network node, routing protocols should save the energy as much as possible. Energy
consumption has a direct influence on network lifetime. From the Quality of Service (QoS)
point of view, in many applications such as real time one, it is necessary to consider
application QoS requirements. In this paper we propose an energy aware routing protocol
for real time traffics in wireless sensor networks. The proposed protocol considers both
energy and delay metrics to find an optimal path with minimum energy consumption and
minimum end to end delay. Simulation results show that the proposed protocol is successful
in low energy consumption and satisfying low end to end delay which makes it suitable for
real time applications.
In the recent years, many researches have been conducted on wireless sensor networks. A
wireless sensor network consists of sensor nodes that communicate with each other using
wireless links. Wireless sensor networks contain hundreds or thousands of sensor nodes
that can both send and forward data (Akyldiz et al., 2002), ( Tubaishat & Madria, 2003) . The
WSNs are used to monitor physical or environmental conditions, such as temperature,
sound, vibration, pressure, motion or pollutants, at different locations. Each node in WSNs
consists of different parts including: sensors, processor unit (usually a small
microcontroller), energy source (usually a battery) and communication unit. The
communication part of a sensor node uses wireless communication devices, to be able to
send and forward data using a wireless link.
During past few years, WSNs have found many different applications. Typical applications
of WSNs include monitoring, tracking, and controlling. Some of the specific applications of
WSNs are: habitat monitoring, object tracking, nuclear reactor controlling, fire detection and
traffic monitoring. Small sensor nodes could also be used for medical applications (Mann,
1997), e.g., for the surveillance of elderly people. In this application, sensor devices monitor
vital function and report them to the family doctor or directly to the ambulance in case of an
emergency like a heart attack. Some sensor nodes could also be implanted into the body in
order to detect diseases like cancer in an early. WSNs are also used in commercial and
industrial applications to monitor data that would be difficult or expensive to monitor using
wired sensors. In the field of home automation (Kidd et al, 1999) sensor nodes could be
located in every room to measure the temperature. Sensor nodes could at the same time
monitor more than only temperature. They could also detect movements within rooms and
report this information to the alarm equipment in case of absent occupants. Sensor networks
have also been used for many real time applications. Each application has unique QoS
requirements (Younis et al., 2004). For example, real time applications need low delay in
data delivery (Akkaya & Younis, 2003). Sensor network’s protocols should run appropriate
algorithm to satisfy application QoS requirements. Routing protocols also should use
appropriate algorithm to find routes with ability to satisfy application QoS requirements.
The sensor nodes in WSNs have many limited sources of energy and computing. The main
constraint of these networks is the amount of energy consumption. The lifetime of a sensor
network depends on its node’s energy. In most of sensor networks there is no way to charge
node’s battery; therefore efficient use of available energy sources is essential. With respect to
all above mentioned points, the protocols in wireless sensor network should consider
energy constraint in all network’s layers. Also routing protocols should use efficient
algorithms that consume energy optimally. Due to the inherent characteristics that
distinguish WSNs from other networks, routing in wireless sensor networks is very
challenging (Qiangfeng et al, 2004). Some of the routing challenges and design issues that
affect the routing process in WSNs are: node deployment, energy consumption without
losing accuracy, data reporting method, node/link heterogeneity, fault tolerance, scalability,
network dynamics, transmission media, connectivity, coverage, data aggregation and
quality of service.
Recently, many new routing protocols have been proposed for the routing in WSNs. These
routing mechanisms have taken into consideration the inherent features of WSNs along with
the application and architecture requirements. The task of finding and maintaining routes in
WSNs is nontrivial since energy restrictions and sudden changes in node status cause
frequent and unpredictable topological changes. To minimize energy consumption, energy
aware routing techniques have been proposed in the different literatures. They employ
some well-known routing tactics such as data aggregation, in-network processing,
clustering, different node role assignment and data-centric methods. In real time
applications, data should be delivered within a certain period of time from the moment it is
sensed, or it will be useless. Therefore, bounded latency for data delivery in real time
applications is too important. However, in many applications, conservation of energy,
which is directly related to network lifetime, is considered relatively more important than
the quality of data sent. As energy is depleted, the network may be required to reduce the
quality of results in order to reduce energy dissipation in the nodes and hence lengthen the
total network lifetime. Hence, energy-aware routing protocols are required to capture this
requirement.
In this paper an efficient energy aware routing protocol for real time traffics in wireless
sensor networks has been proposed. The proposed routing protocol is energy aware so its
main goal is to consume energy optimally. The proposed routing protocol can find the best
route which not only has the optimal energy consumption but also has the minimum end to
end delay. The routing algorithm in the proposed protocol, considers a cost function which
helps the algorithm to assign a cost to each route. This cost function could be determined
based on the application requirements. The proposed algorithm finds the best route
depends on its cost. By using a cost function, the proposed routing algorithm selects an
optimal route with possible lowest cost. The cost function is based on energy consumption
and end to end delay. The end to end delay consists of transmission delay and queuing
delay. In the proposed algorithm, there is an attempt to minimize end to end delay by
minimizing transmission delay. As transmission delay is directly related to the route length,
the minimum transmission delay can be achieved by minimizing route length between
source and sink nodes. The proposed routing protocol uses a neighbor discovery algorithm
to find its neighbors uniquely. As most of routing algorithms need to send data to a specific
neighbor, neighbor discovery is very important. The proposed neighbor discovery
algorithm uses three input parameters includes: node identifier (ID), received signal
strength and a random number. Simulation results show that it can discover the neighbor
uniquely.
The reminder of this paper is organized as follow. The second section, discusses the related
researches in this field. Section 3 describes the proposed energy aware routing protocol in
details. In section 4 the proposed neighbor discovery mechanism is explained. Section 5 is
dedicated to the simulation results. Finally section 6 concludes the paper.
2. Related Works
past few years, many routing protocols have been developed for wireless sensor networks.
As the energy is an important constraint of the WSNs, so the energy aware routing
algorithms are too important. In the rest of this section we review some of the general
routing protocols proposed for wireless sensor networks. Directed Diffusion
(Intanagonwiwat et al, 2000) is a well known routing algorithm for wireless sensor
networks. This algorithm is not complicated and directly diffuses the data related to sensor
nodes. This procedure guarantees high data delivery rate and low delay for
communications. Directed Diffusion consumes more energy to forward and receive
redundant data. Sink sends interests to each network nodes and determines their job. When
a node senses an event, it sends appropriate event related information to sink. SPEED (He et
al, 2003) is a well known algorithm for transmitting real time traffics in wireless sensor
networks. It considers energy consumption in its routing procedure. SPEED is a highly
efficient and scalable protocol for sensor networks where the resources of each node are
scarce. SPEED can be used in both data link and network layers. It is a flat routing
algorithm. By guaranteeing data forwarding rate, it can support real time communications.
It acts locally and uses neighbor information for routing. SPEED uses interesting mechanism
layer for route maintenance and recovery that uses both data link and network layer.
Reactive Energy Decision Routing Protocol (REDRP) (Wang et al, 2007) is another routing
algorithm for WSNs that its main goal is optimal energy consumption. This algorithm
attempts to distribute traffic in the entire network fairly. Using this mechanism, it decreases
total network energy consumption. REDRP is routing reactively, and uses residual node
energy in routing procedure. It uses local information to routing, but nodes have a global ID
which is unique for the entire network. This algorithm is divided into 4 steps. In the first
step, sink sends a control packet to all network nodes. The nodes estimate their distance to
sink relatively by using this packet. Next step is route discovery. Routing is performed on
demand in REDRP. This means that the routes are established reactively. After route
establishment in route discovery step, data is forwarded to sink by using those routes. In
route recovery step if a route is damaged, it will be recovered or a new route will be
established. Real time Power Aware framework (RPTAW) (Toscano et al, 2007) considers
both energy and QoS metrics. This algorithm acts hierarchically. By changing cluster
structure and creating new node which is called Relay Node that its job is forwarding
information from cluster to sink, its goals are achieved. This algorithm claims that by using
data aggregation functions, energy consumption is reduced. Furthermore it can manage
quality of service depending on efficiency of routing protocol used. In (Vidhyapriya &
Vanathi, 2007), (Huifang et al, 2006), (Hassanein & Lou, 2006) and (Shin, 2007), different
reliable energy aware routing protocols for wireless sensor networks have been presented.
3. Proposed Protocol
In this section, we describe the proposed energy aware routing protocol in details. The
proposed protocol uses a flat routing algorithm (Qiangfeng et al, 2004) which is done
proactively. This means that the routes are established before traffic transmission. The
algorithm is run to find the least cost route between source and sink nodes. The proposed
routing algorithm is divided into 3 phases which are: route discovery phase, data
transmission phase and route recovery phase. The last phase is only done when the
topology has been changed. Each node has a unique identifier (ID) which is determined in
the route discovery phase. The nodes also have a routing table which includes 3 fields: ID,
signal strength and route cost. There is a record for each neighbor of a node in its routing
table. The routing table is created in route discovery phase. This table is used in data
transmission phase to send traffic from source to sink. In the following subsections, we
describe the functions of each phase in details.
3.1. Route Discovery Phase

The sink node as the initiator of this phase broadcasts a packet to all its neighbors. This
packet is called Route Discover packet. The structure of Route Discover packet is shown in
figure 1. As shown in this figure, each Route Discover packet consists of three fields which
are: message type, sender ID and best route cost. The message type field determines the type
of packet. The sender ID field determines the value of sender’s ID. The best route cost field
determines the cost of optimal route between sender node and sink.
Message type Sender ID Best route cost

Fig. 1. Structure of Route Discover packet
Usually the value of sender ID field in all Route Discover packets which are sent by the sink
node is equal to zero. As the cost of optimal route between sink node and itself is always
zero, so the value of best route cost field is equal to zero.
After receiving the Route Discover packet, each node follows these steps:
1. The node increments the value of sender ID field in the received packet and
compares the result with its ID. If the result is bigger than the node’s ID, the
received packet is dropped. Otherwise the node’s ID is replaced by the value of the
result. When the node doesn’t have any ID, the node’s ID is equal to: sender ID +1.
If packet is accepted the steps are continued as follows:
2. The node creates a new record for new received packet in its routing table. The ID
field of the routing table is set to the value of sender ID field of the received packet.
3. The forwarding cost of packet which is sent directly from node i to j is calculated
using the following cost function (1):
Cost ij   .F (dist ij )   .G (energy j ) (1)
In function (1), F and G are two functions which their inputs are equal to the distance
between nodes i and j (distij) and the residual energy of node j (energyj), respectively.
Furthermore, α and β are two constant coefficients. By adding the value of best route cost
field exists in the received packet with the value of Costij, the node could be able to obtain
the new value of route cost. Each node to transmit its data toward sink, selects the optimal
route which has least cost. If the new discovered route has a lower cost than the existing
least cost route, the node replace the new discovered route as its best route to the sink. In
this case, the sender of packet is chosen as the next hop node. As the routing strategy is hop
by hop, so each node only stores information about its next hops.
4. To determine the distance between sender and receiver, the signal strength of
received packet is measured. This value is stored in the signal strength field of the
routing table. Using the signal strength, the distance between two nodes could be
determined. Furthermore, using the distance, the transmission delay between two
nodes is obtained. The end to end delay is determined using the transmission
delay.
5. If in steps 2, 3 and 4 any changes occur in the values of node properties, the node
should send a Route Discover packet to its neighbors containing the new value of
the parameters.
In the proposed algorithm, each node receives Route-Discover packet from all its neighbors.
It selects the lowest neighbor’s ID as its ID. When all nodes send the Route-Discover
packets, the value of best cost route field in their routing table is set to the value of least cost
route. At the end of route discovery phase, each node knows the cost of sending data from
itself to the sink node.
To make Route Discovery phase more clear an example is given in the following. Consider a
wireless sensor network with a random topology. Suppose that there is a unique sink node
in the network. The sink node as the initiator of this phase sends Route Discover packet to
all its neighbors. The sink ID field in these packets is set to zero; therefore as describe above,
the value of ID for all neighbors will be equal to 1. In figure 2, the node ID of all 3 sink’s
neighbors will be set to 1. For each neighbor, the cost of route between it and sink node
depends on its distance to sink.
Fig. 2. Sink as the initiator node of route discovery phase, sends first packets
As shown in figure 3.a, suppose that the upper neighbor of sink receives the Route Discover
packet. The node obtains its ID from received packet ID and then sends the Route Discover
packet to its neighbors. As the next hop for this node is the sink node, so the value of route
cost field in the Route Discover packet is equal to the cost between it and the sink node. As
the Route Discover packets are broadcast to all neighbors, so the sink node will also receive
this packet from its upper neighbor, but as the sender ID of the received packet is bigger
than the sink’s ID, the sink node doesn’t process the received Route Discover packet. Now
consider the Square node shown in figure 3.b. When it broadcasts Route Discover packet to
its neighbors, the Diamond node will receive this packet.
Fig. 3.a. The upper node sends Route Discover packet to its neighbors
In figure 3.b the diamond node has already received the Route Discover packet from the
Triangle node. So its current ID is equal to 2. For the Diamond node, the next hop node in
the least cost route toward sink is the Triangle node. So when Diamond node receives the
Route Discovery packet from the Square node, its ID doesn’t change. But the least cost route
between Diamond node and sink node may be changed. The cost of route between Diamond
node and sink node is equal to the sum of cost between Diamond node to Square node and
Square node to sink node.
Fig. 3.b. The Square node sends Route Discover packet to its neighbors
If cost of route from Square node is less than that of Triangle node, the least cost route and
next hop node of Diamond node will be changed. Note that in the Route Discovery phase
the cost of routes is propagated between all nodes using Route Discover packets. This
procedure is continued while all the nodes obtain their least cost route.
3.2. Data Transmission Phase

When a node detected an event, it should send data related to that event to the sink. As
mentioned before, the routes are established in the route discovery phase. All nodes know
their least cost route to the sink. So, using the optimal path the node will be able to send its
data to the sink. Each node knows its next hop node in its least cost route. When a node
detected an event or received any data, it sends them to the sink node via its next hop node.
3.3. Route Recovery Phase

This phase is executed periodically. The length of time periods depends on the node’s
mobility. If a node dies, it will never participate in the routing procedure in the next period.
Therefore, the dead nodes are not belonging to any established route. If the next hop node is
failed, the data are sent using a backup node. All nodes in the network know the cost of
forwarding information through their neighbors. When the least cost route is failed then the
node forwards data using the second least cost route. As the information about all the
possible routes from a node to sink is stored in the node routing table, so it is easy to find
the first and second least cost routes. When the reminding energy of a node is less than a
predetermined threshold, it will inform this situation to all its neighbors. If a node realizes
that its next hope node doesn’t have any sufficient energy, it uses its second least cost route
to send its data
4. Proposed Neighbour Discovery Phase

In this section, we explain the operation of proposed neighbor discovery phase. Most of
routing algorithms need to send data to a specific neighbor. In essence, wireless links are
broadcast links which means, when a node sends a packet, all the nodes placed in its
communication range will receive it. In this situation, every node needs a mechanism that
makes it enable to send data to a particular neighbor so that the other neighbors wouldn’t
process those data.
All energy aware routing protocols need neighbor discovery mechanism. Proposed
approach uses a hop by hop routing algorithm; route to the sink is selected by each node via
choosing next hop, meanwhile different routes are picked out by considering different next
hop nodes. Most of routing algorithms use hop by hop strategy which is more efficient. All
nodes which use hop by hop routing algorithm need information only about their next hop,
which means they just need local information. When an algorithm needs to have a global
view of the entire network, it absolutely must pay much more in contrast with the situation
with only local view. Neighbor discovery algorithms collect local information about node’s
neighbors. To distinguish nodes from each other, we can assign a unique identifier to each
node. This identifier makes enable the other nodes to select one node uniquely. By
considering this deployment, all the node neighbors will receive the data, but only one node
that is identified by the packet destination identifier field will process data. The node’s
identifier could be local or global. When a node’s identifier is global, the node could be
identified by the other nodes uniquely. But as we mentioned before, this type of identifying
is too expensive. When a node uses local identifier, it can only distinguish its neighbors. As
the proposed algorithm needs network’s nodes to distinguish their neighbors uniquely, so it
doesn’t need global identifier and the local identifier is sufficient. In the following, we
propose a new neighbor discovery mechanism for distinguishing node’s neighbors.
In the proposed neighbor discovery mechanism, each node estimates its distance to the
sender using received signal strength. This parameter can be used for distinguishing node’s
neighbors. In the route discovery phase many packets are transmitted between nodes. Using
the signal strength of these packets, the receiver can estimate its distance to the sender node.
Therefore at the end of route discovery phase all nodes know their distance to their
neighbors. As discussed in section 3, in the route discovery phase an ID is assigned to each
node. This ID is not unique in the entire network. The nodes with equal ID have the same
number of hops to the sink. The proposed neighbor discovery algorithm uses both node ID
and received signal strength to distinguish neighbors with a suitable accuracy rate. We
believe that by using the distance between two nodes and the node ID, we can distinguish
the neighbors with a high accuracy. If by using these two parameters, the node couldn’t
distinguish all its neighbors, this means that more than some of its neighbors have the same
distance and ID. In this case, the proposed mechanism uses a random number to discern
them.
When a node detects a collision, this means that it has more than one neighbor with the
same distance and ID. It sends a Collision Recovery packet to the neighbors. In this packet
the sending node advertises that only the nodes which detected any collision should process
it and the other neighbors should ignore it. When the nodes which detected collision receive
this packet, they create a random number between 0 and MAX (usually MAX is a big
number, e.g., 100000) and send it for the node that has sent the Collision Recovery packet,
using Collision Recovery Reply packet. Both sender and receiver, store this random number
in their routing table in an appropriate record. This random number makes distinguishing
action complete. By using distance (signal strength), ID and if needed the random number, it
is possible to distinguish neighbors from each other. When a node wants to send a packet to
one of its neighbors, it should use all of 3 mentioned parameters in the packet. All neighbors
receive the packet, but only the neighbor which can find a match and has the same
properties will process the packet and the other nodes will ignore it. To evaluate the
performance of the proposed neighbor discovery phase, we implemented it in a simulator.
Table 1, shows the simulations results. As mentioned before, the collision is only occurred
when a node has more than one neighbor with the same distance and ID.
Simulation Area Number of Communication Number of

trials (square) nodes range (m) detected collision
1 50*50 200 5 0
2 100*100 200 5 0
3 100*100 500 5 2
4 500*500 1000 20 23
5 10*10 100 1 12
Table 1. Number of collision occurred in different experiments.
The results shown in table 1, confirm that by increasing the density of nodes in the network,
the probability of collision is also increased. We should emphasis here that, by using the
third parameter (random number) as explained earlier, the distinguishing rate may be reach
to 100%. When the mechanism uses the first two parameters (node ID and signal strength)
the overhead is always zero, but when the third parameters is applied, only 2 packets
should be transmitted in the network, that can be disregarded relative to number of packets
transmitted in other phases.
5. Simulation Results
In this section, using computer simulation, we evaluate the performance of the proposed
energy aware routing protocol with that of SPEED protocol. Before evaluating the
performance, we describe the environment of our simulation. After that we analyze the
simulation results and compare the performance of the proposed protocol with that of
SPEED protocol.
5.1. Simulation Environment

We developed a simulation software using C++ language. To compare the performance of
both protocols, we implemented the proposed protocol as well as the SPEED protocol in our
simulation software. The simulated network topology consists of 100 fixed sensor nodes
which are randomly deployed in a 200m  200m area. Each node is able to send data in a
range of 40m. There is one sink node at point (0, 0). The location of sink node can be
changed in many scenarios. We consider many different scenarios to evaluate different
aspect of the proposed algorithm.
5.2. Results Analyze

In figure 4, for both proposed algorithm and SPEED algorithm, the average energy
consumption of all nodes is plotted versus number of events. Less energy consumption
means longer lifetime for the network. Horizontal axis shows the number of events which
are occurred in the network terrain. Events are occurred in a random place in the network.
Vertical axis shows the average energy consumption of the network nodes. The scale of
vertical axis is 0.00005 J. Based on results shown in figure 4, it is obviously observable that
the average energy consumption of the network nodes in the proposed protocol is less than
that of SPEED protocol. When the number of events is less than 200, the energy
consumption of two protocols is nearly equal. But when number of events is more than 200,
the proposed protocol consumes less energy than SPEED. It is necessary to note here that
the main goal of proposed algorithm is to decrease energy consumption.
Fig. 4. The comparison of average energy consumption between two protocols, number of
nodes =100.
In the next trial, we increased the number of nodes to 200. The results are shown in figure 5.
As shown in this figure, when the number of events is less than 500, the average energy
consumption of the SPEED protocol is less than the proposed protocol. Note that the
proposed protocol uses a proactive routing algorithm; it means that the routes are
established in advance before data transmission. So, when the number of events is low, the
average energy consumption of the proposed protocol is more than SPEED.
Fig. 5. The comparison of average energy consumption between two protocols, number of
nodes =200.
In figure 6, for a network with different amount of nodes, the energy consumption of two
protocols is shown. Horizontal axis shows the amount of network nodes and the vertical
axis shows total network energy consumption.
Fig. 6. The total network energy consumption for different amount of network nodes.
Total network energy consumption is calculated as the sum of energy consumption in all
network nodes. It could be seen in figure 6 that for different number of nodes, the total
energy consumption of the proposed protocol is less than SPEED. Furthermore, it is

obviously observable that by increasing the number of nodes, the performance of proposed
protocol is also increased. In figure 7, for both two protocols, the number of dead nodes is
plotted versus the number of events. If the energy of a node is finished, it will be dead.
When a wireless sensor network has high number of alive nodes, it will live longer. Note
that a wireless sensor network with higher number of nodes can perform its functions
better. Figure 7 shows that the number of dead nodes in the proposed protocol is always less
than SPEED.
Fig. 7. Comparing the number of dead nodes between two protocols
Based on results shown in figures 4-7, it is clear that the proposed protocol has better
performance in comparison with the traditional SPEED protocol. Simulation results show
that the average energy consumption of the proposed protocol is lower than that of SPEED
protocol. In the next simulation trials, we evaluate the delay performance of the proposed
protocol. As we mentioned earlier, by decreasing the transmission delay, it is possible to
decrease the end to end delay. Figure 8 shows the simulation results related to delay
performance of both protocols. Results show that the path traversed by packet using
proposed protocol has less delay than that of SPEED protocol. Horizontal axis illustrates the
place where the event has occurred. Note that events occur in points with equal width and
length. For example, number 200 in horizontal axis means that the event has occurred at
point (200,200). The vertical axis shows the path delay which is related to the length of the
route that a packet traverses between source node to the sink. As the queuing delay is
negligible, we ignore it. Results shown in figure 8 clear that the end to end delay of the
proposed protocol is less than that of SPEED protocol. Figure 9 shows the end to end delay,
in the case that the number of nodes in network has been increased to 200 nodes. Based on
results shown in figures 8,9, it is clear that the proposed protocol has lower delay than the
SPEED protocol so it is more suitable for real time applications. As discussed in section 1,
two main objectives of the proposed protocol are to minimize energy consumption and to
choose a route with minimum end to end delay. The simulation results show that the
proposed protocol has achieved to both of its goals.
Fig. 8. End to end delay between source and sink, number of nodes=100
Fig. 9. End to end delay between source and sink, number of nodes=200
5.3. Real Time Traffics

To make the proposed routing protocol more scalable and more suitable for real time
traffics, we used the clustering techniques. In this case, the network is divided to some
clusters. For this purpose, sensor nodes are grouped into clusters by using one of the
clustering techniques (Abbasi & Younis, 2007). Sensor nodes are only responsible for
probing the environment to detect an event. Every cluster has a cluster head that manages
the other members in the cluster. Clusters can be formed based on many criteria such as
communication range, number and type of sensors and geographical location. We assume
that all nodes are stationary and the cluster head is located within the communication range
of all the cluster members. The Routing in Hierarchical routing protocols is divided into two
parts. First, routing between cluster members and cluster head. And second, routing
between cluster heads and the sink. In this section we emphasis on first routing type. To
forward traffic toward sink node, each cluster head should be able to route data to other
cluster heads. The cluster head is responsible to find best route for all its members in terms
of energy consumption and the end to-end delay requirement. We consider two types of
traffics: real time and non-real time. Real time traffics have hard constraint on the value of
end to end delay while non real time traffics don’t have any specific delay constraint. Both
real-time and non-real-time traffic coexist in the network. As delay constraints are associated
only with real-time data, the cluster head is responsible to find the best path for this kind of
traffics so that the end to end delay requirement are meet for real-time traffics. Each sensor
node uses different queues for the two different types of traffic. Furthermore, each node has
a classifier, which checks the type of the incoming packet and sends it to the appropriate
queue. There is also a scheduler, which determines the order of packets to be transmitted
from the queues. The cluster heads use the proposed cost function to find the best path
which not only can meet the delay requirement but also consumes the minimum energy. We
use a modified version of Djikstra routing algorithm. The cluster head is responsible to find
the best route for all of its members. It will select the more suitable route that has enough
resources for transmitting real time traffic from nominee routes with lowest energy
consumption based on modified Djikstra algorithm.. After finding the best route, the cluster
head sends the routing information to all of its members. So, each sensor node in each
cluster knows the best path between itself and its cluster head for transmitting real time
traffic. The cluster heads use an existing routing algorithm to transmit traffic toward the
sink node. Non real time traffics are sent using Gossiping algorithm (Haas et al, 2006). In the
simulation, we set the size of each cluster to 40m*40m. Each cluster consists of some nodes.
In figures 10(a,b,c), for both SPEED and the proposed protocols and for a cluster with 20
nodes, the average energy consumption, the energy consumption per node and the end to
end delay are given, In this experiment, real time traffic is produced in constant rate but
production rate of non real time traffic is variable.. Figure 10(a) shows the average energy
consumption versus number of events occurred in the cluster. It can be seen that the
proposed routing protocol consumes less energy in comparison with the traditional SPEED
protocol. In figure 10(b), the energy consumption of each node is given. In figure 10(c), the
end to end delay of real time traffics are plotted versus different number of non real time
packets. As it can be seen, for the proposed routing protocol, the increasing in non real time
traffic density doesn’t have any serious affect in the end to end delay performance of real
time traffics. In the next simulation trial, we increased the number of sensor nodes in a
cluster to 40 nodes and produced both real time and non real time traffic with variable rate.
In figure 11(a,b,c), the results are shown. Based on results given in figures 10,11, it is clear
that the proposed routing protocol has better delay and energy consumption performance
than the existing SPEED protocol.
a)
b)
c)
Fig. 10. Performance evaluation of two protocols for real time traffics: (a) average energy
consumption, (b) energy consumption of each node and (c) end to end delay (number of
nodes in cluster = 20 nodes)
a)
b)
c)
Fig. 11. Performance evaluation of two protocols for real time traffics: (a) average energy
consumption, (b) energy consumption of each node and (c) end to end delay (number of
nodes in cluster = 40 nodes)
6. Conclusion
Energy aware routing is most challenging issue in wireless sensor networks. Current
research on routing of sensor data mostly focused on protocols that are energy aware to
maximize the lifetime of the network, scalable for large number of sensor nodes and tolerant
to sensor damage and battery exhaustion. In this paper an efficient energy aware routing
protocol was proposed. The proposed routing protocol has two major goals which are low
power consumption and low end to end delay. We evaluated the performance of the
proposed protocol under different scenarios. Simulation results confirmed that the proposed
protocol has more efficient energy consumption in comparison with the traditional SPEED
protocol. Furthermore, the proposed routing protocol can find the optimal path with a low
end to end delay. We believe that, by using data aggregation techniques the higher
performance will be achievable. Also using other cost functions for route selection makes the
proposed protocol suitable for other applications.
7. References
Abbasi, A. A. & Younis, M. (2007). “A survey on clustering algorithms for wireless sensor
networks”, Volume 30, Issues 14-15, 15, Pages 2826-2841
Akkaya, K. & Younis, M. (2003). “An Energy Aware QoS Routing Protocol for Wireless
Sensor Networks.”, ICDCS Workshop
Akyildiz, I. F.; Su, W.; Sankarasubramaniam, W. & Cayirci, E. (2003). ”A Survey On Sensor
Networks.”, IEEE Communication magazine, pp. 102-114.
Haas, Z. J.; Halpern, J. Y & Li, L. (2006). “Gossip-based ad hoc routing”. IEEE/ACM
Transaction on Networking, 14(3):479–491
Hassanein, H. & Jing L. (2006). “ Reliable Energy Aware Routing In Wireless Sensor
Networks”, Second IEEE Workshop on Dependability and Security in Sensor
Networks and Systems, 2006. DSSNS 2006., 24-28 April 2006 Page(s):54 – 64
He, T.; Stankovic, J. A.; Lu, C. & Abdelzaher, T. (2003). “SPEED: A Stateless Protocol for Real
Time Communication in Sensor Networks.”, ICDCS
Huifang C; Mineno, H.; Mizuno, T. (2006). “An Energy-Aware Routing Scheme with Node
Relay Willingness in Wireless Sensor Networks”, First International Conference
Innovative Computing, Information and Control, 2006. ICICIC '06. Volume 1, 30-01
Aug. 2006 Page(s):397 – 400
Intanagonwiwat, C.; Govindan, R. & Estrin, D. (2000). “Directed Diffusion: A scalable and
robust communication paradigm for sensor networks.”, Proceedings of the 16th
Annual ACM/IEEE International Conference Mobile Computing and Networking,
pp. 56-67
Kidd, C. D.; Orr, R.; Abowd, G. D.; Atkeson, C. G.; Essa, I. A.; MacIntyre, B.; Mynatt, E. D.;
Starner, T. & Newstetter, W. (1999). “The aware home: A living laboratory for
ubiquitous computing research.” In Cooperative Buildings, pages 191–198
Mann, S. (1997). “Wearable computing: A first step toward personal imaging.” IEEE
Computer, 30(2):25–32
Qiangfeng, Jiang, Manivannan, (2004) “Routing Protocols for Sensor Networks.”, 1st IEEE
Consumer Communications and Networking Conference, pp. 93-98
Shin, K. Y.; Song, J.; Kim, J. W.; Yu, M. & Mah, P. S. (2007). ” REAR: Reliable Energy Aware
Routing Protocol for Wireless Sensor Networks”, The 9th International Conference
on Advanced Communication Technology, Volume 1, 12-14, Page(s):525 – 530
Toscano, E.; Kaczynski, G. & Bello, L. L. (2007). “RTPAW: a Real Time Power Aware
Framework for Wireless Sensor Networks.”, WIP Proc, of the 13th IEEE Real Time
and Embedded Technology and Applications Symposium, Bellevue, USA, April
2007, pp. 60-63
Tubaishat, M. & Madria, S. (2003). “Sensor Networks: An Overview.”, IEEE POTENTIALS
April/May, pp20-23
Vidhyapriya, R. & Vanathi, P. T. (2007). “Energy Aware Routing for Wireless Sensor
Networks”, Signal Processing, Communications and Networking, 2007. ICSCN '07.
International Conference on 22-24, Page(s):545 – 550
Wang, Y. H.; Hsu, C. P.; Lin, Y. C.; Kuo, C. S. & Ho, H. Y. (2007). “A Routing Method by
Reactive Energy Decision in Wireless Sensor Networks.“, 21st IEEE International
Conference on Advanced Information Networking and Applications Workshops
(AINAW’07)
Younis, M.; Akkaya, K.; Eltoweissy, M. & Wadaa, A. (2004). “On Handling QoS Traffic in
Wireless Sensor Networks.”, Proceedings of the 37th Hawaii International
Conference on System Sciences
Quality of Service Differentiation in WiMAX Networks 753
34
X
Quality of Service Differentiation

in WiMAX Networks
Pedro Neves1, Susana Sargento2, Francisco Fontes1,
Thomas M. Bohnert3 and João Monteiro2
1 Portugal Telecom Inovação
Portugal
2 University of Aveiro / Institute of Telecommunications
Portugal
3 SAP Research CEC Zurich
Switzerland
1. Introduction
Broadband Wireless Access (BWA) based on the IEEE 802.16 standards [IEEE 802.16, 2004]
[IEEE 802.16, 2005], also known as Worldwide Interoperability for Microwave Access
(WiMAX), is gaining momentum as more and more field trials are transformed in
commercial roll-outs. Very much likely, this is certainly a merit of the combined effort of the
IEEE 802.16 standardization community, the WiMAX Forum [WiMAX Forum a, 2008]
[WiMAX Forum b, 2008] and the research community. As of today, standards have matured,
the WiMAX Forum has setup its certification program in order to foster interoperability, and
the research community went through technological details up to an extent, such that many
operators consider the risks associated with commercial deployment predictable.
Yet, WiMAX is still not deemed as globally established. This technology is very new and a
competing system, namely Long-Term Evolution (LTE) [3GPP, 2009], is progressing in a
similar pace. Which technology will finally make it through in the market, or will we have
even several co-existing ones, is, among others, influenced by the research community. The
better a technology is understood, the more likely it will be adopted by manufacturers,
vendors and operators.
Key to scientific consideration is the availability of reliable simulation tools. This applies
equally to WiMAX and any other technology and is embodied by an emerging business of
commercial simulators like OPNET [OPNET] and Qualcomm [Qualcomm]. Unfortunately,
these simulators are fairly expensive and/or their use is strictly licensed. As monetary
considerations and openness frequently prevail, in particular in academia, NS-2 [NS-2], as
open-source alternative, still retains its position.
Naturally, open-source software rarely approaches commercial standards, especially in
terms of completeness and documentation. This is, for example, the case for the public
available WiMAX module for NS-2. This module supports several WiMAX features but
lacks a very essential one, Quality of Service (QoS) support. The lack of this feature
motivates the work presented in this chapter. This chapter describes a novel QoS framework
for WiMAX with efficient service differentiation. The QoS framework is composed by a
packet classification mechanism, as well as by a novel cross-layer scheduling algorithm
based on user’s prioritization and radio resources optimization [Monteiro, 2009].
Furthermore, in order to validate and evaluate the designed solution, a set of QoS oriented
scenarios have been simulated in Network Simulator (NS-2) [NS2-NIST], demonstrating that
the designed model is able to efficiently differentiate users in a competitive environment,
differentiating between the traffic classes defined for WiMAX, mainly in throughput and
delay metrics.
The reminder of this chapter is organized as follows. Section 2 briefly overviews WiMAX,
focusing on the Medium Access Control (MAC) layer functionalities. Section 3 provides an
overview about the WiMAX NS-2 module, focusing on its main features and limitations.
Section 4 describes the proposed QoS model, including the packet classification mechanism
and the scheduler, whereas Section 5 discusses the obtained simulation results for several
scenarios. Finally, Section 6 concludes the chapter.
2. WiMAX High Level Description

Ubiquitous broadband Internet access is an important requirement to satisfy user demands
and support a new set of real time services and applications. WiMAX, a Broadband Wireless
Access (BWA) solution for Wireless Metropolitan Area Networks (WMAN), covering large
distances with high throughputs, is a promising technology for Next Generation Networks.
WiMAX supports both fixed and mobile users, based on IEEE 802.16-2004 [802.16, 2004] and
IEEE 802.16e-2005 [802.16, 2005] standards, respectively. IEEE 802.16 system is connection
oriented and provides Quality of Service (QoS) assurances through service flows and
scheduling services. Therefore, all tasks are based on a connection, uniquely identified by a
16-bit Connection Identifier (CID), and no packets are allowed to traverse the wireless link
without a specific connection allocated. A connection is, by definition, a unidirectional
mapping between the WiMAX Base Station (BS) and the WiMAX Subscriber/Mobile Station
(SS/MS) for transporting a service flow’s traffic. Succinctly, scheduling services specify the
policy used by the WiMAX BS and SS/MSs to manage the poll and grant procedures. Five
scheduling services are defined to meet the QoS needs of the data flows carried over the air
link:
 Unsolicited Grant Service (UGS): designed for real-time service flows that generate
fixed size data packets on a periodic basis, such as VoIP. The service offers fixed
size unsolicited data grants (transmission opportunities) on a periodic basis. This
eliminates the latency and overhead of requiring the SS/MS to send requests for
transmission opportunities, and assures that grants are available to meet the flow’s
real-time needs;
 real-time Polling Service (rtPS): designed for real-time service flows that generate
variable size data packets on a periodic basis, such as video streaming. The service
offers real-time, periodic, unicast request opportunities, which meet the flows real-
time needs and allow the SS/MS to specify the size of the desired grant. In this
case, the SS/MS is not allowed to use any contention request opportunities;
 extended real-time Polling Service (ertPS): designed for real-time services that
generate variable size data packets on a periodic basis, such as VoIP with silence
suppression. This scheduling mechanism is based on UGS and rtPS. Unicast grants
are provided to the MSs in an unsolicited manner, like in UGS, and therefore the
latency of a bandwidth request message is saved. Instead of providing fixed
allocations such as UGS, ertPS provides dynamics allocations;
 non real-time Polling Service (nrtPS): designed for non-real-time service flows that
require variable size data grants on a regular (but not strictly periodic) basis, such
as high bandwidth File Transfer Protocol (FTP). The service offers unicast polls on
a periodic basis but uses more space intervals then rtPS. This ensures that the flow
receives request opportunities even during network congestion;
 Best Effort (BE): designed for traffic where no throughput or delay guarantees are
provided. The SS/MS sends requests for bandwidth in either random access slots
or dedicated transmission opportunities. The occurrence of dedicated opportunities
is subject to network load, and in contrast to the nrtPS, the SS/MS cannot rely on
their presence.
A set of convergence sublayers are defined to map the upper layer packets into the 802.16-
2004 system. The convergence sublayers support packet based protocols, such as Internet
Protocol version 4 (IPv4) and Internet Protocol version 6 (IPv6), as well as cell based
protocols, such as Asynchronous Transfer Mode (ATM). Both point-to-multipoint (PMP)
and mesh modes of operation are supported by the standard, despite the mesh mode of
operation is optional. During the WiMAX SS/MS network entry process, three pairs of
management connections, with distinct QoS levels and hence reflecting three different QoS
requirements, are established:
 Basic connection to transfer short, time-critical MAC management messages;
 Primary management connection to transfer longer, more delay tolerant
management messages;
 Secondary management connection to transfer delay tolerant, standard-based
management messages such as Dynamic Host Configuration Protocol (DHCP),
Trivial File Transfer Protocol (TFTP) and Simple Network Management Protocol
(SNMP).
Besides the aforementioned three pairs of management connections, a broadcast connection

is configured by default and is used to transmit MAC management messages to all the
SSs/MSs. Moreover, a multicast polling connection is used by the SSs/MSs to join multicast
polling groups, allowing them to request bandwidth via polling. Finally, to satisfy the
contracted services, transport connections are allocated for data packets.
With respect to the IEEE 802.16 protocol stack, it defines the Physical (PHY) and the
Medium Access Control (MAC) layers. Internally, the MAC layer is divided in three
sublayers: Service Specific Convergence Sublayer (CS), Common Part Sublayer (CPS) and
Privacy Sublayer (PS). Figure 1 illustrates the IEEE 802.16 protocol stack, focusing on the
MAC CS classification mechanism.
Raw Data Packets (IPv4, IPv6)

CS SAP
CID 1
Service Specific Convergence
Sublayer
Scheduler Input
(MAC CPS)
CID #2
MAC SAP
802.16
CID #3 MAC
Classifier
…. MAC Common Part Sublayer Layer
CID #n
Connections
Security Sublayer
PHY SAP
PHY
Physical Layer Layer
IEEE 802.16 Protocol Stack

Fig. 1. IEEE 802.16 Protocol Stack
The interface between the 802.16 system and the upper layers from the protocol stack is
provided by the CS, receiving the higher layer MAC Service Data Units (SDU) coming
through the CS Service Access Point (SAP) and classifying them to the appropriate
connection. The 802.16 classifier is a set of packet matching criteria applied to each packet. It
consists of some protocol-specific fields, such as IP and MAC addresses, a classifier priority
and a reference to a particular CID. Each connection has a specific service flow associated
providing the necessary QoS requirements for that packet. If no classifier is found for a
specific packet, a specific action must be taken. Since the classifier implementation is vendor
dependent, the chosen decision depends on the algorithm implemented by the vendor – the
packet can be discarded, sent on a default connection, or a new connection can be
established for it, if enough resources are available. Downlink classifiers are applied by the
WiMAX BS and uplink classifiers are applied by the WiMAX SS/MS. Two main types of
CSs are defined within the standard for mapping services to and from the 802.16 system
connections: the Packet Service Specific Convergence Sublayer (Packet CS) and the
Assynchronous Transfer Mode Service Specific Convergence Sublayer (ATM CS). The
Packet CS sublayer is defined for packet-based protocols, whereas the ATM CS is defined to
support cell-based protocols.
The CPS is the second sublayer from the MAC layer. It receives classified packets arriving
from the CS and is responsible for a set of functions, such as addressing, construction and
transmission of the MAC PDUs, scheduling, bandwidth allocation, request mechanisms,
contention resolution, among others. Finally, the PS is the third and last sublayer from the
MAC layer and provides authentication, data encryption and security mechanisms.
Since the IEEE 802.16 standards are focused on defining the PHY and MAC layers for the air
interface, the WiMAX Forum, in particular the Network Working Group (NWG), is
specifying an “All-IP” end-to-end network architecture for IEEE 802.16 [WiMAX Forum a,
2008] [WiMAX Forum b, 2008]. The WiMAX Forum extends the IEEE 802.16 architecture by
defining the Network Reference Model (NRM), which is a logical representation of the
WiMAX network architecture, based on a set of functional entities and standardized
interfaces, known as Reference Points (RPs) – R1 to R8. Using this model, multiple
implementation options for a given functional entity are allowed, maintaining
interoperability across them through the RPs. Three functional entities are defined:
Connectivity Service Network (CSN), Access Service Network (ASN) and the Subscriber
Station (SS)/Mobile Station (SS/MS). The WiMAX NRM is presented in Figure 2.
Fig. 2. WiMAX Network Reference Model
The SS and the MS are responsible for establishing radio connectivity with the WiMAX BS,
for the fixed and mobile standards, respectively. The ASN is generally composed by several
WiMAX BSs connected to one or more ASN-Gateways (ASN-GW); it establishes
connectivity with the CSN. The ASN includes a set of functionalities in order to provide
radio connectivity to WiMAX subscribers. Additionally, it also performs relay functions to
the CSN in order to establish IP connectivity and authentication mechanisms. Finally, the
CSN contains the DHCP, DNS, AAA (Authentication, Authorization, and Accounting) and
MIP servers. Moreover, the CSN is responsible for establishing connectivity with the IP
backbone.
After briefly describing WiMAX, the next section aims to depict the functionalities provided
by default by the WiMAX QoS model implemented by the National Institute of Standards
and Technology (NIST) [NS2-NIST] for the Network Simulator (NS2) [NS2].
3. NIST WiMAX QoS Model

The NS2 IEEE 802.16/WiMAX module [NS2-NIST] was developed by the NIST Seamless and
Secure Mobility Group and henceforth the principal focus is on IEEE 802.16e/Mobile WiMAX.
Nevertheless, the overall architecture is set on top of a basic subset of IEEE 802.16-2004
[IEEE 802.16, 2004] and IEEE 802.16e-2005 [IEEE 802.16, 2005] common functionalities.
Out of the four specified physical (PHY) layers in the combined standard documents, a
multi-carrier air interface using Orthogonal Frequency Division Multiplexing (OFDM) with
256 carriers was adopted, also known as WirelessMAN-OFDM, and Time Division
Duplexing (TDD) was chosen as the duplexing technique. The 2 – 11 GHz licensed band
provides lower transmission rates (75 Mbit/s) compared to the 10 – 66 GHz band, but it
supports both Line-Of-Sight (LOS) and Non-Line-Of-Sight (NLOS) environments. Different
modulations can be configured statically, such as Binary Phase Shift Keying (BPSK),
Quadrature Phase Shift Keying (QPSK), 16-state Quadrature Amplitude Modulation (QAM)
and 64-QAM, allowing the formation of varying robustness and efficient burst profiles.
However, information coding is yet missing and hence the module does not support any
Adaptive Modulation and Coding (AMC) scheme.
The TDD duplexing technique is illustrated in Figure 3, presenting both downlink and
uplink subframes decomposition. The downlink subframe is composed by a Preamble used
for synchronization, followed by the Frame Control Header (FCH). The FCH contains the
Downlink MAP (DL-MAP) and Uplink MAP (UL-MAP) MAC management messages,
indicating the location and burst profile of each downlink and uplink burst, respectively.
Moreover, it contains the downlink and uplink channel descriptors (UCD and DCD
management messages). Following the FCH, starts the downlink data bursts section.
Downlink bursts are transmitted in order of decreasing robustness – QPSK followed by 16-
QAM and finally 64-QAM. The WiMAX SSs/MSs listen to all the bursts that they are
capable to decode, specifically bursts with profiles of equal or greater robustness compared
with the one negotiated with the WiMAX BS during the connection setup phase. Thereafer
the SS analyses the MAC header of each MAC Protocol Data Unit (PDU) inside each burst to
check if the Connection Identifier (CID) belongs to it. At the end of the frame, the Transmit
Transition Gap (TTG) is used to separate the downlink and the following uplink bursts. The
TTG allows the WiMAX SS to switch from receive to transmit mode.
In the beginning of the uplink subframe there are two contention slots. The first contention
slot is used by the WiMAX SSs/MSs for initial ranging (Initial Ranging), whereas the second
contention slot is used by the WiMAX SSs/MSs to send bandwidth request PDUs to the BS
(Request Contention). The remaining transmission slots are grouped by SSs/MSs. Each
SS/MS has a specific slot allocated for uplink data transmission. The Subscriber Station
Time Gap (SSTG) is a time interval used to separate the transmissions of the various
SSs/MSs during the uplink subframe. Finally, the Receive Transition Gap (RTG) is used to
separate the uplink and downlink bursts. The RTG allows the WiMAX SS/MS to switch
from transmit to receive mode.
Frame n-1 Frame n Frame n+1 Frame n+2
Downlink Subframe Uplink Subframe
SSTG
TTG RTG
Broadcast
Preamble
TDM TDM TDM TDM Initial Request

Control Ranging Contention SS1 Data SS N Data
DIUC a DIUC b DIUC c DIUC d …...
DIUC=0 UIUC=2 UIUC=1 UIUC=i UIUC=j
MAC PDU 1 MAC PDU 2 … MAC PDU n
DL-MAP UL-MAP MAC PDU 1 MAC PDU 2 … MAC PDU n
MAC Header MAC Payload CRC
MAC Header MAC Payload CRC

CID
CID
Fig. 3. WiMAX Frame Structure (Downlink and Uplink subframes)
The OFDM specific countermeasure to Inter Symbol Interference (ISI), the cyclic prefix, can
also be configured, i.e. its length. Given these parameters, the module is able to compute the
OFDM symbol duration, packet transmission time per modulation, maximum packet size
per modulation and the number of OFDM symbols. As the implemented architecture is an
extension of the NS-2 wireless networking sub-module, the standard NS-2 channel models
and transmission power levels can be set accordingly to the NS-2 standard tools [NS2]. As
for the PHY layer, the MAC layer supports only a subset of the IEEE 802.16 standard. For
example, currently only the Packet Convergence Sublayer (Packet CS), as detailed in Section
2, is implemented. Although the module can be easily extended, the Packet CS is essentially
a classifier, supporting the IP destination address as the classifier parameter. The connection
oriented nature of IEEE 802.16 between MAC instances has also been implemented in the
NIST model. As the IEEE 802.16 standard defines, each pair of WiMAX BS and SS
establishes three management connections, Basic, Primary and Secondary. One of the
drawbacks of the NIST implementation is that each MAC instance only supports one data
transport connection. Additionally, out of Fragmentation, Packing and Automatic Repeat
Request (ARQ), only the first, Fragmentation, is currently supported. With respect to
mobility, channel scanning, communication parameters negotiation, initial ranging and
registration, the provided implementation adheres largely to IEEE 802.16-2004 and IEEE
802.16e-2005 standards. Periodic ranging used to adjust coding and modulation is left out
for now. Finally, the most crucial missing feature, which motivates this work, is the lack of a
complete 802.16 compliant QoS model. The software has been prepared for future QoS
integration but it was not implemented. Despite scheduling services, service flows and a
basic bandwidth request mechanism for BE (Best Effort) traffic is available, the current
scheduler implements a simple Round-Robin discipline for the scheduler.
4. Proposed QoS Model

4.1 Packet Classification Mechanism
As defined in the 802.16 standard, packets received at the MAC layer, specifically at the
Convergence Sublayer (CS), must be mapped to the correspondent Connection Identifier
(CID), based on a set of packet matching criteria. In order to handle the incoming packets
and the new QoS classes, we have modified the NIST CS module. Besides the existing
connections (Basic, Primary, Secondary and Data connections), additional connections for
UGS (Unsolicited Grant Service), rtPS (real time Polling Service), ertPS (extended real time
Polling Service ), nrtPS (non real time Polling Service) and BE (Best Effort) service classes
have been established. The creation of these new connections required the addition of new
CID ranges, providing each peer node a unique CID for these types of traffic.
To allow the creation of new connections between the WiMAX BS and the SSs/MSs, the
PeerNode class in NIST was changed to have new members of type Connection, for receiving
and sending packets of different traffic classes. These modifications were performed in the
existing DestClassifier class. This new classifier is a subclass of the Service Data Unit
Classifier (SDUClassifier) class, as shown in Figure 4. To improve the packet classification
mechanism, the QoSClassifier class was implemented, also as a subclass of the SDUClassifier
class.
SDUClassifier
+classify(in p : Packet) : int
DestClassifier QoSClassifier
+classify(in p : Packet) : int +classify(in p : Packet) : int
Fig. 4. Classifier Class Diagram
The most important method to implement the classification is the classify() method, called
for all packets, which finds the appropriate PeerNode based on the destination address and
QoS requirements. Thereafter, based on the packet type, the packet is sent to the appropriate
connection queue on the scheduler, as illustrated in Figure 5. For example, if a broadcast
packet is received on the BS classifier, a Broadcast CID is given; if the same packet is
received on the SS/MS classifier, it will be classified with the Secondary Management CID.
On the other hand, if a data packet arrives at the BS/SS classifier, it will be given a new
transport CID and queue allocation correspondent to its traffic type.
Fig. 5. Packet Classifier Diagram
Furthermore, new service classes were introduced in the model, as previously referred –
UGS, ertPS, rtPS, nrtPS and BE. For each one of these service classes, a range of transport
CIDs for the data connections was given. Apart from this association, modifications
throughout the different functions that make use of the service classes were made. For
instance, in the Connection function used by the WiMAX BS to initialize a new connection
and assign the correspondent CID, the support for new connection types was added. In this
case, a new type of connection is distinguished according to its type and respective CID.
4.2 Enhanced Scheduling Algorithm

The implemented BS scheduler enhances the simple Round Robin (RR) algorithm used in
NS2-NIST/WiMAX module by adding a priority scheme – priority Round Robin (PRR).
Instead of equally distributing the available bandwidth between the registered SSs/MSs, the
PRR scheduler prioritizes the most important service classes. Likewise, the SS/MS scheduler
also uses a priority RR algorithm, distributing the available slots in the uplink direction. The
proposed procedure is executed using a priority scheme to distinguish and transmit data
packets in the following order of existing traffic type connections: UGS, ertPS, rtPS, nrtPS
and BE. The new scheduler class – QoSBSScheduler – is also used as a subclass of the existing
BSScheduler, as shown in Figure 6.
Fig. 6. Scheduler Class Diagram
5. Performance Evaluation
This section is devoted to the results and performance evaluation of the implemented QoS
model. In order to evaluate the modifications to the existing NIST model, several simulation
scenarios were implemented to test QoS using distinct network topologies. The obtained
results use performance metrics, such as packet loss, latency, jitter and bandwidth usage,
and also make use of differentiated traffic sources for each service class.
5.1 Simulation Scenario

The tested network topologies consider differentiated traffic going in the uplink direction
from different hosts. Point-to-Point (PTP) and Point-to-Multipoint (PMP) scenarios, as
illustrated in Figure 7, were considered. In the PTP case, four hosts directly connected to one
WiMAX SS establish communication with the WiMAX BS, whereas in the PMP scenario,
four WiMAX SSs communicate simultaneously with the WiMAX BS. Each host’s traffic
represents one connection flow in the uplink direction.
WiMAX Nodes
WiMAX SS #1
BE
UG
rt P S
nrt S
PS
WiMAX SS #2
BE
UGS
WiMAX Nodes
rtPS
Core Network
nrtPS
WiMAX Air Link

Correspondent Node
WiMAX BS
BE
WiMAX SS #3 UG
S
S
rtP
S
WiMAX Nodes n rt P
U E
B
r GS
n r tP S
S
WiMAX SS #4 tP
WiMAX Nodes
Fig. 7. Evaluation Scenario
In order to test the different network topologies, assuring differentiation between the
different service classes, we defined and implemented new traffic sources. As an example,
BE traffic generator contains a variable packet size and interval to emulate FTP/web traffic,
and an UGS traffic generator contains a constant transmission rate. The different values
adopted for these traffic generators are briefly presented in Table 1.
Bitrate Packet Size

Service Class
(Mbps) (Bytes)
BE 1 512 to 1024
UGS 1 300
rtPS 1 200 to 980
nrtPS 1 256 to 1024
Table 1. Service Classes Parameters
5.2 Simulation Results

Initial simulations made use of the PTP topology between the WiMAX BS and the WiMAX
SS#1, presented in Figure 8, with four hosts connected to SS#1 and conveying differentiated
traffic in the uplink direction. From this scenario we have also defined the WiMAX radio
link parameters that would optimize the traffic transmission and subsequent simulation
scenarios. The most important parameters are summarized in Table 2.
Modulation Queue length Bandwidth
64 QAM ¾ 50 Packets 5 Mhz

Table 2. WiMAX Air Link Simulation Parameters
Initially, we tested a scenario in which four nodes, connected to each one of the WiMAX SSs,
generate traffic competing for resources in the uplink direction. Each node has a traffic
source dedicated to a specific WiMAX service class, specifically, UGS, rtPS, nrtPS and BE.
The obtained throughput results are presented in Figure 8.
Throughput vs. Nr of MN
7000
6000
5000
Throughput (kbps)
UGS
4000
RTPS
nrtPS
3000
BE
2000
1000
0
1 2 3 4
Number of Subscribers
Fig. 8. Throughput vs. Number of WiMAX Subscribers Results
As depicted in Figure 8, the throughput values obtained for UGS services are quite
satisfactory, with a reduced latency (Figure 9), jitter (Figure 10) and packet loss (Figure 11),
when compared with the BE service. Therefore, packet differentiation is obtained,
prioritizing the UGS related packets over the BE packets. Furthermore, one can see that with
the increasing number of SSs, the obtained throughput is variable for each service class.
Since the bandwidth is distributed into a higher number of WiMAX SSs, less bandwidth will
be available for each one of them and, consequently, the less prioritized classes will be more
degraded in terms of QoS.
Delay Vs. Nr. of MN
700
600
500
UGS
Dealy (ms)
400
RTPS
nrtPS
300
BE
200
100
0
1 2 3 4
Fig. 9. Delay vs. Number of WiMAX Subscribers Results
Jitter Vs. Number of MN
400
350
300
250 UGS
Delay (ms)
RTPS
200
nrtPS
150 BE
100
50
0
1 2 3 4
Number of subscribers
Fig. 10. Jitter vs. Number of WiMAX Subscribers Results
Analyzing the obtained results for the delay and jitter, presented in Figure 9 and Fig 10,
respectively, it is visible that for the UGS traffic class these values always maintain
reasonable and significantly very low values. Concerning the remaining traffic classes, with
the increasing of subscribers, the delay and jitter values are, as expected, significantly
affected.
With respect to the packet loss results, illustrated in Figure 11, the UGS service class remains
almost unaffected, whereas the remaining service classes progressively drop more packets,
due to the prioritization algorithm.
Packet Loss Vs. Nr of MN
120
100
80
Packet Loss (%)
UGS
RTPS
60
nrtPS
BE
40
20
0
1 2 3 4
Number of subscribers
Fig. 11. Packet Loss vs. Number of WiMAX Subscribers Results
Finally, Figure 12 presents the bandwidth usage for each service class.
Bandwidth Usage vs. Nr of MN
100%
90%
80%
70%
Used Bandwidth
60% BE
nrtPS
50%
RTPS
40% UGS
30%
20%
10%
0%
1 2 3 4
Fig. 12. Bandwidth Usage vs. Number of WiMAX Subscribers Results
It is visible a fair allocation of the available resources within each one of the service classes.
When the number of WiMAX SSs increases, more bandwidth is allocated to the higher
prioritized classes (e.g. UGS), avoiding starvation of the less prioritized classes (e.g. BE).
Summarizing, from the obtained results, one can see that the highest QoS demanding
services achieve higher transmission throughput, penalizing the BE service. This
differentiation is less visible for a single WiMAX SS, but increasing the number of WiMAX
SSs, it is more clear the different treatment given to the packets that belong to different
WiMAX scheduling services. This is explained by the fact that the BS has to distribute the
available bandwidth between four SSs separately, leaving less available bandwidth for each
one.
6. Conclusions
In this chapter we presented an enhancement for the NS2-NIST/WiMAX model in order to
efficiently support QoS. Specifically, a packet classification mechanism and the associated
scheduler, based on priority RR (PRR), have been designed, implemented and tested.
Through the performance evaluation measurements with different topologies, point-to-
point (PTP) and point-to-multipoint (PMP), it was possible to verify the differentiated
behavior of the implemented WiMAX QoS classes. Based on the obtained results, we can
conclude that there is a traffic differentiation visible by the different values obtained for the
QoS parameters (latency, delay, bandwidth usage) in the test scenarios. Moreover, it was
always assured a minimum transmission for all the service classes, although with different
performances due to prioritization. The observed parameters degradation when using more
subscribers is related to the priority RR implemented scheduler, in which less priority
queues may not be served in the case of network overload or congestion.
7. References
3GPP, Evolved Universal Terrestrial Radio Access (E-UTRA) and Evolved Universal
Terrestrial Radio Access Network (E-UTRAN) – Overall Description, TS 36.300,
Stage 2, Release 9, Jun. 2009.
IEEE 802.16, IEEE Standard for Local and Metropolitan Area Networks. Part 16: Air
Interface for Fixed Broadband Wireless Access Systems, IEEE Std. 802.16-2004, Oct.
2004.
IEEE 802.16, IEEE Standard for Local and Metropolitan Area Networks. Part 16: Air
Interface for Fixed Broadband Wireless Access Systems. Amendment 2: Physical
and Medium Access Control Layer for Combined Fixed and Mobile Operation in
Licensed Bands, IEEE Std. 802.16e-2005, Dec. 2005.
Monteiro J., Sargento S., Gomes A., Fontes F., Neves P., “IEEE 802.16 Packet Scheduling with
Traffic Prioritization and Cross-Layer Optimization“, The 1st International
Conference on Mobile Lightweight Wireless Systems (MOBILIGHT), Athens,
Greece, May 2009.
NS2, http://www.isi.edu/nsnam/ns/
NS2-NIST, http://www.antd.nist.gov/seamlessandsecure/download.html
OPNET, http://www.opnet.com
Qualcomm, http://www.qualcomm.com
WiMAX Forum a, WiMAX End-to-End Network Systems Architecture Stage 2: Architecture
Tenets, Reference Model and Reference Points, Release 1, Version 1.2, Jun. 2008.
WiMAX Forum b, WiMAX End-to-End Network Systems Architecture Stage 3: Architecture
Tenets, Reference Model and Reference Points, Release 1, Version 1.2, Jun. 2008.

Trends in Telecommunications Technologies (2010)

Uploaded by

Copyright:

Available Formats

Trends in Telecommunications Technologies (2010)

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

Trends in Telecommunications Technologies (2010)

Uploaded by

Copyright:

Available Formats

I

First published March 2010

Technical Editor: Goran Bajac

Trends in Telecommunications Technologies,

Patras, 17th March, 2010

2. Advanced Modulation Formats and Multiplexing Techniques for Optical

3. A Survey on the Design of Binary Pulse Compression Codes

4. Virtual Multicast 063

5. The Asymmetrical Architecture of New Optical Switch Device 087

8. Telecommunication Power System: energy saving, renewable sources

9. Propagation Models and their Applications in Digital Television Broadcast

10. Interference Modeling for Wireless Ad Hoc Networks 185

12. Directional Routing Protocol in Wireless Mobile Ad Hoc Network 235

13. Free Space Optical Technologies 257

15. Performance analysis of multi-server queueing system operating under control

16. Interdomain QoS paths finding based on overlay topology

17. Dual Linearly Polarized Microstrip Array Antenna 367

19. Layer 2 Quality of Service Architectures 399

20. Secrecy on the Physical Layer in Wireless Networks 413

21. Performance Analysis of Time-of-Arrival Mobile Positioning

22. Mobility and Handoff Management in Wireless Networks 457

23. GPS Total Electron Content (TEC) Prediction at Ionosphere Layer

24. Performance Evaluation Methods to Study IEEE 802.11 Broadband Wireless

26. Building energy efficiency design for telecommunication base

27. Dynamic Space-Code Multiple Access (DSCMA) System: A Double Interference

29. The Development of Crosstalk-Free Scheduling Algorithms for Routing

30. Traditional float charges: are they suited to stationary

31. Neighbor Discovery: Security Challenges in Wireless Ad hoc

32. New Trends in Network Anomaly Detection 715

34. Quality of Service Differentiation in WiMAX Networks 753

A Novel PFC Circuit for Three-phase

Fig. 1. Three-phase single switch PFC circuit by Prasad-Ziogas.

(a) S1 turn-on （b）S1 turn-off

Fig. 4. Conceptual voltage waveform, vAO.

(b) iuiu 100A

Fig. 5. Explanation of distorted input current waveform in conventional method.

2.2 Operation Principle of the Proposed Circuit

Fig. 6. Proposed circuit configuration.

Fig. 7. Waveforms for proposed circuit.

(a) period Ⅰ(t0<t<t1)

(b) period Ⅱ(t1<t<t2)

(c) period Ⅲ(t2<t<t3)

Fig. 8. Operating circuit.

LL: Line inductance 0.25 mH

0.6 Conventional 2mH

0.4 Proposed 3mH, 200μF

Output power P0 [kW]

Fig. 12. Development to buck-converter.

Takahashi, I. Hori, K.(1997). Improvement of Input Current Waveforms of a Single Phase

Advanced Modulation Formats and Multiplexing

2.1 Amplitude Shift Keying

2.2 Phase Shift keying

Fig. 5. Example of QPSK/DQPSK modulated signal

separable. The demultiplexer performs opposite process as multiplexer to separate the

3.1 Time Division Multiplexing

Fig. 8. Example of a TDM system for multipleing two channels

Fig. 9. Example of an ultrafast point-to-point transmission system using OTDM technique

3.2 Wavelength Division Multiplexing

Fig. 10. Implementation of a typical WDM network

3.3 Orthogonal Frequency Division Multiplexing