Bud I Ar Jo Dissertation 2 ND Edition

Adaptive Baseband Processing Techniques
for Cognitive Radio Systems

PROEFSCHRIFT
Ter verkrijging van de graad van doctor

Aan de Technische Universiteit Delft,
Op gezag van de Rector Magnificus Prof. Dr. Ir. J. T. Fokkema,
Voorzitter van het College voor Promoties,
In het openbaar te verdedigen op dinsdag 1 December 2009 om 12.30 uur
door
Ibrahim BUDIARJO
Master of Science in Communications Engineering

Technische Universitaet Muenchen, Duitsland
geboren te Bandung, Jawa Barat, Indonesien.

Dit proefschrift is goedgekeurd door de promotor :
Prof. dr. ir. L. P. Ligthart
Copromotor :
Dr. H. Nikookar
Samenstelling promotiecommissie:
Rector Magnificus voorzitter

Prof. dr. ir. L. P. Ligthart Technische Universiteit Delft, promotor
Dr. H. Nikookar Technische Universiteit Delft, copromotor
Prof. dr. R. Prasad Aalborg Universitet
Prof. dr. ir. E. R. Fledderus Technische universiteit Eindhoven
Prof. dr. ir. I. G. M. M. Niemegeers Technische Universiteit Delft
Prof. dr. ir. S. M. Heemstra De Groot Technische Universiteit Delft
Prof. dr. ir. W. C. van Etten Universiteit Twente
This research has been supported by the AAF project in the framework of the Freeband
programme sponsored by the Dutch Ministry of Economic Affairs.
Work presented in this thesis has been performed at the International Research Centre for
Telecommunications and Radar of Delft University of Technology
Copyright © 2009 by Ibrahim Budiarjo
All rights reserved. No part of the material protected by this copyright notice may be
reproduced or utilized in any form or by any means, electronic or mechanical, including
photocopying, recording or by any information storage and retrieval system, without the prior
permission of the author.
Cover: The illustration of cognitivity with the brain that has the learning capability, collecting
and analyzing information i.e. link budget, spectrum occupancy, radio propagation and
the target QoS to decide the proper radio parameter values to be used by the
Cognitive Radio system. The background is an example of a view from a terminal,
and taken from the pictures collection of Apple Macbook.
ISBN 978-90-9024753-3
Printed in The Netherlands

To the memory of My Father, and to My Mother, Brother and My Fiancee
i
Thesis Summary

Cognitive Radio is a new paradigm in the wireless communications. It puts the intelligence
and awareness dimension to the radio communication system. A Cognitive Radio system will
be aware of the changing condition in its environment. The awareness in this thesis is related
to the spectrum and the channel stipulation. By intelligent learning and understanding from
the environment, a Cognitive Radio device will adapt its transmission parameters to the
changing environment. The objective of a Cognitive Radio system is to have a reliable
communication and efficient spectrum utilization.
Although most of the spectral ranges are already licensed, studies and measurements have
shown that most of the time the spectrum is not fully occupied, and even some bands are
rarely occupied. The congestions happen because of the existing poor spectrum access
technique. This condition raises the opportunity to rent or to access the spectrum while the
licensed user is in idle condition or by having the Cognitive Radio side by side with the
licensed users band occupying the spectrum holes. In this way, the spectrum is utilized in a
more efficient way. The challenging problem is to settle a “win-win” co-existence between
the Cognitive Radio, which is the rental system, with the legacy licensed system. Spectrum
pooling has been proposed in the literature as a technique of sharing the spectrum with the
licensed system by using the Orthogonal Frequency Division Multiplexing (OFDM) as
modulation technique. Some of the OFDM carriers located in the licensed user’s band will be
deactivated in order not to interfere the legacy system access. Due to this flexibility, the
OFDM is considered as a proper modulation technique to be applied to a Cognitive Radio
system. Deactivation of more carriers adjacent to the licensed user’s band, as well as
windowing and several signal processing techniques could reduce the interference
contribution of the OFDM based Cognitive Radio system to the licensed users signal. These
techniques are studied and presented in this dissertation. The spectrum pooling technique
could only work upon the reception of the accurate spectrum occupancy information. The
spectrum occupancy information is derived from a spectrum sensing module. Spectrum
sensing necessitates such a sophisticated module and requires a proper attention. This thesis
does not focus on the spectrum sensing module, but rather it is assumed that the spectrum
ii
occupancy information is available and accurate. The work on spectrum sensing is conducted
by out partner at the Twente University*.
Following the objective of Cognitive Radio to have reliable communications, the observation
or emphasis of the technique is not only on the licensed users side but also on the rental user.
Besides assuring the reliable communications of the licensed system, the target Quality of
Service (QoS) of the Cognitive Radio system should be attained. The practical parameters to
be observed are the Bit Error Rate (BER) and the bit rate of the system. The application of
OFDM with spectrum shaping for the purpose of reducing the interference contribution to the
licensed system can be achieved with the cost of self QoS degradation. Results from literature
have shown that application of adaptive bit loading could enhance the BER of an OFDM
system. By adaptive bit loading, the bits are allocated to each of the OFDM carrier
intelligently according to the channel condition by setting the target BER and bit rate of the
overall system. In this work, we propose to combine the adaptive bit loading with the
spectrum shaping to preserve the Cognitive Radio system’s QoS. The impact of this
combination on the OFDM Peak to Average Power Ratio (PAPR) growth is evaluated through
simulations.
There is a growing interest in replacing the Fourier transform in OFDM with wavelet basis
functions. The technique is termed as the Wavelet Packet Multicarrier Modulation
(WPMCM). Investigations reported in literature have presented the evaluation of WPMCM
and compared the results with OFDM. Following the successful application of a frequency
selective wavelet in the Ultra Wide Band system, in this dissertation we evaluate the
suitability of the frequency selective wavelet in WPMCM combined with the spectrum
pooling concept to be applied to the Cognitive Radio system. As the efficient spectrum
utilization is one of the major objectives of Cognitive Radio, it is reasonable to include
Multiple Input Multiple Output (MIMO) to the Cogntive Radio system. This subject is studied
in this thesis and the performance of MIMO in the OFDM and WPMCM based Cogntive
Radio system is evaluated.
Channel estimation is the crucial module in every OFDM system. We propose an effective
channel estimation scheme based on optimum pilot patterns of conventional OFDM using
virtual pilots and apply it to Cognitive Radio systems. The virtual pilots are derived from the
combination of the linear interpolation/extrapolation between two real pilots with the so
called decision directed method. Without loss of generality we use the Wiener filter as the
channel estimation technique due to its efficient and straightforward method in utilizing the
channel correlation property according to the distance between pilots and data. First we adopt
the Single Input Single Output (SISO) OFDM based Cognitive Radio system, and then we
expand the scheme to the MIMO application.
*
This thesis is part of the Work Package 3 (WP3) Dutch Adaptive Adhoc Freeband (AAF) project. The WP3
will produce 3 PhD theses, one on Spectrum Sensing, one on Baseband Processing (this thesis) and the third
one is on the mapping of the Spectrum Sensing and Baseband Processing into a heterogeneous architecture.
iii
Beside OFDM and WPMCM based Cognitive Radio system, recently Transform Domain
Communication System (TDCS) and Wavelet Domain Communication System (WDCS) have
been introduced as promising modulation techniques for Cognitive Radio application. TDCS
and WDCS have bit rate limitations. As an effort to enhance the bit rate of TDCS and WDCS,
we propose to add an extra embedded symbol to TDCS and WDCS. We analyze the impact of
the embedded symbol on the conventional TDCS and WDCS data detection. The impact of
the embedded symbol on the data detection in a multi-user environment, which is inherently
supported by the conventional TDCS and WDCS, is observed. In addition, we also evaluate
the performance of the TDCS with an embedded symbol in the MIMO system.
As a sort of verification platform for Cognitive Radio we proposed a practical Demonstrator

that involved a spectrum scanning module and baseband processing transceiver module. The
spectrum sensing is employed by the Universal Software Defined Radio Peripheral (USRP)
while the baseband processing transceiver module is applied to an FPGA Development board.
The current made Cognitive Radio verification platform is simple and still has a limited
feature. While further developments are required in this field, details of this effort are also
provided in this thesis.
iv
v
Contents
Thesis Summary i
1 Introduction 1
1.1 Research Background 1
1.2 Research Motivation 3
1.3 Scope and Novelties of the Dissertation 4
1.4 Outline of the Thesis 4
2 OFDM Based Cognitive Radio 7
2.1 Introduction 7
2.2 Spectrum Pooling with OFDM 10
2.3 OFDM Sidelobes Reduction Techniques 16
2.3.1 Sidelobe Suppression by Cancellation Carriers Insertion 16
2.3.2 Sidelobe Suppression by Subcarrier Weighting 18
2.3.3 Sidelobe Suppression by Multiple Choice Sequence 18
2.3.4 Sidelobe Suppression by Adaptive Symbol Transition 19
2.3.5 Sidelobe Suppression by Frequency Domain Additive Signal 20
2.3.6 Sidelobe Suppression by Constellation Expansion 21
2.4 Frequency Hopping 22
2.4.1 Evaluation Using GSM 900 MHz Channel Model 22
2.5 Conclusions 24
3 Baseband Processing for OFDM Based Cognitive Radio 25
3.1 Introduction 25
3.2 Combined Spectrum Pooling, Windowing, Adaptive Bit Loading and Sidelobes
Cancellation Carriers 26
3.2.1 Modulation Detection by Signaling and Blind Detection 28
3.2.2 Adaptive Bit Loading 32
3.2.2.1 Chow Algorithm 33
3.2.2.2 Fischer Huber Algorithm 33
3.2.2.3 Simple Blockwise Loading Algorithm (SBLA) 35
3.2.2.4 Subband Bit Loading based on Fischer-Huber Algorithm 37
3.2.3 Spectrum Pooling and Sidelobes Cancellation Carriers Insertions 37
3.2.4 Combined Adaptive Bit Loading with Spectrum Pooling and Sidelobes
Cancellation Carriers Simulation Results and Analysis 41
3.3 Wavelet Packet Modulation by Frequency Selective Wavelet for Cognitive
Radio 49
3.3.1 Generation of Wavelet Packet Sub-Carriers Bases 51
vi
3.3.2 Generation of Wavelet Packet Dual Bases 53

3.3.3 Design of Best Wavelet Packet Bases for WP-MCM 54
3.3.3.1 Paraunitary Condition 54
3.3.3.2 Compact Support 55
3.3.3.3 Regularity 55
3.3.4 Wavelet Families 56
3.3.4.1 Daubechies 56
3.3.4.2 Coiflet 56
3.3.4.3 Symlet 56
3.3.4.4 Frequency Selective Wavelet 56
3.3.5 Performance Evaluation and Analysis 62
3.4 Conclusions 75
4. Channel Estimation for OFDM Based Cognitive Radio 77
4.1 Introduction 77
4.2 Wiener Filter Channel Estimation 78
4.3 Pilot Pattern for Frequency Hopping OFDM Based Cognitive Radio 84
4.4 Combined Optimum Pilot Pattern and Virtual Pilots for OFDM Based
Cognitive Radio 88
4.5 MIMO OFDM Based Cognitive Radio Optimum Pilot Pattern with Virtual
Pilots Concept 106
4.6 Training and Pilot Pattern for OFDM Based Cognitive Radio Synchronization
and PAPR Reduction 124
4.6.1 Synchronization for OFDM Based Cognitive Radio System 124
4.6.2 Peak to Average Power Ratio Reduction for OFDM Based Cognitive
Radio Channel Estimation 126
4.7 Conclusions 127
5. Single Carrier Cognitive Radio System 129
5.1 Introduction 129
5.2 Transform Domain Communication System (TDCS) with Embedded Symbol 130
5.2.1 Multi User TDCS with Extra Embedded Symbol Detection Analysis 137
5.3 Wavelet Domain Communication System (WDCS) with Embedded Symbol 140
5.4 MIMO Application to Single Carrier Cognitive Radio System 142
5.5 Performance Evaluation and Analysis 145
5.5.1 TDCS Evaluation 145
5.5.2 WDCS Evaluation 157
5.6 Conclusions 159
6. AAF Project Cognitive Radio Demonstrator in IRCTR TU Delft 161
6.1 Introduction 161
6.2 The Hardware Requirements for Baseband Transmission 162
vii
6.3 Spectrum Scanner Module 169

6.4 Cognitive Radio Nodes Demonstrator Design 170
6.5 Conclusions 179
7. Conclusions and Recommendations 181
7.1 Conclusions 181
7.2 Recommendations for Future Work 185
Appendix A - OFDM based Cognitive Radio Frame Design for Frequency Hopping 187
Scenario
Appendix B – Training Symbol Repetition Period Mathematical Proof by Zeros 190
Insertion
Appendix C - Single User CCSK Detection in TDCS with Embedded Symbol 192
Appendix D - The Power to Mean Sidelobe Power Ratio (PMR) of TDCS basis
function 193
Appendix E – Multi User CCSK Detection in TDCS with Embedded Symbol 195
Appendix F – V-BLAST Algorithm 197
Appendix G – Future Work in Wavelet Packet Multicarrier Modulation Based
Cognitive Radio 199
Bibliography 201
List of Abbreviations 212
List of Symbols 214
Publications by the Author 216
Samenvatting 221
Acknowledgements 225
Curriculum Vitae 227
viii
1
Chapter 1
Introduction
1.1 Research Background
The growing demand on wireless communications systems to provide high data rates and the
introduction of new communication applications, e.g. Long Term Evaluation (LTE), IEEE
802.11N and the Ultra Wide Band standard have made the spectrum become congested. This
condition has brought the need for a flexible and efficient use of the spectrum resource. The
regional spectrum allocation policy counteracts the free mobility of radio communication
equipment. The vast majority of the available spectral resources have already been licensed,
so it appears that there is little or no room to add any new services, unless some of the
existing licenses are discontinued. The old spectrum licensing policies need therefore to be
revised. New spectral ranges that are suitable for future radio transmission remain interesting
to be made available to the public even if spectrum parts are already licensed.
Furthermore, recent studies and measurements have shown that vast portions of the licensed
spectra are rarely used, and congestions often happen due to inflexible spectrum utilization
and regulations instead of the non availability of the spectrum.
In order to utilize the unused spectrum efficiently in dynamically changing environments, a

new communication technique is required. It should adapt to the rapidly changing
environment condition while coexist with the existing legacy systems by ensuring minimal or
tolerable interference to the legacy user’s communication. This technology is named as
Cognitive Radio. The idea of Cognitive Radio has been initially introduced by Joseph Mitola
in his Doctoral Thesis [1]. Mitola described how a Cognitive Radio could enhance the
flexibility of wireless services through a radio knowledge representation language. This
concept was further expanded in Mitola’s dissertation.
Simon Haykin gave in his paper [2] the solid definition of Cognitive Radio as an intelligent
wireless communication system that is aware of its surrounding environment (i.e., outside
world), and uses the methodology of understanding-by-building to learn form the
environment and adapt its internal states to statistical variations in the incoming radio
frequency (RF) stimuli by making corresponding changes in certain operating parameters
(e.g., transmit-power, carrier-frequency, and modulation strategy) in real-time, with primary
2
objectives of highly reliable communications and efficient utilization of the radio spectrum.
This definition delivers six keywords related to Cognitive Radio, they are : awareness,
intelligence, learning, adaptivity, reliability, and efficiency. Fette in [3] relates the awareness
scope of a Cognitive Radio system to the transmitted waveform, radio frequency (RF)
spectrum, communication network, geography, locally available services, user needs,
situation and security policy. In adition to the previoiusly mentioned capabilities, the
Cognitive Radio is equipped with reconfigurability. The reconfigurability provides the
adaptation of the radio interface to accommodate variations in the development of new
interface standards, incorporation of new applications, services, and updates in software
technology, and exploitation of flexible heterogeneous services provided by radio networks
[4]. The term cognition itself includes language and communication aspects [5]. The
Cognitive Radio’s language includes the set of signs and symbols that permits different
internal constituents of the radio to communicate with each other [2], [5]. In summary the
main Cognitive Radio tasks include :
• radio-scene analysis which encompasses estimation of interference impact from
temperature of the radio environment and detection of spectrum holes.
• channel identification which encloses the channel state information (CSI) estimation
and channel capacity prediction for use at the transmitter site.
• Transmit power control and dynamic spectrum management.
By including the interaction with RF environment then three tasks can be distinguished and
form a cognitive cycle as depicted in Figure 1.1 [6]. The radio scene analysis will receive
stimuli from the radio environment and further identify the spectrum holes and CSI.
Radio Environment
4 1 5
Signal Data Reception
Transmission Electromagnetic
Spectrum and Channel
Information
3 Radio Scene 2
Analysis
Transmitter Receiver
Side information to Channel State Information,
share channel and Spectrum Holes, Interference
spectrum information temperature, Traffic
Information
Figure 1.1: Cognitive Radio signal transmission cycle [6]

3
In the radio scene analysis block the spectrum and channel information will be evaluated (
line (1) in Figure 1.1) , the analysis will be distributed to the Transmitter and Receiver ( lines
(2) and (3)) and accordingly the Transmitter and Receiver will decide their transmission and
reception technique in interaction with the environment condition (line(4) and (5)).
In this dissertation the scope of “cognition” is limited to have the co-existence between the
Cognitive Radio based rental (unlicensed) users (RUs) and the licensed users (LUs) by means
of reducing the mutual interference between them, while obtaining reliable communications
for RUs. The spectrum owners who are the LUs allow portions of their spectrum to be utilized
for certain periods of time by RUs which apply the Cognitive Radio system in their
transmissions. The underlying mechanism is based on utilizing the actual spectrum occupancy
condition knowing that the allocated spectrum is not occupied most of the time. In fact, some
portions of the spectrum are rarely used [7]. The LUs are ordinary mobile terminals and their
associated base stations. They do not possess much intelligence. The RUs, on the other hand,
should possess the intelligence to sense the spectrum and use whatever resources are available
when they need them. At the same time, the RUs should adjust its spectrum usage when an
LU begins transmission. The RU access should be done in such a way that the neighboring
LUs will not be aware of the RU existence.
1.2 Research Motivation
As one of the Cognitive Radio objectives is to obtain efficient utilization of the radio
spectrum, and the focus of our Adaptive Ad-hoc Freeband (AAF) project is on identification
of free resources in the frequency domain, we limit our observation to the spectrum. The
scope of the spectrum utilization is focused on the co-existence between the Cognitive Radio
based RUs and the legacy LUs. Successful co-existence between RU access and the
neighboring LUs can be accomplished by assuring that the neighboring LUs will not be
interrupted or intervened when a RU wants to get an access to the spectrum hole adjacent to
the LU’s band. This stringent requirement can be technically interpreted as an overlay
demand within an opportunistic RU access mode. The part of the RU spectral energy which is
located at the neighboring occupied LU band should therefore be as low as possible. It must
not interrupt the LU transmission. The mutual interference between the RU and LU should
also be reduced. The mutual interference reduction is a way of preserving reliable
communications for both users. This shaping can be interpreted as sidelobe suppression
technique in frequency domain.
In the RU perspective reliable communications is not connected to an interference avoidance

aspect, but also to the quality of service in terms of target bit rate and maximum allowable bit
error rate (BER). Due to priority onLU access, the RU resources will be limited and may still
affect the LU achievable bitrate, meaning that this resource limitation could also have impact
on the LU BER performance.
4
For these reasons, we investigated some novel baseband signal processing techniques for
Cognitive Radio system to suppress the LU sidelobes, to increase the LU bit rate that
subsequently enhance the spectral utilization efficiency, to counteract fading of the channel to
improve the system BER and to handle the OFDM challenges (e.g peak to average power
ratio (PAPR) and synchronization) in the application of a Cognitive Radio system.
Further, the goal of our AAF project i.e., research and demonstration of a Cognitive Radio
system, which continuously adapts its communications scheme to the available resources, will
be incorporated in this thesis by describing our development of the Cognitive Radio baseband
processing and the AAF demonstrator platform.
1.3 Scope and Novelties of this Dissertation
In correspondence to the motivation of the research, the following novelties and primary
results are delivered in this thesis :
Non Contiguous Orthogonal Frequency Division Multiplexing (NC-OFDM) is
considered as a feasible modulation technique to be applied in Cognitive Radio
systems. A problem arises if the non contiguity delivers inadequate target quality of
service (QoS) e.g. bit rate. Frequency hopping is then considered as an alternative. The
system will hop to a frequency range that could provide the Cognitive Radio with the
desired QoS. With respect to this issue this dissertation contributes to an OFDM based
Cognitive Radio frame design with full active carriers according to the Bluetooth time
division duplex (TDD) for frequent frequency hopping in the 900 MHz channel model
(Chapter 2, Section 2.4).
A combination of spectrum pooling, windowing, adaptive bit loading and sidelobes
cancellation carriers is proposed. Spectrum pooling, windowing and sidelobes
cancellation carriers are incorporated for the purpose of sidelobes reduction (mutual
interference reduction between RUs and LUs). Adaptive bit loading is applied to
preserve the Cognitive Radio based RUs target bit rate and BER (Chapter 3, Section
3.2).
Signal spectrum shaping by a frequency selective wavelet in multicarrier wavelet
packet modulation and a single carrier wavelet domain communication system
(WDCS) is introduced (Chapter 3, Section 3.3 and Chapter 5, Section 5.3).
An optimum pilot allocation strategy for OFDM based Cognitive Radio channel
estimation has been worked out. For that purpose the virtual pilot concept is set up in
order to simplify the filtering process in the channel estimation (Chapter 4). Their
application in MIMO system is also evaluated.
Use of an embedded symbol for higher rate single carrier transform domain
communication system (TDCS) and WDCS based Cognitive Radio is motivated.
Results have shown that by increasing the constellation size of the embedded symbol,
the TDCS signal to noise ratio (SNR) per bit will increase and further will enhance the
legacy data source detection from the cyclic code shift keying data mapping (Chapter
5, Section 5.2).
5
1.4 Outline of the Thesis
The organization of this thesis is as follows :
Chapter 2 provides the insight into the modulation techniques feasible for the Cognitive Radio
system. The spectrum can be accessed in overlay mode or underlay mode. In
overlay mode the signal power is above the noise level while in underlay mode the
condition is opposite. OFDM is the proper multicarrier modulation mode for
overlay spectrum access Cognitive Radio system, while TDCS is the single carrier
approach. The underlay spectrum access Cognitive Radio system can be applied in
systems using the Ultra Wideband Technique. Due to its higher rate and better
performance, the OFDM flexibility and the various degrees of freedom in
modifying its parameters to achieve the successful coexistence between the RU,
which is a Cognitive Radio system and the LU, several OFDM based Cognitive
Radio modulation techniques are analyzed and discussed in this chapter. As several
techniques in suppressing mutual interference between RU and LU are introduced
in this chapter, we come up with the idea to combine some of the available mutual
interference suppression techniques to get lower sidelobes (as low as possible), and
present them in chapter 3.
Chapter 3 introduces the novel combination of spectrum pooling, windowing, adaptive bit
loading and sidelobes cancellation carriers to be applied in an OFDM based
Cognitive Radio system for the purpose of attaining significant mutual interference
reduction between RU and LU while the target quality of service is maintained. An
alternative approach by replacing the Fourier-based transformation in OFDM with
wavelet packet modulation has been investigated. The optimization problem and its
use in the Cognitive Radio application are described. As in this chapter OFDM is
considered as the promising modulation technique for the Cognitive Radio system,
the critical component in OFDM, which is the channel estimation, should be
observed in the cognitive radio context. This issue is studied in the following
chapter.
Chapter 4 proposes a novel pilot allocation technique for OFDM-based Cognitive Radio
channel estimation as an effort to counteract the channel fading to obtain a reliable
quality of communications. Pilot patterns that can cope with the LU existence are
suggested in this chapter. The virtual pilot concept is utilized to aid the channel
estimation process. Further, the application is examined in a MIMO system. The
MIMO OFDM based Cognitive Radio performance evaluation through simulations
reveals its higher BER gain compared to the application of adaptive bit loading into
the SISO OFDM based Cognitive Radio system. Training and pilot designs to
resolve other common challenges like synchronization and high peak to average
power ratio in the context of Cognitive Radio are drawn. The multicarrier
modulation for Cognitive Radio and its critical component have been discussed in
this and the two previous chapters. The single carrier modulation approach can also
6
be used for Cognitive Radio. Therefore, in the following chapter the single carrier
technique is discussed and an optimization technique to this approach is explained.
Chapter 5 describes the efforts in optimizing and enhancing the utilization of the single carrier
modulation termed as TDCS for the Cognitive Radio system. The novel
contribution of the chapter is the application of an embedded symbol to the single
carrier TDCS based Cognitive Radio system. The embedded symbol adds extra
transmitted source bits that can be used as an extra data or side information.
Further, its impact to the sister technique WDCS and their extended application in
a MIMO system are evaluated. The baseband processing for Cognitive Radio has
been discussed and described in chapters 2 till 5. As the implementation aspect of
algorithms is also important, in chapter 6 the implementation efforts of the
developed baseband processing in a demonstrator platform will be described.
Chapter 6 deals with the implementation of the Cognitive Radio baseband processing into a
demonstrator platform which includes control programming of the field
programmable gate array (FPGA) on the selected evaluation board P25M* and
universal software radio platform (USRP† ) .
Chapter 7 summarizes all main results of the thesis, draws overall conclusions and gives some
recommendations for future work.
*
P25M is a development board produced by Innovative Integration equipped with FPGA and DSP module, 4
channels 25 MSamples/sec A/D and 4 channels 50 MSamples/sec D/A.
(http://www.innovative-dsp.com/products.php?product=P25M )
†
USRP is a software radio device with low capacity FPGA and 4 input and 4 output channels; it is controlled by
a computer through the USB connection.( http://www.ettus.com)
7
Chapter 2
OFDM Based Cognitive Radio
2.1 Introduction
In this thesis we assume that the spectrum can be accessed in overlay mode or underlay mode.
In overlay mode the signal power is above the noise level while in underlay mode the signal
power will be below the noise level. In the context of Cognitive Radio the overlay mode
spectrum access is applied side by side (adjacent) to the legacy system, utilizing a spectrum
hole. In underlay mode the signal will overlap with the legacy system, therefore the technique
should be robust to interference by removing the impact of the legacy system signal from the
own signal in order to detect its own signal correctly. Ultra Wide Band is a suitable technique
for underlay spectrum access [8]. Ultra Wide Band is a good technique for short range high
rate and long range low rate communications. In our AAF project we consider long range and
high rate communications; therefore the overlay access mode is the preferable option.
In the overlay mode Orthogonal Frequency Division Multiplexing (OFDM) is considered as

the proper multicarrier modulation technique, and the Transform Domain Communications
System (TDCS) as the single carrier modulation suitable for Cognitive Radio systems. The
reason is that both OFDM and TDCS have the capability in notching a part of their spectrum
energy located at the LU band in order to guarantee the co-existence with the legacy LU.
TDCS has been initially introduced in [9]. TDCS relies on the cross correlation property of its
basis function in detecting its transmitted data. OFDM provides a higher rate and a better
performance compared to the TDCS, while TDCS gives simplicity in shaping its spectrum
energy.
In this chapter we focus our attention on OFDM, while TDCS will be presented in detail in
chapter 5, where we discuss its performance and compare it with the OFDM.
Orthogonal frequency division multiplexing (OFDM) is a multicarrier modulation technique

that splits its total transmit bandwidth into a number of orthogonal subcarriers such that each
subcarrier can transmit independent data from each other. The subcarriers orthogonality is
attained by setting the frequency spacing between two carriers equal to the inverse of its
symbol duration. As the number of subcarriers increases for a certain fixed bandwidth the
symbol duration increases. In this way, a high rate data stream is divided into a number of
8
lower data streams. Due to the long symbol duration, the inter symbol interference (ISI)
impact from the multipath delay channel is reduced. Further, the insertion of a guard interval
(GI) at the beginning of each OFDM symbol will remove the ISI impact completely, as long
as the GI duration is longer than the maximum multipath delay [10].
A simple common OFDM scheme is depicted in Figure 2.1, while the OFDM subsystem with
the spectrum sensing module is in Figure 2.2.
Pilot
OFDM subsystem
Binary symbol
source
Channel Inter- Symbol Pilot OFDM
coder leaver mapper MUX TG
Time and
frequency
Selective
CSI Channel channel
Binary estimator AWGN
sink +
Symbol
Channel Deinter- Pilot IOFDM
de-
decoder leaver DEMUX TG
mapper
Figure 2.1 : OFDM system model
Spectrum Spectrum
ID magnitude
Xn Serial to Add guard Digital

. . Parallel to RF x(t)
Interval + To
parallel . IFFT . serial Module
Windowing Analog
converter . . converter xv
( g(t) ) converter
Multipath
Spectrum
Propagation
magnitude
h(τ,t)
Rn Paralel to . Analog
. Serial to Remove
To RF
serial . FFT . parallel guard
yv Digital Module
+ n(t)
converter . . converter interval
converter y(t)
Figure 2.2 : OFDM subsystem description with spectrum sensing module

9
In the scope of cognitive radio for assuring the co-existence between the RU and LU , the
feasibility of OFDM in achieving this cognitive radio goal can be observed from its power
spectral density (PSD) equation :
2
N FFT (1+α)T u
1
g(t)e− j2π ( f − f n)t
2
X( f ) = ∑ pn X n ∫ dt
N FFT n=1 −(1+α)T u (2.1)
2 N FFT
subject to X ( fLU ) < P Threshold
, and ∑ p =P
n=1
n T
where X(f) is the OFDM spectrum, NFFT is the number of FFT points which refers to the
possible number of total subcarriers, Xn and pn are the modulated data and the allocated
power on carrier n respectively, g(t) is the window function, α is the window roll off factor,
and Tu is the useful OFDM symbol duration. fn is the frequency of carrier n, and PT is the total
transmitted power, fLU is a frequency within the LU band, and PThreshold is the level of
interference power that is still tolerable by the LU.
In order to allow a RU to co-exist with LU, its PSD at the location of any frequency of the LU
should be zero or negligible. The direct interpretation of this statement would be de-activating
of subcarriers at those frequency locations of the LUs. This technique is known as Spectrum
Pooling [11]. The details about this technique will be explained in the following section. The
coexistence effort will depend on the spectrum information accuracy provided by the
spectrum sensing module. Another advantage of OFDM is the use of an FFT module that can
also be utilized in sensing the spectrum.
Spectrum sensing is another critical research issue and requires a proper attention. In this
dissertation we will not explore the techniques applied in spectrum scanner module. This
topic is investigated in detail separately by another PhD candidate in our AAF project. We
assume that the spectrum information is available and accurately estimated. A special
approach is required to gather spectrum information from all available cognitive radio devices
in order to have a really accurate spectrum occupancy information. This is especially the case
when the LU’s geographical position is close to one RU so that the LU is easily detected by
the corresponding RU. In case the LU is far from the other RUs, those RUs would probably
not detect the LU presence. The simple approach could then be applying the logical ‘OR’
operation to the spectrum information vector of all RU transceivers. In our AAF project the
spectrum info is planned to be distributed to all RU (Cognitive Radio) transceivers through a
dedicated control channel. Research has been conducted on use of the Ultra Wide Band
technology as a possible technique for applying such a common control channel [12].
Although the subcarriers of the OFDM based Cognitive Radio located in the LU band are de-
activated, due to the OFDM window properties, part of its spectrum, which is termed as
sidelobes, will be smeared into the LU band. This is the reason why the term PThreshold is
included in (2.1), as a reference for the cognitive radio system designer in shaping the
10
spectrum of the RU signal in such a way that the part of its spectrum in the LU band is
considered to be negligible.
The PSD equation described in eq. (2.1) shows that the OFDM side-lobe power is determined
by the allocated power on each subcarrier, the constellation size of Xn, the composition of the
Xn values, the window roll off factor α, and the window form g(t). The larger the Xn
constellation size, the higher the possibility of large side-lobe power occurrence. A large side-
lobe refers to a higher interference to the licensed system.
Another aspect that determines the side-lobe amplitude is the distance between the arbitrary
frequency point f in the LU band and the carrier position fn. The side-lobe of the carriers
adjacent to the LU band has a major role in interfering the LU. De-activating some of these
carriers will lower the side-lobe magnitude. In the application of adaptive bit loading, the
constellation size of carriers especially the ones close to the LU band should be taken into
account in such a way that the |X(fLU)|2 be constrained to a threshold PThreshold .
In the following sections some available techniques in reducing the OFDM sidelobes for
cognitive radio application will get attention. The sidelobe reduction outcome of all
techniques can be explained by relating them to eq. (2.1).
A novel research output with respect to OFDM based cognitive radio subject will be given in
Section 2.4 of this chapter, where hopping is applied to the cognitive radio system in case the
number of available carriers for cognitive radio transmission is not sufficient to attain the
target BER and bit rate. For the purpose of the hopping, the OFDM frame will be designed
according to the Bluetooth time division duplex (TDD) frame since Bluetooth also uses this
hopping mechanism for its transmission.
2.2 Spectrum Pooling with OFDM
Spectrum pooling has been proposed in [12] and [13] as a strategy where public access is
enabled to these bands without giving significant interference to the actual license owners.
Spectrum pooling enables public access to spectral ranges of licensed frequency bands which
are seldom used by overlaying a secondary rental user (RU) to an existing licensed user (LU).
The LUs are radio systems authenticated to operate under licensed spectral bands. The RUs
are intelligent CR systems that actively scan the landscape of frequency usage and
opportunistically utilize the available and unused resources. The RU relinquishes control over
the resources (here spectrum) as and when the LU starts using them. To identify and utilize
the unused bands, the frequency bands of various radio systems (including licensed and rental
users) are combined to obtain a common spectral pool. Cohabitation of LU and RU systems is
actualized by shaping the transmission waveform of the RU in a way that it utilizes the
unoccupied time-frequency gaps of the LU.
The pioneering work on the subject was conducted by Jondral et al [12] who devised a
spectrum pooling scheme using multi-carrier modulation (MCM). According to this scheme
11
the individual OFDM sub-carriers occupied by the LU and adjacent to LU are de-activated.
Figure 2.3 illustrates a typical scenario of spectrum pooling.
Figure 2.3 : Illustration of spectrum pooling block diagram
Further, in [12] and [13] the extension of each OFDM block with long cyclic prefix and suffix
samples and the application of some windowing to reduce the side-lobes of the subcarrier
channels is proposed. Obviously, this solution occurs at the expense of bandwidth loss
because an excessive time has to be allocated to cyclic extensions which otherwise could be
used for data transmission. The interference reduction method is extended by combining the
windowing technique with de-activating of subcarriers located adjacent to the LU’s band
which provides a kind of shield to the LUs. One commonly used window is the raised cosine
window [12],
⎧ 1 (1 − α )
⎪ 0≤ t ≤ T u
⎪ T u 2
⎪
⎪⎧
(2.2)
⎪ 1
g (t ) = ⎨ ⎨1 + cos ⎢
⎡ π ⎛
⎜t −
T u (1 − α ) ⎞⎟ ⎤⎥ ⎪⎬⎫ T u
(1 − α )
≤ t ≤ T u
(1 + α )
⎪ 2T u ⎩⎪ ⎢⎣ α T u ⎝ 2 ⎠ ⎥⎦ ⎭⎪ 2 2
⎪ 0
Otherwise
⎪
⎪⎩
where g(t) is the window function, α is the roll-off factor, and Tu is the useful OFDM symbol
interval (without guard interval). The sketches of the raised cosine window with different
12
values of roll off factor (α) are depicted in Figure 2.4, while the spectra of an OFDM carrier
with the respective windows are given in Figure 2.5. Increasing the α will reduce the
sidelobes with the expense of longer symbol duration.
0.9
0.8
RAISED COSINE α = 0.5
0.7 RAISED COSINE α = 0.25
Amplitude (g(t))
0.5
Tu
0.4
αT αT
u u
0.3
0.2
0.1
0
0 50 100 150 200
time (t)
Figure 2.4: Raised cosine window with different roll-off factor (α)
1
2
Magnitude Squared |X(f)|
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8
Normalized Frequency (fTu)
Figure 2.5: OFDM carrier spectrum with raised cosine window with different roll-off factor
(α)
13
Other available forms are Bartlett [14],
⎧ 1 (1 − α )
⎪ ,
0≤ t ≤ T u
⎪ T u 2
⎪⎪ 1 1 ⎡ t ⎤ (1 − α ) (1 + α ) (2.3)
g (t ) = ⎨ − ⎢ −
1
⎥,
T u
≤ t ≤ T u
⎪ 2T u T u ⎢⎣ α T u
2α ⎥⎦ 2 2
⎪ 0 O therw ise
⎪
⎪⎩
Better than Raised Cosine (BTRC) [15],
⎧ 1 T (1 − α )
⎪ , 0≤ t ≤ u
⎪ Tu 2
⎪ 1 ( − 2 ln 2 / α ) ⎡ t − ( (1 − α ) / 2 ) ⎤ (1 − α )
⎪ e TU ⎣ TU ⎦
, T u ≤ t ≤T u (2.4)
⎪ T u 2 2
g (t ) = ⎨
⎪ ⎡ ⎛ − 2 ln 2 ⎞ ⎡ T (1 − α )
−t⎥ ⎤
⎤
⎜ ⎟⎢
T u ≤ t ≤ T u (1 + α )
u
⎪ 1 ⎢1 − e ⎜⎝ α T u ⎟⎠ ⎢⎣ 2 ⎥⎦ ⎥
,
⎪T ⎢ ⎥ 2 2
⎪ u ⎣⎢ ⎦⎥
⎪ 0 O th erw ise
⎩
and Flipped Inverse Hyperbolic Secant (farchsech) [16] window,
⎧ 1 Tu (1 − α )
⎪ , 0≤ t ≤
⎪ T u
2
⎪
1 ⎡ ⎛ 1 ⎛ T u (1 + α ) ⎞ ⎞⎤ T (1 − α )
≤ t ≤Tu
⎪ 1 (2.5)
⎪ ⎢1 − arcsech ⎜
⎜α ⎜ −t ⎟ ⎟⎟ ⎥ ,
u
g (t ) = ⎨ T u ⎢⎣ α T uγ ⎝ Tu⎝ 2 ⎠ ⎠ ⎥⎦ 2 2
⎪
⎪ 1 ⎡ 1 ⎛ 1 ⎛ T u (1 − α ) ⎟⎞ ⎟⎞ ⎥⎤ , T (1 + α )
⎪ ⎢ arcsech ⎜ ⎜t −
u
≤ t ≤Tu
⎪T u ⎢⎣α T u
γ ⎜ ⎟
⎝ αT u ⎝ 2 ⎠ ⎠ ⎥⎦ 2 2
⎪⎩ 0 Otherwise
where γ = 2 ln( 3 + 2) / (α T u ) . The shape of the window in the region of T (1−α) ≤ t ≤ T (1+α) u u
2 2
determines how rapidly the OFDM spectrum falls to zero. The PSD will hit zero on the
frequencies in the interval of 1/ Tu. From these windows we learn that the farchsech window
provides the lowest sidelobes ( Figure 2.6). The Figures show the window forms including
their respective spectrum on an OFDM carrier.
14
Time domain Frequency domain

0.01
0
0.009
Normalized Amplitude to NFFT=128
0.008 -10
Magnitude (dB)
0.007 -20
0.006 -30
0.005
-40
0.004
-50
0.003
Rectangular
-60
0.002 Bartlett
Raised Cosine
0.001 -70
Better than Raised Cosine
Farchsech
0 -80
50 100 150 0 0.02 0.04
Samples Normalized Frequency (×π rad/sample)
Figure 2.6: The sketches of different windows with their respective spectrum
According to [17] the duration of the OFDM signal should be 2Tu in order to complete a total
of 2NFFT samples to preserve the orthogonality, and zeros are added in the region outside the
window g(t). If the duration of the OFDM signal is only Tu (1+ α ) then the window will
affect the transmitted signal, and as a consequence an error floor will be formed. Figure 2.7(a)
shows an example of a Bartlett window with an OFDM signal duration Tu(1+α) which affects
the transmitted signal. To avoid the error floor, the applied window must not influence the
signal during its effective period [18]. To fulfill this requirement, the window forms are
expanded, so that the new Tu becomes 2Tu , and α is restricted in the range of
⎛ TGI ⎞
⎜1 − ⎟
⎝ Tu ⎠
0 ≤α ≤ , where TGI is the guard interval duration. The orthogonality is preserved
2
since the PSD hits zeros in the 1/ (2Tu) interval at the expense of a longer duration of the
OFDM signal (2Tu (α+1)). The solution is that the window is truncated to fit the 2Tu OFDM
duration, as depicted in Figure 2.7(b) [6]. The orthogonality is preserved by applying the
rectangular receiver filter with Tu duration, that is implemented by the DFT [18]. In this way,
there is more freedom to shape the window in the T (1−α) ≤ t ≤ T region as long as its PSD at the
u u
symbol boundaries remains low, e.g Gaussian window or half sine window [19].
15
By referring to eq. (2.1) the de-activating carriers adjacent to the licensed band will provide
flexible guard bands that will make the PSD sidelobes of the RU’s OFDM signal at the
licensed band lower. In OFDM bits errors are typically concentrated in a set of severely faded
subcarriers. This is due to the channel propagation condition. For this reason, it is possible
that the quality of service may not be maintained. Meanwhile, the carriers resource is limited
due to de-activation of the RU carriers. Furthermore, the long OFDM symbol duration due to
the windowing reduces the OFDM transmission bit rate. Having adaptive OFDM in CR is
thus a solution. The heavily faded and noisy subcarriers are excluded from transmission to
improve the overall bit error rate (BER) of the system, and the loss of throughput is
counteracted by applying higher order modulation modes to the subcarriers which have better
signal to noise ratios (SNR) or BER.
g(t)
Zeros Zeros
Tu t
αTu αTu
2Tu
(a)
g(t)
2Tu
t
2αTu TGI Tu 2αTu
(b)
Figure 2.7: (a) Window design influencing the transmitted signal (b) Window design not
influencing the transmitted signal [6]
Recently in [20] overlapped OFDM symbol transmission with long symbol duration (known
as offset OFDM) has been introduced to counteract the throughput loss. The scheme is
described in Figure 2.8. The delay between one OFDM symbol and another is designed in
such a way that no inter symbol interference occurs, which means that the next OFDM
symbol should start after the end of the useful data part of the previous OFDM symbol. If
zeros are inserted at the prefix and suffix points (outside the TGI and TU area in Figure 2.7(b)),
the useful data of the next OFDM symbol should start after the last useful data of the previous
OFDM symbol. With this technique the accumulated symbol duration will be shorter
compared to non offset OFDM, hence, the bit rate loss can be reduced.
16
Figure 2.8: Overlapped OFDM symbol with zero ISI; technique applied at the transmitter site
(left) and at the receiver site (right)
2.3 OFDM Sidelobes Reduction Techniques

In this section a sidelobes reduction technique by manipulating the composition of Xn in (2.1)
is described. The results in [21]-[33] show that a significant sidelobe reduction gain can be
obtained.
2.3.1 Sidelobe Suppression by Cancellation Carriers Insertion
In [21]-[23] instead of having de-activated carriers adjacent to the LU band, these carriers are
used to cancel out the sidelobes on the LU band. The concept is to find the proper amplitude
of the cancellation carriers in such a way that the resulting sidelobes be as low as possible.
The optimization can be formulated as :
1 NFFT (1+α)Tu ccN

(1+α)Tu
∑ p (X dt + ∑ p ( Xc ) ∫
− j 2π ( f − fn )t − j 2π ( f − fcc )t
∫
2
X( f ) = ) g(t)e cc g(t)e dt
N FFT n=1
n≠cc
n n
α
−(1+ ) Tu cc=cc1
cc≠n
cc
α −(1+ ) Tu
NFFT
∑ p =P
2
subject to X( fLU ) < PThreshold, and T
n
n=1
(2.6)
where ccN is the total number of sidelobes cancellation carriers, and Xccc is the symbol on the
cancellation carrier position in the range cc1,…,ccN. The power constraint for all subcarriers
including the cancellation carriers remains, meaning that no extra energy is added. The
possible high power required by the cancellation carriers can be compensated from the unused
power due to the carriers de-activation.
17
We introduce an alternative optimization approach by simplifying it to a set of ccN linear

equations to solve ccN cancellation carrier amplitude values. Figure 2.9 provides a simple
example of how the cancellation carriers with this new approach work. The spectrum is
derived from 9 subcarriers (including the cancellation carriers) and uses a rectangular window
which produces a spectrum derived from a sum of sinc functions. One cancellation carrier is
located at the left edge of the RU band adjacent to the LU band and another one is at the right
edge. They are indicated by the dark vertical arrows. If we take a look at the OFDM spectrum
with carriers de-activation (one carrier at the left and one carrier at the right edge) denoted by
the solid curve, we can see that the maximum sidelobe occurs at a frequency point half of the
carrier spacing from the de-activated carrier which, in Figure 2.9 is indicated by the
coordinate position (X:-4.5,Y:-0.1564). The cancellation carriers can be weighted in such a
way that the sum of their spectra will make the OFDM spectrum zero at that particular
coordinate position (e.g. by having a cancellation carrier spectrum coordinate (Xc1=Xc2=-4.5,
Yc1=0.1477, Yc2=0.0087) ).
Original OFDM without Deact.and Sidelobes Canc. Carriers

OFDM with Deactivated Carriers
1.4 OFDM with Sidelobes Cancellation Carriers
Left Cancellation Carrier Spectrum
1.2 Right Cancellation Carrier Spectrum
1
Amplitude (X(f))
0.8
Xc1: - 4.5
0.6
Yc1: 0.1477
Xc2: - 4.5
0.4 Cancellation Carriers
Yc2: 0.0087
0.2
0
X: - 4.5
Y: - 0.1564
-0.2
-8 -6 -4 -2 0 2 4 6 8
Normalized frequency fTs
Figure 2.9: The OFDM spectrum amplitude with the effect of de-activated carriers or
cancellation carriers
The same technique is also applied to the sidelobes reduction on the right side. The resulting
spectrum derived from using the cancellation carriers technique is depicted by the dashed
lines. The sidelobes are lower than the sidelobe of the OFDM with the de-activated carriers.
18
The cancellation carriers can be combined with windowing [24]-[25]. Different windows
provide different sidelobe levels. Combining a window that gives very low sidelobes (e.g. the
farchsech window) with the cancellation carriers technique will lead to a very significant
sidelobe (interference) reduction in the LU band. The results in [24] have shown that the
combination of 2 cancellation carriers on each side adjacent to the LU band and raised cosine
window with roll off factor (α) 0.2 gives more than 30 dB sidelobes reduction compared to
the conventional OFDM.
2.3.2 Sidelobe Suppression by Subcarrier Weighting
Subcarrier weighting is introduced in [26]-[27]. The weighting can be seen as varying the
allocation power pn on each carrier m which, according to (2.1), will affect the spectrum
amplitude. By selecting the appropriate weight for each of the carriers the desired low
sidelobes can be obtained. The sidelobe reduction by means of carriers weighting is attained
at the expense of the bit error rate (BER) versus signal to noise ratio (SNR) degradation [28],
since the weighting will give unequal amounts of transmission power on each of the OFDM
carriers. The weighting method was originally used in [28] to counteract the high peak to
average power ratio (PAPR) problem in OFDM.
2.3.3 Sidelobe Suppression by using Multiple Choice Sequence
In principle multiple choice technique is applied by mapping an OFDM symbol into several
sequences and choose the one which gives the smallest sidelobe to be transmitted. In [29]
several multiple choice sequence techniques are introduced. Among the techniques are the
symbol constellation approach, the interleaving approach and the phase approach. The symbol
constellation approach is applied by mapping the symbol Xn in eq. (2.1) which is derived
from one of the symbols set of QAM or PSK mapping into a different value X’n (l) which is
still derived from one of the symbols set of QAM or PSK mapping. The index l ranges from
l=1,2,…,L, where L is the number of possible alternative values. This means that the bits can
be mapped into L possible symbols. By putting the NFFT symbols of X’n (l) into (2.1), and
observing from all L possible PSDs, the one with the lowest sidelobe is then chosen for
transmission. An example of simple mapping from Xn into X’n (l) is given in [29]. The CS
points of QAM or PSK are numbered as 0,1,….,CS-1, hence the symbol Xn is assigned to the
number In ( I n ∈ {0,1,..., CS − 1}) according to its point position, where CS is the constellation
size. The X’n (l) will be the symbol value on the new number In(l) which is derived from the
addition of the assigned number In with the random number Dn(l) which has L different
possible values [29],
I n (l ) = I n + Dn (l ) . (2.7)
By transmitting the information about the set index information l, the receiver can decode the
received signal and reconstruct the original Xn symbol.
19
By means of the interleaving approach, the NFFT Xn symbols are interleaved by L different
interleaver rules,
X’(l) = П(l)X, (2.8)
where П(l) is the permutation matrix that will do the interleaving, X is the symbol vector, and
X’ is the symbol vector after interleaving. The matrix П(l) has L different possible matrix
values and will be available at the transmitter and receiver. Among L different possible
OFDM spectra using a set of symbols X’(l), the one with the lowest sidelobes is chosen for
transmission. The receiver will use the de-interleaver permutation matrix П-1(l) to recover the
transmitted data upon receiving the side information index l.
In the phase approach each phase of symbol Xn is shifted by a random phase θn(l) with L
different possible values. The phase shifted symbol X’n(l) becomes [29],
X 'n (l ) = X n e
jθ n ( l )
(2.9)
where θn(l) is in the range of between 0-2π. In the vector notation, the phase shifted signal
X’(l) is derived from a point to point multiplication between the original signal X and the
complex number vector derived from ejθ(l). As in the constellation and interleaver approach
also, in the phase approach, with L possible sets for vector θ(l), the one that produces the
lowest sidelobes is chosen and applied to the symbol vector X. The index information l is
transmitted so that the receiver can reconstruct the symbol vector X. Apart from the random
shift, a random amplitude shift can also be applied, but in such cases the BER performance
will be affected. Another approach is to only apply the multiple choice sequence technique to
the carriers adjacent to the LU band, since these carriers mostly affect the sidelobes
magnitude [29].
2.3.4 Sidelobe Suppression by Adaptive Symbol Transition
Sidelobes Suppression by using a time domain approach has been introduced in [30]. A time
domain symbol transition is inserted between two OFDM symbol, and a symbol transition
value is obtained in such a way that the signal spectrum or the Fourier transform of the two
consecutive time domain OFDM symbol (including the symbol transition) will have low
sidelobes. The objective is to minimize the spectrum resulting from two consecutive time
domain OFDM symbols without symbol transition added by the spectrum contributed from
the symbol transition onto the LU band [30]. If the OFDM based cognitive radio signal
contribution on the LU is described as :
⎡ ym ⎤
⎢ ⎥
I = F LU ⎢ a m ⎥ (2.10)
⎢ y m + 1⎥
⎣ ⎦
20
where I is the signal part of the cognitive radio signal in LU band, FLU is the Fourier
transformation matrix for the signal in LU band, ym is the mth time domain OFDM symbol
including the guard interval, am is the additive symbol transition, and ym+1 is the next OFDM
symbol, then the optimization problem will be to find that vector am in making I as small as
possible. The symbol transition scheme is depicted in Figure 2.10.
The symbol transition will not introduce any inter symbol interference (ISI) but the
transmission duration becomes longer, which means that the bit rate is reduced. The search
for the time domain symbol transition values is characterized as a linear least squares problem
with a quadratic inequality constraint [30].
................. y(m) a(m) y(m+1) a(m+1) y(m+2) a(m+2) .................
Figure 2.10: Adaptive symbol transition scheme
2.3.5 Sidelobe Suppression by Frequency Domain Additive Signal
In [31] sidelobe suppression for OFDM based cognitive radio can be achieved by mapping the
data symbol using QPSK modulation mode and add the frequency domain symbol vector X
with random values vector d in such a way that the resulting PSD in the LU region become as
small as possible. The optimization problem is described by
2
1 N FFT (1+α )Tu
∑ p (X + d
− j 2π ( f − f n )t
∫
2
X( f ) = ) g (t )e dt
N FFT n =1
n n n
α
− (1+ ) Tu
, (2.11)
N FFT
∑p
2
subject to X ( f LU ) < PThreshold , and = PT
n
n =1
where dn is the additive signal on subcarrier n. In order not to destroy the cognitive radio BER
performance and keep the total allocated power constant, the value of dn is restricted to a
circle with magnitude | dn |≤ 0.7 [24]. In this way, according to the QPSK gray mapping, the
value of X’n = Xn + dn will not give a decision error at the receiver, except in a severely noisy
and/or faded channel. Simulation results in [31] have shown about 15 dB sidelobes reduction
could be achieved with the cost of 2.8 dB SNR degradation in case the upper limit value of |
dn | is 0.7.
21
2.3.6 Sidelobe Suppression by Constellation Expansion

A most recent technique in OFDM sidelobe suppression has been proposed in [32], where the
M-PSK modulation mapping is expanded to 2xM-PSK mapping in order to have lower
OFDM sidelobe. The original symbol after M-PSK mapping which is located at point a of M
available PSK symbol points is transformed into two possible new points a1 and a2 where the
location of a1 and a2 must not in the same region. The purpose is to have random selection of
points so there will be sidelobe pattern difference compared to the previous sidelobe using the
a point. By combining all possible new OFDM symbol set on the 2x M-PSK points through
all OFDM carriers, the one set which gives the smallest sidelobes will be chosen for
transmission. No additional information is required by the receiver for data detection, since
the receiver has the knowledge of all possible a1 and a2 positions. Therefore, the receiver can
detect the data by calculating the minimum Euclidean distance between the received symbol
and the reference points on 2xM-PSK mapping, and accordingly translates the result to the M
PSK mapping. Figure 2.11 shows an example on how the points from QPSK are expanded to
8-PSK mapping. In this way there are two possibilities of mapping a data symbol, and
accordingly the constellation expansion will add extra randomness to Xn in eq. (2.1) with
various possible sidelobes magnitude. The one which gives the smallest sidelobes will be
chosen to be transmitted. Simulation results in [32] have shown that the sidelobes suppression
technique by the constellation expansion from QPSK to 8-PSK gives only a slight BER
degradation. With this Constellation Expansion technique about 9 dB sidelobes suppression
can be attained [32].
The work has been extended in [33] by combining the constellation expansion technique with
the cancellation carriers inserted on the carriers adjacent to the LU band. Extra sidelobes
reduction has been achieved without having high magnitudes for cancellation carriers, due to
prior sidelobes reduction by the constellation expansion technique. According to [33] the
combination between constellation expansion and one sidelobes cancellation carrier on each
side adjacent to LU band gives around 16 dB sidelobes reduction.
d2
b a b1 a1
a2 c2
c d c1 d1
b2
QPSK 8-PSK
Figure 2.11: Constellation Expansion from QPSK to 8-PSK mapping
22
2.4 Frequency Hopping
Despite the application of spectrum pooling by carriers de-activation in the LU band and
sidelobes reduction by different techniques (windowing, additional carriers de-activation,
cancellation carriers, multiple choice on the transmitted data sequence, additive signal in time
or frequency domain, constellation expansion and combinations between them), a problem
arises if too many carriers need to be de-activated, as there will be not enough resources to
transmit data to fulfil the target bitrate. In [34] frequency hopping is proposed to deal with too
many carriers de-activation in a selected band. While frequency hopping is a proper solution,
another techniques (i.e. Ultra Wide Band), could be an alternative due to its signal power
property below noise level hence it can be overlapped with the LU signal itself. In this thesis,
because of the demand for a Cognitive Radio service which provides high rate transmission at
long distance, we put our attention to the frequency hopping scheme.
2. 4. 1 Evaluation Using GSM 900 MHz Channel Model

We have evaluated the proposed frequency hopping technique with TDD Bluetooth frame
design in simulations. The details about the OFDM frame design for frequency hopping
scenario and the hopping mechanism are available in Appendix A. The evaluations were
presented in [35]. The propagation models in this dissertation are taken from the ETSI GSM
model at 900 MHz [36]. The power delay profile for the rural area model is given in Table
2.1, and for the urban area model is given in Table 2.2. There are 2 alternative taps setting in
each model. The coherence bandwidth for the rural area model of the first alternative taps
setting is 1.6 MHz while the second alternative is 1.2 MHz. In the case of the urban area
model, the coherence bandwidth of the first alternative is 155 kHz while the second
alternative is 159 kHz.
Table 2.1: Propagation model for Rural Area GSM in 900 MHz [36]
Tap Relative Time (μs) Average relative

Number power (dB)
(1) (2) (1) (2)
1 0.0 0.0 0.0 0.0
2 0.1 0.2 -4.0 -2.0
3 0.2 0.4 -8.0 -10.0
4 0,3 0.6 -12.0 -20.0
5 0.4 - -16.0 -
6 0.5 - -20.0 -
23
The first alternative of the two area model suggests that the sampling time of the OFDM
signal should be 0.1 µs while for the second alternative 0.2 µs sampling time is sufficient in
order to have a sample spaced channel model and good channel estimation result. The worst
condition of the propagation model is found in the urban area propagation model where the
maximum delay (τmax) is 5 µs and we have a non constant tap spacing.
Since the frame duration must not exceed 366 µs, Table 2.3 shows the choices of OFDM
parameters that fulfils the 366 µs frame duration constraint. The guard interval (GI) must
exceed the worst case scenario in the urban area channel model with a maximum delay of 5
µs.
Table 2.2: Propagation model for Urban Area GSM in 900 MHz [36]
Tap Relative Time Average relative

Number (μs) power (dB)
(1) (2) (1) (2)
1 0.0 0.0 -4.0 -4.0
2 0.1 0.2 -3.0 -3.0
3 0.3 0.4 0.0 0.0
4 0.5 0.6 -2.6 -2.0
5 0.8 0.8 -3.0 -3.0
6 1.1 1.2 -5.0 -5.0
7 1.3 1.4 -7.0 -7.0
8 1.7 1.8 -5.0 -5.0
9 2.3 2.4 -6.5 -6.0
10 3.1 3.0 -8.6 -9.0
11 3.2 3.2 -11.0 -11.0
12 5.0 5.0 -10.0 -10.0
As depicted in Figure 2.1, the OFDM system requires a channel estimator module to remove
the effect of fading on the received signal. In order to aid the channel estimation process pilots
or trainings need to be inserted to the OFDM frame. The details about frame design including
the pilots or the training pattern, analysis and performance evaluations will be described in
detail in chapter 4.
Table 2.3: OFDM frame design according to TDD Bluetooth frame.
NFFT TS Δf # TG/
(μs) (kHz) OFDM TS
symbols
1024 204.8 4.9 1 1/32
512 102.4 9.7 3 1/16
256 51.2 19.5 6 1/8
128 25.6 39 11 1/4
24
2.5 Conclusions
In this chapter we discussed how OFDM can be considered as the proper technique for the
Cognitive Radio system in realizing the co-existence between the LU and RU. OFDM has the
capability to notch part of its spectrum which is occupied by the LU. This is carried out by de-
activating some of its carriers which fall in the LU band.
By sidelobes reduction techniques some part of the OFDM based Cognitive Radio signal
which are still smearing out into the LU band due to the OFDM power spectrum density
property, can be suppressed.
If too many OFDM carriers must be de-activated due to the LU access, the Cognitive Radio
system target data rate and quality of service (QoS) can not be attained. In this situation the
frequency hopping technique can be applied. Although Ultra Wide Band could be considered
as an alternative technology, frequency hopping is preferable in satisfying the Cognitive
Radio service demand in accommodating long range high data rate transmissions. As the
spectrum scanning module finds a proper band that can accommodate the target data rate and
QoS, the frontend part of the cognitive radio device will set its carrier frequency to that proper
band. An example of the Bluetooth TDD frequency hopping mechanism has been presented.
An OFDM frame according to the Bluetooth TDD slot has been proposed.
25
Chapter 3
Baseband Processing for OFDM-Based

Cognitive Radio
3.1 Introduction
Cognitive Radio according to reference [2] is an intelligent communication system with the
capability of changing its transmission parameters according to the environment condition
with the objectives to attain highly reliable communications and efficient utilization of the
radio spectrum.
The application of Cognitive Radio in this research work is focused on the spectrum
utilization efficiency by operating the Cognitive Radio system as rental user (RU) to co-exist
with the licensed user (LU). OFDM has been described in chapter 2 as the proper candidate to
realize this co-existing aim by de-activating some of its carriers at the LU positions and
applying sidelobes suppression techniques.
In this chapter beside the co-existence between LU and RU, the effort to obtain reliable
communications for the RU system itself is established. Section 3.2 will key out the
combination of spectrum pooling, windowing, adaptive bit loading and sidelobes cancellation
carriers. The purpose of the combination is to obtain the desired target data rate while the
required BER remains. Part of the RU spectrum in the LU band will be seen as interference
by the LU. This interference should be suppressed to a level which can be considered
negligible or tolerable by the LU.
Section 3.3 describes an alternative technique by replacing the Fourier transform in OFDM
with wavelet basis function that forms a wavelet packet multi carrier modulation (WPMCM)
technique. In this chapter a frequency selective wavelet to replace the Fourier transform in
OFDM to form WPMCM is applied§. The WPMCM is combined with the spectrum pooling
§
Initially at the IRCTR Madan K. Lakshmanan and H. Nikookar used a frequency selective wavelet basis
function for single carrier ultra wideband (UWB) modulation and applied pulse position modulation (PPM) as
data mapping technique. The work on WPMCM explained in this thesis was conducted in collaboration with
Madan K. Lakshmanan.
26
concept to have co-existence between the LU and RU. The particular reason why the
frequency selective wavelet is interesting to be applied in WPMCM is due to parameters in
designing the wavelet basis function such as frequency selectivity, regularity order, transition
bandwidth, and filter length that can be set in such a way that the target QoS will be achieved.
Conclusions on this chapter are summarized in Section 3.4.
3.2 Combined Spectrum Pooling, Windowing, Adaptive Bit Loading and

Sidelobes Cancellation Carriers
Conventionally, the modulation scheme in OFDM is constant over all data subcarriers for a
given mode. A critical situation that may occur in OFDM within this mode is when bit errors
are concentrated in a set of severely faded subcarriers, while in the rest of the OFDM
spectrum often no bit errors are observed. Such event raises the idea of having adaptive
OFDM, where subcarriers that will exhibit high bit error probability are excluded from
transmission. It will improve the overall BER with a potential slight loss in system
throughput. The potential loss in throughput due to the exclusion of faded subcarriers is
counteracted by employing higher order modulation modes on subcarriers which have a very
small BER. The data should be concentrated more on the sub channels with higher frequency
response gain, hence a higher SNR, and noisy sub channels should carry little or no data.
In transmitting such a preferred OFDM symbol, the transmitter needs to estimate the expected
channel condition at that time. The channel prediction may use the knowledge of the past
channel quality estimation. Such adaptive system can only operate efficiently in slow varying
channel conditions. If the communications between two stations is bidirectional and the
channel can be considered reciprocal (important for using Time Division Duplex (TDD) and
Frequency Division Duplex (FDD)), Tx/Rx station can estimate the channel quality based on
the received OFDM symbols and adapt the parameters of the local transmitter to this
estimation. This approach is named open loop adaptation, since there is no feedback between
the receiver of a given OFDM symbol and the choice of the modulation parameters. For the
case of a non-reciprocal channel, the stations cannot determine the parameters for the next
OFDM symbol’s transmission from the received symbols; the receiver has then to estimate
the channel quality and this perceived channel quality information has to be used in the
transmitter for the inverse link. This approach is called closed loop adaptation. The two cases
are described in Figure 3.1. The solid lines represent the data transmission, while the dashed
lines represent the inherent side information about the signal mode i.e. modulation, coding
parameters. The BER in each subcarrier is determined by the fluctuations of the current
frequency-domain channel transfer function Hn with the aid of channel transfer function
estimation provided by information extracted from the transfer of the pilot symbols. More
accurate measures of the channel transfer function can be gained by means of decision-
directed or time-domain training sequence based techniques. Another crucial thing to be
considered is the delay between the channel quality estimation and the actual transmission of
the OFDM symbol in relation to the maximum channel Doppler frequency. It will influence
the adaptive system’s performance. This delay in the closed-loop adaptive system is generally
27
longer than for an open-loop adaptive system, hence the Doppler frequency of the channel is
more critical in the closed-loop adaptive system.
(a)
transfer the transfer the
(b)
Figure 3.1: Signaling scenarios in adaptive modems [37] for (a) Open Loop Adaptation and
(b) Closed Loop Adaptation
Based on the channel fading information derived from the channel estimation process in
OFDM, the constellation size of each sub-carrier is determined. In order to demodulate the
transmitted data properly, the information of modulation mode in each carrier is required by
the receiver. This information can be derived by estimating it blindly [38],[39], or from the
transmitted signalling symbols [38].
28
The next subsection will describe how the signaling and blind estimation for the purpose of
modulation detection are applied including their pros and cons, followed by another
subsection discussing several available adaptive bit loading algorithms.
3.2.1 Modulation Detection by Signaling and Blind Detection
If the channel quality estimation and parameter adaptation have been performed at the
transmitter of a particular link, then according to the open loop adaptation, the resulting set of
parameters has to be communicated to the receiver in order to successfully demodulate and
decode the OFDM symbol.
In the closed loop case, the receiver determines the requested parameter set to be used by the
remote transmitter, the same amount of information has to become available in the remote
transmitter for the inverse link. If the signaling information is corrupted then the receiver can
not correctly decode the OFDM symbol corresponding to the incorrect signaling information.
Adaptive OFDM systems have to react to the frequency-selective nature of the channel, by
adapting the parameters across the subcarriers. The resulting signaling overhead can become
significantly higher, and can be prohibitive. In order to overcome these limitations, efficient
and reliable signaling techniques have to be employed for practical implementation in
adaptive OFDM systems.
If some flexibility in choosing the transmission parameters is sacrificed within an adaptation

scheme, the amount of signaling can be reduced. Alternatively, blind parameter detection
schemes can be devised, which require little or no signaling information. With respect to the
effects of transmission parameter adaptation in OFDM systems the following issues should be
considered: data buffering and latency due to varying data throughput, effects of co-channel
interference, and bandwidth efficiency.
For example in order to keep the system complexity low, the modulation mode is varied on
sub-band to sub-band basis. The total set of OFDM subcarriers are split into blocks of
adjacent subcarriers which means divided into several sub-bands. The same modulation
scheme is employed for all subcarriers in a sub-band. This will simplify the task of signaling.
The simplest way of signaling the modulation mode employed in a sub-band is to replace one
data symbol by an MPSK symbol, where M is the number of possible modulation modes,
hence the reception of each of the constellation points directly signals a particular modulation
mode in the current sub-band. For four modulation modes and assuming perfect phase
recovery, the probability of a signaling error ps(γ), when employing one signaling symbol, is
the symbol error probability of QPSK. The expression for the modulation mode signaling
error probability according to [37] is
p s (γ ) = 1 − (1 − Q ( γ ) )
2
(3.1)
29
where Q(.) is the probability Gaussian Function (x ) =

1
∞
−
t2
. The signaling error
Q
2π ∫e
x
2
dt
probability can be reduced by employing multiple signaling symbols and maximum ratio
combining of the received signaling symbols Rs,n in order to generate the decision variable R’s
prior to the decision [37], i.e.
Ns
R ' s = ∑ R s , n Hˆ s*, n (3.2)
n =1
where Ns is the number of signaling symbols per sub-band, the quantities of Rs,n are the
received symbols in the signaling subcarriers and Ĥs,n represents the estimated values of the
frequency-domain channel transfer function (CTF) at the signaling subcarriers. With the
assumption of perfect channel estimation and constant values of the channel transfer function
across the groups of signaling subcarriers, the signaling error probability for Ns signaling
symbols can be expressed as [37]:
p ' (γ , N ) = 1 − (1 − Q ( N γ ) )
2
s s
(3.3)
s
The signaling symbols for a given sub-band can be interleaved across the entire OFDM
symbol bandwidth, to benefit from frequency diversity in fading wideband channels.
Another way of signaling is by transmitting the signaling symbols from QAM mapping as
shown in Figure 3.2. where the allowable constellation size is restricted to the set of {0, 2,4,16
and 64}. The inner rectangle which is the decision area for the so called 0 modulation mode
3d
(no symbol is transmitted) can be made smaller up to a distance of about from the other
4
reference symbols; d is the distance between the outer points and the no symbol point (zero).
This is due to the fact that a decision error (e.g no symbol transmitted but either BPSK or
another modulation mode is decided) is just caused by noise (no contribution from channel
fading). By this technique the signalling error probability can be reduced since the minimum
Euclidean distance between the signalling points is high; this distance equals according to
Figure 3.2 to 7 2 (distance between no data is transmitted (0) and other modulation modes).
Blind detection algorithms estimate the employed modulation mode directly from the received
data symbols, therefore avoiding any loss of data capacity due to signaling subcarriers. There
are two algorithms for this blind detection. One is based on SNR estimation and the other one
is based on incorporating error correction coding. By SNR estimation, the receiver has no a
priori knowledge of the modulation mode employed in a particular received sub-band but
estimates this parameter by finding the closest distance between the received data symbols
after the fading removal ( Rn/Ĥn ) in that sub-band to all possible value of Rˆ n ,m in the chosen
modulation modes M for each subcarrier index n in the current sub-band. The index m defines
the index of possible points on the constellation map, and ranges from 1,…, M . The error
30
energy em between the ideal constellation phasor positions and the received phasors is
calculated for each modulation mode via [40]
2
⎛ ⎞
em = ∑ ⎜ R n − Rˆ n , m ⎟ , (3.4)
ˆ
n ⎝ Hn ⎠
Figure 3.2: An example of signaling symbol points for adaptive bit loading with allowable
M= {0, 2, 4, 16, and 64}
The modulation mode Mm, which minimizes em, is then chosen for the demodulation of the
sub-band. It is mentioned in [40] that there are two ways to conduct blind modulation
detection. They are pattern recognition and decision theoretic approaches. The pattern
recognition approach needs a relatively large number of samples available to make a decision.
This requires a large memory during computations. Therefore signaling is considered to be a
better alternative than the pattern recognition approach. There are three methods known for
use in the decision theoretic approach.
The first one is based on the average Euclidean distance, which is calculated as [40] :
31
em =
∑ Rs s
− Rˆ s , m
m = BPSK , QPSK ,... etc
S
(3.5)
Rs is the received sample on index sample s, R̂ is the hypothesis of the received sample, m is
the index of the modulation, and S is the number of samples used for averaging. The scheme
which minimizes the average Euclidean distance e is chosen for demodulation.
The second method is based on the Kullback-Leibler (K-L) distance algorithm. The K-L
distance is defined as the distance between any two probability distribution functions, f and g.
In the method, f is the probability distribution of errors from the received symbols, and g is
the distribution for the approximate model. The distance between the two distributions is a
measure of dissimilarity between them. It should be noted that the directed distance from f to
g is not the same as that from g to f. The K-L distance can be expressed as [40] :
L ⎛ p ⎞
I ( f , g ) = I ( p, π ) = ∑ p i log ⎜⎜ π i ⎟⎟ (3.6)
i =1
⎝ i ⎠
where pi is the ith true probability outcome, π1 to πL correspond to elements of the

approximating probability distribution. L is the number of possible modulation modes. Both,
Σpi and Σπi are equal to 1. f and g correspond to pi and πi, respectively. The notation I(f, g)
according to information theory is the mutual information between random variables f and g
and in this context is referred as the information lost when g is used to approximate f.
Since in adaptive modulation, different modulation schemes are used between specific SNR
bounds for the purpose of optimum performance, the probability density function (PDF)
approximation can be used to determine the SNR and then the modulation scheme can be
estimated at the transmitter. Assuming that for a set of K SNRs, the probability distributions
of errors, i.e the PDFs g1, ..., gK for K SNRs, are known, therefore K can be seen as the
number of samples (data for which their modulation modes have to be determined) to be
observed. The distribution of the errors of received noisy symbols (i.e. model f) can now be
calculated and allow us to estimate the K-L distances to all approximating PDFs g1 to gK. By
minimizing the K-L distance, the resulted SNR and transmitted modulation scheme can be
obtained.
The third method uses the mean-squared-error (MSE) between the elements f and g,
calculated from [40] :
MSE k = ∑ ( p i − π k ,i ) ,
L
2
1≤ k ≤ K (3.7)
i =1
32
By using this method, only a simple measure of the distance and not a weighted directed
distance is obtained. This is why K-L distance is well suited for SNR estimation. Two sets of
distributions may have the same MSE, but may result from different SNRs.
3.2.2 Adaptive Bit Loading
Adaptive bit loading is an intelligent bit allocation technique. The allocation is applied based
on the channel condition. More bits are allocated to the good channel and less bits to the bad
channel. Each bit loading algorithm has its own optimization criterion. The common required
information by the bit loading algorithms are the channel state information (H), the noise
variance (σn2) and interference variance (σI2) that refers to the signal to interference plus noise
ratio (SINR) on each carrier. This information can be acquired by sending periodic training
pilot symbols in an OFDM frame. When a channel estimation module produces the channel
estimates, the estimated noise ( Nˆ n ,l ) and interference ( Iˆn ,l ) on carrier n of OFDM symbol l
can be derived by subtracting the received signal (Yn,l) on carrier n of OFDM symbol l from
the multiplication of the channel transfer function estimate (Ĥn,l) and the pilot (XPn,l ) at that
same position, i.e.
Nˆ n,l + Iˆn,l =Y n,l − ( Hˆ n,l X nP,l ) = (( H n ,l X nP,l ) + N n ,l + I n,l ) − ( Hˆ n,l X nP,l )

= ( H n ,l − Hˆ n ,l ) X nP,l + N n ,l + I n ,l (3.8)

Δ n ,l
= Δ n ,l X nP,l + N n ,l + I n ,l
where Hn,l is the actual channel state, Δn,l is the channel estimate mismatch error, Nn,l and In,l
are the original noise and interference term on carrier n of OFDM symbol l. In this way the
multiplication of the channel mismatch error Δn,l and pilot XPn,l is treated as another source of
noise/interference. The noise and interference variances can be calculated from the collection
of the estimated noise and interference terms on carrier n of all training symbols in an OFDM
frame. If the channel states are perfectly estimated at the receiver, the value of Δn,l will be
zero.
The bit loading can be applied carrier-wise or group-wise. In the carrier-wise bit loading the
bits are allocated to a particular carrier according to the estimated channel state, noise and
interference variance for that carrier. In the group-wise bit loading the carriers are grouped
where each group consists of L carriers. The grouping can be based on the index numbering of
the carriers or derived from the sorting of the channel state gains (from the lowest up to the
highest one). The carriers with low channel state gain will be grouped with other carriers with
low channel state gains, while a carrier with high channel state gain will be grouped with
other carriers having similar channel gains. The bits to each group will be allocated according
33
to the average estimated channel state, noise and interference variance for that group. In the
following sub sections some existing bit loading algorithms will be overviewed and evaluated.
3.2.2.1 Chow Algorithm
This algorithm focuses on the rate distribution knowing the capacity of the subcarriers. The
goal is not to send as many bits as possible, but to send a fixed data rate most reliably by
maximizing the performance margin. The optimization problem can be described as follows
[41]:
Nc Nc
max γ m arg in subject to ∑ R n = R t arg et , ∑ p n = P budget , (3.9)
n =1 n =1
where γmargin is the system performance margin (in dB) that defines the maximum allowable
noise quantity on the system, Rn is the number of allocated bits in subcarrier n, Rtarget is the
targeted total number of bits per OFDM symbol, pn is the power allocation on subcarrier n,
and Pbudget is the total allocated power per OFDM symbol. Chow algorithm tries to maximize
the allowable amount of noise while it still can achieve the targeted bit error rate (BER). For
each subcarrier the bits are allocated based on the following equation [41]:
⎡ ⎤
⎢
= log 1 +
SNR n ⎥
R n 2 ⎢
Γ +γ ⎥
, (3.10)
⎣ m arg in ⎦
SNRn is the SNR in subcarrier n (in dB). Γ is the SNR gap (in dB) which is a constant which
estimates the difference between the channel capacity and the actual capacity usage by the
transmission scheme is defined as: Γ = (Q (SER/ 4))2 / 3 where SER is the expected symbol
−1
error rate, and Q(x) is the Gaussian probability function. γmargin is updated iteratively until the
total number of bits per OFDM symbol satisfies the target rate. Rn is quantized to RQn
according to the allowable constellation size of the modulation (i.e., 2 bits for 4-QAM, 4 bits
for 16-QAM, 6 bits for 64-QAM, and 8 bits for 256-QAM). The complete algorithm is given
in [41].
3.2.2.2 Fischer Huber Algorithm
This algorithm tries to guarantee that all subcarriers have the same SNR, considering that the
overall error rate is dominated by the subcarrier with highest error rate. The rate and power
are distributed in such a way that the error probability on each subcarrier is minimized. In a
34
QAM modulated system the optimization problem is minimizing the probability of symbol
error on subcarrier n (i.e., pn), which is defined as [42]:
⎛ d n ⎞⎟
p n = K n Q⎜⎜ 2 ⎟
(3.11)
⎝ 2 σn ⎠
where pn is the probability of error on subcarrier n, Kn is the number of nearest neighbors, dn

is the minimum distance between constellation points, and σ2n is the variance of the noise on
subcarrier n. The minimum BER is achieved when all subcarriers have the same error
probability. The number of bits per subcarrier per symbol Rn can have any real value. The
optimum bit distribution which maximizes the SNR under the condition that it is constant for
all subcarriers equals [43] :
⎡ ⎛ K −1 2 ⎞ ⎛ 2 ⎞⎤
1⎢ ⎜ σ k ⎟ ⎜ σn ⎟ ⎥ ,
Rn = 2⎜ ∏
R + log 2⎟
− K log 2⎜
(3.12)
⎢ ⎜ H ⎟⎟⎥
T 2
K ⎜ k =0 H ⎟
⎣ ⎝ k ⎠ ⎝ n ⎠⎦
where RT is the target rate, K is the number of active carriers, σ2n is the noise variance on
subcarrier n, and Hn is the channel gain on subcarrier n.
If Rn becomes negative for any n, the n-th subcarrier is excluded and the algorithm is applied
again over a reduced K subcarriers set. This is done iteratively until all remaining subcarriers
have a positive rate.
Rn is also quantified into RQn as described in the Chow algorithm. After all the subcarriers
have their allocated bits according to eq. (3.12), but if the target rate is still not obtained, then
addition or subtraction of bits is conducted based on the difference between Rn and RQn until
the total rate reaches the targeted bit rate.
The set of quantified rates must be chosen such that it minimizes the mean square error
between the optimum rate distribution Rn and the quantized distribution RQn, with the
constraint that Σn RQn = Rtarget. Because of the quantization procedure, the SNR is no longer
constant at the different subcarriers, as obtained for the continuous distribution of Rn. A
constant SNR can be guaranteed by adapting the power for the remaining subcarriers [43]:
σ
2
p =C n
2
2
RQn
, n∈D (3.13)
H
n
n
where D is the subset of active subcarriers, C is a power constant such that Σn pn = PT is

satisfied (PT equals the total transmitted power). If the noise is white, i.e., σ2n = σ2 for any n,
35
the algorithm depends on the channel gains |Hn| only. If the transmitter knows the channel
coefficients, the algorithm can be performed without requiring knowledge of the noise level at
the receiver.
3.2.2.3 Simple Blockwise Loading Algorithm (SBLA)
The given number of Nc subcarriers is divided into nB blocks of b adjacent subcarriers. For
block i, the algorithm determines a modulation level mi, such that at a given bandwidth
efficiency E ( in units [bit/subcarrier]), code rate R and given number of blocks nB the
following equation is satisfied [44]:
R n B −1
∑ mi = R.m = E . (3.14)
nB i =0
nB -1
1
where m = ∑ m . One key property of the SBLA is assigning the same modulation level
i
n B i =0
to adjacent subcarriers within one block. Due to the observation that in general the channel
gains and consequently the SNRs of neighboring subcarriers are (highly) correlated, no
noticeable performance degradations are expected by introducing a block of subcarriers,
which are smaller than the coherence bandwidth of the channel.
A second key property of SBLA is the low computational complexity, which is mainly based
on the usage of a predetermined SNR grid. The SNR grid contains a list of differences of the
SNR thresholds, required to switch between the modulation levels mi of the supported QAM
modulation alphabets for a specific block.
The choice of modulation level mi for a specific block of subcarriers is based on the (mean)
SNR for this block i.e., SNRi. The predetermined SNR grid (containing fixed interval sizes for
each modulation level mi) is used, as given for example in Figure 3.3. The absolute position
on the SNR axis, i.e. the absolute SNR threshold for each desired bandwidth efficiency E is
determined by shifting the grid according to the average modulation level m and the mean
SNR (SNRmean) for the given channel. (i.e. the mean SNR over all subcarriers). After first
allocation of bits per subcarrier which has been performed by usage of an SNR grid without
any iteration, it might be necessary to add or subtract bits in order to meet the desired
bandwidth efficiency E (i.e. data rate). The detailed steps of this method are as follows [44]:
• Initialization of the total number of subcarriers (Nc), the number of subcarriers (b) per
cluster, the average number of bits per subcarrier (m) over all subcarriers, and the
inverval sizes for the predetermined SNR grid according to the targeted bit error rate
BERtarget. The total number of bits to be distributed by the algorithm in one OFDM
symbol then follows as Rtarget = mNc.
36
• Determine the SNR for each block of subcarriers (SNRi) and the mean SNR (SNRmean)
over all subcarriers.
• Shift the SNR grid according to the average number of bits per subcarrier m and
SNRmean.
• Assign a modulation scheme (number of bits mi) to each subcarrier block i according
to the position of SNRi in the SNR grid.
• Calculate the total number of allocated bits as Rtotal = Σi mi.b
• As long as Rtotal ≠ Rtarget, add or subtract bits as in the Fischer algorithm and increase
or decrease Rtotal accordingly.
• Perform a scaling of the transmit power (equal for all subcarriers), depending on the
number of unused subcarriers.
Determination of the SNR grid is done by applying the OFDM transmission technique, the
frequency selective radio channel is transformed into K frequency non selective channels.
Thus, K parallel AWGN channels with different SNRs result. Therefore K can be seen as the
number of SNR level or the number of modulation modes applied on this adaptive bit loading
technique. With the aim to obtain the same BERtarget on all subcarriers, the BER curves for the
AWGN channel can be used to derive the SNR grid. An example of the SNR grid is shown in
Figure 3.4.
Figure 3.3 : Allocation of modulation schemes. The grid is shifted according to the average
number of bits per subcarrier (m=3 as example) and the mean SNR [44]
Figure 3.4 : An example of SNR gap for applying the simple block wise loading algorithm
[45]
37
3.2.2.4 Subband Bit Loading based on Fischer-Huber Algorithm
Equation (3.14) can be modified in order to have the bit allocation per subband. As given in
[46] the equation becomes:
R = R TOT
+
1
. log
⎛
⎜
⎜
2 ⎜
(H ) i
2 N ' B ⎞⎟
⎟
2 ⎟
(3.15)
i
N ' B N ' B
⎜ ∏ k ∈ψ H k ⎟
⎝ ⎠
where Hi denotes the channel gain of the i-th subband, which is the average value of the
channel gains of the subcarriers affiliated to that subband. ψ contains the set of indices of the
active subbands (subcarriers with non-negative values for Ri ), N’B is the number of active
groups in the set. The grouping of subcarriers can be based on their indices (successive
subband grouping) or based on the order of the channel gain (sorted subband grouping).
3.2.3 Spectrum Pooling and Sidelobes Cancellation Carriers Insertions
In previous chapter the windowing design was described. The window according Figure 2.7
(b) is applied in order to reduce the bit rate loss. The scheme can be seen as having
overlapped OFDM. Figure 3.5(a) depicts the overlapping time domain OFDM signal (with
duration 2Tu(α+1)) with overlap length 2αTu.
Another overlapped OFDM scheme is depicted in Figure 3.5 (b). The overlap length is
arranged in such a way that the useful signal in the area of TGI and Tu of one OFDM symbol
will not be overlapped with the useful signal of the following OFDM symbol. For this reason,
the value of the signals outside the TGI and Tu areas of the OFDM symbol (see Figure 3.5(b))
are filled with zeros.
Another approach to reduce the long symbol duration is by applying the sidelobes
cancellation carriers technique proposed in [21]-[23] which can avoid the long symbol
duration due to windowing. The sidelobes cancellation carriers are carriers that will cancel the
peak sidelobes at some optimization points in the LU band. The principle of sidelobes carriers
de-activation is described in Figure 3.6. By looking at the optimization area, we can choose
the number of side-lobe peaks that we want to nullify. Accordingly, the number of peaks to be
nullified will determine the number of sidelobes cancellation carriers required.
38
g(t)
2Tu
2αTu 2αTu
TGI Tu
(a)
g(t)
Tu
(Tu - TGI)/2
αTu (Tu - TGI)/2

TGI Tu TGI Tu
2αTu
(b)
Figure 3.5 : Window design (a) without influence to the transmitted signal in the period of
0 ≤ t ≤ T u (1 − α) with shorter overlap period and (b) with longer overlap period
2
In [21] and [22] the optimum number of cancellation carriers is found by observing the
resulting sidelobes from the OFDM power spectrum density with cancellation carriers through
simulations. The results revealed that more than 2 cancellation carriers on each border side
with respect to the LU band will not add significantly to the sidelobes reduction. The
sidelobes cancellation carriers will be used in the replacement scheme of de-activated carriers.
Choosing l peak points to be nullified requires a set of l linear equations that involves l
variables. The example in Figure 3.6 shows the problem of determining the amplitude of the 4
cancellation carriers (Xccc1, Xccc2, Xccc3, Xccc4 ). The amplitude value of these 4 cancellation
carriers will be obtained by solving 4 linear equations according to the general optimization
equation in eq. (3.16) for each frequency of the sidelobe peak i (i.e., fo(i)),
No
∑ [ Xc G ( f (i) − f )] + X
cc=cc1
cc cc o cc dd ( fo (i)) = 0 , i = 1,..., No (3.16)
where No is the number of optimization points ( in our case is 4), fcc is the frequency location
of the cancellation carrier cc, Gcc(f) is the window spectrum ( G(f) is derived from the Fourier
39
transform of the window function g(t)) of the the cancellation carriers cc on frequency f (if
rectangular window is used, Gcc(f) will be a sinc function), and Xdd(fo(i)) is the OFDM
spectrum without cancellation carriers (i.e. OFDM spectrum by replacing the cancellation
carriers with de-activated carriers) on the side-lobe peak i .
Cancellation Cancellation
Normalized Spectrum Amplitude (X(fTu))
Carriers Carriers
Xccc1 Xc Xc Xccc4
cc2 cc3
X (fo(2)) Xdd(fo(3))
dd
fo(1) fo(4)
fo(2) fo(3)
X (fo(1)) Xdd(fo(4))
dd
Optimization Optimization
Area Area
( LU Band ) ( LU Band )
Normalized Frequency (fTu)
Figure 3.6: Cancellation carriers technique by determining the side-lobe peak points to be
eliminated in the optimization area
Due to carriers de-activation in the LU band there will be some unused power that can be
allocated to the cancellation carriers in case high cancellation carriers amplitudes occur. The
remaining power can be allocated to the data signal in order to have a better data detection at
the receiver. The optimization problem in eq. (3.16) takes all cancellation carriers into
account. In case the distance between the cancellation carriers at one edge and the side-lobe
peak points at the other edge is large, then the influence of the cancellation carriers side-lobes
at that side-lobe peak point can be neglected (e.g. Xc1(fo(3)- fc1)≈ Xc2(fo(3)- fc2) ≈ Xc3(fo(2)- fc3)
≈ Xc4(fo(2)- fc4) ≈0 ). Therefore, the optimization can be separated by solving the equations
with variables involving the subset of cancellation carriers at one edge, and then solving the
equations with variables involving the second subset of cancellation carriers at the other edge.
The side-lobes reduction by the cancellation carriers can be seen in Figure 3.7. The spectrum
is derived from a rectangular windowed OFDM spectrum with 16 FFT points, 9 active
carriers located in the middle that makes 4 carriers on the left and 3 carriers on the right are
40
de-activated. Each active carrier is allocated with a BPSK symbol. Two cancellation carriers
are located at the left edge, and two are at the right edge. The peaks at the optimization points
are eliminated, thus lower side-lobes are achieved.
OFDM Spectrum with Deactivated Carriers

Left Cancellation Carrier 1 Spectrum
Left Cancellation Carrier 2 Spectrum
Right Cancellation Carrier 1 Spectrum
Right Cancellation Carrier 2 Spectrum
OFDM Spectrum with Cancellation Carriers
Spectrum Amplitude (X(fTu))
Left Cancellation Carrier 2 Right Cancellation Carrier 1
Right Cancellation Carrier 2
Left Cancellation Carrier 1
Area Area
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Normalized frequency fTu
Figure 3.7: The rectangular windowed OFDM spectra with sidelobe suppression by using
two cancellation carriers on each edge
Fig. 3.8 depicts the comparison between the side-lobes of a combined rectangular windowed
OFDM with de-activated carriers, cancellation carriers, and cancellation carriers combined
with de-activated carriers and other kind of windows. The OFDM parameters in Figure 3.8
are the same as the ones in Figure 3.7. The duration of the other windows (non-rectangular) is
2Tu(α+1), hence the spectrum is orthogonal in the frequency period of 1/ (2Tu) and the useful
signal is not deteriorated by the window. The sidelobe of the combined Tu duration
rectangular window with the sidelobes cancellation carriers in the optimization area is
considerably lower than for the other form of windows. By observing the outcomes of having
lower side-lobes and shorter symbol duration, the combination of OFDM with de-activation
of carriers in the LU band and sidelobes cancellation carriers located adjacent to LU band is a
promising modulation technique to be applied in Cognitive Radio.
Without the sidelobes cancellation carriers, the application of combined adaptive bit loading
and spectrum pooling is restricted by the allowable side-lobes power (PThreshold ) in the LU
41
band. Inserting the sidelobes cancellation carriers will ease the application of adaptive bit
loading since each sidelobes cancellation carrier will make sure that the OFDM based CR
sidelobes will be reduced. Hence the adaptive bit loading algorithm can be optimally utilized.
Therefore, the sidelobes increase in the LU band due to the allocated bits will be counteracted
by the insertion of cancellation carriers.
OFDM Spectrum with Deactivated Carriers

OFDM Spectrum with Cancellation Carriers
OFDM with Bartlett Window 2 T
0.35 u
OFDM with Raised Cosine Window 2T
u
0.3 OFDM with BTRC Window 2T

u
Normalized spectrum amplitude (X(fT u))
0.25
0.2
0.15
0.1
0.05
-0.05 Area Area
-0.1
-10 -8 -6 -4 -2 0 2 4 6 8 10
Normalized frequency fTu
Figure 3.8: Spectrum side-lobe comparisons between rectangular windowed OFDM with de-
activated carriers, cancellation carriers, and, cancellation carriers combined with different
kind of windows (α =0.25) and de-activated carriers
3.2.4 Combined Adaptive Bit Loading with Spectrum Pooling and Sidelobes
Cancellation Carriers Simulation Results and Analysis
In the initial work we applied combined adaptive bit loading with Spectrum Pooling. The
spectrum pooling was applied carefully in such a way that the mutual interference between
the LU and the Cognitive Radio system is very low (negligible). The simulation parameters
are given in Table 3.1. We hereby assume perfect synchronization between transmitter and
receiver, and the receiver can detect the modulation mode on each carrier correctly.
42
Figure 3.9 shows the performance comparison between several adaptive bit loading schemes
and the Better Than Raised Cosine (BTRC) window with α =21/128, under the condition of
perfect channel estimates. It is shown that the Fischer algorithm is superior to other bit
loading algorithms. At a BER=10-2, around 5 dB gain is then achieved [47], [48]. Since the
window shape will not affect the BER performance of the OFDM system (see Figure 3.5(a)),
it can be assumed that the the gain remains for other window forms such as Raised Cosine and
Bartlett.
Conv. OFDM - ideal est - BTRC

Fischer - ideal est - BTRC
Chow - ideal est - BTRC
-1
10 SBLA - ideal est - BTRC
Fischer groupwise - ideal est - BTRC
BER
-2
10
0 5 10 15 20
SNR (dB)
Figure 3.9 : BER vs SNR (dB) comparisons of Adaptive Bit Loading algorithms using Better
Than Raised Cosine Window form according the design in Figure 2.3(b) with α =21/128
(TGI /Tu) and allowable modulation modes : BPSK, 4-QAM,16-QAM, 64-QAM, and 256-
QAM
In the case of an imperfect channel estimate after Wiener filtering, the adaptive bit loading
performance will degrade. The results show that the SNR degradation of the Fischer
algorithm with Wiener filtering is about 7 dB at a BER of 10-2 compared to the case with
perfect channel knowledge as depicted in Figure 3.10 [47]. With respect to the CR sidelobes,
its impact to the CR BER will be manifested if the guard band between LU and CR is not
really implemented; hence, the I term in eq. (3.10) will exist and should be treated as extra
noise that (according to the available bit loading algorithm) will be counteracted by reducing
the bit allocations on the carriers contaminated by those interferences for the purpose of
maintaining the targeted BER.
43
Table 3.1: Simulation Parameters
OFDM Parameters Channel Parameters

Carr. freq. 5.5 GHz Max. channel delay 5 μs
# of carriers 256 # of Taps 11
Carr. spacing 31.25 kHz Taps distance 0.5 μs
Guard time 5.25 μs #. of Paths 12
Bit Rate 3 Mbps Veh. speed 100 Km/hr
Chn. coding OFF Fading Model Rayleigh
An imperfect channel estimate denotes an imperfect frequency shift recovery by the channel
estimation module due to the Doppler effect, resulting into an intercarrier interference (ICI)
effect. The higher the sidelobes the larger the ICI which will degrade the BER performance.
By decreasing the sidelobes the impact of ICI will be reduced. The phenomenon can be
analyzed from the received signal on subcarrier k (Yk) of an OFDM symbol after fading
removal by utilizing the estimated channel transfer functions (CTFs):
N FFT j 2 πf s t ( 1+ α ) Tu
Yk = ∑ X n H n je2 πfˆ t ∫ g (t ) e− j 2 π ( f k − f n ) t dt
Hˆ n e
s
n =1 − ( 1+ α ) Tu
( 1+ α )T u
N FFT
= ∑ X n An e j 2 πΔf st ∫ g (t ) e− j 2 π ( f k − f n ) t dt
n =1 − ( 1+ α ) Tu
N FFT ( 1+ α ) Tu
= ∑ X n An ∫ g (t ) e− j 2 π ( f k − f n − Δf s ) t dt
n =1 − ( 1+ α ) Tu
( 1+ α )Tu Tu
j 2 πΔf s t
N FFT ( 1+ α )
= X k Ak ∫ g (t ) e dt + ∑ X n An ∫ g (t ) e− j 2 π ( f k − f n − Δf s ) t dt
n =1
T
− ( 1+ α ) u − ( 1+ α )T u

n≠k

Desired Signal
ICI Term
(3.17)
where g(t) is the window function, NFFT is the number of FFT points (number of subcarrier),
Hn, Ĥn, and fn are the CTF, estimated CTF and frequency of subcarrier n respectively , fs is
the carrier frequency shift due to the channel fading (can be seen as Doppler effect), fˆs is the
estimated carrier frequency shift, Δ f = f − fˆ , and An= Hn/ Ĥn. The first term in eq. (3.17)
s s s
is the desired signal while the second term is the ICI contribution. By observing eq. (3.17) we
44
can see that the window function g(t) plays a role in determining the ICI quantity. Therefore
by proper selection of a window g(t) with low sidelobes, the ICI effect can be minimized.
The BER results of the OFDM-based CR system employing combined adaptive bit loading,
spectrum pooling and cancellation carriers will have the same tendency as shown in Figure
3.9, where carrier-wise bit loading using the Fischer-Huber algorithm outperforms the other
algorithms since the useful signal will not be affected by the window or by the sidelobes
cancellation carriers. The carriers close to the border with the LU band would be the most
interfered carriers by the LU access due to the LU signal side-lobes, but this condition will be
anticipated by adaptive bit loading algorithm and by taking into account the interference term
as described in eq. (3.10). The bits allocated to these carriers (close to the border with the LU
band) will be small, and a small constellation size is more robust against interference and
noise.
Fischer - ideal estimate

Fischer - Wiener Filter
Chow - Wiener Filter
-1
Groupwise Fischer - Wiener Filter
10 Groupwise sorted Fischer - Wiener Filter
BER
-2
10
0 5 10 15 20
SNR (dB)
Figure 3.10: BER vs SNR (dB) for Adaptive OFDM after a Rect Wiener Filter Channel
Estimation; allowable modulation modes: BPSK, 4-QAM,16-QAM, 64-QAM, and 256-QAM
An important issue in applying combined adaptive bit loading with spectrum pooling and
cancellation carriers into an OFDM-based CR is the interference quantity (sidelobes) that the
CR system will cause to the neighboring LU band. We have discussed that the adaptive bit
loading will take care of the constellation size of carriers located at the border with the LU by
considering the LU sidelobes quantity on those carriers. The worst condition is when the
interference by the LU to the OFDM based CR carriers at the border stays undetected; further
also, if the sidelobes contributions of all OFDM-based CR carriers on one reference peak in
the LU band have the same phase, and each of them has maximum symbol values that will
add together constructively and so making a high sidelobe peak.
45
In order to further explain the impact of adaptive CR, we start from a normalized spectrum
amplitude example of an OFDM-based CR with 256 carriers. 36 carriers in the middle
(carriers 111 -146) are occupied by the licensed user. Totally 40 carriers are de-activated, that
means carriers 1-108 and 149-256 are active carriers. If 2 sidelobes cancellation carriers are
inserted on each side of the border with the LU band then they will be located on carriers
109,110, 147 and 148. If only 1 sidelobe cancellation carrier is inserted then only carriers 109
and 148 are used. As shown in Figure 3.8 the maximum sidelobe peak will normally occur
within a normalized frequency distance of 0.5 to the closest sidelobes cancellation carrier on
the LU band. (e.g on position 110.5 for the 2 sidelobes cancellation carriers insertion and on
position 109.5 for the 1 sidelobe cancellation carrier insertion). The worst case will be if all
the carriers will apply 256 QAM. The 256 QAM modulation gives a high error vector
magnitude (EVM) compared to the lower constellation sizes and as a consequence gives the
highest impacts on the BER performance [49]. This is an important reason why this
constellation size is not used in any practical standard. However, in this chapter we only try to
observe the impact of such worst case situation on our OFDM based CR signal. Putting
fTu=110.5 and also fTu=109.5 into eq. (2.1) and having g(t) as rectangular window, then the
maximum peak will occur if the odd numbered carriers produce a negative symbol QAM
point with the highest amplitude ( in 256 QAM this is the symbol on the -15 –j15 point)
while the even numbered carriers produce a positive symbol QAM point with the highest
amplitude (in 256 QAM this is the symbol on the 15 + j15 point). The sidelobes in the region
of fTu=109.5 untill fTu=147.5 for the OFDM with de-activated carriers, 1 sidelobe
cancellation carrier and 2 sidelobes cancellation carriers cases are depicted in Figure 3.11.
The roll-off factor α=0.2.
As a consequence of applying the adaptive bit loading, there is the possibility that the
sidelobes cancellation carriers require high power as can be seen in Figure 3.11. This power
requirement can be compensated as the unused power due to carriers de-activation in the LU
band can be allocated to those sidelobes cancellation carriers. Further, by combining the
sidelobes cancellation carriers with time domain windowing as suggested in [22] and [23] (e.g
Raised Cosine, BTRC, or farcsech ), the amplitude of the sidelobes cancellation carriers can
be further reduced. Figure 3.11 depicts the impact of the raised cosine window in avoiding
high amplitude of sidelobes cancellation carriers compared to sidelobes cancellation carriers
combined with rectangular window. The results in [24] have shown that the combination of 2
cancellation carriers on each side adjacent to the LU band and a raised cosine window with a
roll-off factor α=0.2 gives more than 30 dB sidelobes reduction compared to the conventional
OFDM. In comparison with our observation in the worst case scenario, the combination
between 2 sidelobes cancellation carriers on each side adjacent to the LU band and the raised
cosine window with the same roll-off factor value gives a sidelobes reduction more than 30
dB, even more than 40 dB compared to the rectangular window with carriers de-activation as
depicted in Figure 3.11(b).
With respect to the peak to average power ratio (PAPR) problem in OFDM, the sidelobes
cancellation carriers contribution to the PAPR growth is described as given in eq. (3.18),
46
⎛ ⎛ N FFT −1 ⎞
2
⎞
⎜ 1 2 π kn N cc 2 π kcc ⎟
⎜ + ∑ Xc cc e N FFT ⎟
⎜⎜ n∑
j j
max k ⎜ X n e N FFT ⎟⎟ ⎟
⎜ N FFT
=0
⎝ n ≠ cc
cc = cc 1
⎠ ⎟
PAPR ( x [ k ]) = ⎝ ⎠
⎡ N FFT −1 2 π kn
2
⎤
⎢ 1 ⎥
∑ X e
j
E x ⎢ n N FFT
⎥
⎣⎢ N FFT
n=0
⎦⎥ (3.18)
⎛ ⎛ N FFT −1 ⎞
2
⎞
⎜ 1 2 π kn N cc 2 π kc ⎟
⎜ ⎟
⎜⎜ n∑ ∑ Xc e
j j
max k ⎜ X n e N FFT + cc N FFT ⎟ ⎟
⎜ N FFT
=0
⎝ n ≠ cc
cc = cc 1 ⎟
⎠ ⎟
= ⎝ ⎠
⎡ N FFT −1 2 π kn
2
⎤
⎢ 1 ⎥
∑ X e
j
E x⎢ n N FFT
⎥
⎢⎣ N FFT
n=0
⎥⎦
where x(k) is the time domain OFDM signal for the k th sample, Xccc is the symbol value on
the sidelobes cancellation carriers position cc, Ex is the averaging operator, cc1 is the first
sidelobes cancellation carrier, Ncc is the total number of sidelobes cancellation carriers being
inserted, while the rest of the symbols have been explained before (in eq. (2.1)). By observing
eq. (3.18) we guess that the sidelobes cancellation carriers will not give a significant
additional PAPR growth. [ The peak power in the time domain signal is the result of
accumulated coherent phase of the multiplication between the frequency domain OFDM
2 πkn ⎛ 1 N FFT ⎞
symbol (Xn) and the factor e j N FFT ⎜ xk = ∑
N FFT n =0
X n e j 2π nk ⎟ ]. We predict that less than 2 %
⎝ ⎠
peak power contribution comes from the sidelobes cancellation carriers ( assuming NFFT=256,
Ncc=4). It is possible that this 2% contribution raises a higher PAPR. The evaluation in [23],
based on simulations, shows that the sidelobes cancellation carriers give only a slight PAPR
increase.
To check this juncture we have simulated random symbols from a 256 QAM constellation
mapping (in order to demonstrate the effect of adaptive bit loading) applied in combination
with sidelobes cancellation carriers and windowing. The OFDM parameters remain the same
with the parameters used in producing Figure 3.11. The complementary cumulative
distribution function (CCDF) graphs of the PAPR for adaptive OFDM with de-activation
carrier and rectangular window, with addition 2 cancellation carriers on each edge, or 2
cancellation carriers combined with raised cosine windows are given in Figure 3.12. The
Figure indicates an insignificant impact by the sidelobes cancellation carriers and its
combination with low sidelobes time domain windowing, in this case the Raised Cosine
window was used to the PAPR growth.
47
Adaptive OFDM with carriers deactivation

Adaptive OFDM with 1 canc. carrier + Rect. Window
40 Adaptive OFDM with 2 canc. carriers + Rect. Window
Power Spectrum Density ( |X(f)| ) (dB)
Adaptive OFDM with 2 canc. carriers + Raised Cosine Window

Optimization area for
2
20
sidelobes reduction
-20
-40
-60
-80
107 111 115 119 123 127 131 135 139 143 147 150
Normalized Frequency (fTs)
(a)
Adaptive OFDM with carriers deactivation
Adaptive OFDM with 1 canc. carrier + Rect. Window
40 Adaptive OFDM with 2 canc. carriers + Rect. Window
Power Spectrum Density ( |X(f)| ) (dB)
Adaptive OFDM with 2 canc. carriers + Raised Cosine Window
20
Optimization area for
2
sidelobes reduction
0
-20
-40
-60
-80
108.5 109 109.5 110 110.5 111 111.5 112 112.5 113 113.5 114
Normalized Frequency (fTs)
(b)
Figure 3.11 : Spectrum sidelobe comparison between the rectangular windowed OFDM with
de-activated carriers; 1 sidelobe cancellation carrier, 2 sidelobes cancellation carriers at the
border with the LU band and 2 sidelobes cancellation carriers with a raised cosine windowed
OFDM in combination with adaptive bit loading in the worst case scenario with α=0.2 , (a)
the complete optimization area and (b) left part of the optimization area
48
0
10
Prob(Symbol PAR > value)
-1
10
-2
10
-3
10
-4
10
-5
10 AOFDM w/ Carriers Deact + Rect. Window
AOFDM w/ Carriers Deact + 2 Cancell. Carriers + Rect. Window
AOFDM w/ Carriers Deact + 2 Cancell. Carriers + Raised Cos. Window
-6
10
5 6 7 8 9 10 11 12
Peak to Average Ratio (dB)
Figure 3.12 : PAPR CCDF comparison of a) Adaptive OFDM with a carriers de-activation
and Rectangular window; b) combination of Adaptive OFDM with a carriers de-activation
and 2 cancellation carriers adjacent to the LU band and using a Rectangular window, and
c) combination of Adaptive OFDM with a carriers de-activation and 2 cancellation carriers
adjacent to the LU band and using a Raised Cosine window
It has been discussed in section 2.2 that in order to have a window that will not influence
(deteriorate) the bit rate, the useful signal duration should be 2Tu; a longer signal will reduce
therefore the transmission bit rate. The windowing strategy according to Figure 3.5 (b) tries to
reduce the total OFDM frame duration as low as possible, therefore this window would be a
promising alternative to attain the targeted bit rate.
PAPR is a common problem in every OFDM system, therefore, the current available
conventional OFDM-PAPR reduction techniques can be applied in the Cognitive Radio
context. According to the analysis, the OFDM based CR bit rate and sidelobes reduction
capability will be the trade-off optimization problem in the CR system.
49
3.3 Wavelet Packet Modulation by Frequency Selective Wavelet for

Cognitive Radio**
The Fourier transform in OFDM uses Cosine and Sinusoidal waveforms as its transformation
function. These Cosine and Sine can be replaced by wavelet basis function. Hence a new
modulation technique termed as wavelet packet multi carrier modulation (WP-MCM) is
formed. In this way, there is a degree of freedom in shaping the spectrum of the wavelet based
OFDM. In the context of Cognitive Radio, the spectrum shaping is crucial. The Cognitive
Radio spectrum should be shaped in such a way that it will not disturb the legacy system
signal. The co-existence between the Cognitive Radio and the legacy system should be
preserved. The flexibility in shaping the wavelet-based OFDM spectrum makes wavelet based
OFDM interesting to be explored further.
The two fundamental tasks of the WP-MCM cognitive system are – 1. Radio spectrum
analysis to gather intelligence on spectrum holes and occupied bands, and 2. WP-MCM
based adaptive data transmission on idle bands. At the transmitting end, an incoming high-rate
serial data stream is split into M lower-rate parallel streams. The data in each parallel branch
is then upsampled by M and used to modulate M subcarriers. Meanwhile, the spectrum
estimator gauges the channel and performs a radio scene analysis to estimate LU frequency
bands and detect spectrum holes. Based on the spectrum estimates, the cognitive modules
dynamically de-activate those sub-channels of the WP-MCM system that lie in and around the
spectrum of the LU. The idea is to dynamically sculpt the CR transmission signal so that it
has no or very little time-frequency components competing with the LU. In this way the CR
can seamlessly blend with the LU operation. The sub-carriers are then modulated and scaled
to the desired energy level to obtain the WP-MCM transmission signal. The transmitted signal
in the discrete domain, y[k], is composed of successive modulated symbols, each of which is
constructed as the sum of M sub-carriers, φm[k], individually modulated. It can be expressed
in the discrete domain as:
M −1
y[k ] = ∑ ∑ (a
s m =0
s ,m ϕ m[ k − sM ]) (3.19)
where as,m is a constellation encoded sth data symbol modulating the mth waveform. To
reduce the error probability, the sub-carriers are made orthogonal. The orthogonality of the
subcarriers is defined as [50]:
< ϕ m [ k ], ϕ n [ k ] >= δ [ m − n ] (3.20)
**
This work is the result of collaboration with M.K. Lakshmanan. The theoretical content of this section is
selected from the papers [6], [62], [65] and the paper from M. K. Lakshmanan and H. Nikookar, “Optimum
Wavelet Packet Construction for Multicarrier Based Cognitive Radio system”, published in Proc. Int. Symp. on
Wireless Personal and Multimedia Communications (WPMC), Finland, Sept. 2008.
50
where, < ⋅ > represents an inner product operation and δ [⋅] is the dirac-delta function. In
j 2π mk
OFDM, the discrete functions φm[k] are complex basis functions g[k] e M limited in the time
domain by the window function g[k]. In the WP-MCM scheme, the subcarrier waveforms are
obtained by successively decomposing a couple of elementary paraunitary filters [50]-[52].
The nature of the subcarriers therefore depends on the properties of the underlying filters
used; therefore filters with suitable properties have to be developed. More on this will be
presented in the following subsection. The steps to generate the WP-MCM transmission
signal are elucidated in Figure 3.13.
To detect the transmitted data, the received signal r[k ] is de-convolved with a receiver sub-
~ ~
carrier waveforms, ϕm[k] . The sub-channels ϕm[k] are the duals of φm[k] and their
orthogonality can be expressed as
~
< ϕm [k ],ϕn [k ] >= δ [m − n] (3.21)
^
The estimate of each subchannel a s ,m is derived from
^ ~
a s,m = (r[k]*ϕm[k − sM ]) ↓M (3.22)
Here the down arrow represents decimation by M while * is the convolution operation. These
estimates are then de-mapped and parallel to serial converted to identify the transmitted data.
WP-MCM Module Cognitive Module

Design of Paraunitary Filter Radio Spectrum Analysis
Banks And Spectrum Pooling
Generation of Wavelet Packet Information on Idle and

Sub-carriers and their duals Active Licensed User Bands
Deactivate affected WP-MCM Sub-Carriers
Construct WP-MCM Transmission Waveform

with NO time-frequency components competing
with Licensed User
Figure 3:13: Fundamental Blocks of WP-MCM based Cognitive Radio system

51
3.3.1 Generation of Wavelet Packet Sub-Carrier Bases
The subcarrier signal waveforms in traditional MCM implementations, such as OFDM, are
sine / cosine basis functions. In WP-MCM the sub-carrier waveforms are derived from poly-
channel tree structures built by cascading multiple two-channel filter banks like the one
shown in Figure 3.14.
Analysis Filter Banks Synthesis Filter Banks
H 2:1↓ 1:2↑ H’
G 2:1↓ 1:2↑ G’
Figure 3.14: Two channel filter bank analysis with analysis filters H and G (low and high
pass, respectively) and synthesis filters H’and G’ (low and high pass, respectively)
A two-channel filter bank consists of a set of 4 perfect reconstruction filters (2 high pass and 2
low pass) which allow the decomposition and reconstruction of a signal without amplitude or
phase or aliasing distortion [53]. The two-channel filter bank has the property of splitting the
signal into two lower resolution versions – namely the coarse (low pass) and the detail (high
pass). When the decomposition into coarse and detail components is continued iteratively, it
leads to the generation of wavelet packet bases. When the perfect reconstruction filters used
satisfy an additional property known as the para-unitary condition (explained later in section
3.3.3), they lead to wavelet packet bases with impulse responses that are mutually orthogonal
to one-another and to their duals [53]-[55].
The wavelet packet sub-carriers (to be used at the transmitter end) are generated through a
multi-channel filterbank consisting of cascaded two-channel filters applying the synthesis
filters (H’ and G’). This represents an inverse discrete wavelet packet transformation
(IDWPT) and consists of binary interpolation (up-sampling) by 2, filtering and recombination
at each level. The number of iterations J determines the number of subcarriers M generated
and the relationship is given as M ≤ 2 J .The time domain representation of the wavelet packet
bases ψ i [k ] is obtained through a simple convolution rule.
ψ i [k ] = w(k )* w(k / 2)*....* w(k / 2J −2 )* w(k / 2J −1 );

where,0 ≤ i ≤ 2J − 1 (3.23)
⎧⎪h '(k ),for lowpass branches
and, w(k ) = ⎨
⎪⎩ g '(k ),for highpass branches
52
Here h’ and g’ stand for the impulse responses of the low and high pass synthesis filters,
respectively.
3.3.2 Generation of Wavelet Packet Dual Bases
The wavelet packet duals (to be used at the receiver end) are obtained from multi-channel
filter bank analysis too, though the processes are reversed. The duals are obtained from the
analysis filters (H and G) through the analysis procedure which consists of filtering,
decimation (down sampling) by 2 and decomposition at each stage. This process is called
discrete wavelet packet transformation (DWPT). First the signal is passed through a half-band
high and low pass filter. The half-band low pass filter removes all frequencies that are above
half of the highest frequency, while the half-band high pass filter removes all frequencies that
are below half of the highest frequency of the signal. Such a half-band filtering halves the
resolution, but leaves the scale unchanged. The signal is then sub-sampled by two since half
of the number of samples is redundant, according to the Nyquist’s rule. This decomposition
halves the time resolution since only half the number of samples then comes to characterize
the entire signal. Conversely, it doubles the frequency resolution, since the frequency band of
the signal spans only half the previous frequency band effectively reducing the uncertainty by
half. This procedure is iteratively repeated till the desired degree of resolution. The duals
~
ψ i [k ] are derived from:
~
ψ i [k] = f (k)* f (2k)*....* f (2J −2 k)* f (2J −1 k);
where,0 ≤ i ≤ 2J −1
(3.24)
⎧⎪h(k),for lowpass branches
and, f (k) = ⎨
⎪⎩g(k),for highpass branches
In (3.24) h and g denote the impulse responses of the low and high pass analysis filters,
respectively. Figure 3.15 illustrates the derivation of 8 wavelet packet bases and their duals
from a cascaded level-3 tree structure. Accordingly by following the rules of obtaining the
wavelet packet subcarrier bases in (3.23) and the duals according to (3.24), the Mallat
reconstruction and decomposition filterbanks [56] are depicted in Figure 3.16.
Due to the convolution process between the parallel data stream with the filter bank, the
resulting output of the time domain signal per WP-MCM symbol will be longer than the
number of the subcarriers. Therefore during the transmission, the WP-MCM symbols will
overlap. The overlapping distance should be designed in such a way that the total number of
samples in one WP-MCM frame will be equal to the multiplication between the number of
subcarriers and the number of WP-MCM symbol per frame.
53
2↑ H’
2↑ H’
2↑ G’
2↑ H’
2↑ H’
2↑ G’
2↑ G’
2↑ H’
2↑ H'
2↑ G’
2↑ G’
2↑ H’
2↑ G‘
2↑ G’
(a)
H 2:1↓
H 2:1↓
G 2:1↓
H 2:1↓
H 2:1↓
G 2:1↓
G 2:1↓
H 2:1↓
H 2:1↓
G 2:1↓
G 2:1↓
H 2:1↓
G 2:1↓
G 2:1↓
(b)
Figure 3.15 Generation of wavelets. (a) A level-3 tree gives 8 wavelet packet bases for the
transmitter and their (b) duals for the receiver. The up arrows represent interpolation by 2, the
down arrows represent decimation by 2. H’ and G’, denote the frequency responses of the low
and high pass reconstruction filters, while H and G denote the frequency responses of the low
and high pass decomposition filters, respectively
54
8 ψ0 (k ) ψ′′0 (k ) 8
8
ψ1 (k ) ψ′′1 (k ) 8
8 ψ7 (k ) ψ′′7 (k ) 8
Reconstruction Decomposition
Figure 3.16: An Mallat representation of 8-subcarriers Reconstruction and Decomposition

filter banks
3.3.3 Design of Best Wavelet Packet Bases For WP-MCM
The attributes of a multicarrier modulation system greatly depends on the set of waveforms
used. The property of the waveforms in turn depends on underlying filter banks used. Many
considerations go into the design of a wavelet system including properties such as
orthogonality, compact support, symmetry, and smoothness. Here we shall discuss a few
important ones.
3.3.3.1 Para-unitary Condition
A rational transfer function A(z) is said to be para-unitary when it obeys the relation
~ ~ ~
A( z ) A( z ) = 1 . Here A( z ) is the para-conjugate of A(z) and is given as A( z ) = A† ( z −1 ) where the
superscript † denotes the conjugation of the coefficients.
The paraunitary condition is essential for many reasons. Firstly, it is a prerequisite for
generating orthonormal wavelets [53]. Second, it automatically ensures perfect reconstruction
of the decomposed signal [53], i.e. the original signal can be reconstructed without amplitude
or phase or aliasing distortion, if the filter banks used satisfy the paraunitary condition. Only
para-unitary filters are considered in the thesis and for such solution pairs the high pass and
low pass filters share the relationship [53]-[55]:
g[L −1− n] = (−1)n h[n] (3.25)
where L is the length of the filters. Further, para-unitary filters automatically satisfy the
perfect reconstruction criterion [55] with the decomposition and reconstruction filters being
55
the complex conjugate time reversed versions of one another i.e.
h'[n] = h†[−n] and g'[n] = g†[−n] (3.26)
Filters satisfying this condition are commonly used in signal processing, and are known as the
Quadrature Mirror Filters (QMF). A nice aspect of these relations is that it is enough to design
a single filter, either the low or high pass filter alone.
3.3.3.2 Compact support
This property ensures that the wavelet is of finite duration and the filter banks used to derive
the wavelets have a finite number of non-zero coefficients [57].
Paraunitary and compact support properties are necessary and sufficient conditions for the
wavelets to be realized. However, they may not always guarantee the generation of smooth,
regular and well shaped wavelets. Quite often the wavelets can be irregular or even fractal
shaped [58]. Therefore to ensure smoothness or regularity of the wavelets the additional
property of regularity is important.
3.3.3.3 Regularity
This property is a measure of smoothness of the wavelet. The regularity condition requires
that the wavelet be locally smooth and concentrated in both time and frequency domain. It is
normally quantified by the number of times a wavelet is continuously differentiable. The
simplest regularity condition is the “flatness” constraint which is stated on the low pass filter.
A LPF is said to satisfy Kth order flatness if its transfer function H(z) contains K zeroes
located at the Nyquist frequency ( z = −1 or ω = π ). Parameter K is called the regularity order
and for a filter of length L it satisfies the relation 0 ≤ K ≤ L / 2 .
Wavelets are defined by the wavelet function ψ(t) (i.e. the mother wavelet) and scaling
function φ(t) (also called father wavelet) in the time domain. Another way to determine the
regularity of the wavelets is in terms of the number of vanishing moments of the wavelet and
scaling functions [57] and it uses the dual vanishing moments to determine the convergence
rate of the multiresolution projections. The lth moments of the wavelet and scaling functions,
mw (l ) and ms (l ) , respectively, are defined in continuous time domain by:
mw (l ) = ∫ t lψ (t )dt and ms (l ) = ∫ t lϕ (t )dt (3.27)

56
3.3.4 Wavelet Families
In this work we shall largely deal with the Daubechies family and its variants.
3.3.4.1 Daubechies
The Daubechies wavelets are a family of orthonormal wavelets with compact support and
highest degree of smoothness. It was derived by Ingrid Daubechies [54] who used all degrees
of freedom K to generate a wavelet family of maximum regularity for a given filter length L,
or to generate the minimum L for a given regularity [57]. This she did by imposing the
maximum number of zero moments to the wavelet function in the vanishing moments’
condition (equation (3.27)).
3.3.4.2 Coiflet
Coiflets wavelets are a variation of the Daubechies wavelets. They are so named because it
was derived by I. Daubechies at the recommendation of R. Coifman who suggested the
construction of orthonormal wavelet basis with vanishing moment conditions for both wavelet
and scaling functions (unlike Daubechies wavelets where only the wavelet functions have
zero moments). The wavelet function has 2L moments equal to 0 and the scaling function has
2L-1 moments equal to 0.
3.3.4.3 Symlet
The symlet family of wavelets is another variant of the Daubechies family which are nearly-
symmetrical (as opposed to being symmetrical). These modifications were also proposed by I.
Daubechies and the properties of the two wavelet families are similar.
3.3.4.4 Frequency Selective Wavelet
Choosing the right wavelet is a delicate and some times even an overwhelming task. In theory
any time and frequency limited function can be utilized. However in practice, the wavelet
bases cannot be arbitrarily chosen and instead have to satisfy a number of specific
requirements. In general the choices to make concern with the system of representation
(continuous or discrete), properties of the wavelets desired (orthogonality/ bi-orthogonality,
regularity/smoothness, frequency selectivity), the application in hand and the context of use
[58]. A framework that accounts for these requirements must first be defined and the wavelet
should be selected in a principle approach through optimization of the wavelet design
parameters.
With regard to the applicability to WP-MCM systems, the desirable properties may be listed
as follows:
• The wavelet bases must be time-limited.
57
• The bases must be well confined in frequency.

• The wavelet packet bases and their duals must be orthogonal (or at least linearly
independent) to each other to enable perfect reconstruction.
• The bases must be orthogonal (or at least linearly independent) to each other in order
to have a unique demodulation capability.
• The bases must enable the system to handle channel effects and other distortions.
• The system must be easily realizable and must permit application of fast algorithms.
And in the filter bank domain the objective of the design procedure translates to a
construction of filters with the characteristics that they:
• have finite impulse response (FIR)
• are maximally frequency selective
• allow orthonormal expansion and perfect reconstruction of discrete-time signals
• satisfy the para-unitary condition
• satisfy a desired flatness/regularity condition
Amongst these properties the para-unitary and regularity properties are mandatory to the
design of the filter banks. In addition to these properties the criterion that needs specific focus
is the property of frequency selectivity.
In an ideal scenario the filter banks used to generate the wavelets have zero transition bands
B, i.e. zero difference at the transition between pass and stop band frequencies (refer Figure
3.17). Under such an ideal scenario the wavelet packet bases derived from a level-i
decomposition have confined spectral footprints with bandwidth (1/2i) times that of the
Nyquist frequency. However, available wavelet families are derived from filter banks that
have a wide transition band and hence the resultant wavelet sub-carriers have a dispersed
spectrum with footprints spilling into neighboring regions. The wider the transition bandwidth
the greater the dispersion of the carrier’s spectral footprint and therefore the greater the
difficulty in isolating those sub-carriers that fall in the region of the licensed user. This greatly
reduces the efficiency of the system. It is therefore important to design filter banks that have
narrow transition bands.
The design procedure comprises of defining a low pass FIR filter, satisfying the regularity,
para-unitary and frequency selectivity conditions, expressed in the form of an impulse
response h(n) or a transfer function H(z) or a difference equation. For a filter of length L this
means essentially solving L unknown filter coefficients from L linear equations. Of these L
linear equations, L/2 equations come from the para-unitary constraint, K equations come from
the regularity or flatness constraint and the remaining L/2 – K conditions offer the room for
flexibility to establish the desired wavelet property such as frequency selectivity. The larger
the value of L/2-K, the greater the degree of freedom for frequency selectivity and the greater
the loss in regularity. There is therefore a trade-off between frequency selectivity and
regularity. Wavelets such as the Daubechies family are maximally flat with regularity order
K=L/2 and hence they are not frequency selective.
58
Figure 3.17: Filter Characteristics in frequency. In the figure ωp and ωs, denote pass and stop
band frequencies, respectively, [0, ω p] is called the pass-band, [ωs,π] is called the stop-band
and [ωp, ωs] is the transition band. δ is the tolerance or ripple
To implement frequency selective filters, the design parameters are stated in the frequency
domain in terms of the desired magnitude response of the LPF as shown in Figure 3.17. In the
figure ωp and ωs, denote pass and stop band frequencies, respectively, [0, ω p] is called the
pass-band, [ωs,π] is called the stop-band and [ωp, ωs] is the transition band. δ is the maximum
value of the tolerance or ripple. The design goal is to generate the filter with a desired
transition band even while the maximum error δ in the pass/stop-band is minimized.
The fundamental theory on the design of frequency selective filter banks was developed by
Rioul and Duhamel [59]-[60]. They devised the procedure to design maximally frequency
selective filter banks under a given set of constraints using the Remez exchange algorithm.
The Remez exchange algorithm is an optimization algorithm that is commonly used in the
design of FIR filters. It is popular because of its flexibility and computational efficiency. It is
known also as the Parks-McClellan algorithm; it works by converting the filter design
problem into a problem of polynomial approximation [61]. The algorithm is an application of
the Chebyshev alternation theorem that constructs the polynomial of best approximation to a
desired function under a set of constraints. Through a minimax approximation the scheme
seeks to arrive at a lth order approximation polynomial function C(x) that best approximates a
desired filter polynomial function B(x) (in our case the LPF H(z) that satisfies the design
specifications) in the given interval such that the absolute maximum error is minimized. The
error is defined here as the weighted difference between the desired filter polynomial function
and the approximation polynomial function and is given by
E(x) = W(x)(B(x) - C(x)) (3.28)

59
E(x) and W(x) are respectively the error and weighting polynomial functions. All polynomial
l
functions are of the form ∑ p i x i , with coefficients p i ; the degree of the polynomial function
i =0
equals l. Chebyshev proved that such a polynomial C(x) exists and that it is unique [61]. He
also gave the criteria for a polynomial to be a minimax polynomial. The algorithm states that
in the interval of consideration, the necessary and sufficient condition that C(x) is a unique
mini-max polynomial solution of degree l is that there are at least (l + 2) points at which the
error function E(x) attains the absolute maximum value δ with alternating sign, i.e.
E( xi ) = −E(xi +1 ) = ± max x∈I {| E( x) |} = (−1)i δ (3.29)
for x1 < x2 < ... < xl + 2 in the desired interval I. Parks and McClellan proved that this approach
could be used to derive a filter of a given length with minimal ripple. The right set of extremal
points xi is arrived through an iterative procedure. In each iteration an interpolation problem is
solved and the reference set of extremal points is updated. Rioul and Duhamel deduced that
L / 2 − K + 1 extremal points in the pass-band are necessary and sufficient to characterize a
unique and optimal solution [59].
The procedure starts by choosing an arbitrary set of L / 2 − K + 1 points in the given interval.
These L / 2 − K + 1 points help to form L / 2 − K + 1 linear equations. The filter coefficients are
obtained by solving the L / 2 − K + 1 linear equations in a way that the error at the
L / 2 − K + 1 points is equal in magnitude and alternating in sign. It cannot be guaranteed after
the first step that the solution satisfies the minimax condition for the error function. It means
that the magnitude of the error need not be the absolute maximum magnitude in the interval of
consideration. In order to find the minimax solution, the second step of the algorithm seeks
new set of L / 2 − K + 1 points that approach the L / 2 − K + 1 points of the minimax solution. The
new set is determined by locating those points where the slope of the error function E(x) is
zero. Once these points are identified, the old set of L / 2 − K + 1 points is exchanged with the
new points. This process is iteratively performed till the desired set of points (that satisfy the
minimax solution) is obtained. The algorithm has converged when the set of extremal points
remains unchanged. Once the right set of extremal points is identified, the optimum error and
filter can be obtained. The exact details of how these equations are solved to obtain the low
pass filter H(z) using the modified Remez exchange algorithm can be found in [59]-[60]. A
flow chart of the algorithm is shown in Figure 3.18. From the low pass filter H(z), the high
pass filter G(z) and the reconstruction filters (H’(z) and G’(z)) can be obtained by applying
eqs. (3.27) and (3.28).
As mentioned earlier, in the construction of maximally frequency selective filter banks three
design parameters have a role to play namely, the length of the filter L, the regularity order K,
and the transition bandwidth B. All three parameters affect the system performance. As an
example we here derive a frequency selective filter bank using the following specifications –
60
length of filter L = 30, regularity order K = 12, degree of freedom N = L/2 – K = 3, transition
bandwidth B= 0.1.
Read design specifications

L, K, B
Initialize extremal points xi
Calculate the optimum

error on the extremal points
Interpolate to obtain
Update extremal points Error Function E (x) Remez
Exchange
Algorithm
Determine new extremal points
by identifying locations
where slope of E (x) vanishes
Yes Have the extremal

points changed ?
No
End of Algorithm-
Best Approximation obtained
Figure 3.18: Flowchart of the Remez exchange algorithm
Out of all combinations the LPF transfer function is so chosen that all zeros lie inside the unit
circle. Such choice makes the LPF a minimum phase filter and hence ensures stability. Figure
3.19(a) shows the distribution of the zeros in the complex plane of the chosen minimum phase
LPF H(z) solution. An FIR filter of length 30 has 29 zeros and for regularity order K = 12, the
zero located at z = -1 has a multiplicity of 13. The other zeros are located on or inside the unit
circle. For both K = 12 and K = 13, the zero located at z = -1 has a multiplicity of 13. This is
because when L/2 - K is odd, the optimal solution for regularity order K + 1 have the same
number of alternations as the optimal solution for regularity order K, [62]. Fig. 3.19(b) shows
the frequency response of the minimum phase LP-HP filter pair.
And Figure 3.20 (a) shows the wavelet packet sub-band pulses derived from a 3-stage
cascading of maximally frequency selective wavelets (with specifications length L=30,
regularity K=12 and transition band B=0.1) and their corresponding 8-channel spectrum
(Figure 3.20 (b) ). More on the frequency selective wavelets is explained in the next section.
61
Low Pass Filter - Pole/Zero Plot

1
0.8
0.6
0.4
Imaginary Part
0.2
13
0
-0.2
-0.4
-0.6
-0.8
-1
-1 -0.5 0 0.5 1
Real Part
(a)
Frequency Response
1.5
Low Pass Filter

1 High Pass Filter
Reconstruction - Low Pass Filter
Amplitude
Reconstruction - High Pass Filter
0.5
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Normalised Frequency
(b)
Figure 3.19: (a) Distribution of the zeros in the complex plane of the chosen minimum phase
LPF H(z) solution. For regularity order K = 12, the zero located at z = -1 has a multiplicity of
13. The other zeros are located on or inside the unit circle; (b) Frequency response of the
maximally frequency selective para-unitary filter pair where the curve that gives the pass band
on the low frequency range is the low pass filter while the one that gives the pass band on the
high frequency range is the high pass filter.
62
1
0
-1
0 50 100 150 200 250 300 350
1
0
-1
Norm alised Am plitude

0 50 100 150 200 250 300 350
1
0
-1
0 50 100 150 200 250 300 350
1
0
-1
0 50 100 150 200 250 300 350
1
0
-1
0 50 100 150 200 250 300 350
1
0
-1
0 50 100 150 200 250 300 350
1
0
-1
0 50 100 150 200 250 300 350
1
0
-1
0 50 100 150 200 250 300 350
Samples
(a)
0.9
Normalised Magnitude
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalised Frequency (xπ rad/sample)

(b)
Figure 3.20: Wavelet Packet bases for MCM carriers (eight carriers). (a) Sub-band pulses in
time domain. (b) Spectra of carriers. The wavelet packet bases are derived from maximally
frequency selective para-unitary filter banks with specifications L=30, K=12, B=0.1
3.3.5 Performance Evaluation and Anaysis
We start out evaluation on the WPMCM performance for a CR system operating with 128
equally spaced carriers. The LU bandwidth is considered to be comparable to 32 carriers
(1/4th) of the CR system and located in the middle of the CR spectral band (Figure 3.21). The
128 wavelet packet carriers are derived from a level-7 cascaded tree. To simplify the
evaluation of the system performance, the channel is assumed to be affected by Additive
63
White Gaussian Noise (AWGN) with no distortions such as intersymbol interference (ISI) or
inter carrier interference (ICI). The modulation scheme used is Quadrature Phase Shift Keying
(QPSK). Lastly, the cognitive modules at the transmitter and receiver are taken to be always
aware of the transmission signal characteristics, including details of the active and de-
activated carriers. The synchronization between transmitter and receiver is assumed to be
perfect.
The frequency ordering of wavelet packet carriers is not in a succession but rather in binary
Gray code order. This is because the output of any 2-channel analysis is the result of
low/high pass filtering followed by down sampling. Down sampling generates two new filter
results with half the number of elements. In addition to this it also results in mirroring of the
high pass components. This switches the order of low and high pass components in a
subsequent decomposition. When the wavelet packet algorithm is recursively applied, the
frequency ordering of the resultant carriers follows the Gray code sequence [59]. This factor
must be considered during the identification of affected carriers.
Cognitive Radio System

(128 sub-carriers)
Licensed User
(32 sub-carriers)
.... ....
CR Active sub-carriers
CR deactivated sub-carriers in the band of LU
CR deactivated sub-carriers adjacent to LU
Figure 3.21: Cognitive Radio (CR) and Licensed User (LU) Characteristics
The length of the filter has a direct impact on the system performance, i.e. the longer the filter
length the greater the possibility to configure the other filter parameters, K and B, to arrive at
the optimal combination. However, it is important that the filter length is kept within
practically realizable limits. The key in the approach is to identify the best combination of L,
K and B such that the transition band is less and the ripples are tolerable. During the
simulation studies we considered a set of about 700 permutations of L, K and B to arrive at the
optimum combination. The ranges of L were between 20 and 50; B varied from 0.05 and 0.15
while K was in the region L/2-6 to L/2. From the set considered the best performance was
obtained for the combination L=50, K=19, B=0.1. In the following simulations we shall use
these specifications to derive the filter banks and wavelet packets.
64
We next have evaluated the performance of wavelet packet modulation with a frequency
selective filter within a cognitive radio context and have compared the results with OFDM-
based cognitive radio. The comparisons are presented in [63]. Figure 3.22 gives the BER
performance curves of the WP-MCM based CR system in the scenarios: absence of LU (only
AWGN), presence of LU without any carrier de-activation and presence of LU with carrier
de-activation. From the plots it is quite clear that the presence of LU affects the CR
performance. When the CR transmission characteristics are communicated around, the LU
with carriers in and around the region of interference removed, and the CR system
recuperates. Best results are obtained when 40 or more of the CR carriers are removed. When
44 carriers are removed seamless co-existence with LU is possible with a performance almost
matching the AWGN case (when LU is not present).
-1
10
-2
10
-3
10
B E R
-4
10
AWGN channel - NO LU
-5
10 With Interference - No Removal
With Interference - Removal 32
-6
10 With Interference - Removal 44
-7
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 3.22: Performance of WP-MCM based CR system with frequency selective wavelet in
the presence of LU. The carriers of CR in the region of the LU are removed to improve
system performance and promote co-existence of CR and LU
It is highly imperative that the CR access to free/idle bands of LU does not cause any harmful
effects on the LU communication. To gauge this effect, we present in Figure 3.23 the
performance of the LU system under CR. The relevant BER curves are for operation of the
LU under the conditions: absence of CR (AWGN channel), presence of CR without any
carrier de-activation and presence of CR with carrier de-activation. As in the earlier case, the
de-activation of CR carriers is indeed necessary to prevent any loss in LU performance. For
the scenario considered here, a noticeable improvement in the performance is obtained when
40 or more of the CR carriers are removed.
65
It will be interesting to see how the WP-MCM construction matches with traditional OFDM
implementations. For fairness of comparison both the OFDM and WP-MCM scheme use the
same set of transmission parameters.
As illustrated in Figure 3.24, the performance of both the OFDM and WP-MCM based CR
systems worsens (WP-MCM marginally more tolerant than OFDM) in the presence of LU and
perform better when their carriers in and around the LU spectral bands are removed. For
OFDM the best results are obtained when 36 or more carriers are removed while for WP-
MCM the number is about 40 carriers. The advantage of WP-MCM is that once its
transmission signal is steered away from the LU bands, it comfortably outperforms its OFDM
counterpart. For example, for a given BER of 10-4 with 44 carrier removals, WP-MCM gives
a gain of about 2 dB in comparison to OFDM.
0
10
-1
10
-2
10
B E R
-3
10
AWGN channel - NO CR
-4 With Interference - No Removal
10
-5
10 With Interference - Removal 40
-6
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 3.23: BER performance of LU in the presence of the WP-MCM based CR system. The
carriers of CR in the region of the LU are removed to enable RU co-existence with LU
The LU performance worsens in the presence of both WP-MCM and OFDM CR systems and
(as in the previous case), removal of CR carriers brings relief (Figure 3.25). OFDM needs a
lower number of carriers to be removed in comparison to WP-MCM to restore the system
performance to the levels of AWGN.
Figure 3.26 depicts the effect of removal of carriers on the system performance. The SNR is
fixed at 8 dB. From the plots we may note that though OFDM requires less carrier removals,
WP-MCM based CR gives a better SNR gain. Another point to note is that removing the
carriers does promote co-habitation of the CR and LU systems but this has a saturation point
beyond which removing anymore carriers has no more benefit.
66
0
10
-1
10
-2
10
-3
10
WP-MCM:NO LU
BER
WP-MCM:No Removal
-4
WP-MCM:Removal 32
10
WP-MCM:Removal 36
WP-MCM:Removal 40
WP-MCM:Removal 44
-5
10 OFDM:NO LU
OFDM:No Removal
OFDM:Removal 32
-6
10 OFDM:Removal 36
OFDM:Removal 40
OFDM:Removal 44
-7
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 3.24: BER performance comparison of OFDM and WP-MCM based CR systems in the
presence of LU
0
10
-1
10
-2
10
B E R
-3
10
WP-MCM:No Removal
WP-MCM:Removal 32
WP-MCM:Removal 36
10
-4
WP-MCM:Removal 40
WP-MCM:Removal 44
OFDM:No Removal
-5
OFDM:Removal 32
10
OFDM:Removal 36
OFDM:Removal 40
OFDM:Removal 44
-6
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 3.25: Influence of WP-MCM and OFDM based CR systems on the operation of LU
67
-1
10
WP-MCM:CR System
WP-MCM:LU System
OFDM:CR System
OFDM:LU System
-2
10
B E R
-3
10
-4
10
-5
10
30 35 40 45 50 55 60
Number Of Carriers Deactivated

Figure 3.26: OFDM versus WP-MCM based CR: Number of carriers de-activated and their
effect on system performance. The BER curves are plotted for a SNR of 8 dB
The use of another kind of wavelet basis function to WP-MCM has been evaluated and
compared with the frequency selective wavelet. Here we equate the performance
improvements brought by a maximally frequency selective filter bank with a few known
wavelets. The wavelets considered are Daubechies-15, Coiflet-5 and Symlet-15. For the
fairness of comparison, all filter banks are taken to be of the same length (L=30). Table 3.2
depicts the considered wavelets characteristic.
Table 3.2: Filter Bank Characteristics
Name Orthonormal Length Regularity

Daubechies Yes 30 15
Coiflet Yes 30 15
Symlet Yes 30 15
Frequency Selective Yes 30 12
The system performance has a direct correlation to the frequency selectivity of the filter bank
used to derive the carriers. This effect is exemplified in Figure 3.27 (a), where the frequency
responses of the wavelet families has been plotted, and in Figure 3.27(b) which shows the
corresponding MCM spectra of the lowest 4 carriers. The Daubechies-15 and Symlet-15
responses almost match and hence the responses of Symlet-15 are not visible. The BER
performance curves of the CR and LU systems have been plotted in Figures 3.28(a) and
3.28(b), respectively. The maximally frequency selective filter banks have the narrowest
68
transition band and hence they clearly outperform the other wavelet families while Coiflet-5 is
much worse because of its very wide transition band.
0.8
Normalised Magnitude
Daubechies-15
Coiflet-5
0.6 Symlet-15
Frequency Selective
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalised Frequency (x π rad/sample)
(a)
100
Frequency Selective
Coiflet
90 Symlet
Daubechies
80
70
Ma g n i t u d e
60
50
40
30
20
10
0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
ω (r a d i a n s)
(b)
Figure 3.27: (a) Frequency responses of the LPF-HPF pairs of various wavelets. (b) Wavelet
Packet bases as MCM carriers (only first 4 carriers shown).The wavelets considered are
Daubechies-15, Symlet-15, Coiflet-5, Maximally Frequency Selective wavelet with L=30,
K=12, B=0.1.
69
0
10
AWGN channel - NO LU
With Interference - No Removal
FS; With Interference - Removal 40
sym15;With Interference - Removal 40
-1 db15;With Interference - Removal 40
10
coif5;With Interference - Removal 40
BER
-2
10
-3
10
-4
10
0 2 4 6 8 10 12 14
SNR (dB)
(a)
0
10
-1
10
-2
10
BER
-3
10
-4
10
With Interference - No Removal
FS; With Interference - Removal 40
10
-5 sym15;With Interference - Removal 40
db15;With Interference - Removal 40
coif5;With Interference - Removal 40
-6
10
0 2 4 6 8 10 12 14
SNR (dB)
(b)
Figure 3.28: Comparison of different wavelets: (a) BER performance of the WP-MCM based
CR system in the presence of LU. (b) BER performance of LU in the presence of the WP-
MCM based CR system. The carriers of CR in the region of the LU are removed to enable it
to co-exist with LU. FS denotes the maximally frequency selective wavelet, db denotes
Daubechies, sym – symlet, coif stands for coiflet
70
Frequency Selective Daubechies

100 100
80 80
Magnitude
Magnitude
60 60
40 40
20 20
0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
ω (radians) ω (radians)
Symlet Coiflet
100 100
80 80
Magnitude
Magnitude
60 60
40 40
20 20
0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
ω (radians) ω (radians)
Figure 3.29: Comparison of the spectrum of different WP-MCM based CR system with
carrier removals in the region of the LU. The rectangle marks the LU spectral footprint
Figure 3.29 shows spectrum comparisons of different WP-MCM based CR systems with
carrier removals in the region of the LU. The frequency selective wavelets ensure a neat
separation of the CR and LU footprints while all other wavelets have residual infringing
components. To validate this further we refer to Figure 3.30 which depicts the effect of
removal of carriers on the system performance. The SNR is fixed at 8 dB. From the plots we
note that the usage of maximally frequency selective filters requires less carrier removals and
gives a better SNR gain.
71
-2
10
FS
db15
sym15
coif5
BER
-3
10
-4
10
30 35 40 45 50 55 60
Number Of Carriers Deactivated

Figure 3.30: BER performance of different wavelets: Number of carriers de-activated and
their effect on system performance at 8 dB SNR. FS denotes the maximally frequency
selective wavelet, db denotes Daubechies, sym – symlet, coif stands for coiflet.
As the objective of cognitive radio is to obtain efficient spectrum utilization, it is logical that
adding MIMO to the system will fulfil this goal even further. By applying the Vertical Bell
Labs Space Time (V-BLAST) architecture at the receiver the BER of the system will be
enhanced due to channel diversity exploitation [64]-[65].
We have evaluated the performance of the CR system under various WP-MCM-MIMO

configurations [66]. The WP-MCM-MIMO Architecture is depicted in Figure 3.31. The WP-
MCM-MIMO system operates under a LU with 40 of its 128 carriers (in and around the
region of LU) are de-activated. We also consider AWGN as the channel model. The data
symbols are derived from BPSK mapping. In the fading channel model with Doppler effects
the impact of multipath delay cannot be counteracted by the use of a cyclic prefix like in
OFDM. This is due to the overlapping transmission between consecutive WP-MCM symbols.
Hence both inter-symbol interference (ISI) and inter–symbols inter-carriers interference (ISCI)
will occur. The fading in WP-MCM has to be cancelled by a time-domain equalization
scheme [50]. At the moment we have not deeply investigated the time domain equalization for
WP-MCM. The use of MIMO in AWGN channel will not give a diversity gain from the
channel but the performance gain will be derived from the increasing number of receive
antennas. This increasing number will give extra redundancy and the effect of code repetition,
so that the data detection at the receiver will be improved. The results in Figure 3.32 show
that for all combination of transmit and receive antennas, as long as the number of receive
antennas is increased, the BER performance will be improved.
The carriers are derived from maximally frequency selective filters with L=30, K=12 and
B=0.1. To better understand the operation, we classify the MIMO configurations as follows:
72
a) First we consider the case with the same number of antennas in both the transmitter
and receiver ends (Figure 3.32a). The formations considered are 1x1, 2x2, 3x3 and
4x4. The pattern of the BER curves is un-ambiguous. With increase in the number of
transmit-receive pairs, the BER improves. This is due to the array gain in the AWGN
channel.
b) Effect of receiver antennas – Increasing the number of receiver antennas significantly
improves the system performance. This is illustrated in Figure 3.32b where the
number of transmit antennas is kept constant at 2 while increasing the number of
receive antennas to 3, 4, 5 and 6 respectively. For a BER of 10-4 the improvements
with respect to the 2x2 configuration are 2 dB for 2x3, 3dB for 2x4, 4dB for 2x5 and
5dB for 2x6.
c) Figure 3.32c shows the curves for a few other MIMO transmit-receive antenna
combinations. In all MIMO configurations the performance of the CR setup is
efficient and allows co-existence with LU.
TRANSMITTER Spectrum
Estimation
Intelligence on Spectrum holes

and Interference Zones
Wavelet Packet Based Adaptive
Modulation and MIMO Tx 1
Inverse Transmitted Data
WPMCM 1
Digital Source D
Tx 2
Constellatation M Inverse
Mapping
U WPMCM 2
X Radio Channel
Inverse Tx M
WPMCM M
Pilot Channel to transmit Service

Received
Request as well as share transmission
Data
details
RECEIVER
Rx 1
WPMCM 1
V-BLAST Rx 2 Interference
Reconstructed Constellation Signal WPMCM 2 Source
Data Demapping
Processing
Rx N
WPMCM N
Figure 3.31: MIMO transceiver architecture for a WPMCM based spectrum pooling system
73
0
10
SISO - 1x1
MIMO - 2x2
-1 MIMO - 3x3
10
MIMO - 4x4
-2
10
-3
10
BER
-4
10
-5
10
-6
10
-7
10
0 2 4 6 8 10 12 14
SNR (dB)
(a)
-1
10
MIMO - 2x2
MIMO - 2x3
MIMO - 2x4
MIMO - 2x5
-2
MIMO - 2x6
10
-3
10
BER
-4
10
-5
10
-6
10
0 2 4 6 8 10 12 14
SNR (dB)
(b)
74
SIMO - 1x2
MIMO - 1x3
MIMO - 2x3
MIMO - 2x4
-2 MIMO - 3x4
10 MIMO - 3x5
-3
10
BER
-4
10
-5
10
1 2 3 4 5 6 7 8 9
SNR (dB)
(c)
Figure 3.32: Performance of WPMCM MIMO based CR system for various combinations of
transmit-receive antennas. a) Balanced Case when the number of transmit and receive
antennas are the same; b) Effect of changes in the receiver configuration. c) A few other
MIMO configurations; MIMO stands for Multiple input-multiple output, SISO for single
input-single output and SIMO for single input-multiple output. The carriers were derived from
Frequency selective filters with L=30, K=12 and B=0.1. The CR is a 2x2 MIMO
Due to the deployment of the V-BLAST architecture at the receiver, the minimum mean
square error (MMSE) technique in detecting the transmitted signal is used in an optimum
way. It combines the signal linearly to get an MMSE between the estimated signal and the
transmitted signal. The optimum gain can be derived when the number of transmit antennas is
equal or smaller than the number of receive antennas [67].
It will be interesting to see how the WP-MCM construction matches up to traditional OFDM
implementations. Both the OFDM and WP-MCM scheme use the same set of transmission
parameters – 128 carriers each with 40 carriers in the interference band removed. The used
wavelet belongs to the family of maximally frequency selective filters. The curves (Figure
3.33) are for the MIMO configurations 1x1, 2x2 and 3x3. For all cases considered the WP-
MCM setup performs comparably well with its OFDM counterpart.
75
WPMCM SISO - 1x1

WPMCM MIMO - 2x2
-1
WPMCM MIMO - 3x3
10 OFDM SISO - 1x1
OFDM MIMO - 2x2
OFDM MIMO - 3x3
-2
10
BER
-3
10
-4
10
1 2 3 4 5 6 7 8 9 10
SNR (dB)
Figure 3.33:. BER performance comparison of OFDM and WP-MCM based CR systems in
the presence of LU. MIMO stands for Multiple input-multiple output, SISO for single input-
single output
3.4 Conclusions
This chapter has described the novel combination of CR interference avoidance technique
through spectrum pooling, sidelobes reduction via combined windowing and sidelobes
cancellation carriers insertion, and intelligent bit allocation in order to attain the targeted bit
rate while the target BER is preserved.
Choosing the window with low sidelobes will ease the application of sidelobes cancellation
carriers insertion. Due to the low sidelobes, high amplitudes of the sidelobes cancellation
carriers can be avoided. In terms of BER performance, the window with low sidelobes will
suppress the impact of inter carrier interference due to receiver imperfection in estimating the
channel state information as an effort to remove the channel fading influence on the receive
signal.
The introduced combination technique tries to solve three optimization problems. The most
crucial one is the co-existence between our CR system with the LU system. We therefore
design signals in such a way that the impact of CR transmission on the LU communication
quality of service (QoS) is negligible. This can be accomplished physically by avoiding the
band occupied by LU and suppress a part of the CR signal that smeared into the LU band.
76
The second one is the CR system bit rate, and the third is the CR system BER. The second
and third optimization problem can be seen as the QoS aspect of the CR system. The three
optimization problems will interact to each other, meaning that it will lead to a trade-off
problem.
Interference (sidelobes) suppression of the CR signal in the LU band by windowing and

sidelobes cancellation carriers is accomplished with the expense of the LU bitrate
degradation. The application of subcarrier weighting or the use of a frequency domain
additive signal in lowering the sidelobes (as described in chapter 2) will negatively affect the
CR BER performance. This BER degradation however can be compensated by the application
of robust channel coding techniques such as the low density parity check codes.
Another novel technique proposed in this chapter is the combination of the spectrum pooling
concept with the wavelet packet multicarrier modulation (WP-MCM) using frequency
selective filter wavelet basis function as replacement for the Fourier transform in the OFDM
system. The signal sidelobes can be shaped by modifying the length of the wavelet filter,
frequency selectivity, regularity order and the transition bandwidth of the filter.
By combining the WP-MCM concept with spectrum pooling, co-existence between the rental
CR system with the LU can be achieved. Parameters (frequency selectivity, regularity order
and transition bandwidth) modification will lead to a trade-off problem between the BER
performance (due to the frequency selectivity option) and the sidelobe suppression.
Due to the overlapping property of the time domain WP-MCM signal, it is considered that the
WP-MCM would not need the guard interval in avoiding inter symbol interference. The
simulation results in the AWGN channel show that although the transmitted signal (in time
domain) of one WP-MCM signal overlaps with its neighboring signals, the receiver can still
reconstruct the transmitted signal.
MIMO as an effort to enhance efficient spectrum utilization has been deployed in the WP-
MCM based Cognitive Radio system. The performance in an AWGN channel has been
evaluated. The performance gain of MIMO in AWGN channel is derived from increasing the
number of receive antennas. Spatial-Multiplexing gain on transmitting different data on
different antenna can be achieved by having equal or a larger number of receive antennas
compared to transmit antennas.
Further investigation on equalization for WP-MCM is required, and its deployment in a

MIMO-WP-MCM system should be explored. MIMO has been evaluated in an OFDM
system. The results show that MIMO with a V-BLAST receiver architecture has improved the
OFDM BER performance by exploring the channel diversity. For this reason it is expected
that the MIMO application will give performance improvement in WP-MCM, while
equalization is required in removing the fading effect in WP-MCM. This last aspect is crucial
in the application of V-BLAST in the MIMO-WP-MCM.
77
Chapter 4
Channel Estimation for OFDM Based

Cognitive Radio
4.1 Introduction
OFDM has been considered as a robust multicarrier technique to counteract the inter symbol
interference due to the multipath channel. By the insertion of the cyclic prefix or zero padding
as the guard interval (GI) [68]-[69], the inter symbol interference (ISI) due to multipath
channel can be avoided.
Still for coherent data detection by demapping (QAM or PSK), information about the channel
state is required and therefore should be estimated by the OFDM receiver. Due to this reason
the channel estimation module is a crucial element in an OFDM system as an effort to remove
the fading effect in the received signal. The channel estimation can be applied blindly by
utilizing the received data signal [70]-[72] or by the aid of pilots which are transmitted
symbols, known by the receiver, in terms of their values and positions. Pilot insertion can be
seen as a method of sampling the channel condition within the period of data transmission.
Blind estimation requires extra complex computations and the results are less accurate
compared to the pilot-aided channel estimation. Therefore, in this thesis we consider the pilot
aided channel estimation to be applied in the OFDM system.
According to the OFDM frame duration, the pilot spacing can be arranged to cope with the
channel condition. Pilot insertion in the form of training signal which is a stream of pilots
dedicated to one OFDM symbol is suitable for short packet length where the frame duration is
much shorter than the coherence time of the channel. Hence, one OFDM symbol dedicated as
training should be sufficient to estimate the channel within the frame duration [10].
In a long frame duration the pilots are multiplexed into the data stream before transmission.
The pilots will be spread among the data either in the frequency direction (within the OFDM
symbol unit) or in the time direction (within the OFDM subcarrier unit). In the Cognitive
Radio (CR) context, there will be de-activated carriers and some carriers which are dedicated
to reduce the sidelobes of the rental OFDM based CR spectrum in the licensed user (LU)
78
band. This condition raises a new challenge in the design of pilot patterns in the OFDM
frame.
The Wiener filter interpolation as channel estimation technique has been described in [73]-
[75]. The technique directly utilizes the correlation of the channel in time and frequency based
on the estimated statistical property of the channel [73]-[75]. The overview of Wiener filter
channel estimation, design and criteria will be explained in section 4.2
Previously in section 2.4 we have described how the frequency hopping technique with TDD
Bluetooth frame design is applied if OFDM subcarriers need to be de-activated due to the LU
access. The Wiener filtering technique in this frame design including the performance
evaluation will be presented in section 4.3.
Section 4.4 proposes novel works dedicated to the technique in designing optimum pilot
patterns by adding the virtual pilot concept for OFDM channel estimation in the context of
cognitive radio. The method in estimating the channel state information for the OFDM based
CR system is described.
Further, since adding multiplexing gain to the CR system affirms one of the CR objectives [2]
in improving the spectrum utilization, in section 4.4 we expand the optimum pilot pattern in
SISO-OFDM-based CR system for use in to a MIMO system. Analysis and evaluations of the
proposed system are reported in section 4.5.
As pilots can also be utilized for addressing other OFDM challenges such as the
synchronization and peak to average power ratio (PAPR) problems, in section 4.6 we will
overview how these pilots can be designed in the cognitive radio context for this purpose.
Conclusions on this chapter are given in section 4.7.
4.2 Wiener Filter Channel Estimation

In principal the Wiener filter is constructed based on the orthogonality between the estimation
error and the initial channel estimates to reach the minimum mean square channel estimation
error. If the estimation error ε n,k is defined as [75]:

ε n, k = H n, k − H n , k (4.1)

where H n ,k and H n ,k are the true and final estimated channel transfer function (CTF) (after
the Wiener Filtering process) on subcarrier n of OFDM symbol k respectively, then 2
dimension (2-D) Wiener filter are obtained by applying the orthogonality principle in linear
mean square estimation [75],
79
{
E εn,k Hˆ n*',k ' = 0, } ∀{n' , k '}∈ P (4.2)
The symbol P in (4.2) defines the set of pilot positions within an OFDM frame while Hˆ n ',k ' is
the initial channel estimate (before filtering) on the pilot position which is derived from the
least square method dividing the received signal on the pilot position with the corresponding
pilot value. It is calculated from:
Y n ',k ' = N n ', k ' + I n ',k ' , ∀{n ' , k '} ⊂ P

Hˆ n ', k ' = H n ', k ' + (4.3)
S n ', k ' S n ', k '
Where Sn,k, Nn,k, and In,k are the transmitted symbol, noise and interference components on
carrier n at OFDM symbol k. The estimated (CTF) is derived from the matrix multiplication
between the Wiener filter ( ω ) and the set of pilots. It is described in a component-wise
multiplication forms by:

H n ,k = ∑ ω n ', k ', n , k Hˆ n ', k ' (4.4)
{ n ', k '}∈P
Combining eqs. (4.1), (4.2) and (4.4) will lead to the so called Wiener Hopf equation [75]:
{

{ n ', k '}∈P
}

{
E H n , k Hˆ *n '',k '' = ∑ ω n ', k ', n , k E Hˆ n ',k ' Hˆ *n '',k '' , ∀{n' ' , k' ' }∈ P } (4.5)
θ n − n '',k − k '' φ n ' − n '',k ' − k ''
The symbol θn−n'',k −k '' is the channel time-frequency correlation function. Symbol φn'−n'',k '−k ''
is the autocorrelation function on the pilot positions which can be written as :
+ σ +σ
2 2
φn '−n '',k '−k '' = θn'−n '',k '−k '' I N

2
δ n'−n '',k '−k '' (4.6)
E{| Sn ',k ' | }
The second term on the right side of eq. (4.6) is the inverse of the average OFDM frame
signal to interference plus noise ratio where σI2 and σN2 are the interference and noise
variance, respectively. By combining (4.5) and (4.6) we can find the matrix for generating the
Wiener filter coefficients (ω) by multiplying the channel correlation matrix (θ) and the
inverse of the autocorrelation matrix of the channel on the pilot positions (Ф) [75]:
−1
ω =θ Φ
T T
(4.7)
80
If the receiver does not have perfect information regarding the actual channel correlation
function, then the 2-D (frequency-time) Wiener filter will be generated with a mismatch
correlation function. It will then produce the channel estimation mean square error also,
according [75]:
MSEn,k = E εn,k { }= E{(H − ω~ hˆ)(H − hˆ ω~ )}

2
n, k n, k
T *
n, k
H
n, k
*
= E{H }− E{H hˆ }ω~ −ω ~ E{hˆH }+ ω

~ E{hˆhˆ }ω
~
2 H * T * T H *
n, k n, k n, k n, k n, k n,k n, k (4.8)
= E{H }− θ ω ~ −ω~ θ +ω ~ Φω ~ 2 T * T * T *
n, k n, k n, k n, k n,k n, k n, k
~
where ωn,k is the Wiener filter coefficients matrix with the mismatch model to estimate the
CTF on subcarrier n of OFDM symbol k while ĥ is the matrix of the initial channel estimates
on the pilot positions. The term mismatch model means here the real statistical model of the
channel is not equal to the estimated channel model statistics.
According to references [73]-[74] the 2-D (time-frequency) channel correlation function

( θn−n '',k −k '' ) is the result of a multiplication between the time channel correlation function
( θ n−n '' ) and the frequency correlation function ( θ kΔ−tk '' ), i.e.
Δf
θn−n'',k −k '' = θΔn−fn '' θΔk−tk '' (4.9)
and accordingly the autocorrelation function would be derived from [75]:
σn
2
{ }δ
Δf Δt
φn'−n'',k '−k '' = θ θ
n ' − n '' k ' − k '' + 2 n ' − n '',k ' − k '' (4.10)
E Sn',k '
Consequence of the above considerations is that the 2-D Wiener filtering can be simplified by
implementing two 1-D filters sequentially. As an example if the first filtering is applied in
frequency direction, it can be followed by the time direction filtering. In this way the channel
estimates after first filtering are derived from [75]

H n[ 1, k] = ω [n1 ] hˆ k ' (4.11)
The matrix ω is given by the first filter coefficients set and vector hˆ k ' represents the subset
[1]
n
of pilots of the initial estimates derived from (4.3) at OFDM symbol k’. The mean square
error after this first filtering becomes:
81
MSE n[ 1, k] = E H n , k { 2
}− θ T
n
~ [ 1 ]* − ω
ω n
~ [ 1 ]T θ * + ω
n n
~ [ 1 ]T Φ ω
n
~ [ 1 ]*
n
(4.12)
The final estimated CTFs are produced by the second filtering ,

H n,k = H n[ 2,k] = ω[k2 ]T hˆn[1' ] (4.13)
The matrix ω is the second filter coefficients to be used for the time direction channel
[ 2 ]T
k
estimation. Vector hˆ n[ 1' ] represents the subset of pilots that are obtained from the first filtering
as calculated in (4.11). Due to the second filtering the mean square error for 2x1-D filtering is
obtained from (4.14).
MSEn,k = MSEn[ 2,k] = E H n,k { 2

}− E{H hˆ
n ,k n }
[1] H ~[ 2 ]*
ωk − ω k { n }
~[ 2]T E hˆ [1] H * + ω
n,k k { n n }
~[ 2]T E hˆ [1]hˆ [1]H ω
~[ 2]*
k
(4.14)
Simulations results in [71] show that the mean square channel estimation error (channel MSE)
performances between 2-D and 2x1-D filtering are only slightly different while 2x1-D
filtering gives less complexity than the to 2-D filtering.
Δt
The estimated channel correlation in time θ k−k '' is derived from the inverse Fourier transform
of the estimated Doppler power density, while the estimated channel correlation in frequency
θ Δn−fn '' is derived from the Fourier transform of the estimated power delay profile. The pilots
are arranged in a frame in such a way that it will fulfil the sampling theorem [75] :
1 1 (4.15)
d ≤ d ≤
t
2f D filter T S
f
τ filter ΔF
where fD filter and τfilter are the estimated Doppler frequency and maximum channel delay filter
parameters for estimating the channel correlation function in time and frequency, Ts is the
OFDM symbol time including the guard interval duration, dt is the distance between pilots in
time, ΔF is the carrier spacing and df is the distance between pilots in frequency.
Optimum sampling can be obtained if the sampling rate of the channel in time and frequency
are the same. This condition is termed as the so called balanced design. By observing (4.15),
the balanced design is obtained if :
82
1
f D filter
Ts d t
=
2τ filter
ΔF d f (4.16)
This balanced design will be optimal if the power delay spectrum and power Doppler
spectrum have the same shape [69]. The rectangular form of the power delay spectrum and
power Doppler spectrum will then produce the frequency (θΔF) and time (θΔt) correlation
function [75] :
sin(2πτ filter ΔF (n − n'' ))
θΔF (n − n'' ) =
2πτ filter ΔF (n − n'' )
(4.17)
sin(2π f D filterTS (k − k '' ))
θΔt (k − k ) = ''
2π f D filterTS (k − k '' )
The autocorrelation ( in time or frequency) between the channel in pilot positions is defined as
[75] :
1
φi ' − i " = θi ' − i " + (4.18)
γc
where γc is the average SNR on all carriers. Correspondingly, the Wiener filter coefficients
for frequency (ωΔF) and time (ωΔt) direction filtering will be derived as :
−1
ωΔF = θΔF Φ ΔF
T T
ωΔt =θ Φ Δt
T T −1
Δt . (4.19)
The pre computed filter coefficients should be designed according to the worst case channel
condition that may occurs. Therefore, the estimated filter delay τfilter should be chosen equal to
the maximum expected excess delay of the radio channel τmax, while the estimated filter
Doppler frequency fD filter should be chosen equal to the maximum expected mobile radio
channel Doppler frequency [73],[75]. Optimum channel estimates will be derived if all
available pilots in time or frequency direction filtering are used to estimate one CTF. The
filter will require a lot of memory due to the large size of filter coefficients. Another approach
is to use some parts of pilots in estimating CTF and refers to the so called sliding window
method. The window will slide until all CTFs on one direction filtering are estimated. An
example of the sliding window frame for filtering in frequency direction is depicted in Figure
4.1.
83
Figure 4.1: Frequency direction filtering sliding window format [76]
The symbol foffset is the distance that define the position of filter coefficients that gives the
Minimum Mean square error (MMSE) (filter region with strongest correlation values), dfreq is
the distance between pilots in frequency, Nf is the number of pilots to be used to estimate one
CTF, Nfsym is the maximum number of CTFs which in this case is the number of subcarriers
that can estimated by the window in one slide. If an optimum channel estimation technique
can be used, (which means all available pilots on one direction filtering are utilized), then
Nfsym in Figure 4.1 will be equal to the number of total subcarriers.
Figure 4.2 depicts an example how the sliding window is applied in estimating the CTFs of
one OFDM symbol. The CTFs on the low subcarriers index edge are estimated by using the
filter coefficients on the sliding window within a (foffset+ dfreq) x Nf region. The CTFs at the
center part of the OFDM symbols are estimated by using the filter coefficients in the dfreq x Nf
region, while the filter coefficients within ( dfreq + (foffset +1))x Nf region are used to estimate
the CTFs at the high subcarriers index edge. The box with the letter C contains the filter
coefficients in the region of dfreq shown in Figure 4.1.
Figure 4.2: Wiener Filter channel estimation by using a sliding window [76]
84
4.3 Pilot Pattern for a Frequency Hopping OFDM Based Cognitive Radio
In section 2.4 we have discussed how the frequency hopping technique using the TDD
Bluetooth frame is able to be applied in OFDM-based Cognitive Radio. Different possible
OFDM frame configurations within this technique are given in Table 2.3.
If 2x 1-D filtering would be applied as the channel estimation technique in this OFDM
system, then 2 OFDM symbols will be required for this purpose, while 1-D channel
estimation requires only 1 OFDM symbol. The filtering in 1-D in this case is applied only in
frequency direction. By observing Table 2.3, choosing 128 subcarriers with 11 OFDM
symbols per frame is the best OFDM frame configuration option since it gives the highest
throughput. The bit rate achieved with a 4-QAM constellation on each carrier for the case
2x1-D channel estimation is around 3.7 Mbits/sec without channel coding. The frame design
itself is depicted in Figure 4.3 where the frame is considered as a short package. The symbol
Δf refers to the carrier spacing of the OFDM symbol while Ts is the OFDM symbol duration
including the guard interval duration. For the case of 1-D channel estimation only one
training/pilot symbol is required, hence it gives a higher throughput compared to 2x-1D, but
there might be a BER degradation. We evaluate both filtering techniques through simulations.
(a) (b)
Figure 4.3: OFDM frame design for short packet transmission with (a) 1D channel estimation,
and (b) 2x1D channel estimation
85
The channel model used in the simulations is a WSSUS (wide sense stationary uncorrelated
scattering) channel [77], with a power delay profile based on the GSM channel model for
rural and urban area [36] provided in Table 2.1 and 2.2. A frequency hopping channel is
added by updating the Doppler frequency of the channel each time after transmitting an
OFDM frame. The carrier frequency used is between 800-900 MHz. The hopping distance is
at multiples of 10 MHz. The number of subcarriers is 128, the guard interval length is 6.4 µs,
and 16 QAM is chosen as modulation. A frame consists of 11 OFDM symbols, and the
vehicular speed is 120 km/h. Since the frame duration is constrained to be 366 µs , the
distance between pilot / training symbols even fulfils a 4 times oversampling assuming a
vehicular speed of 300 km/h and fc 900 MHz. The frame duration with those parameters are
also much shorter than the coherence time of the channel. Rural and urban area propagation
models are examined. For the sake of simplicity of simulations, it is assumed that the
synchronization is perfect. Two scenarios are examined, updated and non updated Wiener
filters. In the updated Wiener filter scenario, the transceiver will update its filter coefficients
while knowing the current operating carrier frequency which means that the correlation
matrix in time will be updated automatically. The update will only affect the performance of
2x1-D Wiener filter since in the 1D channel estimation only the correlation matrix in
frequency is required.
Updating the correlation matrix in time will add computation cost and extra processing time
since the new correlation matrix has always to be saved in memory. Figure 4.4 shows that
2x1-D Wiener filter performs about 1 dB better than 1-D Wiener filter. Updating the
correlation matrix in time in the range 800 – 900 MHz hopping distance does not improve the
BER performance of 2x1-D Wiener filter significantly as shown in Figure 4.5. Updating the
correlation matrix in time refers to updating the estimated normalized maximum Doppler
frequency ( f D filter Ts ). Due to the OFDM duration of the system and the vehicular speed, the
deviation error between the updated f D filter Ts and a non updated f D filter Ts is not significant,
therefore the performance degradation is also not significant. By assuming a constant speed
120 km/h, and Ts= 32μs the deviation of the maximum Doppler happens if the real fc is 900
MHz while the estimated f D filter Ts still uses 800 MHz as the carrier frequency, hence the
deviation of f D filter Ts will be ((fc1- fc2 )v/c)Ts =((900MHz-800MHz)120km/h / 3.108 m/s)32 μs
=0.00035 which is quite small. Figure 4.6 shows that 1-D Wiener filter gives a slight BER
degradation (around 0.5 dB) compared to the updated 2x1-D Wiener Filter. Therefore 1-D
Wiener filter is preferable compared to the 2x1-D Wiener Filter. The 1-D Wiener filtering
according to Figure 4.3 is applied in frequency direction only. It estimates the CTFs on one
OFDM symbol and assumes that these CTFs will not change significantly for the period of an
OFDM frame. Meanwhile in the 2x1-D the whole CTFs in one OFDM frame must be
estimated. The Wiener filtering in frequency direction is applied first by the sliding window
technique applied onto two OFDM training symbols. Further the time direction filtering is
applied on each subcarrier position. The window slides from the first OFDM symbol until the
last one. It is obvious that the estimation process using 2x1-D filtering is much more complex
86
than estimation based on the 1-D Filtering. According to the complexity and performance
results the 1-D filtering is the promising channel estimation technique in the OFDM frame
with short durations. Furthermore by having the 1-D channel estimation, the last OFDM
training symbol can be replaced by data. In this way the bitrate will be higher compared to the
2x1-D channel estimation. The BER performance is lower bounded by the filter matched to
the power delay profile and power Doppler Spectrum of the GSM channel model. In case the
CR system can access the spectrum longer than the TDD frame duration, the distance between
training symbols can be increased. Figure 4.7 shows that the BER gap between the 1-D
Wiener filter and the 2x1D filter for longer frame duration is larger compared to the case of
short frame duration as depicted in Figure 4.6. The distance between training symbols in long
frame duration is increased by a factor of 2 of the one in short packet duration. The 1-D
Wiener filter can not cope with the change of the channel in time. This is the reason why the
1-D channel estimation is applied only to the system with frame duration which is shorter than
the channel coherence time.
OFDM - 1D Wiener Rect - Rural Area chn

OFDM - 1D Wiener Rect - Urban Area chn
OFDM - 2x1D Wiener Rect - Rural Area chn
OFDM - 2x1D Wiener Rect - Urban Area chn
-1
10
BER
-2
10
0 2 4 6 8 10 12 14 16
SNR (dB)
Figure 4.4: BER performance of OFDM with 1-D and 2x1-D non-updated Wiener filter
87
OFDM - 2x1D Wiener Rect - Rural Area chn

OFDM - 2x1D Wiener Rect - Urban Area chn
OFDM - 2x1D udt.Wiener Rect - Rural Area chn
OFDM - 2x1D udt.Wiener Rect - Urban Area chn
-1
10
BER
-2
10
0 2 4 6 8 10 12 14 16
SNR (dB)
Figure 4.5: BER performance of OFDM with 2x1-D updated and non updated Wiener filter
-1
10
BER
OFDM - 1D Wiener Rect - Rural Area

OFDM - 1D Wiener Rect - Urban Area
OFDM - 2x1D udt.Wiener Rect - Rural Area
OFDM - 2x1D udt.Wiener Rect - Urban Area
OFDM - 1D Wiener Matched - Rural Area
OFDM - 1D Wiener Matched - Urban Area
OFDM - 2x1D udt.Wiener Matched - Rural Area
-2 OFDM - 2x1D udt.Wiener Matched - Urban Area
10
0 2 4 6 8 10 12 14 16
SNR (dB)
Figure 4.6: BER performance of OFDM with 1-D and updated 2x1-D Wiener filter
88
-1
10
BER
OFDM - 1D Wiener Rect - long packet - Rural Area

OFDM - 1D Wiener Rect - long packet - Urban Area
OFDM - 2x1D Wiener Rect - long packet - Rural Area
OFDM - 2x1D Wiener Rect - long packet - Urban Area
OFDM - udt. 2x1D Wiener Rect - long packet - Rural Area
-2 OFDM - udt. 2x1D Wiener Rect - long packet - Urban Area
10
0 2 4 6 8 10 12 14 16
SNR (dB)
Figure 4.7: BER performance of OFDM with 1-D, 2x1-D, and updated 2x1-D Wiener filter
while assuming long packet duration
4.4 Combined an Optimum Pilot Pattern and Virtual Pilots for OFDM-
Based Cognitive Radio
The main concern in estimating the channel by using pilot symbols in the context of Cognitive
Radio environment is where to place the pilot symbols if there is a frequency band that is
occupied by the LU. The main requirement of the pilot pattern design is that the designed
pilot pattern will still be able to estimate the channel optimally with the presence of LU;
meanwhile the performance of LU in its spectrum band should not be deteriorated by the
presence of these pilot symbols. The arrangement of the pilot symbols within an OFDM frame
themselves still has to satisfy the criterion as given in eq. (4.15).
In spectrum sharing for CR, the interference is avoided by de-activating subcarriers falling in
the Licensed User’s band and further interference reduction is applied by de-activating more
sub-carriers adjacent to LU band, complemented with time domain windowing and long
symbol duration. These interference reduction methods have already been presented in
chapter 2 and 3.
In this work we assume that the transmitter and receiver of our system have the knowledge of
the presence of LU via the information obtained from spectrum sensing. Our receiver and
transmitter will receive the information about the frequency bands that are occupied by the
89
LU or in the context of OFDM modulation, the subcarriers that are to be de-activated in order
not to interfere with the Licensed User.
The approach to solve this problem is dea-ctivating several subcarriers (including the pilot
symbols) that overlap with the frequency bands used by the LU. Most degradation in the
estimation occurs where the de-activated band edges are at the pilot positions. This is because
the channel estimates at the position after these pilots depend on the pilots which are further
away in frequency and therefore less correlated. In order to solve this problem we shift the
pilot that supposed to be on the edge of the de-activated subcarriers to one subcarrier position
most nearby.
This can be applied by simply exchanging the position of the pilot symbols with selected data
symbols. This process is done by exchanging the pilot subcarrier with the data subcarrier
which lies either previous or next to the original pilot subcarrier. By doing this we can still be
able to keep the pilot symbols to be used in the channel estimation process and to sample the
channel up to the border with the LU band; however we will have to sacrifice one data
subcarrier and losing some throughput. The proposed pilot pattern for the case of wideband
interference is depicted in Figure 4.8.
We assume that extra effort in mitigating the Cognitive Radio system sidelobes (by adding
sidelobes cancellation carriers as described in Chapters 2 and 3) can be accommodated by
treating them on the positions of the de-activated carriers. In this way the shifted pilots will
not be on those positions. The CTFs on sidelobes cancellation carrier positions will not be
estimated.
Figure 4.8: Comb/rectangular pilot pattern in the presence of a wideband licensed user
90
By observing Figure 4.8, we can see that the filtering by a sliding window can be done in two
regions, above the LU region and below LU region. Examining eqs. (4.18), (4.19) and Figure
4.1 we know that the filter is generated by assuming constant distances between pilots.
However, due to the pilot shifting shown in Figure 4.8, the distance between shifted pilot and
the regular pilot is changed; therefore two extra filters should be designed. One to estimate the
CTFs in the vicinity of the shifted pilots in the region above the LU band and the other one for
the CTFs in the vicinity of the shifted pilots in the region below the LU band.
A hexagonal sampling pattern is considered to be the most efficient two-dimensional

sampling scheme. This scheme requires 13.4% fewer samples than the rectangular sampling
to represent the same circularly band limited continuous signal [78]. In [79] and [80] the
application of a hexagonal pilot pattern in OFDM was evaluated and the simulations revealed
that the hexagonal pattern gives a better BER compared to the Rectangular pilot pattern. In
[81] the analysis was made in designing the optimum pilot pattern for OFDM channel
estimation. Through simulations it was verified that the hexagonal pilot pattern is BER indeed
outperformed the rectangular pilot pattern. The complete design requirements for the
hexagonal pilot pattern can be found in [81].
There is another consideration assuming that the channel estimation will be optimum if the
pilots are allocated in a random scattered way i.e. putting them only on the subcarrier and
OFDM symbol positions that have really good SNR. In this way, the channel estimation mean
square error can be reduced. The disadvantage is that if the sliding window technique is
applied, this pattern randomness will require several sliding window filters, because a sliding
window filter requires an invariant pilot distance. Therefore, in this dissertation the hexagonal
pilot pattern is preferable compared to the random pattern despite we know that optimum
filtering (using a complete set of pilots) will favour the application of the random pilot pattern
since in that case, only one filter is used to do channel estimation. However, using a complete
set of pilots in estimating one channel transfer function results into extra computational
complexity.
If the hexagonal pilot pattern is adopted in the Cognitive Radio system combined with the
pilot shifting method, then we will have the pattern as depicted in Figure 4.9.
Still in the form shown in Figure 4.9 there will be a non-constant distance between pilots, and
due to the hexagonal pilot pattern and de-activated carriers in some OFDM symbols, there
will be no pilots at the edges. Therefore several filters are required to estimate the CTFs.
In order to counteract this problem we propose to include the virtual pilot concept to simplify
the filtering process by only utilizing one Wiener filter set. Inserting the virtual pilots between
two pilots will give a two-times pilot oversampling effect to the channel estimates. The new
pattern becomes as depicted in Figure 4.10.
The virtual pilot concept has been previously introduced in [82]. Virtual pilots are no pilots
91
allocated by the transmitter to the OFDM frame for the purpose of channel estimation, but
rather are imaginary pilots located at certain data positions. Through a signal processing
technique these virtual pilots will give almost the same impact as the real pilots in providing
initial channel estimates. These estimates will further be utilized by the filtering or channel
estimation process in producing the final channel estimates. Virtual pilots are located in
between 2 consecutive pilots, which are actually at the data positions. It should be emphasized
that no actual pilots are transmitted at these positions, but the received data at those locations
are used for generating virtual pilots to enhance the channel estimation performance. The
initial channel estimates at the virtual pilot positions are derived by interpolating linearly the
pairs of the initial estimates at the actual pilots position (derived from eq. (4.3)) close to the
position of the virtual pilots. The interpolation is carried out by assuming that the channel
does not change significantly (slowly changed) between a pair of pilots, which denotes that
the pilot spacing in frequency direction should be smaller than the channel coherence
bandwidth while in time the pilot spacing is smaller than the channel coherence time.
Figure 4.9: Hexagonal pilot pattern in the presence of wideband licensed user
92
Figure 4.10: Hexagonal pilot pattern with integrated virtual pilot concept in the presence of a
wideband licensed user
Based on these assumptions we try to predict the value of the initial channel estimates in
between two neighboring pilot symbols that can be seen as the initial channel estimates at the
virtual pilot symbol positions. The application of virtual pilots starts from the initial estimate
of the channel derived from (4.3). The value of preliminary initial channel estimates on a
virtual pilot position is obtained from a simple linear interpolation method, described by:
1
[
Hˆ n ', k ' = Hˆ n ', k ' −1 + Hˆ n ', k ' +1
2
] or 1
[
Hˆ n ', k ' = Hˆ n ' −1, k ' + Hˆ n ' +1, k '
2
] (4.20)
where Ĥn’,k’-1 , and Ĥn’,k’+1 indicate the initial estimates of the previous and the next pilot
symbols in time, meanwhile Ĥn’-1,k’ , and Ĥn’+1, k’ indicate the initial estimates of the previous
and next pilot symbols in frequency. The virtual pilots located at the edge of the frame are
estimated by extrapolating pair of pilots in the vicinity of the virtual pilot position; it can be
expressed as:
[ ] 1 ⎡N +I N +I ⎤
Hˆ n',k ' = Hˆ n'+1,k ' + Hˆ n',k '+1 = [H n'+1,k ' + H n',k '+1] + ⎢ n'+1,k ' n'+1,k ' + n',k '+1 n',k '+1 ⎥ .
1
2
1
2 2 ⎢⎣ S n'+1,k ' S n',k '+1 ⎥⎦
(4.21)
93
Adding the concept of virtual pilots resembles twice oversampling of the initial pilot pattern.
The block diagram of the OFDM with channel estimation using virtual pilots at the receiver is
depicted in Figure 4.11. The decision directed method will estimate the initial channel
estimates at virtual pilot positions by predicting the transmitted symbol at those virtual pilot
positions utilizing the results from eq. (4.20) or eq. (4.21). The aim of the decision directed
method is to reduce the effect of interpolating the noisy initial channel estimates derived from
eq. (4.20) or eq. (4.21). The process is described in Figure 4.12. X̂ v , Xˆ 'v , Yv , Ĥv and Ĥ’v are
the estimated transmitted symbol before de-mapping, estimated transmitted symbol after
using the decision directed method (after mapping), the received symbol, the preliminary
initial channel estimate from simple linear interpolation/extrapolation and the final initial
channel estimate at the virtual pilot position v after using the decision directed method,
respectively. It is worth mentioning that we assume that the virtual pilot positions be always
constant in between two real pilots or at the edges while in reference [82] the virtual pilot
position varies according to the minimum Euclidean distance between the results before using
the decision directed method and after, and the threshold selection. Our method has the
advantage that by having constant virtual pilot positions the filtering process especially in the
application of a sliding window, can be greatly simplified.
In the case the decision directed estimation module can perfectly estimate the transmitted
symbol at the position of the virtual pilots, the initial estimates at the virtual pilot positions
will be the same as if the transmitter sends the real pilots at those positions (virtual pilots).
Extra deviation from the initial channel estimates at the virtual pilots comes from the decision
errors in estimating the transmitted symbol at the virtual pilot positions. The received signal in
frequency domain at the virtual pilot positions can be envisioned as:
H v . X v + N v + I v = .⎛⎜ 1 + H v − Hˆ v ⎞⎟ + N v + I v
Yv = ˆ ˆ Xv ⎜ ˆ ⎟ ˆ
Hv Hv ⎝ Hv ⎠ Hv
⎧ − Hˆ v N +I ⎫
= X v + ⎨Hv .Xv+ v v ⎬
ˆ Hˆ v ⎭
⎩ Hv
(4.22)
where Hv, Nv and Iv are the respective CTF, noise and interference on virtual pilot position v.
The decision error in estimating Xˆ 'v occurs when:
H v − Hˆ v . Nv + I v > D
Xv + , (4.23)
Hˆ v ˆ Hv 2
with D as the minimum distance between the constellation points of symbol Xv. The deviation
94
error increases as ( H v − Hˆ v ) is significant and the constellation size of Xv increases. The

maximum deviation takes place at the largest constellation point Xv,max. The effect of a
decision error in estimating the symbol at the virtual pilot position will be an additional
deviation beside noise and interference contribution to the final initial channel estimates
compared to eq. (4.3),
ˆ = H v . X v + Nv + I v = .⎛⎜ 1 + X v − Xˆ 'v ⎞⎟ + Nv + I v
H 'v Hv ⎜ ⎟
Xˆ 'v Xˆ 'v ⎝ Xˆ ' v ⎠ Xˆ 'v
(4.24)
⎧ − Xˆ 'v N +I ⎫
=Hv + ⎨ X v . H v + v v ⎬.
ˆ Xˆ 'v ⎭
⎩ X 'v
If the channel state between two pilots is slowly varying and assuming the noise plus
interference averaging advantage in estimating Ĥv (as described in eq. (4.21)), then most
likely ( H v − Hˆ v ) will be close to zero. Moreover, in a dynamic spectrum access concept the
sub-carriers in the interference band, which in this context is the LU band, will be de-
activated. The sidelobe of the interferer signal will distort the CR signal, but the impact will
be decreased because of the larger distance between the OFDM based CR sub-carrier to the
LU band. By applying the mechanism of de-activating more carriers adjacent to LU band, or
by inserting the cancellation carriers as described in chapters 2 and 3, the interference effect
to the CR signal will become very low and perhaps negligible. Therefore, according to eq.
(4.23), the decision error in estimating the transmitted symbol at the virtual pilot position will
be dominated by the influence of the noise term Nv and interference term Iv which are divided
by Ĥv . If Ĥv is large, the decision error probability in estimating Xˆ 'v will be reduced. If the
decision directed method can estimate the transmitted symbol accurately, the initial channel
estimates at the virtual pilot position described by eq. (4.24) will be equal to eq. (4.3) because
then Xv = Xˆ 'v . This condition will be comparable to the scheme where real pilots are inserted
at the virtual pilot positions, and the initial channel estimates are derived by the least square
method (dividing the received pilot symbols by the reference pilot symbols). This analysis
will be further validated in the performance evaluations in section 4.4.1.
Virtual pilot
Channel
Estimation by
Interpolation
and applying
Decision
directed
method
Figure 4.11: OFDM channel estimation with virtual pilots processing in the receiver
95
Figure 4.12: Decision directed method in estimating the virtual pilots
The additional complexity in the inclusion of the virtual pilot concept to an OFDM system is
comparable to the number of virtual pilots included to the OFDM frame. Since the virtual
pilots normally exist in between two real pilots, the number of extra computations required in
a free environment (without LU access or de-activated carriers) equals:
⎛⎡ ⎤ ⎞⎛⎡ ⎤ ⎞ ⎛⎡ ⎤ ⎞⎛⎡ ⎤ ⎞ ⎛⎡ ⎤ ⎞⎛⎡ ⎤ ⎞

O = ⎜ ⎢ N c ⎥ − 1 ⎟.⎜⎜ ⎢ N s ⎥ − 1⎟⎟ + ⎜ ⎢ N c ⎥ − 1⎟.⎜⎜ ⎢ N s ⎥ − 1⎟⎟.M + ⎜ ⎢ N c ⎥ − 1 ⎟.⎜⎜ ⎢ N s ⎥ − 1⎟⎟ , (4.25)
⎜⎢ d f ⎥ ⎟ ⎢ dt ⎥ ⎜⎢ d f ⎥ ⎟ ⎢ dt ⎥ ⎜⎢ d f ⎥ ⎟ ⎢ dt ⎥
⎝
⎠ ⎝
⎠ ⎝ ⎠ ⎝ ⎠
⎝ ⎠ ⎝
⎠
a b c d
where O is the number of computations, Nc is the number of active subcarriers, Ns is the

number of OFDM symbols per frame, df is the distance between pilots in frequency, dt is the
distance between pilots in time, and M is the constellation size of the data at the virtual pilot
positions. The term a in eq. (4.25) indicates the number of virtual pilots in frequency, b is the
number of virtual pilots in time. The multiplication between a and b refers to the number of
interpolation/extrapolation required. The term c is the number of mathematical operations
required in calculating the Euclidean distance between a symbol derived from eq. (4.22) and
the reference points on the respective constellation data mapping (M-QAM or M-PSK) used at
the virtual pilot positions, while the term d is the number of division operations between the
received signal at virtual pilot positions and the estimated symbol after using the decision
directed process as described in eq. (4.24). Since the application of the virtual pilots is to
simplify the filtering process, the position of the virtual pilots should always be in the middle
of two pilot positions. Therefore the distance between two pilots should be even.
If the number of de-activated sub-carriers is large enough compared to the LU band the
mutual interference from or to the LU will be significantly low and can be considered
negligible. It means that the interference contribution denoted by symbol I in eq. (4.3) and
eqs. (4.21)-(4.24) will be close to zero. This analysis is needed in case a residual interference
due to sidelobes of LU exists. The residual interference will be considered as an extra or
additional noise that will degrade the SINR and will add an extra deviation into the
transmitted data at the virtual pilot positions as described in eq. (4.22) and to the final initial
channel estimates as depicted in eq. (4.24).
The decision directed method requires the received signal information at the virtual pilot
positions. In the OFDM-based CR system sub-carriers located at the LU band and adjacent to
it are de-activated for the purpose of mitigation of the mutual interference. Since the received
96
signal at the de-activated sub-carriers positions (LU band) is not available, the decision
directed method cannot be applied to those positions. For cases where initial channel
estimates located at the de-activated sub-carrier positions are required, the channel estimates
will be obtained by linear interpolation / extrapolation.
By observing eq. (4.8), eqs. (4.17)-(4.19) a higher correlation value will render a lower
theoretical channel estimation mean square error (MSE). By having a two-times pilot
oversampling effect due to the application of virtual pilots, the Wiener filter MSE in
mismatched model error case given in eq. (4.8), can be reduced. This is achieved at the cost of
complexities for applying the linear interpolation/extrapolation between two real pilots with
the addition of using that decision directed process.
The channel estimation is applied separately to the region before as well as after the de-
activated band. If a large number of LUs exists, more sub-carriers need to be de-activated. If
the positions of the LUs are scattered, then more edge effects in the channel estimation occur
that will further degrade the channel estimation performance. The edge effect can be avoided
if the LU band is less than the distance between pilots in frequency since the initial channel
estimates at the de-activated pilot positions can still be nicely performed by the interpolation
method. If many carriers need to be de-activated due to scattered LUs, then optimal filtering
(using all available pilots in each filtering direction) will be feasible to be applied as long as
the memory-saving capability of the Cognitive Radio device suffices.
4.4.1 Performance Evaluation and Analysis
In practice the actual power delay profile and Doppler power spectrum are not perfectly
known at the receiver; the Wiener filter assumes a rectangular power delay profile and
Doppler power spectrum, which creates a model mismatch. The simulations ran using a
channel model based on the statistical discrete time model wide sense stationary uncorrelated
scattering (WSSUS) which combines multipath fading and Doppler frequency effects [77].
The simulation parameters are provided in Table 4.1. According this Table there will be a
maximum Doppler frequency of 82 Hz. No channel coding and interleaving are applied. Other
channel models can be used to validate the results of the use of virtual pilots for OFDM-based
CR channel estimation. Since the virtual pilot is located in between two real pilots, the virtual
pilot concept will be effective for slowly varying channels. This means the coherence
bandwidth and coherence time of the channel which refer to the delay spread and maximum
Doppler frequency will determine the effectiveness of the virtual pilot concept.
Synchronization is an important issue in every OFDM system. The problem can be solved by
the aid of training symbols or the cyclic prefix characteristic of the OFDM GI. The process in
the context of Cognitive Radio will be explored in section 4.6. For the simplicity of
simulation, we assume perfect synchronization between transmitter and receiver.
We begin by evaluating the performance of the proposed pilot pattern according to Figure 4.8.
Figure 4.13 depicts the channel mean square error (MSE) and BER results of the proposed
97
rectangular pilot pattern with shifted pilots due to LU access and conventional OFDM without
carriers de-activation (Normal (3,3) Pilots, meaning that the distance between pilots in
frequency and time is 3). Two LU widths are considered, with a 10 carriers de-activation
(Modified(3,3) Pilot Pattern scheme (10 subcarriers de-activated)) and a 61 carriers de-
activation (Modified(3,3) Pilot Pattern scheme (61subcarriers de-activated)). No significant
performance degradation occurs related to the number of de-activated carriers. The error floor
is formed due to the model mismatched used between the true power delay profile and
Doppler power spectrum and the estimated ones in designing the Wiener filter. The mismatch
error can be termed as the Wiener filter error.
Significant degradation can be observed if a number of wideband LU sets of carriers are

scattered and therefore the filtering is applied separately in each free region (bands not
occupied by LU). Due to this condition, as depicted in Figure 4.1, filters at the edges that
give higher MSE compared to the middle filter (in the dfreq region that gives MMSE), will
often be used. The simulation results depicted in Figure 4.14 show the expected behaviour,
where each LU with 10 carriers de-activation is scattered in the spectrum. The results in
Figures 4.13 and 4.14 have been presented in conference paper [83].
Table 4.1: Simulation parameters
Bandwidth 1.28 MHz

Number of subcarriers 213
De-activated subcarriers 61
Subcarrier spacing 5 KHz
FFT length 256
Guard interval 43.75 μs
OFDM symbol duration (inc. cyclic prefix) Ts’= 243.75 μs
OFDM symbols per frame 46
Modulation for pilot symbols BPSK
Modulation for data symbols 4-QAM
Normalized pilot spacing in frequency df = 3 or df = 4
Normalized pilot spacing in time dt = 3
Fading model Rayleigh
Maximum Doppler frequency fDmax = 82 Hz
Maximum delay τmax = 9.375 μs
RMS delay spread τrms = 3.4 μs
# of channel taps 4
Number of paths per tap 10
Power decrement per tap 1 dB
Tap spacing 3.125 μs
Channel coding and interleaver OFF
98
Figure 4.13: Performance of the modified pilot pattern (Eb/N0 in dB) in the presence of a
wideband Licensed User (Left: MSE, Right: BER)
Figure 4.14. Performance of the proposed pilot pattern (Eb/N0 in dB) with an increasing
number of scattered wideband Licensed Users with 10 subcarriers de-activated for each
Licensed User (Left: MSE, Right: BER)
99
Next we evaluate the performance of the pilot pattern according to Figure 4.10 where virtual
pilots are utilized to aid the channel estimation process. As a start, in order to observe the
effectiveness of the virtual pilot concept, we evaluate an OFDM system without the existence
of LU access. Therefore, at the beginning no carriers de-activation is applied.
Figure 4.15 shows the MSE performance comparisons of the regular rectangular, hexagonal,
rectangular and hexagonal with virtual pilots from the interpolation/extrapolation process (no
decision directed) in the absence of a LU band. At high SNR the hexagonal pattern with
virtual pilots outperforms the other three patterns significantly. The rectangular pattern with
virtual pilots outperforms the normal rectangular and hexagonal pattern at high SNR, while at
low SNR their performances are close to each other. The results have been presented in
reference [84].
By adding the decision directed method, the use of virtual pilots has shifted the MSE of the
hexagonal pattern downwards. Both the rectangular and hexagonal pattern with decision
directed virtual pilots converge close to each other as shown in Figure 4.16. At an MSE of 10-
3
the use of the decision directed virtual pilot gives about 5 dB SNR gain compared to the use
of virtual pilots from interpolation/extrapolation only. The two figures (Figure 4.15 and
Figure 4.16) demonstrate that the use of the virtual pilots has reduced the Wiener filter MSE
(mismatch errors) which is indicated by the reduction of the error floor for high SNR. The
channel estimation MSE is thus influenced by the noise and mismatch errors.
0 Rectangular pattern
10 Hexagonal pattern
Rectangular pattern with interpolation of virtual pilots
Hexagonal pattern with interpolation of virtual pilots
-1
10
MSE
-2
10
-3
10
-4
10
0 10 20 30 40
SNR(dB)
Figure 4.15: Channel estimation MSE vs SNR(dB) comparisons of regular rectangular,

regular hexagonal pilot patterns, as well as rectangular and hexagonal pilot pattern with
virtual pilot symbols from the interpolation/extrapolation process
100
0 Rect. pattern with interp. of virt. pilots

10 Hex. pattern with interp. of virt. pilots
Rect. pattern with comb. interp. and dec. directed of virt. pilots
Hex. pattern with comb. interp. and dec. directed of virt. pilots
Rect. pattern with 2x oversampling
-1
10
MSE
-2
10
-3
10
-4
10
0 10 20 30 40
SNR(dB)
Figure 4.16: Channel estimation MSE vs SNR(dB) comparisons of rectangular, hexagonal

pilot pattern with virtual pilots derived from interpolation./extrapolation, their combinations
with the decision directed method and with a two-times oversampling rectangular pattern
At high SNR the noise contribution to MSE becomes negligible and only the mismatch errors
contribution remains. Figure 4.17 shows the BER improvement by using virtual pilots and the
decision directed method in the hexagonal pattern compared to the BER results without
decision directed. As seen from this figure an 8 dB gain is derived at a BER of 2x10-3. This
shows that the decision directed method improves the channel estimation performance
considerably, which consequently enhances the BER performance when compared to the
regular hexagonal pattern. To ensure that the combined virtual pilots with hexagonal pilot
pattern is indeed powerful, we have also evaluated its performance using the low pass filtering
interpolation channel estimation method [85]. The results are depicted in Figure 4.18.
Although the low pass filtering exhibits a larger interpolation error compared to the Wiener
filter, the application of virtual pilots has enhanced its interpolation performance.
The performance is lower bounded by the two-times oversampling rectangular pilot pattern,
and they are close to each other. This means that the proposed hexagonal pilot pattern with
virtual pilots and the decision directed method can perform almost the same as the two-times
oversampling rectangular pilot pattern without adding more pilots. In other words, by doing
under-sampling with hexagonal pilot pattern combined with virtual pilots using the decision
directed method, the same channel MSE performance as without doing under-sampling pilot
allocation, can be obtained while the number of pilots transmitted is reduced.
101
0 Rectangular pattern
10 Hexagonal pattern
Hex. pattern w/ Combined interp. and dec. directed of virt. pilots
Rectangular pattern with 2x oversampling
-1
10
-2
10
BER
-3
10
-4
10
-5
10
0 10 20 30 40
SNR(dB)
Figure 4.17: BER vs SNR(dB) comparisons of rectangular, hexagonal pattern, combined

hexagonal pattern with virtual pilots from interpolation/extrapolation and decision directed
method, and the two-times oversampling rectangular pattern
0
10
-1
10
-2
10
MSE
-3
10
-4
10 Hex. patt. using low pass filter interp. without virtual pilots
Hex. patt. using low pass filter interp. with virtual pilots
Hex. patt. using Wiener filter interp. without virtual pilots
-5
10 Hex. patt. using Wiener filter interp. with virtual pilots
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.18: Channel estimation MSE vs SNR(dB) comparisons of hexagonal pattern, with
virtual pilots from interpolation/extrapolation and the decision directed method for low pass
filter interpolation and the Wiener filter channel estimation
102
By referring to eq. (4.22), in a slowly varying channel, the deviation of estimating the
transmitted symbol on the virtual pilot positions (due to inaccurate initial channel estimates)
from a simple linear interpolation/extrapolation is negligible, and the decision error is
dominated by the noise component. The condition shown in Figure 4.16 where the use of the
combined linear interpolation/extrapolation and decision directed virtual pilots to the
hexagonal pattern has gained almost the same MSE performance as the two-times
oversampling rectangular pattern refers to the circumstance that seldom decision errors occur
in estimating the symbol on the virtual pilot positions. Thus eq. (4.24) will be the same as eq.
(4.3) what implies the term as if we transmit real pilots on the virtual pilot positions. By also
observing eq. (4.24) we have simulated the MSE performance of the use of virtual pilots to
the hexagonal pattern and varying the constellation size M of the data symbols; that can be
interpreted as varying the constellation size of the symbols on virtual pilot positions.
According to eq. (4.24) increasing M will decrease the deviation by the noise part. Figure 4.19
confirms the analysis where at low SNR the highest M gives the lowest MSE. We have to
emphasize at this point that reference [82] fails to address this issue as in that reference the
pilot spacing between real pilots is decreased when allocating data with higher M.
On the contrary, the BER curve in Figure 4.20 shows that symbols with higher M are more
vulnerable to noise which follows the property of QAM or PSK symbols; therefore the
highest M gives the poorest BER performance. An exception is the case where the BER
performance of virtual pilots M =4 outperforms M=2; this is (although the BER in M=4 is
more vulnerable to noise compared to M=2) because according eq. (4.24) M=4 has less
channel estimation deviation error. The overall errors (bit errors due to noise and channel
estimation errors) for M=4 and M=2 are shown by the BER results in Figure 4.20. The virtual
pilots with constellation size M=4 is more optimal when compared to M=2.
In Cognitive Radio spectrum sharing context of our simulations, there are 61 carriers de-
activated (carrier no. 81 – 141) due to LU occupancy. The LU itself occupies 40 carriers in
the middle; this means 10 guard bands on each side of LU are provided. In this way the
mutual interference between LU and our Cognitive Radio system will be negligible. The
distance between pilots in frequency is 4 sub-carriers, while in time it is 3 OFDM symbols.
The sliding windows in time and frequency for CTFs estimation are designed using 4 pilots
per window filter slide. Observing the frequency direction in Figure 4.9, we can see that the
hexagonal pattern renders 2 sets of pilot patterns, which consequently refers to a two sets
filter coefficients requirement. The proposed hexagonal pilot pattern with virtual pilot
symbols simplifies the implementation of hexagonal pilot pattern filtering due to the new
structure that resembles the rectangular pilot patterns structure. According to the parameters
in Table 4.1 there will be 640 virtual pilots per frame required to aid the filtering. Due to the
existence of de-activated carriers and pilot shifting, we cannot use eq. (4.25) to calculate the
number of extra operations (interpolation/ extrapolation + decision directed) required before
filtering is applied, but the number of virtual pilots in a frame and the M of the data at the
virtual pilot positions suffice.
103
0
10
M=2
M=4
M=8
-1 M=16
10 M=32
MSE
-2
10
-3
10
-4
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.19: Channel estimation MSE vs SNR(dB) comparison of combined hexagonal
pattern with virtual pilots from interpolation/extrapolation and the decision directed method
with different constellation size (M) of the symbol at the virtual pilot positions
0
10
M=2
M=4
M=8
-1 M=16
10 M=32
BER
-2
10
-3
10
-4
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.20: BER vs SNR(dB) comparison of combined hexagonal pattern with virtual pilots
from interpolation/extrapolation and the decision directed method with different constellation
size (M) of the symbols at the virtual pilot positions
104
Having M=4 means that 3840 extra computations are required in the application of virtual
pilots, where 640 computations are dedicated to linear extrapolation/interpolation, 2560
computations for calculating the Euclidean distance, and 640 division operations between
received data at virtual pilot positions and the estimated symbol after the decision directed
process. The performance comparison is done in the form of MSE, which is the mean square
difference between the channel estimates and the real CTFs which according to reference [86]
is the appropriate measure to compare the performance of different channel estimation
techniques. The MSE results can not be translated directly as a one to one relationship to the
bit error rate (BER) [86]. This is due to the fact that in certain cases (although MSE occurs)
MSE does not deliver the decision errors onto the demapping (QAM or PSK) process. In the
presence of LU, we consider the LU band 200 KHz wide which is equal to about 40 OFDM
sub-carriers. Without loss of generality, the band is assumed to be located in the middle of the
CR band. By de-activating 61 sub-carriers the CR system provides about 10 sub-carriers (50
KHz) guard band with respect to the LU band; hence, the mutual interference is considered to
be negligible. Figure 4.21 shows the performance of the hexagonal pilot pattern with the use
of virtual pilots from interpolation/extrapolation and its combination with the decision
directed method in the presence of LU. The decision directed method provides at MSE of
2x10-3 about 2 dB SNR gain compared to the use of virtual pilots only derived from the
interpolation/extrapolation method.
In the presence of LU, more edge effect occurs, since the estimation using the sliding window
is divided into several regions. As the number of LUs increases and scattered within different
spectrum regions, the edge effect will occur more often. If the sliding window technique is
used as filtering method (i.e. in the situation a large number of LUs are scattered in the whole
band), the channel estimation performance will be degraded. Therefore, for higher numbers of
LUs the optimum Wiener filter (i.e. utilizing all pilots within an OFDM symbol) can be
applied in the simulations as long as the required memory to save the filter coefficients is
sufficient. Figure 4.22 shows the performance of the optimum Wiener filter with a hexagonal
pilot pattern using virtual pilots derived from interpolation, or from combined interpolation
and the decision directed method, and the one without virtual pilots. With the optimum
Wiener filter, and adding the decision directed method it provides about 2.5 dB gain at an
MSE 2x10-3 compared to estimating the virtual pilot using only interpolation, while the
hexagonal pattern without virtual pilots never reaches that MSE value (2x10-3) due to the
error floor from the mismatch error between the estimated shape and the actual channel power
delay profile and Doppler power spectrum shape. The optimum Wiener filter always
outperforms the filtering using the sliding window, because more pilots are utilized in the
estimation of one CTF. There are two factors that influence the channel estimation error, they
are noise and the mismatch in estimating the channel power delay profile and Doppler power
spectrum. Through all simulation results the impact of the inclusion of the virtual pilots on
our system can be seen at high SNR where the mismatch error is minimized and little noise
contribution exists. The impact of noise on the BER performance can be mitigated by using
channel coding and interleaving. In this way the gain of the virtual pilots to our system can be
105
seen in the lower SNR region. The results in Figures 4.16-4.22 have been presented in
reference [87].
0
10
Hex. pattern with interp. of virt. pilots
Hex. pattern with comb. interp. and dec. directed of virt. pilots
-1
10
MSE
-2
10
-3
10
0 10 20 30 40
SNR(dB)
Figure 4.21: Channel estimation MSE vs SNR(dB) comparison of the hexagonal pilot pattern
with virtual pilot symbols derived from interpolation/extrapolation and its combination with
the decision directed method in the presence of wideband LU
0
10
Hex. pat. w/ interp.of vir. pilots and opt. Filter
Hex. pat. w/ interp. + dec. directed of vir. pilots and opt. Filter
Hex. pat. opt. Filter without vir. pilots
-1
10
MSE
-2
10
-3
10
0 10 20 30 40
SNR(dB)
Figure 4.22: Channel estimation MSE vs SNR(dB) of the optimum Wiener Filter using the
hexagonal pilot pattern and virtual pilot symbols derived from interpolation/extrapolation and
its combination with the decision directed method; comparison with the one without virtual
pilots in the presence of wideband LU
106
4.5 MIMO OFDM Based Cognitive Radio Optimum Pilot Pattern with the
Virtual Pilots Concept
Since one of the major objectives in Cognitive Radio is the efficient spectrum utilization,
MIMO is considered as possible technique to reach this goal [2]. Extending our previous work
on single input single output (SISO) OFDM channel estimation by use of the Wiener filter
and applying the optimum hexagonal pattern (in the Cognitive Radio context with the aid of
virtual pilots), in this section we propose the application of this hexagonal pilot pattern with
the aid of virtual pilots for the MIMO system. Without loss of generality, we consider a
specific 2x2 MIMO system. The V-BLAST† algorithm is included in the receiver as the
technique to separate the data from each transmit antenna, and to detect the received signal
according the maximum likelihood detection. The general model of the MIMO OFDM
channel estimation with the V-BLAST receiver for the Cognitive Radio system is depicted in
Figure 4.23.
In our 2x2 MIMO system two spatial streams are transmitted. Due to multipath propagation
the signal travels along different paths. There will be 4 independent links in the MIMO
channel. The time-domain signals on the two received antennas are defined as:
y1 = s1 ∗ cirT1R1 + s2 ∗ cirT2 R1 + n1
(4.26)
y2 = s1 ∗ cirT R + s2 ∗ cirT + n2
1 2 2 R2
where y1 and y2 are the signals received by receive antennas 1 (R1) and 2 (R2) respectively. s1
and s2 are the symbols from transmit antennas 1 (T1)and 2 (T2) respectively. The cirT1R1 ,
cirT1R2 , cirT2 R1 , cirT2 R2 are the channel impulse response (CIR) between T1 & R1 , T1 & R2 ,
T2 & R1 and T2 & R2 , respectively, while n1 and n2 are the noise terms in each of the receive
antennas which are independent of each other.
The channel transfer function (CTF) matrix will be:

⎛ CTFT1R1 CTFT2 R1 ⎞
CTF2 Χ 2 = ⎜ ⎟ (4.27)
⎜ CTFT R CTFT2 R2 ⎟⎠
⎝ 1 2
The CTF is derived from the Fourier transform of the CIR. Accordingly, the frequency
domain received signals of the two received antennas will be:
Y1 = S1CTFT R + S 2 CTFT + N1
2 R1
1 1
(4.28)
Y2 = S1CTFT R + S 2 CTFT + N2
1 2 2 R2
†
V-BLAST (Vertical Bell Laboratories Space Time) algorithm is a detection algorithm on the received signals
in a MIMO system. First detection is applied on the strongest signal, then it removes the contribution of the
strongest signal on the received signal for the detection of the second strongest signal, and further continues the
same process until the weakest signal has been detected.
107
where Y1 and Y2 are the frequency domain signals received by antennas R1 and R2
respectively. S1 and S2 are the frequency-domain symbols from transmit antennas T1 and T2
respectively.
Spectrum
Sensing
Spectrum
Occupancy
Info
Source PSK/QAM Pilots + Data S/P IFFT P/S D Tx1

bits Mapping Multiplexer GI M
Insertion U Tx2
X
Pilots
TxM
(a) Transmitter
Spectrum
Occupancy
Info
Pilot Channel estimation QAM/PSK D

GI
FFT subtraction for for link Tx 1…Tx M – Demapping E
Rx1 Removal
Tx 1...Tx M Rx1 Tx 1 T
E
C
Rx2 GI FFT Pilot Channel estimation T
Removal V-BLAST E
subtraction for for link Tx 1…Tx M –
signal D
Tx 1...Tx M Rx2
processing
B
I
GI T
RxN FFT Pilot Channel estimation QAM/PSK
Removal S
subtraction for for link Tx 1…Tx M – Demapping
Tx 1...Tx M RxN Tx M
(b) Receiver
Figure 4.23: System model of the MIMO OFDM Based Cognitive Radio at (a) transmitter
and (b) receiver side
We propose the pilot grid of the two transmitters. The grid must be orthogonal to each other
so the channel state of each link can be properly estimated. The receiver has to recognize the
pilots dedicated to each transmitter. We adopt the hexagonal pattern frame in the SISO system
108
[87]. The hexagonal pattern can be separated into two rectangular (comb-type) patterns. One
rectangular pattern is dedicated to one transmit antenna while the other one is dedicated to the
second transmit antenna. In this way at the receiver the received frame will be the
combination of the two rectangular patterns which form a desired hexagonal type pilot
pattern. In both patterns the pilot distance (either in time or frequency) has to fulfill the
sampling theorem as described in eq. (4.15).
The pilot pattern for the first and second transmitters are depicted in Figure 4.24. The black
boxes denote the pilot positions, the X boxes are virtual pilot positions, and the boxes signed
by 0s are entailed to have pilot orthogonality between the first and the second transmit
antenna pilot pattern. Data symbols are allocated to the white boxes. To calculate the number
of pilots in a certain direction (frequency or time) we use the general formula:
⎛ N sym − A0 ⎞
N p = 1+ ⎜ ⎟ (4.29)
⎝ d ⎠
where Np is the number of total pilots in a certain direction (frequency or time), Nsym is either
the number of active carriers for calculating the number of pilots in frequency or the number
of OFDM symbols per frame for calculating the number of pilots in time. The parameter A0 is
the position of the first pilot symbol and d is the spacing between the pilots in frequency or
time. The zeros or nulling are applied to reserve those positions for the pilots of the other
transmitter and thus to prevent any interference or addition of pilots while adding / combining
those two frames together at the receiver.
We can notice from the pattern in Figure 4.24 that the position of the virtual pilots for both
transmit antenna frames is the same. This reduces the processing complexity at the receiver.
X X 0 X 0 X
X 0 X 0 X X
X X 0 X 0 X
X 0 X 0 X X
X X 0 X 0 X
X 0 X 0 X X
0 X 0 X
X X
(a) (b)
Figure 4.24: Pilot pattern for (a) the first and (b) the second transmit antennas; ‘X’ denotes the
virtual pilot position, ‘0’s are inserted to preserve orthogonality between pilots dedicated to
transmit antennas 1 and 2, black boxes are the pilot positions, and white boxes are data
positions
109
The resulting received frame at the receiver is as shown in Figure 4.25. The virtual pilots are
estimated from two neighboring pilots [87]. This means that virtual pilots in between two real
pilots are determined by linear interpolation. This is applied based on the assumption that the
channel doesn’t change or changes slowly between pairs of pilots. The virtual pilots at the
edges of the frame are estimated by simple extrapolation of closeby pilots. The final pattern
will resemble as if twice pilot oversampling is applied. This gives a better estimate of the
channel. The decision directed process applied in a SISO system [87] has not been included
to the system. This is because the overlapped received signal (addition of signal from transmit
antennas 1 and 2), as described in eq. (4.28), at the virtual pilot positions will lead to high
decision errors.
Figure 4.25: Received frame pattern on both receive antennas (Black boxes are the pilots
while the white boxes are the data)
The initial channel estimate at the boxes signed by 0s are derived only from linear
interpolating / extrapolating of adjacent pairs of initial estimates ( i.e. previously derived
initial channel estimates at virtual pilot positions ), which can be seen as a linear interpolation
involving 4 initial channel estimates at real pilot positions. Finally the 0 boxes will become
virtual pilots as well. Due to the use of the virtual pilot concept to the proposed pilot pattern,
the outcome of the final pattern will resemble the regular comb type pilot pattern. The pattern
on antenna 2 (depicted in Figure 4.24(b)) so provides more virtual pilots compared to antenna
1 (depicted in Figure 4.24(a)). This regularity is important to obtain a proper and simple
filtering process. The receiver of each receive antenna will separate the pilots from the data.
The pilots dedicated to the first transmit antenna will be separated from the rest of the pilots
which are dedicated to the other antenna. In this way the channel states of all links can be
estimated.
In the case where the initial channel estimates at the virtual pilot positions derived from linear
interpolation or extrapolation gives large deviation error, then there will be an additional
110
mismatch contribution to the filtering process. The impact can be seen as the number of filter
coefficients is increasing. In this case the eq. (4.4) can be written as :

H n ,k = ∑
{ n ', k '}∈P
ω n ', k ', n , k Hˆ n ', k ' + ∑ω
{ v }∈V
n ', k ', n , k Hˆ v
= ∑
{ n ', k '}∈P
ω n ', k ', n , k Hˆ n ', k ' + ∑ω
{ v }∈V
n ', k ', n , k ( H v + ΔH v ) (4.30)
where V is the set of the virtual pilot positions. The deviation from linear interpolation or
extrapolation (ΔH) comes from the noise average contribution as described by eq. (4.21) and
the interpolation/ extrapolation mismatch. By observing eq. (4.30) we can see that by sliding
window filtering, and increasing the filter size, due to the proposed pilot pattern, it will
consequently increase the number of virtual pilots involved in estimating the CTF and that
will accumulate an extra error. In addition to this error, the mismatch model used in designing
the filter (ω) will sum up the channel estimation error. This analysis will be verified in the
performance evaluation given in subsection 4.5.1. According to references [78]-[81] the
hexagonal sampling pattern is considered to be the most efficient two-dimensional sampling
scheme, since it requires 13.4% fewer samples than rectangular sampling in order to represent
the same circularly band limited continuous signal. The previously proposed MIMO 2x2
pattern adopts the SISO hexagonal pattern.
Further we compare the results with previous MIMO pattern with the new pattern for each
transmit antenna as depicted in Figure 4.26. The pattern adopts the SISO rectangular pattern
as shown in Figure 4.26 (a), resulting into 2 different transmit antenna hexagonal pilot
patterns depicted in Figure 4.26 (b) and (c). Zeros of the pattern are for the purpose of having
orthogonality between the pilot dedicated to transmit antennas 1 and 2. It can be observed
from the second MIMO pattern (pattern in Figure 4.26 (a)) that it requires more real pilots
compared to the first MIMO pattern (pattern in Figure 4.25). Two virtual pilot allocation
techniques have been evaluated. The first one is depicted in Figure 4.27 where the virtual
pilots are only allocated on the zeros, and the second one is shown in Figure 4.28 where
virtual pilots are allocated on the zeros and between the zeros and the real pilots. The first
MIMO pilot pattern with the second MIMO pilot pattern will be compared in subsection 4.5.1
in terms of their channel estimation mean square error (MSE) and bit error rate (BER). The
second MIMO pilot pattern includes the two virtual pilot allocation techniques.
The three MIMO pilot pattern techniques (described in Figures 4.24, 4.27 and 4.28) will be
compared, and the one that gives a better performance will be chosen as pattern for the
Cognitive Radio system, where some carriers will be de-activated due to the licensed user
access. The three pilot pattern techniques have been simulated. The results indicate that the
first pattern (described in Figure 4.24) gives a better channel estimation performance than the
second pattern ( Rectangular to Hexagonal) either with less virtual pilots or with more virtual
pilots (Figure 4.27 and 4.28); therefore the first pattern will be utilized in the Cognitive Radio
111
system. The results will be further discussed in the performance evaluation and analysis in
subsection 4.5.1.
freq.
time
(a)
Zeros
Pilot of T1
freq. freq.
Pilot of T2
time time
(b) (c)
Figure 4.26: The second proposed Pilot pattern for MIMO 2x2 derived from (a) the
Rectangular Comb Pilot pattern to Hexagonal Pilot pattern on (b) transmit antenna 1 (T1) and
(c) transmit antenna 2 (T2)
112
freq.
V V
V V V
V V
V V V
V V
V V V
V V
(a) time
freq. V V V
V V
V V V
V V
V V V
V V
V V V
(b) time
Pilot of T1 Pilot of T2 V Virtual Pilot Positions on Zeros
Figure 4.27: The second proposed Pilot pattern for MIMO 2x2 derived from the Rectangular
Comb Pilot pattern to Hexagonal Pilot pattern with less virtual pilots on (a) transmit antenna 1
(T1) and (b) transmit antenna 2 (T2)
113
freq.
V V
V V V V V
V V V
V V V V V
V V
V V V V V
V V V
V V V V V
V V
V V V V V
V V V
V V V V V
V V
(a) time
freq. V V V
V V V V V
V V
V V V V V
V V V
V V V V V
V V
V V V V V
V V V
V V V V V
V V
V V V V V
V V V
(b) time
Pilot of T1 Pilot of T2 V Virtual Pilot Positions on Zeros
V Virtual Pilot Positions on Non Zeros
Figure 4.28: The second proposed Pilot pattern for MIMO 2x2 derived from the Rectangular
Comb Pilot pattern to Hexagonal Pilot pattern with more virtual pilots on (a) transmit antenna
1 (T1) and (b) transmit antenna 2 (T2)
114
In the presence of licensed users, the shifted pilot scheme in Figure 4.9 will be applied, and
the virtual pilot concept will be added to have a regular distance between initial channel
estimate positions for the simplicity of the Wiener filtering operation. The new pattern is
time
X X 0 X 0 X
freq.
X 0 X 0 X X
X X
X X X
X X0 XX X0 X
X
X LU
0 BandX 0 X LU Band
X
X X X X
X X0 XX X0 X
X
X 0 X 0 X X
0 X 0 X
X X
(a) (b)
Figure 4.29: Pilot pattern for (a) the first and (b) the second transmit antenna in the presence
of LU; ‘X’ denotes the virtual pilot position, ‘0’s are inserted to preserve orthogonality
between pilots dedicated to transmit antennas 1 and 2, the black boxes are the pilot positions,
and the white boxes are data positions
4.5.1 Performance Evaluation and Analysis
We consider the wide sense stationary uncorrelated scattering (WSSUS) channel model in our
simulations since it relates the power delay profile, number of multipaths, and Dopler
frequency of the channel in a simple and direct way. The simulation parameters are equal to
the SISO evaluation provided in Table 4.1. By adopting the hexagonal pilot pattern in the
SISO system as depicted in Figure 4.25 and allocating pilots for both transmit antennas
according to Figure 4.24, the distance between pilots in time is increased by a factor of two.
This leads to degradation of the performance of the estimator. However applying the virtual
pilots will make the effect of having the original pilots distance in time (dt=3). Having more
virtual pilots at the edges of the frame (edge effect) causes more performance degradation.
This is because the correlation at the edges is less and each time a virtual pilot is interpolated
according to eq. (4.20) or eq. (4.21) there is some interpolation error which accumulates and
causes a bigger MSE. In the Wiener filtering process either in time or frequency the sliding
window utilizes 4 pilots to estimate the CTFs in one slide.
Figure 4.30 shows the MSE of the four links of the first 2x2 MIMO system utilizing the
virtual pilots, compared with the SISO system using a rectangular pilot pattern with df=4,
dt=6 and df=4, dt=3 (two-times pilot oversampling in time). The SISO with df=4, dt=6 forms
115
an MSE floor in the order of 10-1, the SISO with two-times pilot oversampling in time has an
error floor of around 5x10-2. The proposed MIMO 2x2 pilot pattern with the virtual pilot
concept will do a two-times pilot oversampling in time and in frequency; the results in Figure
4.30 show that the proposed pattern outperforms the two-times pilot oversampling in time by
obtaining an error floor 10-2. Due to the deviation error from linear interpolation/extrapolation
as described in eq. (4.30) the performance cannot reach the performance for the two-times
pilot oversampling in time and in frequency.
Figure 4.31 depicts the advantage of the V-BLAST algorithm in exploiting the channel
diversity in detecting the transmitted bits where it forms a BER floor of around 6x10-3.
Meanwhile the SISO scheme with the same amount of real pilots (without virtual pilots)
forms a BER floor of around 6x10-2, the SISO with two-times oversampling forms a BER
floor 2x10-2, while the SISO with two-times oversampling in time and frequency forms a BER
floor of around 10-4. The results in Figure 4.30 and 4.31 have been disseminated in reference
[88].
The impact of increasing the Wiener filter size to the channel MSE and the BER can be seen
in Figure 4.32 and 4.33 respectively. Figure 4.32 shows that the MSE of the proposed MIMO
pattern with 4x4, 8x8 and 16x16 form almost the same level of error floor. The deviation from
the linear interpolation/extrapolation at the virtual pilot positions and the mismatched design
of the Wiener filter add together into the overall channel MSE as described in eq. (4.30) and
eventually impacts the BER outcome. Since in the SISO case without the presence of virtual
pilots the deviation does not exist, it means that as the Wiener filter size is increased a
significant performance improvement can be seen.
0
10
-1
10
-2
10
MSE
-3
10
SISO Rect.
SISO Rect 2x Ovsmpl. in Time
SISO Rect 2x Ovsmpl in Time and Freq. Dir.
-4
10 MIMO Link 11 w/ Virtual Pilots
MIMO Link 21 w/ Virtual Pilots
MIMO Link 12 w/ Virtual Pilots
-5 MIMO Link 22 w/ Virtual Pilots
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.30: MSE vs SNR (dB) comparison between SISO Rectangular, Hexagonal and
MIMO pilot patterns with a 4x4 Wiener Filter size
116
0
10
-1
10
-2
10
BER
-3
10
-4
10 SISO Rect.
SISO Rect 2x Ovrsmpl. in Time Dir.
SISO Rect 2x Oversampl. in Time and Freq. Dir.
-5
MIMO 2x2 w/ Virtual Pilots
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.31: BER vs SNR (dB) comparison between SISO Rectangular, Hexagonal and
MIMO pilot patterns with a 4x4 Wiener Filter size
0
10
-1
10
MSE
-2
10
-3
10 MIMO w/ Filter Size 4x4
MIMO w/ Filter Size 8x8
SISO Rect. w/ Filter Size 4x4
-4
10 SISO Rect. w/ Filter Size 8x8
SISO Rect. 2x ovsmp. w/ Filter Size 4x4
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.32: MSE vs SNR (dB) comparison between SISO Rectangular, Hexagonal and
MIMO pilot patterns with increasing number of filter coefficients (number of pilots used per
sliding window)
117
0
10
-1
10
BER
-2
10
-3
10 MIMO w/ Filter Size 4x4
-4
10 SISO Rect. w/ Filter Size 8x8
SNR(dB)
SISO Rect. 2x oversamp. w/ Filter Size 4x4
0 5 10 15 20 25 30 35 40
Figure 4.33: BER vs SNR (dB) comparison between SISO Rectangular, Hexagonal and the
proposed MIMO pilot patterns with increasing number of filter coefficients (number of pilots
used per sliding window)
The performance comparison among the first proposed MIMO pattern (Figure 4.24), the
second MIMO pattern with less virtual pilots (Figure 4.27) and more virtual pilots (Figure
4.28) at a high maximum Doppler frequency (fDmax = 82 Hz) are shown in Figure 4.34 and
4.35, while the performance comparison at low Doppler frequency (fDmax = 2.05 Hz) are
depicted in Figure 4.36 and 4.37.
The results in Figures 4.34 to 4.37 show that the first proposed MIMO pattern (Figure 4.24)
outperforms the other two techniques (Figure 4.27 and 4.28). The first proposed MIMO
pattern does not only give better channel estimation MSE but also requires less real pilots
compared to the second proposed MIMO pattern. Having less pilots refers to a higher data
rate. The virtual pilot concept can optimally be utilized in slowly fading channels. The
comparison of the proposed MIMO pattern in high Doppler and low Doppler confirms this
fact. The best performance is given by the Wiener filter 16x16 of the first proposed MIMO
pattern. At high Doppler frequency this pattern forms an error floor in the order of 9x10-3
while at low Doppler frequency the error floor is in the order of 2x 10-3. The MSE
performance gap between the first proposed MIMO pilot pattern (Figure 4.24) compared to
the other two techniques (Figure 4.27 and 4.28) is more significant at low Doppler frequency
than at high Doppler frequency as depicted in Figures 4.34 and 4.36. This phenomenon
happens since the first proposed MIMO pilot pattern emphasizes the virtual pilot utilization of
the first and second transmit antennas in time. Therefore, as the Doppler frequency is low the
channel is slowly varying in time, and the aid of the virtual pilots in this direction will be
significant. Meanwhile, in the other two techniques the virtual pilot utilization is emphasized
118
in frequency. Therefore, the impact of the virtual pilot utilization in the two other techniques
depicted in Figure 4.27 and 4.28 becomes significant for the channel with lower delay spread.
Figure 4.38 shows for all links the MSE on the 2x2 MIMO system with three MIMO pilot
patterns with an 8x8 Wiener filter for the high maximum Doppler frequency (fDmax = 82 Hz).
In the first pattern (described in Figure 4.24), more pilots are allocated to the first transmit
antenna compared to the second antenna; therefore the MSEs on links 11 and 12 can be better
compared to links 21 and 22 respectively. The MSE performance of the second pattern with
less and more virtual pilots of all links are close to each other due to the same number of
allocated pilots on transmit antennas 1 and 2.
Hex. > Rect. w/ 4x4 Filter

0 Hex. > Rect. w/ 16x16 Filter
10 Rect. > Hex. w/ 4x4 Filter df=4
Rect. > Hex. w/ 8x8 Filter df=4
-1
Mean MSE
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.34: Mean MSE vs SNR (dB) comparison of all links between the first proposed
MIMO pilot pattern (Hexagonal > Rectangular), and the second proposed MIMO pilot pattern
(Rectangular > Hexagonal) with less virtual pilots (df=4) and more virtual pilots (df=2) for
different number of filter coefficients (number of pilots used per sliding window) at high
maximum Doppler (fDmax = 82 Hz)
119

0
-1
BER
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.35: BER vs SNR (dB) comparison of all links between the first proposed MIMO
pilot pattern (Hexagonal > Rectangular), and the second proposed MIMO pilot pattern
different number of filter coefficients (number of pilots used per sliding window) at high
maximum Doppler (fDmax = 82 Hz)
-1 Rect. > Hex. w/ 8x8 Filter df=2
10
Mean MSE
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
different number of filter coefficients (number of pilots used per sliding window) at low
maximum Doppler (fDmax = 2.05 Hz)
120
0
-1
BER
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
different number of filter coefficients (number of pilots used per sliding window) at low
maximum Doppler (fDmax = 2.05 Hz)
Hex. > Rect. w/ 8x8 Filter Link 11
0 Hex. > Rect. w/ 8x8 Filter Link 21
10 Hex. > Rect. w/ 8x8 Filter Link 12
Hex. > Rect. w/ 8x8 Filter Link 22
Rect. > Hex. w/ 8x8 Filter d =4 Link 11
f
f
f
f
-1
10 Rect. > Hex. w/ 8x8 Filter d =2 Link 11
f
Mean MSE
f
f
f
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.38: Mean MSE vs SNR (dB) comparison of each link between the first proposed
(Rectangular > Hexagonal) with less virtual pilots (df=4) and more virtual pilots (df=2) for 8
filter coefficients (8 pilots are used per sliding window) in high maximum Doppler (fDmax = 80
Hz)
121
In the presence of the licensed user, some carriers will be de-activated. Like our evaluation in
the SISO system, in this simulation scheme 61 carriers are de-activated (carrier no. 81 – 141)
due to LU occupancy. The LU itself occupies 40 carriers in the middle that renders 10 guard
bands on each side of the LU band. As too many carriers are de-activated the mutual
interference between LU and our Cognitive Radio system becomes negligible. The distance
between pilots in frequency is 4 sub-carriers, while in time it is 3 OFDM symbols. Figure
4.39 and 4.40 show the performance of the first proposed MIMO pilot pattern in the presence
of LU (Figure 4.29) with a 4x4, 8x8, and 16x16 Wiener Filter size at high maximum Doppler
frequency, while the performances at low maximum Doppler frequency are depicted in
Figures 4.41 and 4.42.
Beside the effectiveness of the virtual pilot concept in slowly varying channel as presented in
the comparisons of Figures 4.34 to 4.37, the results in Figures 4.39 to 4.42 show how the
active licensed user can deteriorate the MSE performance of the proposed MIMO pilot pattern
in a Cognitive Radio system. A particular result that can be observed from these Figures is
that the MSE performance in the case of the proposed MIMO pilot pattern without the
presence of an active licensed user (while having some carriers de-activated) is close to the
MIMO system without carriers de-activation performance.
0
Hex. > Rect. w/ 4x4 Filter , Carr. Deact, No LU
Hex. > Rect. w/ 8x8 Filter Carr. Deact, No LU
-1
Hex. > Rect. w/ 4x4 Filter Carr. Deact, Active LU
10 Hex. > Rect. w/ 8x8 Filter Carr. Deact, Active LU
Mean MSE
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.39: Mean MSE vs SNR (dB) of the first proposed MIMO pilot pattern (Hexagonal >
Rectangular) without carriers de-activation, with carriers de-activation without the presence of
a licensed user (LU), and with carrier de-activation and active LU at high maximum Doppler
(fDmax = 80 Hz)
122

0
-1
BER
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.40: BER vs SNR (dB) of the first proposed MIMO pilot pattern (Hexagonal >
a licensed user (LU), and with carrier de-activation and an active LU at high maximum
Doppler (fDmax = 80 Hz)
0
10
-1 Hex. > Rect. w/ 16x16 Filter Carr. Deact, No LU
10
Mean MSE

-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.41: Mean MSE vs SNR (dB) of the first proposed MIMO pilot pattern (Hexagonal >
a licensed user (LU), and with carrier de-activation and an active LU at low maximum
Doppler (fDmax = 2.05 Hz)
123

0
-1
BER
-2
10
-3
10
0 5 10 15 20 25 30 35 40
SNR(dB)
Figure 4.42: BER vs SNR (dB) of the first proposed MIMO pilot pattern (Hexagonal >
a licensed user (LU), and with carrier de-activation and an active LU at low maximum
Doppler (fDmax = 2.05 Hz)
There are two factors that influence the accumulation of MSE :

1) the edge effects due to the filter design according Figure 4.1 and the sliding window
channel estimation technique depicted in Figure 4.2
2) the deviation error by the virtual pilots that will be accumulated, as the number of
filter coefficients is increased (described in eq. (4.30)).
The MIMO system without carriers de-activation will experience less edge effects compared
to the MIMO system with carriers de-activation. Depending on the size of the carriers de-
activation band, the MIMO system with carrier de-activation will have less virtual pilots
compared to the MIMO system without carrier de-activation. By having the current number of
de-activated carriers and other simulation parameters given in Table 4.1, the curves in Figures
4.39 to 4.42 are the results of the two accumulated channel deviation errors (edge effects and
virtual pilots). It means that the number of virtual pilots in the region of de-activated carriers
has managed to balance the error due to the edge effects of the MIMO system with de-
activated carriers.
124
4.6 Training and Pilot Pattern for OFDM Based Cognitive Radio
Synchronization and PAPR Reduction
Beside channel estimation, synchronization and peak to average power ratio (PAPR) are
common problems in an OFDM system. Therefore the Cognitive Radio system that uses
OFDM as modulation technique will also face these common problems. In this section the
impact of synchronization and PAPR reduction will be explored and analyzed within the
Cognitive Radio context.
4.6.1 Synchronization for an OFDM-Based Cognitive Radio System
Carrier frequency, sampling frequency and frame offset are the common problems that an
OFDM system normally experience. In this section we overview suitable synchronization
techniques to compensate the three offset problems to be applied in an OFDM-based CR
system.
A particular point to be addressed here is the impact of the carriers de-activation to the
synchronization technique. The guard interval (GI) can be utilized to estimate the carrier
frequency and frame offset [89]–[92]. The guard interval is a replica of the last NGI time-
domain OFDM symbols. In this way by using the cross-correlation method the beginning of
an OFDM symbol and the carrier frequency offset can be detected. In [93] the frame offset
estimation is improved by quantizing the time-domain OFDM symbol into a 4-QAM symbol
constellation. The in-phase (I) or quadrature(Q) symbol values which are equal or bigger than
zero are assigned by a value of 1, while the rest are -1. Further averaging the cross correlation
value of the quantized time-domain OFDM symbol for the entire OFDM symbols in a frame
will add to the robustness of the frame offset detection.
The frequency offset can be detected by utilizing the cross correlation amplitude between the
time-domain OFDM symbol with its shifted version. The shift length is chosen as the period
between the time-domain OFDM symbol with its replica, which practically in every OFDM
symbol is the resemblance of the distance between the symbol on GI and its replica on the
useful OFDM symbol. Beside utilizing the GI, training symbols can be inserted to the OFDM
frame [94]. According to the IFFT property, half of the time-domain OFDM symbol can be
replicated to the rest half of the time-domain signal by assigning zeros to the frequency
domain OFDM symbols on the carriers with odd index number, while the index is starting
from 0 (e.g. index k=0,…, NFFT-1). The scheme is depicted in Figure 4.43.
The replication scheme itself can be analytically proven by expanding the condition of the
OFDM time domain symbol xn = x N FFT as given in eq. (4.31),
+n
2
125
xn = x N FFT
+n
2
⎛N ⎞
j 2π k ⎜ FFT + n ⎟
N FFT −1 j 2π kn N FFT −1 ⎝ 2 ⎠
1 1
N FFT
∑
k =0
X ke N FFT
=
N FFT
∑
k =0
X ke N FFT
N FFT −1 j 2π kn
1
=
N FFT
∑
k =0
X ke N FFT
e jπ k
left eq. equals to right eq. , only if e jπk = 1
(4.31)
N FFT
f
0 0 0 0 0 0 0 0
N F FT /2 IFFT
t
a b c d e f g h a b c d e f g h
Repetition
Figure 4.43: Training symbol design for synchronization purpose
According to eq. (4.31) the condition can be satisfied for any kind of symbol Xk values on
even index subcarriers ( k=0,2,4,…, NFFT-2), while the value of Xk on the odd index
subcarriers (k=1,3,5,…, NFFT-1) must be zero. The detail proof is available in Appendix B.
In a Cognitive Radio context there are two aspects to be considered for designing the training
symbols for synchronization. The first one is the requirement of having replica of the time
domain training symbol within the distance of NFFT /2 by assigning zeros on the odd index
carriers in the frequency-domain training symbols. The second aspect concerns the
application of carriers de-activation on the LU band in order to avoid mutual interference
between the Cognitive radio system and the LU by putting zeros on the location of the LU
band in the frequency-domain training symbols. The designed training symbol will not harm
the LU transmission, and in the Cognitive Radio system itself the mutual interference can be
avoided hence more accurate synchronization would be attained. By cross-correlating these
training symbols with its reference at the receiver, the noise and fading effect that can degrade
126
the cross-correlation value will be mitigated. The cost of this technique is a reduction in bit
rate due to the training symbol.
The sampling frequency offset can be estimated from the phase difference between two pilots
positions in one OFDM symbol and two pilots positions in another OFDM symbol [95]. This
technique is improved in reference [96] by applying equalization to mitigate the fading effect
on the sampling frequency offset and by iterative averaging over all pilots. Since the
synchronization problem can be solved with the insertion of pilots and training symbols, the
relation with the Cognitive Radio context would only be in interference avoidance by de-
activating the pilots or training symbols on the LU band. In this way the time, frequency and
sampling frequency offset can be more accurately estimated.
4.6.2 Peak To Average Power Ratio Reduction for an OFDM-Based Cognitive Radio
Channel Estimation
The peak to average power ratio (PAPR) reduction for an OFDM-based Cognitive Radio can
be included by using a sidelobe suppression technique. This extra constraint on PAPR (see eq.
(2.1) in chapter 2) means that the frequency-domain symbol composition (data and pilots)
also should minimize the PAPR. The sidelobe suppression constraint becomes:
2
1 N FFT (1+ α ) Tu
∑ p X
− j 2π ( f − f n ) t
∫
2
X(f) = g (t )e dt
N FFT n =1
n n
− (1+ α ) Tu
2 N FFT (4.32)
subject to X( f LU
) < PThreshold , ∑ p
n =1
n
= PT and
⎛ N FFT −1 2
⎞
⎜ ⎟
∑
j 2 π nk / N
max Xne FFT
k ∈{0,..., N −1} ⎜ n=0 ⎟

⎝ ⎠
FFT
min PAPR = min

⎧ N FFT −1 2
⎫
⎪ ⎪
E ⎨ ∑ Xne
j 2 π nk / N
FFT
⎬
⎪⎩ n = 0 ⎪⎭
In section 3.2.4 the PAPR has been analyzed and evaluated with respect to the application of
adaptive bit loading and sidelobes cancellation carriers. In this section the PAPR is analyzed
in relation to the pilot insertion. In reference [97] the PAPR reduction is applied by utilizing a
choice of pilot values inserted in OFDM symbols in such a way that the pilots can reduce the
PAPR. Pilots can be chosen from several sets of orthogonal pilot sequences. The pilots that
most satisfy the constraints in eq. (4.32) will be chosen for transmission. No extra information
about the pilot sequence is required. Since the pilot sequences are orthogonal to each other,
the receiver can detect the transmitted pilot sequence by cross-correlating the received pilot
sequence with the reference pilot sequences and choose the sequence with the highest cross
127
correlation value (maximum likelihood method). A drawback of this orthogonal pilot

sequence is that the pilots should always be available in each OFDM symbol. Some of the
carriers in each OFDM symbols will therefore be as pilots, so the data rate will be reduced.
In the Cognitive Radio context some OFDM carriers will be de-activated due to the licensed
user occupancy. The scheme can be seen as non-contiguous OFDM. In reference [98] the
impact of the OFDM carriers de-activation to the PAPR has been investigated. The results
show that the probability of high PAPR increases with the number of total subcarriers in the
case of constant active subcarriers, therefore the PAPR reduction is more important in the
case of large number of de-activated subcarriers compared to conventional OFDM (without
carriers de-activation).
4.7 Conclusions
In this chapter the Wiener filter has been extensively used in the Cognitive Radio application.
The Wiener filter is chosen since it directly relates the index of the subcarriers and OFDM
symbol to the correlation property of the channel in frequency and time. The sliding window
technique is utilized to avoid a high memory requirement in storing the Wiener filter
coefficients. Frequency hopping and spectrum pooling contexts have been discussed. In the
frequency hopping scenario the sliding window is applied in one region, while in the spectrum
pooling context one active subcarriers group will be located in a different region than another
active subcarriers group (separated by subcarrier holes) because of carriers de-activation in
the licensed user’s band. The proposed OFDM frame design according to the Bluetooth TDD
duration refers to the short frame duration; therefore a training symbol is adequate in
estimating the channel transfer functions in an OFDM frame as long as the frame duration is
shorter than the coherence time of the channel.
The Wiener filter interpolates the channel transfer function by utilizing a noisy channel
sample at the pilot positions. The pilots are spread among an OFDM frame and fulfill the
sampling theorem. The distance between pilots in frequency should be lower than the
coherence bandwidth of the channel while the distance between pilots in time should be lower
than the channel coherence time.
The virtual pilot concept is employed with the purpose of improving the performance of the
Wiener filter by adding pilot oversampling. The virtual pilots are not real pilots, and are
located between real pilots. The virtual pilots are derived from linear
interpolating/extrapolating the noisy channel samples at real pilot positions. The decision
directed method will enhance the accuracy of the estimated channel transfer function value at
the virtual pilot positions. This method could only be applied efficiently in a SISO system. In
the MIMO system the signals from all transmit antennas will contribute to the received signal
of one receive antenna; therefore a direct application of the decision directed method to the
estimated channel transfer function (resulting from linear interpolation/extrapolation at the
virtual pilot positions) will give a higher probability of decision errors and that will lead to a
128
high channel estimate deviation error. An extra mechanism is thus required to separate the
signal from each transmit antenna in order to make the decision directed technique properly
utilized. For this reason, in this chapter the initial channel estimates at the virtual pilot
positions for MIMO system are only derived from linear interpolation/ extrapolation.
A literature study has shown that the hexagonal pattern is the optimum pilot pattern to sample
2 dimensional circularly bandlimited signals. Adding the virtual pilot concept to the
hexagonal pilot pattern, and combining the linear interpolation/extrapolation and the decision
directed technique in predicting the initial channel estimates at virtual pilot positions for the
SISO system provides a significant channel estimation improvement. It gives almost the same
performance as the two-times pilot oversampling. Further, in Cognitive Radio context, the
virtual pilot concept simplifies the Wiener filtering by the sliding window technique.
The three pilots allocation techniques for MIMO system have been proposed. Without loss of
generality a 2 transmit antenna system was used. One pattern adopts the Hexagonal SISO
pattern and separate the pilots of the first and the second transmit antenna in time. The other
patterns adopt the Rectangular SISO pattern and separate the pilot allocation of the first and
the second antenna with emphasis on frequency. The new pattern forms a hexagonal pattern
on each of the transmit antenna. Given the channel model in Table 4.1, the simulation results
show that among the three patterns the MIMO pilot pattern that adopts the SISO Hexagonal
pattern outperforms the other two patterns. The impact of delay spread to the channel
correlation in frequency is more accommodated by the MIMO pattern that adopts the SISO
hexagonal pattern. This is due to smaller distance between the real pilots of each of the
transmit antennas. The aid of virtual pilots in time becomes significant in a slowly varying
fading channel. Therefore, the channel MSE performance gap between the pattern that adopts
the SISO Hexagonal pattern and the ones that adopts the SISO Rectangular pattern is more
significant in a channel with low Doppler frequency compared to a channel with high Doppler
frequency.
With respect to the utilization of virtual pilots for MIMO channel estimation in Cognitive
Radio context, the accumulated MSE comes from the deviation error in the virtual pilots and
the edge effects of the Wiener filter. The more licensed users access the spectrum in a
scattered manner, the more edge effects will occur. More licensed users denote more carrier
de-activations that eventually reduces the number of virtual pilots; therefore there should be a
trade-off between the number of edge effects and the number of de-activated carriers at those
virtual pilot positions that contribute to the final channel MSE.
Peak to average power ratio (PAPR) and synchronization are common problems in the OFDM
system. Literature has given solutions to these challenges in conventional OFDM systems.
Combining the spectrum pooling concept (carriers de-activation) with the available
synchronization and PAPR reduction techniques can accommodate these common OFDM
aspects in a Cognitive Radio context.
129
Chapter 5
Single Carrier Cognitive Radio System
5.1 Introduction
Beside the multi-carrier approach another method of utilizing the spectrum efficiently in
cognitive radio context is by shaping the waveform of a single carrier transmitted signal
dynamically over a spectral region. The shaping is done so that it can adapt to a changing
electromagnetic environment and it has synthesized waveform features in frequency domain
which do not interfere with the occupied spectrum. The fundamental idea is to avoid a preset
frequency use and operate dynamically over a chosen spectral region. The Transform Domain
Communication System (TDCS) is introduced in references [99],[100] as a candidate for a
single carrier technique in Cognitive Radio. In principle the transmitted signal is shaped in
such a way that its bandwidth does not overlap with the LU band.
We have expanded the work in TDCS by adding an extra embedded symbol (in line with our
Cognitive Radio objective) for enhancing efficient spectral utilization [2]. The scheme,
analysis and performance evaluations are described in section 5.2.
In references [100] to [102] the Fourier transform as tool to shape the spectrum in TDCS is
replaced by the wavelet basis function. The technique is called Wavelet Domain
Communication System (WDCS). Following the extension of our research on TDCS, we also
propose to implement the extra embedded symbol in the WDCS system. The scheme,
analysis, limitation and evaluations are given in section 5.3.
Further, the MIMO application to the single carrier modulation Cognitive Radio system will
be explored in section 5.4.
The conclusions will be drawn in section 5.5.

130
5.2 Transform Domain Communication System (TDCS) with Embedded

Symbol
The transmitter block diagram of the TDCS system is depicted in Figure 5.1. TDCS is a single
carrier transmission where its bandwidth can be divided into smaller subbands which are
similar to subcarriers in an OFDM system. In this way, it is easier to locate the part of the
band occupied by the LUs and subsequently not to put energy in that region. Information
regarding the spectrum occupancy of the licensed users are distributed to each CR device.
Upon receiving the information about the region of the LU band, a vector map
A=[A0,..,Ak,..,ANFFT-1] is produced by the Spectrum Magnitude module where k defines the
index of subband. Zeros are placed on the subband indexes where the LU band is located, and
ones are placed on the other subband indexes. Further, the vector map A is multiplied with
a random phase vector ejθ that will produce the vector Bb.. The multiplication with scaling
factor C has the purpose of realizing the desired or targeted energy used for transmitting a
TDCS symbol. After multiplication with the scaling factor C, the vector B will be produced.
Applying an IFFT to B will produce the time-domain TDCS basis function b(t). After
modulating b(t) with data d(t) and Guard interval (GI) insertion, followed by windowing, the
to be transmitted signal s(t) is formed. The GI is added to avoid inter symbol interference
(ISI) due to the multipath channel effect. Windowing can be added in order to reduce the
sidelobes of the transmitted signal on the LU band. TDCS gives more degrees of freedom in
choosing a window to lower the sidelobes of its spectrum. Unlike in OFDM, the spectrum of
the window of TDCS does not have to make a zero on the subband spacing interval as long as
its sidelobes are very low, e.g a half sine window [17]. Inter-subband interference will not
degrade the data detection at the receiver due to the robustness of the autocorrelation method.
Windowing can be replaced by de-activating more subbands adjacent to the LU band (adding
more zeros to vector A at the sub bands adjacent to the LU band) since it will not reduce the
transmission bit rate.
Spectrum
ID
Spectrum Tx
magnitude s(t)
b(t) b(t)
A Bb B
GI
Scale IFFT Buffer Mod insertion +
e jθ C Windowing
Random d(t)
phase data
Figure 5.1: Transmitter architecture of TDCS [99]

131
Channel T
GI FFT
removal
fading
removal IFFT ∫ (.)dt Decision d’(t)
r(t)
0
rule data
Channel Conj c(t)
estimation CTF
IFFT
Spectrum Spectrum Random

ID magnitude phase
A ejθ
Figure 5.2: Receiver architecture of TDCS with Channel Estimation module
The impact of the zeros insertion to vector A can be observed from the TDCS power spectral
density (PSD) equation,
2 2
T T
(1+α ) (1+α )
2 1 N FFT −1 2
d (t ) p(t )e − j 2π ( f − f k )tdt
PSD( f ) = ∫ s(t )e− j 2π ftdt = ∑ CAk e jθk ∫
T
−(1+α )
T N FFT k =0 −(1+α )
2
2
(5.1)
where T is the useful TDCS signal duration, p(t) is the window function, α is the roll off
factor of the window, fk is the frequency on subband k and θk is the phase on subband k
produced by the random phase module, C is the scaling factor, d(t) is the modulated data and
NFFT is the number of FFT points used in the IFFT process (it can be considered as the
number of subbands) [98].
If a rectangular window is used (p(t)=1) and α is zero, the result of the integral can be
replaced by Tsinc((f - fk)T). A zero amplitude at sub-band k will remove the PSD on that sub-
band position, and due to the spectrum orthogonality by this window the power contributions
from the other subbands at that carrier position are also zero. Except at the frequencies not
located at multiples of the subband spacing, the sidelobes coming from the sinc function of all
active TDCS carriers will give power contributions. Since the sidelobes from the sinc function
are deteriorated corresponding to the distance with the subband position, de-activating more
subbands will lower the sidelobes. Therefore, there will be no significant signal power from
132
the TDCS based CR at the region of TDCS subbands which are occupied by the LUs. The
time-domain signal b(t) is modulated with the transmitted data d(t) using a Cyclic Code Shift
Keying (CCSK), Pulse Amplitude Modulation (PAM) or Pulse Position Modulation (PPM)
scheme. In PAM the bits are mapped onto a real number (either positive or negative). The
higher the constellation size (CS) of the PAM the higher will be the maximum amplitude of
the PAM signal. In the context of Cognitive Radio and if PAM is applied, the PSD constraint
should be taken into account and therefore a low constellation size is preferable. In CCSK the
signals are cyclic shifted by the decimal number representing the composition of the bits
transmitted with a length of bits equal to log2 CS. The shift length depends on CS and the
decimal representation of the bits. The bit rate of the TDCS system with CCSK or PAM can
be calculated from,
lo g CS
R T D C S − C C SK o rP A M
= 2
(5.2)
T s
where TS is the TDCS symbol duration including the guard interval (GI). The GI is inserted to
avoid the inter symbol interference (ISI) which is introduced by the multipath channel. In the
PPM the bits are mapped as shifted pulses [8]. It requires longer symbol periods for higher
constellation sizes; it is therefore not advantageous to be applied in the TDCS if high rates are
needed. The maximum CS that can be applied in CCSK is restricted to the number of
subbands (NFFT ). The nice property of CCSK is that the BER improves as the CS increases
[102], [103], therefore the optimum constellation size for CCSK modulation is equal to the
number of subbands (NFFT). Considering these facts CCSK is considered as the most optimum
modulation in TDCS.
In comparison with OFDM, if the scaling factor C is 1 the PSD of OFDM in eq. (2.1) has
larger sidelobes (with a factor of NFFT) compared to the PSD of TDCS in eq. (5.1). OFDM
has a factor of NFFT higher bitrate compared to TDCS since the bit rate of OFDM is
N F F T .lo g C S
2
while the TDCS bit rate is described by eq. (5.2). Increasing the scaling
T s
factor C will increase the required energy to transmit the TDCS symbol but it will also
increase the SNR of TDCS, hence, the TDCS BER performance will be improved.
The outputs of the random phase modules at the transmitter and at the receiver must be equal,
as the latter will be used as reference at the receiver for data detection by using a correlation
process. The spectrum occupancy information available at the transmitter should also be
available at the receiver for accurate data detection. After the GI removal, the signal is
transformed into frequency domain, followed by fading effect cancellation. In order to detect
the transmitted signal, the received signal is transformed again into time-domain by the IFFT
and is then de-correlated with the reference signal c(t). De-correlation for symbol detection
can be simplified by taking the maximum absolute value of the inverse Fourier transform of
133
the product of the received signal and the conjugate of the reference signal in the frequency-
domain [103].
In an attempt to increase the bit rate of TDCS, we describe in this section the proposed
embedded symbol to TDCS-CCSK. The embedded symbol can be derived from PAM, QAM
or PSK modulation. The proposed transmitter and receiver architectures of the system are
depicted in Figure 5.3.
(a)
(b)
Figure 5.3: (a) Transmitter and (b) Receiver architecture of TDCS with extra embedded
symbol
134
There are two bit sources available. The first one is for the embedded symbol, and the second
one is for the CCSK modulation. We choose to have NFFT as the CS value of the CCSK
modulation since it is the optimum one, while the embedded symbol constellation size will be
investigated further in the performance evaluation subsection.
At the transmitter one embedded symbol will be multiplied with a vector derived from the
point to point multiplication between the vector produced by the random phase θ and the
vector produced by the spectrum magnitude A. Further, legacy processes in producing the
TDCS signal are applied. At the receiver, CCSK data detection is applied first. After
removing the fading effect in frequency-domain, the CCSK data detection is applied by de-
correlating the time-domain received signal with the time-domain reference signal. De-
correlation can be simplified by taking the maximum absolute value of the inverse Fourier
transform of the product of the frequency-domain conjugate received signal and the reference
signal [103]. The received discrete time-domain signal y at time instant n after fading
removal can be expressed as:
1 N FFT − 1 ⎡ ⎤ 2 π nk
+ N k ⎥ e j N FFT
− 2 πτ k
jθ k
yn = ∑ ⎢ Ak e jϕ j
Se e N FFT
N FFT k =0 ⎣ H k⎦ (5.3)
1 N FFT − 1⎡ ⎤ 2 π nk
+ N k ⎥ e j N FFT
− 2 πτ k
= ∑ ⎢ A ' k Se jϕ j
e N FFT
N FFT k =0 ⎣ H k⎦
where S, φ, Nk, Hk are the magnitude, phase of the embedded symbol, noise and channel gain
on subband k respectively, and A’k = Ak ejθk, and τ is the data from source 2 for CCSK
modulation. The reference signal c at time instant n is defined as :
cn =
1 NFFT−1
NFFT k=0
jθk j
2πnk
∑ Ake e NFFT = [ ]
1 NFFT−1 j 2πnk
∑ A'k e NFFT
NFFT k=0
(5.4)
The correlation function R of signals y and c becomes

NFFT−1
Ri = ∑ yn .c*n−i
n=0
NFFT−1 1 NFFT −1 ⎡ j
−2πτk
N ⎤ j 2πnk 1 NFFT −1 * j −2π(n−i )l
= ∑ ∑ ⎢ A'k Se e
jϕ NFFT
+ ⎥ e NFFT
k
∑ A' l e NFFT
n=0 NFFT k =0 ⎢ H ⎥ N
⎣ k⎦ FFT l =0
135
⎡ N FFT−1N FFT−1⎛ 2 π( nk −τk −nl +il ) 2 π( nk −nl +il )

⎞⎤
⎜
⎢ ∑ ∑ A'k A' l Se e * jϕ
j
N FFT
+ A' l * N k
e
j
N FFT ⎟⎥
⎜ Hk ⎟⎥
N FFT −1
1 ⎢ k =0 l =0,l ≠k ⎝ ⎠
= ∑ 2 ⎢ ⎥
n=0 ( N FFT ) ⎢ N FFT −1 ⎛ 2 π( i −τ ) r 2 πir
⎞ ⎥
+ A'*r N k e NFFT ⎟
j j
⎢+ ∑ ⎜⎜ A'r Se e
2 jϕ N FFT
⎟ ⎥
⎢⎣ r =0 ⎝ Hk ⎠ ⎥⎦
(5.5)
The first term in the final eq. (5.5) is negligible due to random phase property of the TDCS
[97]. Eq. (5.5) can be approximated as eq. (5.6). The complete derivation is available in the
Appendix C.
N FFT −1 1 ⎡ N FFT −1 ⎛ j
− 2 πτ r
N ⎞ j N2π ir ⎤
∑ ⎢ ∑ ⎜ A 'r Se e N FFT + A '* r
2 jϕ
Ri ≈ r
⎟e FFT ⎥
( N FFT ) ⎢ r = 0 ⎜⎝
2
H ⎟
r ⎠ ⎥⎦
n=0
⎣
⎡ ⎤
⎢ N FFT −1 j 2 π ( i −τ ) r 2 π ir ⎥ (5.6)
⎢ Se jϕ ∑ e N FFT + ∑ A '* r N r e N FFT ⎥
N FFT −1
1 j
=
( N FFT ) ⎢
r =0

r =0
H r
⎥
⎢⎣ signal term noise term
⎥⎦
The derivation in eq. (5.6) fully agrees with reference [103] in doing CCSK detection. The
position where the maximum correlation occurs (τ) will be selected as the estimated
transmitted data on source data 2. A measure of peak power to mean sidelobe power ratio
(PMR) of the basis function b(t) can be employed to determine the robustness of CCSK data
detection [103]. The PMR is the ratio between the square of the peak cross-correlation of the
basis function and the square of the average remaining cross-correlation values (termed as
sidelobes). If we consider the transmitted data on source data 2 is zero (τ=0) then the PMR of
the proposed TDCS will be represented by eq. (5.7), while in case TDCS-CCSK is without
embedded symbol the S e jϕ in eq. (5.7) is replaced by 1.
⎡ 2
⎤
⎛ 1 ⎛ 2 jϕ
NFFT −1
* Nr
⎞ ⎞
E ⎢⎢ ⎜ ∑ ⎜⎜ A'r Se + A' r ⎟⎟⎟⎟ ⎥
⎥
⎜N
E ⎡⎣R02 ⎤⎦ ⎢⎣ ⎝ FFT
r =0 ⎝ Hr ⎠⎠ ⎥⎦
PMR = =
E ⎡⎣Ri2,i≠0 ⎤⎦ ⎡ ⎛ NFFT −1 ⎛ 2 jϕ ⎞ j 2NπFFT
(ir )
⎞
2
⎤ (5.7)
* Nr
NFFT −1
1
E⎢⎜ ∑ ∑ A'
⎜⎜ r Se + A' ⎟⎟e ⎟ ⎥
⎢ ⎜ i=1 NFFT r =0 ⎝
r
Hr ⎠ ⎟⎠ ⎥
⎣⎝ ⎦
S2e j 2ϕ +σN2 / H
≈
σN2 / H
136
Nr
In eq. (5.7) it is assumed that the has a Gaussian distribution with variance σN/H2.
H r
However, despite Nr is Gaussian distributed and if Hr is also Gaussian distributed it does not
mean that N r is necessarily Gaussian distributed. The emphasis in giving the PMR equation
H r
is to show how the embedded symbol can enhance the PMR of TDCS-CCSK. The details and
complete derivation of eq. (5.7) is provided in Appendix D.
The data from CCSK detection represents the estimated symbol cyclic shift that was taking
place at the transmitter to transmit the data for CCSK modulation. This estimated shift is then
further used to shift the reference signal, followed by the transformation into frequency-
domain by the FFT module.
The received signal in frequency-domain (before CCSK data detection) is then divided by the
shifted reference signal (in frequency-domain) which can be described as dividing the Yr in
− 2 πτ r
eq. (5.6) by A 'r e j N FFT
; hence, before the decision process the estimated embedded symbol
d̂ equals
−2πτr
+ Nr
j
jϕ NFFT 2πτr
A'r Se e j
NFFT
dˆ = Yr = Hr = Sejϕ + Nr e (5.8)
−2πτr −2πτr
j
NFFT
j
NFFT
A'r Hr
A'r e A'r e
After taking the average value per NFFT samples and detection by PAM, QAM or PSK de-
mapping, the bits from source data 1 are reconstructed. Another method is by utilizing the
detected data on source 2 (τ) and multiplying the product between the (in frequency-domain)
2 πτr
received symbol and reference basis function (Ur=FFT(y).FFT*(c) in eq. (5.6)) with e j N FFT ,
where r ∈ {0,..., N FFT − 1} . In this way the transformation of the shifted reference basis
function into frequency-domain can be avoided. In order to improve the source data 1 BER
performance further, channel coding and interleaving can be applied before the mapping
process, meanwhile the de-interleaving and decoding are applied after de-mapping. For
simplicity these techniques will not be included here.
The new TDCS bit rate becomes:
log CS .CS EM
(5.9)
R NEWTDCS
= 2
'
T s
where CSEM is the embedded symbol constellation size.

137
From spectrum point of view, the PSD of TDCS-CCSK with extra embedded symbol yields
2
T
(1+ α )
1 N FFT − 1 2 (5.10)
PSD ( f ) = Se jϕ ∑ Ak e jθ k ∫ p (t ) e − j 2 π( f − f k )tdt
N FFT k =0 T
− (1+ α )
2
By observing eq. (5.10) it is reasonable to consider PSK as most appropriate embedded

symbol modulation mode among the three mapping types (PSK, PAM, and QAM). The PSK
symbols are spread over a unit circle with magnitude one. Therefore embedding the PSK
symbol to the TDCS with CCSK modulation will not change the TDCS transmission power,
while embedding PAM and QAM requires more power. The extra amount of power required
will depend on the constellation size of the PAM/QAM embedded symbol. Moreover, the
embedded symbol will affect the sidelobes of the proposed TDCS. Because PSK symbol
magnitude is one, according to eq. (5.10) the sidelobes amplitude of the proposed TDCS will
not change. PAM and QAM symbols with constellation size bigger than 2 have an average
magnitude larger than one and therefore the sidelobes amplitude of the proposed TDCS will
be affected. PAM modulation in this case will be the worse option since for the same
constellation size (larger than 2) its maximum amplitude is bigger than the QAM mode.
Increasing the embedded symbol’s constellation size will enhance the symbol- to- noise ratio
on each of the TDCS subband. According to reference[103] the symbol-to noise ratio
enhancement will lead to a symbol error probability improvement and eventually to a bit error
rate improvement. Increasing the constellation size means allocating more bits to the
transmitted signal and as a consequence more energy is practically allocated. The amount of
allocated energy is linearly dependent on the number of allocated bits. Therefore, the SNR
gain obtained by increasing the CSEM compared to the conventional TDCS-CCSK
corresponds to
SNR Gain = 10 log 10 (log 2 (CS EM ) )dB (5.11)
where log2(CSEM) refers to the number of bits allocated to the embedded symbol.
5.2.1 Multi-User TDCS with Extra Embedded Symbol Detection Analysis
It is described in references [9],[99] that the random phase generator can be designed to
support multiple user transmission. We have analyzed that the embedded symbol will not
destroy the CCSK detection on each user. It can be learnt from the following derivations.
138
If the received signal on the receiver of user 1 is :

1 NFFT −1 ⎡ ⎤ j 2π nk
u 1
yn,1 = N ∑ ⎢ Ak,1e S1e e FFT + ∑ Ak,ue

jθk ,1 jϕ1 j
−2πτ1k U
jθk ,u H k jϕu j
−2πτu k
N
N S
1 u
e e NFFT +
1 e
k
⎥ NFFT
FFT k =0 ⎢ ⎣ u =2 Hk H k ⎥⎦ (5.12)
1 NFFT −1 ⎡ ⎤ j 2π nk
u 1
∑ k,1 1 e jϕ1 j
−2πτ1k
∑
U
H jϕu j
−2πτu k
N
= ⎢ A ' S e NFFT + A 'k ,u
k
1 u
S e e NFFT +
1 e
k
⎥ NFFT
NFFT k =0 ⎢
⎣ u=2 Hk H k ⎥⎦
where Hku is the channel transfer function of user u on TDCS subband k. The symbols with
index attribute u in eq.(5.12) belongs to user u, while symbols with index attribute 1 belongs
to user 1. U denotes the total number of users. We use the same basis reference signal c as
described in eq. (5.4), but we put attribute 1 to it in order to refer it as reference signal for user
1.
The correlation on the user 1’s receiver becomes:

NFFT−1
∑ y .c
*
Ri,1 = n,1 n−i,1
n=0
NFFT−1 1 NFFT −1⎡ −2πτ1k u −2πτuk 1

⎤ j 2πnk 1 NFFT −1
= ∑ ∑
j
jϕ1 NFFT
∑
U
H jϕu
j
N ∑
j
−2π (n−i)l
1 e l,1e
⎢A'k,1 Se
1 e + A'k,u
k
S
1 u
e e NFFT
+ k
⎥ NFFT A'*
NFFT
n=0 NFFT k=0 ⎢⎣ u=2 Hk Hk ⎥⎦ NFFT l=0
⎡NFFT−1 NFFT−1⎛ 2π (nk −τ1k−nl +il ) u 2π (nk−τuk−nl +il ) 1 2π (nk−nl +il ) ⎤

⎞
+ ∑A'k,u A'*l,1 H1k Suejϕu e + A'*l,1 N1k e NFFT ⎟⎥
j U j j
⎢ ∑ ∑ ⎜ A'k,1 A'*l,1 S1ejϕ1e NFFT NFFT
NFFT−1 1 ⎢ k=0 l=0,l≠k ⎜⎝ u=2 Hk Hk ⎟⎥

⎠
= ∑ 2
⎢ ⎥
n=0 (NFFT ) ⎢ NFFT −1⎛ ⎞ ⎥
2π (i−τ1 )r u 2π (i−τu )r 1 2πir
⎢+ ∑ ⎜ A'r,1 S1e jϕ1e NFFT + ∑A'r,u A'*r,1 H1r Suejϕu e NFFT + A'*r,1 N1r e NFFT ⎟
j U j j
2
⎥
⎢⎣ r=0 ⎝⎜ u=2 Hr Hr ⎠⎟ ⎥⎦
(5.13)
The first term of eq. (5.13) is negligible due to the random phases and multi-user orthogonal
basis function properties of the TDCS [9],[99], [100]. Therefore eq. (5.13) can be simplified
ito (details are in the Appendix E),
N FFT −1 ⎡ NFFT −1 ⎛ −2πτ1r u −2πτ u r 1
⎞ j N2π ir ⎤
Ri ,1 ≈ ∑
1
⎢ ∑ ⎜ A 'r ,1 S1e e
2 jϕ1
j
NFFT
U
* H
+ ∑ A 'r ,u A ' r ,1 1 Su e e
r jϕu
j
NFFT N
+ A ' r ,1 1 ⎟e FFT ⎥
* r
( NFFT )2 ⎢ r =0 ⎜ Hr H r ⎟⎠ ⎥⎦
n =0
⎣ ⎝ u =2
⎡ jϕ NFFT −1 j 2πN(i −τ1 )r U u 2π (i −τ u ) r

⎤
⎢S1e ∑ e
1 FFT
NFFT −1
H
+ ∑ Su e ∑ A 'r ,u A ' r ,1 1r e NFFT ⎥
jϕu *
j
⎢
⎢

r =0

u =2 r =0
Hr
⎥
⎥
1 signal term
multiple access interference term
≈ ⎢ ⎥
( NFFT ) ⎢ NFFT −1 1 π
⎥
* Nr
2 ir
j
⎢+ ∑ A ' r 1
e NFFT ⎥
⎢ r =0 Hr
⎥
⎢⎣ noise
term ⎥⎦
(5.14)
139
According to [100], if each user has a different dedicated generator polynomial in producing
the random phase then the multiple access interference term in eq. (5.11) will be small.
Therefore, if the noise term does not exist, the receiver could still estimate the shift from the
CCSK modulation of each user (τu) correctly by utilizing the reference basis function of each
user (A’u=[A0,u ,..., Ak,u,.., AN FFT −1,u ]) and by choosing index i for the maximum value in
correlation of each user (Ri,u).
If the receiver has a base station kind of capability, then the basis function of all users will be
available at the receiver. Equation (5.14) in general form can be written as :
⎡ jϕ NFFT −1 1 2π (i−τ1 )r 2π (i−τ2 )r

⎤
⎢S1e ∑ A'r,1 A' r,cu cu e
1 * H r
j
NFFT jϕcu
NFFT −1
+... + Scue ∑ A'r,cu A' r,cue NFFT ⎥*
j
1 ⎢
⎢ r=0 Hr r=0 ⎥
⎥
Ri.cu ≈ U 2π (i−τU )r cu 2πir
⎢
(NFFT ) +... + S e jϕU
NFFT −1
∑ A ' A'* H r
e
j
NFFT
+
NFFT −1
∑ A '* Nr e NFFTj ⎥
⎢ U r ,U r ,cu cu r ,cu cu ⎥
⎢ r=0 Hr r=0
H r

⎥ (5.15)
⎢⎣ noise term ⎥⎦
1
≈ ⎡⎣ψ1Ε1,i cu +... +ψcuΕicu,cu +... +ψuΕiu,cu +... +ψUΕUi ,cu +ηcui ⎤⎦
(NFFT )
where Ri,cu is the cross correlation between the received signal after fading removal with the
shifted basis function of user cu on shift distance i and where ψu = S u e jϕu ,
2π (i −τu ) r 2 π ir
N FFT −1 u
E
i
= ∑ A'r ,u A'* H r
j
e N FFT and η i =
N FFT −1
'* N
cu
r
j
N FFT
.
u ,cu r ,cu cu ∑ Ar ,c u e
r =0 H r
cu
r=0
H
cu
r
By utilizing the basis function of all users and calculate Ri,cu of all users, τu of all users wil be
derived. The vector-matrix form of eq. (5.15) is
RcuNFFT x 1 = EcuNFFT xUψcuU x 1 + ηcu NFFT x 1 (5.16)
The matrix Ecu is derived by utilizing the obtained τu of all users and the basis function
vectors A’u of all users. Accordingly, the value of embedded symbols of all users before
decision by QAM/PSK/PAM de-mapping (maximum likelihood) can be derived by
1 U
ψ= ∑
U cu=1 (E cu
−1 cu
R ) 1 U cu
U cu=1
−1
= ∑ ψ + Ecu ηcu ( ) (5.17)
where ψ = [ψ 1 , ... , ψU ] = [S 1e j ϕ , ... , SU e j ϕ 1 U

] are the estimated embedded symbols of all users.
140
5.3 Wavelet Domain Communication System (WDCS) with Embedded

Symbol
A variant of TDCS is the wavelet domain communication system (WDCS) [100]-[102] where
the Fourier transform operation in TDCS is replaced by a wavelet transformation. The
wavelet transform has been known having lower sidelobes compared to Fourier based
modulation. Wavelets are used in this scheme to identify and establish an interference-free
spectrum. The advantages of using wavelet are [100]:
1. Increased adaptation over a larger class of interfering signals,
2. Finer high-frequency resolution,
3. Allow implementation of CS-ary orthogonal signaling
The WDCS uses a packet-based transform to estimate the electromagnetic spectrum [93].
Through the use of adaptive thresholds and notches, sub-bands containing the interference are
effectively canceled. From this estimate, a unique communication basis function A in the
transform domain is generated so that no (or very little) energy-bearing information is
contained in the areas occupied by primary users. These functions are then multiplied with a
pseudo-random(PR) phase vector e jθ to generate Bb. The PR code is used to randomize the
phase of the spectral components. The resulting complex spectrum is then scaled with C to
provide the desired energy in the signal spectrum. A time domain version b(t) of the basis
function is then obtained by performing an inverse wavelet transfom. The wavelet basis
function can be a Coifflets, Daubechies, Haar, Symlets or Remez filter as applied earlier in
WP-MCM. Finally, the basis function is modulated with data using PAM, or PPM, or CCSK
modulation and then transmitted. The block diagram of the WDCS transmission process is
At the receiver the detection is preceded by correlating the received signal with the reference
signal in-time domain. The receiver structure of WDCS is almost similar to the TDCS, the
differences are the FFT and IFFT modules in TDCS that are replaced by discrete wavelet
transform (DWT) and inverse discrete wavelet transform (IDWT), respectively.
S p e c tru m ID
a n d S p e c tru m
M a g n itu d e
( W av elet
T ra n sfo rm
E s tim a tio n &
S h a p in g ) Tx
In v e rse S (t)
A Bb B W a v e le t b (t)
S c a le T ra n sfo rm M od
e jθ C
R andom d (t)
P hase d a ta
Figure 5.4: WDCS Transmitter blocks [100]

141
Spectrum
ID
Spectrum
magnitude Tx
A B b(t) b(t)
Bb Inverse
PAM /QAM Wavelet
d1(t) / PSK Scale Buffer CCSK
data 1 jθ C Packet
Mapping e Transform
Random d2(t)
phase data 2
(a)
T
Equalizer ∫ (.)dt Decision d2’(t)

r(t)
0
rule data 2
Conj c(t)
CCSK
Wavelet
Packet Transform Inverse Wavelet
Packet Transform Wavelet
Packet Transform
Spectrum Spectrum Random
R( f ) ID magnitude phase ( . )-1
jθ
A e
PAM /QAM
Mean / PSK d1’(t)
Demapping data 1
(b)
Figure 5.5: (a) Transmitter and (b) Receiver architecture of WDCS with extra embedded
symbol
The application of the embedded symbol to WDCS is presented in Figure 5.5. According to
reference [52] the impact of fading in the modulation technique using the wavelet basis
function would be properly removed by time-domain equalization. Therefore, different with
the TDCS scheme (Figure 5.3) the frequency-domain channel estimation and fading removal
is replaced by time-domain equalization. Further, the CCSK detection in WDCS with extra
embedded symbol system is applied in time-domain, while the embedded symbol detection
142
remains conducted in frequency-domain. The frequency domain embedded symbol detection

is done due to the interference avoidance reason. By neglecting the signal at the LU band
position (Spectrum Pooling technique), the embedded symbol will not be affected by the
interference from the LU signal.
5.4 MIMO Application in a Single Carrier Cognitive Radio System
Following the gain of applying the MIMO system with V-BLAST receiver architecture in the
TDCS system and using our experience reported in [105] we here extend the work by adding
to the TDCS-CCSK an extra embedded symbol for higher rate transmission. The comparison
in performance of the MIMO V-BLAST and the conventional TDCS will be described in next
section. The MIMO system with V-BLAST receiver has been proven to be robust in
counteracting the fading channel. It exploits channel diversity to improve the system BER
performance through space time-coding of the estimated channel state information on all
channel links. Furthermore, MIMO also increases the system bit rate. In the CR context the
bit rate and BER enhancement by adding MIMO with V-BLAST receiver to our TDCS-
CCSK with an extra embedded symbol refer to an efficient spectrum utilization improvement.
In this way our proposed system tries to fulfill the CR objectives of highly reliable
communications and efficient spectrum utilization [2] with extra complexity as a
consequence.
The transmitter and receiver architectures of the proposed system are depicted in Figure 5.6,
where M is the number of transmit antennas and N is the number of the receive antennas. At
the transmitter after the guard interval (GI) insertion, the transmitted signal is de-multiplexed
to be transmitted to a selected transmit antenna. The transmitted signal on each transmit
antenna is different to each other; as a result the data rate increases M times compared with
the single antenna rate. In order to preserve the amount of power used for transmission, the
total power is distributed among transmit antennas. For the sake of simplicity, in our system
we apply an equal distribution. The power spectral density (PSD) of the TDCS-CCSK with
embedded symbol in a MIMO system becomes
2
T
(1+α )
1 1 2
M jϕ N FFT −1 − j 2π ( f − f k )tdt
PSD( f ) = ∑ Sme m ∑ Ak e jθk ,m ∫ T p(t )e
N FFT =1
m M k =0 −(1+α )
2
(5.18)
where Sm and φm are the magnitude and phase of the embedded symbol on transmit antenna
m; T is the useful signal duration; p(t) is the window function; α is the roll off factor of the
window; Ak and θk,m are the spectrum occupancy info ( 0 if it is occupied by LU, and 1 if it is
available) and the phase produced by the random phase module on sub-band k of antenna m
143
1
respectively. A factor is included in eq. (5.18) because the total power is distributed
M
equally to all M transmit antennas. All transmit antennas have the same spectrum occupancy
info A.
If a rectangular window is used (p(t)=1) and α is zero, the result of the integral can be
replaced by Tsinc((f -fk)T). A zero amplitude at sub-band k will remove the PSD at that carrier
position, and due to the spectrum orthogonality of this window the power contributions from
the other carriers on frequency fk are also zero. Since all transmit antennas utilize the same
spectrum occupancy info A, there will be no signal power from the TDCS based CR in the
region of TDCS sub-bands which are occupied by the LUs. The overall sidelobes will be the
average of the total sum of each transmit antenna sidelobes.
The data rate of the TDCS with embedded symbol in MIMO system becomes
M.log ( CS.CSEM )
REM _TDCS _ MIMO = 2
(5.19)
T s
where CS is the constellation size of CCSK modulation, CSEM is the embedded symbol
constellation size and Ts is the TDCS symbol duration including the GI.
(a)
144
(b)
Figure 5.6: (a) MIMO Transmitter and (b) V-BLAST Receiver architecture of TDCS with
extra embedded symbol
At the receiver, the received signal of each receive antenna is transformed into frequency-
domain by using an FFT and then processed by the V-BLAST signal processing module
where its fading effect will be removed, space-time coded and later the correlation is
performed for the detection of data from the data2 source. The results from the data2 source
detection will be used to detect the transmitted data from source data1.
V-BLAST is a promising implementation of a multiple input and multiple output (MIMO)

system [61],[62]. At each symbol the V-BLAST algorithm detects the strongest layer of the
transmitted signal, cancels the effects of this strongest layer from each of the received signals,
then continue to detect the strongest on the remaining layers, until the weakest one is detected.
The V-BLAST algorithm reconstructs the transmitted signal by removing the fading effect of
the channel in frequency domain. The received signal at each receiver antenna n from each of
the transmit antenna m equals [105]:
M
Y n = ∑Hnm X m + η .
m=1
n
(5.20)
In matrix form the received signal vector is derived as :
Y = HX + η , (5.21)
where Y=[y1,y2,…yN] is the received signal vector, X= [x1,x2,…xM] is the transmitted signal
vector, M is the number of transmit antennas, N is the number of receive antennas, η=[η1, η2,…,
145
ηN] is the noise vector and the channel matrix H has coefficients Hnm, where each coefficient
describes the channel link between transmit antenna m and receive antenna n [105].
⎡H 11 H 12
" H 1M
⎤
⎢ ⎥
⎢H 21 H nm
" H 2M ⎥ (5.22)
⎢ # # % # ⎥
⎢ ⎥
⎣⎢ H N 1 H N 2
" H NM ⎦⎥
With the assumption that the information about the channel transform functions (CTFs) are
available at the receiver, the channel matrix H can be obtained. The V-BLAST algorithm
detects the signals from all the transmit antennas sequentially according to the signal strength.
The V-BLAST implementation is described in Appendix F.
5.5 Performance Evaluation and Analysis
5.5.1 TDCS Evaluation
Table 5.1: Simulation parameters
TDCS Parameters
Carrier Frequency 5.5 GHz
Number of Subcarriers 128
Carrier Spacing 31.25 KHz
Guard Interval Duration 5.25 μs
Channel coding OFF
Channel Parameters
Maximum Channel Delay 5 μs
Number of Taps 6
Distance between Taps 1 μs
Number of Paths 12
Power decay between Taps 1 dB
Vehicular speed 100 Km/hr
Fading Model Rayleigh Fading
The TDCS and channel parameters are provided in Table 5.1. The fully loaded signal
bandwidth is 4 MHz. In the beginning the simulations ran in a system with one transmitter
and one receiver (no multiple access). In the simulations with a Rayleigh fading channel, the
delay spread is 1.6282 µs. If we use the criteria for determining the coherence bandwidth as
the bandwidth over which the signal envelope correlation function be larger than 0.5, the
channel coherence bandwidth will be about 123 KHz. The requirement to have flat fading on
each subband (carrier) is fulfilled by having a subband spacing of 31.25 kHz . The LU band is
assumed to be located in the middle of our TDCS band, occupying 26 subbands (equal to a
146
bandwidth of 812.5 kHz); we also assume that this information is available at the transmitter
and the receiver. Here we will not explore the effect of windowing, therefore we de-activated
40 subbands (1.25 MHz) which are considerably wider than the LU band so that the
sidelobes to the LU’s band and also the interference from LU to the TDCS system are
negligible. We have simulated and compared the CCSK performance of conventional TDCS-
CCSK with CS=128 for different number of phase points to be used by the random phase
generator in generating the phase vector θ. The phase is derived from a unit circle with
uniform space. The comparison is depicted in Figure 5.7. The choice of CS=128 is made with
the objective of obtaining an optimum TDCS bit rate and BER according to the number of
available subband/carriers, which is 128.
0
10
Conventional TDCS 8 points
-1
10
BER
-2
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.7: BER vs SNR (dB) comparison of the conventional TDCS-CCSK with different
number of phase points on a unit circle in the AWGN channel
Table 5.2: Bit rate OF TDCS-128CCSK with different kind embedded symbols constellation
size
Embedded symbol # of bits / Bit

modulation mode TDCS rate
symbol (Kbps)
BPSK 8 215
QPSK 9 242
16-PSK 11 295
64-PSK 13 349
256-PSK 15 403
147
The simulation results show that there is no performance difference as the number of points in
the unit circle is increased, therefore in our next simulation results the random phase will be
generated from the constellation of 8 phase points of a unit circle with uniform space of 45
degrees between one point to the next point. Table 5.2 shows the total bitrate of the proposed
TDCS-CCSK with different embedded symbol’s constellation size.
A. Analysis in AWGN Channel
We examine the BER of the embedded symbol (data1) and the data for CCSK modulation
(data2) separately. The overall BER is derived by adding the sum of bit error from data1 and
data 2 divided by the total number of bits. In the context of dynamic spectrum access we do
not want to interfere the LU system, therefore we apply PSK mapping as the embedded
symbol so the power spectrum density of TDCS-CCSK with extra embedded symbol will not
be changed
Figure 5.8 shows the BER performance of the embedded symbol which refers to the detection
of data from source 1 in an AWGN channel. According to the theoretical properties of PSK
[106], higher constellation size is more vulnerable to noise, therefore the BER performance
will be degraded. The exception of embedded BPSK detection shown in Figure 5.8 could be
explained after reviewing the BER results on CCSK detection because the embedded symbol
detection is applied after CCSK detection. This means that the accuracy of the CCSK
detection will also influence the embedded symbol detection. Further, by observing Figure 5.9
we can see that the TDCS with 128 CCSK BER performance has almost the same curve
pattern as the embedded BPSK BER performance shown in Figure 5.8.
0
10
-1
10
BER
TDCS-CCSK w/ embedded BPSK

TDCS-CCSK w/ embedded QPSK
-2
10 TDCS-CCSK w/ embedded 16-PSK
TDCS-CCSK w/ embedded 64-PSK
-3
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.8: BER vs SNR (dB) performance on data1 source (embedded symbol) of the
combined TDCS-CCSK with PSK embedded symbol in an AWGN channel
148
This symptom occurs because this embedded BPSK detection is equal to the data detection of
TDCS with binary antipodal PAM modulation (without CCSK) that previously has been
investigated in reference [98]; the results show that its performance is slightly better than the
TDCS with 128 CCSK.
Figure 5.9 shows the effect of the embedded symbol to the CCSK detection of the TDCS
system. At a BER of 10-2 the SNR gain of increasing the constellation size of the embedded
symbol derived from the simulation results agrees with SNR gain defined in eq. (5.11). The
CCSK detection SNR gain with 4, 16, 64, and 256 PSK are 3, 6, 7.8, and 9 dB
((10.log10(log2[4,16,64,256]) dB) respectively.
0
10
-1
10
-2
10
BER
-3
10
Conventional TDCS-CCSK
-4
10 TDCS-CCSK w/ embedded QPSK
-5 TDCS-CCSK w/ embedded 256-PSK
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.9: BER vs SNR (dB) performance on data2 source (CCSK symbol) of the combined
TDCS-CCSK with PSK embedded symbol in an AWGN channel
As described in Figure 5.3(b), due to a system design where the CCSK detection is applied
before the embedded symbol detection, the accuracy of the CCSK detection will influence the
BER performance of the embedded symbol detection. Upon the perfect detection of the
CCSK symbol, the embedded symbol BER performance will resemble its theoretical error
probability (PSK, QAM or PAM) where lower modulation modes refer to a better BER
performance. If the CCSK detection is not perfect then the CCSK BER performance shown in
Figure 5.9 where higher modulation modes gives lower BER, will have impact on the BER
performance of the embedded symbol; hence, the theoretical properties of PSK, QAM or
PAM embedded symbol will be changed.
Therefore, the results in Figure 5.8 denotes a degradation of the theoretical BER performance
of the PSK modulation modes due to imperfectness of the CCSK detection that delivers
149
QPSK as the most optimum embedded symbol modulation mode applied in the TDCS-CCSK
system.
By considering the advantage/weakness of one modulation mode in term of BER at source

data1 and data 2, Figure 5.10 shows the overall BER of the TDCS system with the PSK
embedded symbol. For a SNR below 11 dB the BER of TDCS with embedded 64 PSK still
outperforms the conventional TDCS-CCSK. For a SNR higher than 13 dB the TDCS-CCSK
with embedded 64 PSK is outperformed by the conventional TDCS-CCSK; therefore, in this
region the constellation size limit of the embedded symbol using QAM and PSK mapping is
16. The embedded symbol with 4 PSK gives the best overall BER performance at high SNR,
while for low SNR 16 PSK is a better alternative. These facts will be compared with the
results in case a Rayleigh fading channel is assumed.
We have observed the performance of the TDCS with QPSK embedded symbol in multi user
AWGN environment. For simplicity reasons we have considered 2 users in the simulations.
Figure 5.11 depicts the BER performance of both users at the source1detector, source2
detector and the overall BER. The BER performance curves of both users have the same
behaviour as shown in Figures 5.8 to 5.10 for the TDCS-CCSK with embedded QPSK case.
This means that the embedded symbol does not destroy the CCSK symbol detection in a multi
user AWGN environment.
0
10
-1
10
-2
10
BER
-3 Conventional TDCS-CCSK
10
-4
-5
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.10: Overall BER vs SNR (dB) performance of the Combined TDCS-CCSK with
PSK embedded symbol in an AWGN channel
150
0
10
-1
10
BER
-2
10
TDCS w emb. QPSK, CCSK detection on user1

-3 TDCS w emb. QPSK, CCSK detection on user2
10
TDCS w emb. QPSK, emb. sym. detection on user1
TDCS w emb. QPSK, emb. sym. detection on user2
TDCS w emb. QPSK, overall detection on user1
-4 TDCS w emb. QPSK, overall detection on user2
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.11: BER vs SNR (dB) performance of the Combined TDCS-CCSK with QPSK
embedded symbol in a 2 users environment assuming an AWGN channel
B. Analysis in case of a Rayleigh Fading Channel
The Rayleigh fading channel model is derived according to the wide sense stationary
uncorrelated scattering (WSSUS) model [73] using parameters given in Table 5.1. The
Doppler effect is included in the model. We assume that the transmitter and receiver are
perfectly synchronized, and the channel transfer functions (CTFs) of all links are perfectly
known at the receiver. For simplicity the frequency-domain equalization for all simulations
used zero-forcing equalization.
Figure 5.12 shows how Rayleigh fading has severely destroyed the overall BER performance
of TDCS-CCSK with embedded symbol. Although we assume perfect knowledge of CTFs in
the simulation, the division between the noise and CTF has altered the correlation property of
the TDCS basis function, because we use zero forcing equalization technique; hence, a BER
degradation occurs as depicted in Figure 5.12. Therefore, we add MIMO with the V-BLAST
receiver architecture to the system to mitigate the fading effect.
In a multi-user environment with Rayleigh fading channel the overall performance of the
TDCS-CCSK with embedded symbol degrades. According to eq. (5.14) and eq. (5.15) the
fading channel gain can disturb the correlation property of the TDCS basis function that
eventually has an impact on the CCSK detection and further on the embedded symbol
detection. Figures 5.12 and 5.13 show how the fading channel destroys the TDCS
performance. The performance in a MIMO system with V-BLAST receiver architecture
(which will be explored later) gives a clear view on the BER degradation of TDCS-CCSK
with embedded symbol in a multi-user environment with a Rayleigh fading channel.
151
-0.4
10
-0.5
10
-0.6
10
BER
Conventional TDCS-CCSK
-0.7
10 TDCS-CCSK w/ embedded BPSK
-0.8
-0.9
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.12: Overall BER vs SNR (dB) performance of the combined TDCS-CCSK with PSK
embedded symbol in a Rayleigh fading channel
-0.4
10
-0.5
10
BER
-0.6
10 TDCS w emb. QPSK, CCSK det. on user1
TDCS w emb. QPSK, CCSK det. on user2
TDCS w emb. QPSK, emb. sym. det. on user1
TDCS w emb. QPSK, emb. sym. det. on user2
-0.7
10 TDCS w emb. QPSK, overall det. on user1
TDCS w emb. QPSK, overall det. on user2
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.13: BER vs SNR (dB) performance of the Combined TDCS-CCSK with QPSK
embedded symbol in a 2 users environment and a Rayleigh fading channel
In an earlier phase we searched for the optimum configuration of MIMO system with the most
promising modulation mode for the extra symbol to be embedded to the MIMO TDCS-CCSK
system. For this purpose we have selected QPSK as the embedded symbol modulation mode,
as the results in Figure 5.10 show, this mode gives the best overall BER.
152
In reference [105] we investigated the influence of MIMO V-BLAST to the conventional

TDCS-CCSK where the embedded symbol was not included, and we compared their
performance with MIMO OFDM with the V-BLAST receiver architecture. For a scaling
factor C=1, figures 5.14 to 5.16 depict some of the results given in reference [105].
According to that reference MIMO with a ratio between transmit and receive antennas of ½
and lower gives optimum BER results. Further simulation results show that OFDM MIMO
architecture gives a better SNR gain as compared to the application of adaptive bit loading in
OFDM. In subsection 3.2.4 we found out that the adaptive bit loading with the Fischer
algorithm in OFDM and perfect channel estimates provides a 5 dB SNR gain. From Figure
5.14 we learn that in order to have a BER of 10-2, conventional OFDM requires 14 dB SNR
while in Figures 5.15 and 5.16 (by increasing the number of receive antenna compared to the
transmit antenna) will give more than 6 dB SNR gain.
0
10
-1
10
BER
-2 OFDM based CR SISO

10
OFDM based CR MIMO 2x2
TDCS based with CCSK CR SISO
TDCS based with CCSK CR MIMO 2x2
-3 TDCS based with CCSK CR MIMO 3x3
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.14: BER vs SNR (dB) performance of CR TDCS with CCSK, and OFDM-based
with BPSK, in a MIMO V-BLAST architecture with a balanced antenna design
Figures 5.15 and 5.16 show that having a ratio of ½ between transmit and receive antenna
gives a significant performance improvement to TDCS. It even outperforms OFDM at high
SNR.
Now we investigate the TDCS-CCSK with embedded QPSK and different MIMO
compositions. Figure 5.17 depicts the overall BER of the scheme. The notation m x n denotes
the MIMO system with m transmit antennas and n receive antennas. The results show that the
configuration with more receive antennas than transmit antennas gives more promising
results, especially when the ratio between transmit and receive antennas is ½ and lower.
153
0
10
-1
10
-2
10
BER
-3
10
-4
10 OFDM based CR MIMO 3x4
-5 TDCS based with CCSK CR MIMO 3x4
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.15: BER vs SNR (dB) performance of CR TDCS with CCSK, OFDM-based with
BPSK, in a MIMO V-BLAST architecture and a difference of one between the number Tx
and Rx antennas
0
10
-2
10
BER
-4
10

-6
10 OFDM based CR MIMO 2x4
-8
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.16: BER vs SNR (dB) performance of CR TDCS with CCSK, OFDM-based with
BPSK, in a MIMO V-BLAST architecture with 2 Tx antennas and equal or higher number of
Rx antennas
154
0
10
-1
10
-2
10
Embedded QPSK- SISO 1x1
BER
Embedded QPSK- MIMO 1x2

10
-3 Embedded QPSK- MIMO 1x3
-4
10 Embedded QPSK- MIMO 3x3
-5 Embedded QPSK- MIMO 3x6
10
0 2 4 6 8 10 12 14
SNR
Figure 5.17: Overall BER performance vs SNR (dB) of the SISO and MIMO TDCS-CCSK
with embedded QPSK in a Rayleigh fading channel and several Tx-Rx antennas configuration
Next we made simulations for the MIMO 2x4 antennas configuration. Figure 5.18 shows
QPSK as most optimum embedded system modulation mode, in terms of BER performance.
The BER tendency and characteristics of the results of the MIMO 2x4 TDCS-CCSK with
PSK embedded symbol in a Rayleigh fading channel shown in Figure 5.18 resemble the
performance in an AWGN channel, as depicted in Figure 5.10. Therefore, we investigate the
BER performance on source data 1 and source data 2 separately.
0
10
-1
10
BER
-2
10
TDCS - MIMO 2x4

TDCS with Embedded BPSK - MIMO 2x4
-3
10 TDCS with Embedded QPSK - MIMO 2x4
TDCS with Embedded 16 -PSK - MIMO 2x4
TDCS with Embedded 64-PSK - MIMO 2x4
-4
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.18: Overall BER performance vs SNR (dB) of the MIMO 2x4 TDCS-CCSK with
embedded PSK and assuming a Rayleigh fading channel
155
Figure 5.19 shows that the performance of the CCSK detection in MIMO 2x4 TDCS-CCSK
with PSK embedded symbol is in line with the finding we had in the AWGN channel. As the
embedded symbol constellation size increases the CCSK correlation property is enhanced.
The embedded symbol detection results in Figure 5.20 show similar BER characteristics as
the BER results we have found in the AWGN channel, depicted in Figure 5.8.
0
10
-1
10
-2
10
BER
-3
10
TDCS - MIMO 2x4
-4
10 TDCS with Embedded QPSK - MIMO 2x4
TDCS with Embedded 16 -PSK - MIMO 2x4
-5 TDCS with Embedded 256-PSK - MIMO 2x4
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.19: BER performance vs SNR (dB) on source data2 (CCSK symbol) of the MIMO
2x4 TDCS-CCSK with embedded PSK symbol in a Rayleigh fading channel environment
In Figure 5.19, at a BER of 10-2 of the CCSK detection in 2x4 MIMO TDCS-CCSK, the SNR
gain achieved by increasing the embedded symbol constellation size in a Rayleigh fading
channel environment is equal to the SNR gain obtained by increasing the embedded symbol in
an AWGN channel (3, 6, 7.8, and 9dB for 4-,16-,64- and 256-PSK respectively).
Considering the trade-off between the overall bit rate and overall BER performance (that can
be observed from Table 5.2, Figures 5.10 and 5.18), we come to the conclusion that 16 PSK is
the optimal embedded symbol modulation mode.
The BER results of MIMO 2x4 TDCS-CCSK with QPSK embedded symbol in a 2 user
environment with Rayleigh fading (Figure 5.21) indicate how the channel fading can change
the cross correlation detection robustness of the TDCS basis function in a multi-user
environment. If we compare the CCSK detection performance at BER 10 -3 , the single user
MIMO 2x4 TDCS-CCSK with QPSK requires about 13 dB SNR (Figure 5.19) while the
multi-user part requires 14 dB SNR (Figure 5.21). That means that the multi-user environment
in a Rayleigh fading channel gives 1 dB SNR degradation. Another observation is that the
overall TDCS-CCSK (with QPSK embedded) performance in a single user Rayleigh fading
channel at BER 10 -2 requires 11 dB SNR (Figure 5.18), while in the multi-user case it
requires about 12 dB SNR (Figure 5.21).
156
0
10
-1
10
BER
-2
10
TDCS - MIMO 2x4
TDCS with Embedded QPSK - MIMO 2x4
-3 TDCS with Embedded 16 -PSK - MIMO 2x4
10 TDCS with Embedded 64-PSK - MIMO 2x4
-4
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.20: BER performance vs SNR (dB) on source data1 (embedded symbol) of the
MIMO 2x4 TDCS-CCSK with embedded PSK symbol in a Rayleigh fading channel
environment
0
10
-1
10
BER
-2
10
MIMO 2x4 TDCS w emb. QPSK, CCSK det. on user1

-3
10 MIMO 2x4 TDCS w emb. QPSK, CCSK det. on user2
MIMO 2x4 TDCS w emb. QPSK, emb. sym. det. on user1
MIMO 2x4 TDCS w emb. QPSK, emb. sym. det. on user2
MIMO 2x4 TDCS w emb. QPSK, overall det. on user1
-4 MIMO 2x4 TDCS w emb. QPSK, overall det. on user2
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.21: BER performance vs SNR (dB) on source data1 (embedded symbol) of the
MIMO 2x4 TDCS-CCSK with embedded QPSK symbol in a 2 users environment
experiencing Rayleigh fading
157
5.5.2 WDCS Evaluation
As applied in the TDCS evaluation, in the WDCS with embedded symbol system we also
examine the BER of the embedded symbol (data1) and the data for CCSK modulation (data2)
separately. The overall BER is derived by adding the sum of bit errors from data1 and data 2
divided by the overall number of bits. For the same reason as in the TDCS system, in the
context of dynamic spectrum access we do not want to interfere the LU system; therefore we
apply the PSK mapping as the embedded symbol; so the power spectrum density of WDCS-
CCSK with extra embedded symbol will not be changed. The “Frequency Selective Wavelet”
that has been used as basis function for WPMCM in section 3.3 is also employed as basis
function in this WDCS. The Frequency Selective Wavelet is more flexible compared to other
wavelet basis function since its parameters such as filter length, regularity order and the
transition bandwidth can be varied according the desired targeted performance. In this
evaluation we consider a filter length of 30, regularity order of 12 and transition bandwidth of
0.1. We evaluate the performances only for the AWGN channel. The synchronization between
transmitter and receiver is considered to be perfect.
Figure 5.22 shows the BER performance of the embedded symbol. Different from Figure 5.8,
the property of PSK mapping is clearly seen in the embedded symbol BER results. BPSK
gives the best BER performance, while increasing the constellation size will deteriorate the
BER performance. This difference will be explained after analysis of the CCSK detection
results.
The CCSK detection BER performances are depicted in Figure 5.23. The BER tendency has
the same behaviour for TDCS-CCSK detection as depicted in Figure 5.9. As the embedded
symbol constellation size increases, the CCSK BER improves. The 9 dB gain for 256-PSK as
embedded symbol compared to BPSK and depicted in the TDCS case in Figure 5.9 is also
By observing Figure 5.9 and 5.23 we see that the CCSK detection in the WDCS system
outperforms the CCSK detection in TDCS system. It requires 14 dB SNR for the conventional
TDCS-CCSK to reach the BER 10-2 while for the WDCS-CCSK system with a frequency
selective wavelet it requires only 10.2 dB SNR. It has been explained in the TDCS evaluation
that the performance of the CCSK detection will influence the embedded symbol detection.
Therefore due to its better BER performance property, the frequency selective WDCS-CCSK
could preserve the embedded symbol BER performance according to its error probability
behaviour (i.e. BER degrades as the constellation size increases).
The overall BER is shown in Figure 5.24. The overall BER is derived from dividing the total
number of errors on the embedded symbol detection and the CCSK detection divided by the
total number of transmitted bits.
158
0
10
-1
10
-2
10
-3
10
BER
10
-4 WDCS-CCSK w/ embedded BPSK
WDCS-CCSK w/ embedded QPSK
WDCS-CCSK w/ embedded 16-PSK
-5 WDCS-CCSK w/ embedded 64-PSK
10
-6
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.22: BER vs SNR (dB) performance on source 1 (embedded symbol) of the combined
WDCS-CCSK with PSK embedded symbol assuming an AWGN channel environment
Conventional WDCS-CCSK
1
10 WDCS-CCSK w/ embedded BPSK
0
10 WDCS-CCSK w/ embedded 64-PSK
-1
10
-2
10
BER
-3
10
-4
10
-5
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.23: BER vs SNR (dB) performance on source 2 (CCSK symbol) of the combined
WDCS-CCSK with PSK embedded symbol assuming an AWGN channel environment
159
1
10
0
10
-1
10
-2
10
BER
Conventional WDCS-CCSK
10
-3 WDCS-CCSK w/ embedded BPSK
10
-4 WDCS-CCSK w/ embedded 64-PSK
-5
10
0 2 4 6 8 10 12 14
SNR (dB)
Figure 5.24: Overall BER vs SNR (dB) performance of the combined WDCS-CCSK with
PSK embedded symbol assuming an AWGN channel environment
5.6 Conclusions
In this chapter the feasibility of the transform domain communications sytem (TDCS) as
single carrier modulation technique for Cognitive Radio system has been studied. An effort to
combat the TDCS bit rate limitation has been employed. A novel application of an embedded
symbol as extra data source has been proposed.
In the Cognitive Radio context sidelobes refer to the amount of interference to the LU access.
The sidelobe level of a signal are found in its power spectrum density (PSD). According to
PSD observations, OFDM gives a factor of (1/ NFFT) larger sidelobes compared to TDCS, if
the TDCS scaling factor C is 1. If the scaling factor C is equal to N FFT TDCS sidelobe will
have the same magnitude as OFDM sidelobes. The TDCS BER performance will be improved
but the TDCS bitrate remains and it is lower than the OFDM bitrate. Increasing the scaling
factor C denotes stepping up the SNR of TDCS, hence, the TDCS BER performance will be
improved.
Therefore, by putting the power spectrum density constraint as criterion in selecting the
embedded symbol modulation mode, the PSK modulation is considered as the appropriate
mode since embedding the PSK symbol will not increase the sidelobes of the TDCS signal.
160
The impact of the embedded symbol to the multi user data detection has been analyzed. The
embedded symbol does not deteriorate the CCSK multi-user detection in an AWGN channel.
However, the Rayleigh Fading channel disturbs the orthogonality property of each user’s
TDCS basis function. Therefore, the CCSK BER performance on each user degrades.
Following the successful application of the embedded symbol in the TDCS-CCSK system, the
application of the embedded symbol in WDCS-CCSK system has also been explored. Based
on the good performance of the frequency selective wavelet basis function in the Multi-carrier
wavelet packet modulation, in this chapter we evaluated the application of the frequency
selective wavelet in WDCS-CCSK. The results have shown that the WDCS-CCSK with
frequency selective wavelet outperforms the TDCS-CCSK system.
In the design of TDCS-CCSK or WDCS-CCSK with embedded symbol, the CCSK detection
is applied prior to the embedded symbol detection. Therefore the CCSK detection accuracy
will influence the accuracy of the embedded symbol detection. Results have shown that
increasing the constellation size of the embedded symbol will improve the CCSK detection,
while the property of the PSK modulation denotes that the BER will be deteriorated as the
constellation size decreases. Imperfections in CCSK detection will disturb the error
probability property of PSK (increased BER as constellation size increased). Since the
WDCS-CCSK gives a better CCSK detection accuracy, it could preserve the BER property of
the PSK modulation mode.
MIMO V-BLAST has been well known as an effective technique in utilizing channel
diversity to enhance the BER performance of a wireless system. In addition to this MIMO
increases the bit rate of a wireless system. We evaluate the feasibility and the effectiveness of
the MIMO V-BLAST in the proposed TDCS-CCSK with embedded symbol system. The
results have shown a significant improvement in data detection for a multi-user environment
experiencing Rayleigh fading.
161
Chapter 6
AAF Project Cognitive Radio Demonstrator

in IRCTR TU Delft
6.1 Introduction
The goal of AAF is: “Research and demonstration of a Cognitive Radio system, which
continuously adapts its communications scheme to the available resources”. The focus of the
research is on identification of free resources in the frequency domain and adaptive OFDM-
based transmission. Important challenges include stability of the system, identification of
interference to other systems and meta-communications. All are addressed as part of the
research. A verification platform will prove the consistency and comprehensiveness of the
solution. The available demonstrator allows us to investigate the potentials of co-existence
between the licensed system and the rental Cognitive Radio based system.
In this chapter, the design and implementation of a Cognitive Radio test bed is proposed. The
test bed can be used to validate to which degree the Cognitive Radio system is non-interfering
to primary users. The system can emulate primary and secondary users and accordingly the
performance of various adaptive radio transmission schemes can be evaluated. The CR test
bed comprises two basic parts, the radio (to generate, transmit, receive and process wireless
signals) and a cognitive brain which learns from the environment and performs rational
processes and predicts probable consequences and remembers past successes and failures.
The baseband processing in this thesis itself is the radio that generate, transmit, receive and
process the wireless signals.
The AAF requirements are considered in creating a working hardware demonstrator.

Accordingly the hardware requirements are specified. Based on the requirements, criteria
have to be developed to enable of making decision on selection of the vendors and the boards.
In this chapter the Cognitive Radio development verification platform in IRCTR TU Delft§
within the AAF project will be described. Section 6.2 gives the insight into the hardware
§
The works on the verification platform at IRCTR TU Delft were performed by student assistants : Shreyas Bhargav
Raghunathan, Maarten v. d. Oever, Zhen Qin, R. D. Mohammady under the author’s guidance and Przemysław Pawełczak,
as well as technician Fred v. d Zwan and Faculty members Dr. H. Nikookar, Dr. R. Hekmat and Prof. Dr L. P. Ligthart.
162
selection according to the hardware specification that fulfils the desired Cognitive Radio
requirements. Among several available boards we come up with the decision of choosing
P25-M and USRP as the development boards for the IRCTR TU-Delft Cognitive Radio
verification platform.
Section 6.3 describes the observed scheme configuration involving the collaboration of USRP
and P25M. The adaptive spectrum selection will be shown, and the proposed future expected
scenario is presented. Summary of this chapter is given in section 6.4.
6.2 The Hardware Requirements for Baseband Transmission

A digital signal processing (DSP) board for the processing of the signals is required in a
Cognitive Radio verification platform. A DSP board is required because a regular computer
would not be able to process the high bandwidths. The DSP board must be able to fulfill
various applications required.
The most important parts on the board are the decoding and encoding of the analogue signal.
The encoding and decoding are done by Digital to Analogue Converters (DAC) and Analogue
to Digital Converters (ADC). If these components are too slow then it is impossible to decode
or encode the desired bandwidth for the transmission.
According to [34] the requirements in Table 6.1 can be distilled.
Table 6.1: The AAF project parameter requirements
Parameter Value
Topology Multi-hop
Modulation scheme ODFM
Minimum frequency (RF) ƒmin 400 MHz
Maximum frequency (RF)ƒmax 900 MHz
Transmit power 30 dB(m)
Minimum channel SNR 15 dB
Minimum bandwidth B min 1 MHz
Maximum bandwidth B max 10 MHz
Maximum number of carriers 128
Maximum Doppler frequency ƒd 15 Hz (at 900MHz, 18 km/h)
125 Hz (at 900 MHz, 150 km/h)
250 Hz (at 900MHz, 300 km/h)
163
The desired total bandwidth is set to 10 MHz. Figure 6.1 depicts an illustration of the system
output, where [C1, C1,…,CNFFT] are the position of each subcarrier.
Figure 6.1: OFDM subcarriers with 10 MHz bandwidth
The 10 MHz bandwidth (B) delivers a minimum sample rate (fs) for the analogue to digital
converter of 20 MHz. The relation between B and fs is defined as :
fs
B= (6.1)
2
Modern filter techniques allow for generation a smooth signal with a low output frequency.
The DAC must be able to pre-form a digital to analogue conversion at the same rate as the
ADC.
The performance of ADCs can be measured by its Figure of merit (FoM). The FoM takes
into account the converter’s power dissipation P[W], its resolution (specified in
Effective Number of Bits (ENOB)) and its sampling rate fs [Hz]. This FoM is given as [34]:
P
FoM = ENOB
[pJ / conversion - step] (6.2)
2 . fs
Nowadays, having a FoM of 2pJ/conversion-step for an ADC is considered to be reasonable.

The ENOB that an ADC must support can be estimated by the dynamic range of the band of
interest i.e., ΔdB.
164
Assuming a 6 dB per bit dependence for AD converters, the minimum number of desired bits
can be determined from this dependence and the dynamic range ΔdB. Next formula is used
especially for OFDM Systems [34]
ΔdB + 4.77
b≅ (6.3)
6.02
where b is the effective number of bits. The dynamic range is determined from the carrier
spacing Δf , and the minimum channel frequency f min.
f
ΔdB = 20log(1 + min ) (6.4)
Δf
The carrier spacing is defined as :
1
Δf = (6.5)
1
N FFT
B
where B is the signal bandwidth and NFFT is the total number of subcarriers. By using the
maximum bandwidth (Bmax) 10 MHz, the carrier spacing would be,
1
Δf = = 78.125 kHz (6.6)
1
128 ⋅
10 ⋅ 106
Therefore, the dynamic range value equals:
890 ⋅ 106
Δ dB = 20log(1 + ) = 81.13 dB (6.7)
1
1
128 ⋅
10 ⋅ 106
Putting this dynamic range ΔdB value in eq. (6.4) will yield minimal number of desired bits of
the AD converter
81.13 + 4.77
b= = 14.26 bits (6.8)
6.02
Therefore, a minimum of 14 bits converter would be suitable to fulfil the 10 MHz signal
bandwidth. To assure that each OFDM subcarrier will experience flat fading, the carrier
spacing should be much larger than the Doppler frequency (fD) [34]
f D << Δf (6.9)
165
By referring to Table 6.1, having an 78.125 KHz carrier spacing will satisfy eq. (6.9). It will
accommodate the expected 250 Hz maximum Doppler frequency. Further, the requirement
according to [34] described in (6.10)
B
N FFT << (6.10)
fD
will be fulfilled as because
⎛ 10 ⋅ 106 ⎞
128 << ⎜ ⎟ = 40 x 10
3
(6.11)
⎝ 250 ⎠
The evaluation boards have Field Programmable Gate Arrays (FPGA) on board. An FPGA is
a device with programmable connections and logic. The FPGA can be used for the signal
processing. The FPGA is programmed in Very High Speed Digital Hardware Description
Language (VHDL).
The FPGA must be able to process the data and transfer it to the DSP chip. The FPGA
capability depends on the number of available gates and clock rate. The maximum data rate of
the samples that is offered to the FPGA is 25 MHz of data. It is offered by the ADC. The
FPGA is able to process the data pipelined there, with a required minimum clock speed of 25
MHz.
Some of the boards also have a DSP processor on it. The DSP processor is a processor
dedicated for digital signal processing. The FFT is going to be implemented on the DSP
processor. This is a major advantage over the FPGA. The processors can be programmed
using C/C++. The available memory on the DSP processors is important, and also the number
of bits the DSP can handle. Usually the number of bits of the DSP is equal or bigger than the
number of bits of the ADC. If the DSP processor is used for the FFT, then the result matrixes
need to be stored. The required memory is quantified as :
FFT table = 2 ⋅ 20 ⋅ 128 ⋅ 16 ⋅ 2 = 163840 bits (6.12)
The FFT table size consists of an FFT table and the invFFT table (this is the reason for the
factor of 2 in eq. (6.12)), with 20 symbols for each of the 128 carriers that consist of 16 bits
real and 16 bits imaginary numbers. The result in eq. (6.12) denotes that 163.840 bits
memory space is required to save the FFT tables.
A dedicated control channel is required to provide information about the state of the link. This
is necessary when a node experiences a lot of interference and receives corrupted data. A
166
proposed solution for the control channel is Tmotes ** . The disadvantage of Tmotes (or
standard 802.15.4) is its short distance communications. For the demonstrator it would be
convenient if the board has a Multiple Input Multiple Output (MIMO) capability. MIMO
provides a higher bitrate. MIMO requires an additional ADC and DAC installed on the Board.
The extra ADC and DAC can be used as side channel. If a board contains only one additional
ADC or DAC then they will be used for side channel.
For the boards implementation usually a computer and programming software is necessary.
For the programming of the FPGAs a JTAG connector is required. The FPGA needs to be
programmed because the ADC and DAC is usually connected to the FPGA. When a board is
selected, it is preferable that the board supports 32 or 64 bit PCI, because these standards are
supported by common computers. When it comes to varying the number of subcarriers and
the signal bandwidth, Table 6.2 and 6.3 show the expected hardware requirements that have to
be satisfied. Table 6.2 shows the required hardware specification if the signal bandwidth is 10
MHz and the number of subcarriers is 128.
Table 6.2: Hardware specifications if B= 10 MHz and NFFT=128
Variable value
Samples frequency 25 MHz††
ΔdB 81.13 dB
Inter frequency spacing 78125 Hz
ADC resolution‡‡ 14,26 bits
Doppler shift tests Passed
FFT tables memory 163.840 bits
Table 6.3: Hardware specifications if B= 10 MHz and NFFT=64
Variable value
Samples frequency 25 MHz§§
ΔdB 75.11 dB
ADC resolution*** 13,26 bits
Doppler shift tests Passed
**
Tmotes are small embedded devices which are capable of measuring several features in the surrounding
environment such as temperature and light. In addition to this they are equipped with a radio transmitter which
can be used to communicate between Tmotes.
http://www.daimi.au.dk/~spony/adv_courses/NPaI/report/node4.html
††
20MHz + 5MHz over sampling
‡‡
The ADC resolution is also valid for the DAC
§§
***
167
Lowering the FFT size on the same signal bandwidth will reduce the required number of ADC
bits therefore the required memory size will be reduced. Table 6.4 shows the hardware
specification if the signal bandwidth is 2.5 MHz and the number of carriers is 128.
Table 6.4: Hardware specifications if B= 2.5 MHz and NFFT=128
Variable value
Samples frequency 6 MHz†††
ΔdB 93.17 dB
ADC resolution‡‡‡ 16,26 bits
Doppler shift tests +§§§
From Table 6.4 it can be concluded that the sample frequency has decreased but the number
of required bits has increased. This means that the processing speed can be a lot slower.
After excluding some high end boards, other options needed to be taken into consideration.
The selection of another high end board was not preferred. After browsing the Innovative
Integration site the P25M board was the closest one which fits with the requirements given in
Table 6.1. Therefore P25M has been selected [107]. The development board is depicted in
Figure 6.2.
Figure 6.2: P25M development board
†††
‡‡‡
§§§
+ means passed the Doppler shift tests
168
The specifications of the P25M are listed in Table 6.5. With these specifications the device is
capable to meet the requirements concerning the ADC and DAC.
Table 6.5: Specifications of Innovative Integration P25M
ADC component DAC component

Number of channels 4 Number of channels 4
Sample rate 25 MSPS Max generation rate 50 MSPS
Max Bandwidth 12 MHz Max Bandwidth 25MHz
Bits width 16 Bits Width 16
Connection To FPGA Connection To FPGA
Power dissipation unknown Power dissipation unknown
The onboard FPGA is a Xilinx Spartan 3 with 1 million gates. The DSP chip of the P25M is
a Texas Instruments with clock speed of 300 MHz and 32 Mega Bytes of RAM memory. The
processor is a 32 bit floating point processor.
The TMS320C6413 DSP processor takes about 500 clock cycles to calculate an FFT. This
means that the TMS320C6413 can process 6000 FFT`s per second. That is not enough for the
demonstrator.
The P25M fits in a standard PCI slot. The PCI slot is a 64 bit slot. This means a 64 bit
computer needs to be used in for the project.
The P25M is capable to meet the bandwidth requirements. By having 4 channels, the P25M
can support MIMO as well. Besides the results of the DSP processor the P25M can keep up
with the requirements and is relatively low priced. That makes the P25M to be a serious
candidate for the use in the demonstrator.
169
6.3 Spectrum Scanner Module

In scanning the spectrum, the universal software radio peripheral (USRP****) is chosen as the
suitable device. USRP is a software radio device, connected to a host computer via the USB
port with 480 Mb/s data rate. It can send and receive signal up to 16 MHz of RF bandwidth.
The USRP contains an FPGA that can be reprogrammed, 4 high speed ADCs, 4 high speed
DACs, and lot of auxiliary analog and digial IO to do integration into a larger system.
USRP can accommodate up to 2 transceiver daughterboards, therefore it is capable of having

2x2 MIMO. Table 6.6 describes the specifications of the USRP.
Table 6.6: USRP Specifications
ADC DAC
Number of channels 4 Number of channels 4
Sample rate 64 MSPS Max generation rate 128 MSPS
Max Bandwidth 16 MHz Max Bandwidth 16MHz
Bits width 12 Bits Width 14
SFDR 85 dB SFDR 83 dB
SFDR stands for Spurious Free Dynamic Range. SFDR is defined as the ratio of the RMS
value of the carrier frequency (maximum signal component) at the input of the ADC or DAC
to the RMS value of the largest noise or harmonic distortion component at its output.
USRP is equipped with 4 digital downconverters with programmable decimation rates, 2

digial upconverters with programmable interpolation rates, high speed USB 2.0 interface (480
Mb/s); modular architecture supports wide variety of RF daughterboards; auxiliary analog and
digital I/O support complex radio controls such as RSSI and AGC.
High sample rate processing takes place in the FPGA, while the lower sample rate processing
happens in the host computer. The two onboard digital downconverters (DDCs) mix , filter,
and decimate (from 64 MS/s) incoming signals in the FPGA. Two digital upconverters
(DUCs) interpolate baseband signals to 128 MS/s before translating them to the selected
output frequency. The DDCs and DUCs combined with the high sample rates also greatly
simplify analog filtering requirements.
The baseband processing for transmission and reception is applied in a host computer
connected to the USRP through the USB interface. The spectrum sensing itself would be
considered as reception processing, and it is performed based on energy detection.
****
Information about USRP is available at http://www.ettus.com/
170
The USRP mainboard can be connected to several daughterboard options. Daughterboards are
the front end sub system part that will transmit or receive the signal on the desired operating
carrier frequency.
Figure 6.3 depicts the USRP development board including its daughterboard.
Figure 6.3: Inside the USRP
6.4 Cognitive Radio Nodes Demonstrator Design

The basic single node Cognitive Radio configuration is depicted in Figure 6.4. The evaluation
system involves 2 USRP modules, 2 laptop PCs as host computers for the USRP, and a
desktop computer with the P25M board installed. The results of the configuration according
Figure 6.4 has been presented as a demonstrator paper in reference [108].
The system applied cooperative sensing, the spectrum sensing from USRP1 is combined with
spectrum sensing result from USRP2 through a Samba Linux network. USRP1 acts as server
that will give the decision about final spectrum occupancy information, and further sends this
information to the baseband processing module on the P25M. Until now, the P25M is not
connected to any front end interface for transmitting the signal to air. But the final goal of
such a demonstrator is transmitting of adaptive CR signals in air.
The demonstrator aims to be a proof-of-concept Dynamic Spectrum Access (DSA) system.

Specifically, a simple multi-carrier OFDM-based CR system with Spectrum Pooling (SP) has
171
been implemented in the P25M platform. SP is applied by windowing and carriers de-
activation, where cancellation carriers can be inserted. Channel estimation has been applied
by using the Least Squares method. In a later phase the developed Wiener filter with optimum
pilot pattern and virtual pilot concept will be implemented. In the Least Squares method
several OFDM symbols are transmitted as pilots.
Linux GNU Radio

USRP 1 Laptop
PC1
(Server)
PC
Samba Linux
Network
P25M
Linux GNU Radio

USRP 2 Laptop
PC2
(Client)
Figure 6.4: Single Cognitive Radio unit with 2 spectrum sensing elements (USRPs) and one
baseband processing sub-unit
The block design of P25M is represented in Figure 6.5. In the beginning we implement a
simple OFDM-based cognitive radio scheme with spectrum pooling. The spectrum pooling
will be simply the time domain windowing combined with adaptive de-activation of carriers
located in the occupied licensed user’s band. The Block diagram of the OFDM demonstrator
is depicted in Figure 6.6. Figure 6.7 shows the algorithm of TDCS at the transmitter side,
while the receiver side is depicted in Figure 6.8.
The P25M performs the following tasks;

Implementation of Adaptivity by the Adaptive bit loading algorithm.
Baseband transmission and reception of the data bits (coding and modulation).
172
Figure 6.5: P25M Block design
Figure 6.6: Signal flow of IRCTR-AAF Cognitive Radio Demonstrator

173
Spectrum
Sensing
Memory
DAC
Spectrum
occupancy
Matrix
Real Real Real Real
Scaling Guard
IFFT Buffer Interval
and sorting Shifter
Insertion
Imag Imag Imag Imag
Random Phase
Generator
Data bits
Figure 6.7: Transmitter architecture of TDCS
Channel
Estimation
Real Real Real Real

Remove Channel
Guard FFT Fade IFFT Desicion Data
Interval remover
Imag Imag Imag Imag
Imag
Real
Conjugate
Imag
Real
Reference Phase
Figure 6.8: Receiver architecture of TDCS with Channel Estimation module

174
An example of cooperative spectrum sensing results is shown via a graphical user interace
(GUI) on a linux computer, depiced in Figure 6.9.
Figure 6.9: An example of GUI spectrum scanner with 3 MHz bandwidth , frequency range
2.402-2.405 GHz
We have implemented the spectrum pooling technique described in section 2.2 to the
baseband processing in P25M. The window is rectangular, 128 subcarriers were designed.
Figure 6.10 (a) shows the subcarrier occupied by the Cognitive Radio (CR) system using
P25M baseband processing, and accordingly Figure 6.10 (b) gives the power spectrum density
of the CR OFDM signal.
The VHDL simulation results of P25M are presented in Figure 6.11. There is a timing offset
between the results before IFFT module (data[5:0] in Figure 6.11) and after FFT module
(data_out[5:0] in Figure 6.11) on the simulations due to FFT and IFFT processing time. Each
symbol is represented by 6 bits ([5:0]).
Figure 6.12 depicts the expected future observation of a Cognitive Radio networks with three
nodes. Each node will have its own spectrum scanning capability. A node will act as the
server that gives a decision about the spectrum occupancy information. This information will
be distributed to the three nodes.
175
(a) (b)
Figure 6.10: (a) Carriers occupied by the Cognitive Radio System; (b) Cognitive Radio power
spectral density
176
Figure 6.11: VHDL simulation results of the OFDM based Cognitive Radio System on FPGA
177
USRP
WLAN side
channel
USRP
USRP
Figure 6.12: Outlook of three Cognitive Radio nodes with spectrum sensing capability on
each node
The following conditions of the network should be satisfied :

Each SU (Secondary User) is a PC equipped with an USRP and a P25M DSP board.
The RFX400 transceiver daughter boards are installed on the USRP main board and
are in charge of the dynamic spectrum access.
P25M runs the OFDM transmission and adaptive bit loading algorithms under PC’s
control.
A dedicated control channel based WLAN is used for SU’s to cooperate with each
other.
An SU communicates with other SU’s by sending its result of spectrum sensing.

178
The USRP block design is depicted in Figure 6.13. The PC-based USRP is a device which
allows us to create a software radio and can be programmed for the following tasks:
Gathering local spectrum sensing data. Each node listens to the activities in the
currently designated band via an RF interface built on and controlled by the USRP.
The data is then processed by the USRP and awaiting for P25M access.
Networking. Meta communication has been investigated by using an Ultra Wide Band
Technique [109]. In this demonstrator the meta communications is carried out by an
WLAN side channel, to exchange spectra availability information. Real data
communication between nodes is taken care of by USRP’s with a wireless interface.
Figure 6.13: USRP block design

179
6.5 Conclusions
In this chapter a demonstrator as verification platform of a Cognitive Radio system has been
proposed. The sensing is conducted by the Universal Software Radio Peripheral (USRP)
device, while the signal processing for transmission is applied in an P25M evaluation board.
The initial system has been build by combining 2 USRP devices as sensing module. Each of
the USRP is connected to a computer for sensing the signal processing purpose. One
computer will act as server that will do cooperative sensing by combing the sensing
information from the 2 USRP devices.
It is expected that the system can work with 10 MHz signal bandwidth. The ADC and the
DAC component of the transceiver module should fulfil the nowadays FoM of 2pJ per
conversion step. Accordingly, among several available evaluation boards on the market, the
P25M board is selected.
At the initial stage, the minimum baseband processing i.e. the spectrum pooling and
windowing have been implemented in the P25M. It is foreseen that complex algorithms such
as channel estimation, synchronization and PAPR reduction can be incorporated in the
platform.
The information about the spectrum occupancy is circulated through the WLAN channel. It is
expected that in a real application this information should be broadcasted to each Cognitive
Radio node in a side channel i.e. UWB.
Extra evaluation and design are required. It is expected that in future the P25M will be
connected to the front end (RF) module. The feasibility of the USRP as front end module and
the possibility of USRP to be applied as transceiver module are still under investigation.
180
181
Chapter 7
Conclusions and Recommendations
Cognitive Radio is an intelligent communication system. It has the capability to learn from its
environment and react accordingly by changing its transmission parameters to obtain reliable
communications and efficient spectrum utilization. Since one of the Cognitive Radio
objective is to obtain efficient utilization of the radio spectrum, and the focus of our Adaptive
Ad-hoc Freeband (AAF) project is on identification of free resources in the frequency domain,
the specific scope of Cognitive Radio is the effective spectrum use. The intelligence and
learning process by observing and analyzing the channel and spectrum conditions and further
react to it by adjusting the transmission parameters will aid the radio system to access the
spectrum effectively while reliable communications can still be achieved. In this thesis,
special attention has been given to efficient spectrum utilization in the Cognitive Radio
context by having the co-existence between the rental (Cognitive Radio) system with the
legacy (licensed) system side by side while each of the systems has their quality of service
(QoS) preserved.
7.1 Conclusions
The successful co-existence between the rental system with the legacy system side by side can
be realized by having low mutual interference between the two systems. Chapter 2 introduces
Orthogonal Frequency Division Multiplexing (OFDM) as a proper modulation technique to be
applied in the Cognitive Radio system due to its flexibility in the notching part of its spectrum
on the licensed user band by de-activating some of its carriers. In order to apply the carrier de-
activation, spectrum occupancy information is required to locate the position of de-activated
carriers in frequency-domain. In this dissertation, this information is assumed to be available.
The spectrum occupancy information is obtained from the spectrum sensing module which is
not the focus of this thesis. Spectrum sensing is another critical research issue and requires a
proper attention. Simple sensing can be applied by energy detection in a certain frequency
band. The project partner in Twente university conducted the spectrum sensing related
research issue. The principal investigator is Marnix Heskamp. In order to have really accurate
spectrum occupancy information, a special mechanism is required to gather spectrum
information from all available cognitive radio devices. This issue is even more critical
especially in the case that the LU’s geographical position is close to one RU so that it is easily
182
detected by the corresponding RU while that LU is far from other RUs that probably cannot
detect its presence. A simple mechanism would be the application of the logical ‘OR’
operation to the spectrum information vector of all RU transceivers. In the AAF project the
spectrum information is planned to be distributed to all RU (Cognitive Radio) transceivers
through a dedicated control channel. Techniques for mutual interference reduction between
the OFDM-based rental (Cognitive Radio) system and the licensed system has been studied in
this thesis. Among them are combined windowing and guard band handling (adjacent carriers
de-activation), sidelobes cancellation carriers, frequency-domain additive signal, adaptive
transition symbol, multiple choice sequence and constellation expansion. If due to the
licensed user access, too many carriers of the OFDM-based Cognitive Radio system need to
be de-activated, then the available carrier resources cannot accommodate the targeted
Cognitive Radio system QoS. In this case frequency hopping is suggested. Ultra Wide Band
could be an alternative due to its signal power property below the noise level hence it can be
overlapped with the LU signal it self. Because the demand of the Cognitive Radio service is to
provide high data rate at long distance transmission, the use of frequency hopping scheme is
preferable. The frequency hopping has to collaborate with the spectrum sensing module. As
the spectrum scanning module indicates that a certain amount of bandwidth is available at a
certain carrier frequency (fc), then the front end module will hop to that frequency, stabilizes
itself and then transmits data.
Following the objectives of Cognitive Radio to obtain reliable communications and the
emphasis of efficient spectrum utilization on the co-existence between the rental system with
the legacy system, in chapter 3 we have proposed OFDM with adaptive bit loading, as the
technique to preserve the rental system QoS, to be combined with the sidelobes reduction
technique using windowing and cancellation carriers as a means to reduce the mutual
interference between the rental and the legacy systems. Simulation results have shown that
with perfect channel estimation, a 5 dB SNR gain could be achieved by applying the adaptive
bit loading with the Fischer algorithm compared to the conventional OFDM; the combination
of a raised cosine window with roll off factor 0.2 and 2 cancellation carriers on each side
resulted into more than 30 dB sidelobes reduction. The possibility of peak to average power
ratio (PAPR) growth due to the sidelobes cancellation carriers insertion has been analyzed.
The evaluation shows that the sidelobes cancellation carriers give insignificant impact to the
PAPR growth. Furthermore, the Fourier transform in OFDM can be replaced by the wavelet
basis function. Literature has indicated that as wavelet basis functions give lower sidelobes
compared to Fourier-based OFDM; in this thesis we proposed the use of a frequency selective
wavelet basis function instead of the Fourier transform in OFDM combined with the spectrum
pooling concept (de-activation of carriers in the licensed user band) for Cognitive Radio
application. The technique is termed as the wavelet packet multicarrier modulation
(WPMCM). The parameters of the frequency selective wavelet basis function such as the
filter length, regularity order and the transition bandwidth can be set to fulfill the desired
performance outcome i.e. the sidelobe amplitude. The results in an AWGN channel showed
that at certain frequency selective wavelet parameters the WPMCM based Cognitive Radio
outperforms the OFDM based Cognitive Radio in term of BER performance. As MIMO has
183
been well known as technique to increase specral efficiency, its advantage in the Cognitive
Radio application was observed. The combination technique between windowing,
cancellation carriers and adaptive bit loading will give a spectral efficiency enhancement but
at the expense for extra complexity in the implementation.
Channel estimation is crucial and is considered as the most stringent module in the OFDM
system. The channel estimation module is even more important in the Cognitive Radio
context. The channel estimation would be the Cognitive Radio awareness element on the
changing channel environment and its fading condition. An example of its use is in the
application of adaptive bit loading. In adaptive bit loading the bits are allocated to each
subcarrier according to the estimated fading state on the corresponding subcarrier. In this
thesis the pilot-aided channel estimation is chosen to be applied in the Cognitive Radio
system due to its better performance and less complex implementation compared to the blind
channel estimation. A Wiener filter utilizes the distance between pilots in time and frequency
in an OFDM frame and relates them with the channel correlation function in time and
frequency to design the filter coefficients for channel estimation purpose in a straightforward
way. Therefore, without loss of generality, in this thesis this filter was chosen for the
Cognitive Radio system. In the application of frequency hopping, the OFDM frame design
according to the Bluetooth TDD duration was proposed. Bluetooth was evaluated, as this
technology uses frequency hopping technology in its transmission. Through simulations the
frame design was evaluated in 900 MHz in a rural area and urban area. The results showed
that one training symbol can effectively accommodate the channel estimation on one frame.
The 900 MHz channel was explored due to the available information regarding its channel
power delay profile characteristics in the ETSI documentation. Adopting the efficient
implementation of a hexagonal pilot pattern in a conventional OFDM system, in chapter 4 we
proposed the combination of hexagonal pilot pattern, pilot shifting and virtual pilots concept
to be applied in the Cognitive Radio context, where some of the carriers will be de-activated
due to the licensed user access. The proposed virtual pilot concept itself takes advantage of
the slowly varying channel condition between two real pilot positions and accordingly
predicts the initial channel estimates at the virtual pilot positions by linear
interpolating/extrapolating the initial channel estimates at the real pilot positions combined
with the decision directed method. The results showed that the utilization of the virtual pilots
in a hexagonal pilot pattern within the Cognitive Radio context increased the channel
estimation performance by more than 10 dB compared to the Hexagonal pattern without
virtual pilots. Extensive work was done with respect to the optimal pilot pattern for a
Cognitive Radio system by adopting the optimum single input single output (SISO)
Hexagonal pilot pattern combined with the virtual pilot concept for Cognitive Radio system
into its multiple input multiple output (MIMO) system version. Without loss of generality a
2x2 MIMO system was considered. Since the received signal at each of the receive antenna
gets contributions from signals of all transmit antennas, the application of the decision
directed method in the virtual pilot concept could lead to a severe decision error. Therefore
the decision directed method was removed from predicting the initial channel estimates at the
virtual pilot position. The stresses of two virtual pilot utilization concept was made to the
184
MIMO pilot pattern for the Cognitive Radio system. The first pattern utilizes the virtual pilots
at a longer distance between pilots in time. The impact of the virtual pilots will be
significantly caught at low maximum Doppler frequencies. The second pattern emphasizes the
use of the virtual pilots at a longer distance between pilots in frequency. Therefore, a
significant impact of the virtual pilots could be seen on the channel model with low delay
spread. Further, the second pattern could be extended to have more virtual pilots for the
intention of having an oversampling gain. Given the simulation parameters with fixed power
delay profile, and different maximum Doppler frequency, the first pattern outperformed the
second pattern including the one with more virtual pilots. The gap between them was even
larger at low maximum Doppler frequencies.
In chapter 5 an attempt was made to increase the single carrier transform domain
communications system (TDCS) bit rate by embedding an extra symbol as another bit source.
The embedded symbol has not only increased the TDCS bit rate but also improved the cyclic
code shift keying (CCSK) detection as the legacy data source depending on the constellation
size of the embedded symbol. As the embedded symbol constellation size increases the CCSK
BER reduces. Analysis and simulations showed that the embedded symbol does not destroy
the multiple access capability of TDCS in an AWGN channel environment. A Rayleigh
fading channel disturbs the cross correlation of the TDCS multi-user basis function.
Simulation results indicated how the CCSK and embedded symbol BER were degraded. A
further attempt to enhance the TDCS performance was made by employing MIMO. The
results displayed how MIMO TDCS could combat the fading effect. The simulation results
indicated, by having a ratio of ½ between the transmit and receive antenna, the BER
performance of TDCS-CCSK with embedded symbol in a Rayleigh fading channel resembles
a performance as in an AWGN channel. Due to the system design, the embedded symbol
detection was applied after CCSK detection. Therefore, the error in CCSK detection
propagated to the embedded symbol detection. For to this reason, the BER on the embedded
symbol was influenced by the tendency that a higher embedded symbol constellation sizes
gave a better CCSK detection while the embedded symbol detection was deteriorated. These
two tendencies put a trade-off to the final embedded symbol error rate. The results showed
that in the TDCS system the embedded symbol QPSK detection performs better than the
BPSK detection. The application of the embedded symbol was also evaluated in the wavelet
domain communication system (WDCS). In the WDCS the Fourier transform that forms the
basis function into time domain signal is replaced by the wavelet packet basis function. In this
thesis, we proposed the usage of the frequency selective wavelet in the WDCS with an
embedded symbol. The results demonstrated that the CCSK detection in WDCS is more
accurate than in TDCS. For this reason, the WDCS could preserve the PSK modulation
behavior (higher constellation renders higher BER) in the embedded symbol detection. In
terms of power spectral density, and scaling factor 1 the single carrier TDCS/WDCS based
Cognitive Radio has a factor 1/NFFT lower sidelobes compared to multi carrier
OFDM/WPMCM counterpart. Increasing the scaling factor of TDCS/WDCS increases its
sidelobes and also its SNR at the expense of a larger amount of required transmit power.
185
In chapter 6 a prototype Cognitive Radio verification platform was presented. The P25M was
selected for the development board, while USRP was considered to aid the P25M in providing
the spectrum occupancy information. The initial platform includes 2 USRP as a means to
provide the cooperative sensing effect. Each of the USRPs is connected to a computer that
conducts the signal processing, while the P25M is embedded also to a computer. One USRP-
computer node acts as a server where the decision about the final spectrum occupancy
information is made. The connection between the two USRP-computers and the computer
(which contains the P25M board for the spectrum occupancy information transfer) is
employed through a local area network (LAN). Progress on this prototype is still ongoing. It is
planned that the developed baseband processing in this thesis will be implemented in the
development board (P25M). Furthermore the transceiver front end component will be
appended to the P25M and the system will be evaluated from the signal on air.
7.2 Recommendations for Future Work
In this thesis research has been done on Fourier-based modulation such as OFDM and TDCS
based Cognitive Radio. Sidelobes reduction techniques in OFDM have been overviewed.
Without loss of generality the sidelobes reduction and fading mitigation applied to OFDM
could also be applied to the TDCS system. Channel estimation in OFDM-based Cogntive
Radio has been explored. The same channel estimation technique in OFDM is expected to be
applied in the TDCS system too. Bit rate enhancement in OFDM is applied intelligently by
adaptive bit loading, while in TDCS extra embedded symbol is the proposed technique.
Limited subjects have been explored to the Fourier-based modulation counterpart, which is
wavelet based modulation such as WPMCM and WDCS. The evaluations have been limited
to the AWGN channel. Literature describes that time domain equalization is a suitable
technique to remove the fading effect in the wavelet based modulation technique (WPMCM
and WDCS).
Our evaluations have shown that frequency domain equalization is not feasible to be applied
to WPMCM system. We predict that this problem is caused by the overlapping property of the
transmitted WPMCM signal. It seems like the time domain equalization is the realistic
solution in counteracting the channel fading effect to the signal.
By applying a suitable equalization technique the use of frequency selective wavelet in

WPMCM and WDCS based Cognitive Radio system could be explored further in a Rayleigh
fading channel, including the embedded symbol application to the WDCS. The work can
further be extended to MIMO applications since the channel state information will be
provided by the equalization module.
Another possible future work with respect the WPMCM implementation is the application of
the tree extension on the subcarriers with good channel conditions to compensate the loss of
186
data source due to de-activated carriers. It is expected that extending the tree on a subcarrier
with good channel condition will give higher rate and good BER. The scheme is depicted in
Appendix G.
187
Appendix A
OFDM based Cognitive Radio Frame

Design for Frequency Hopping Scenario
A frame design is chosen with small bandwidth, and the frame duration is based on the TDD
Bluetooth frame. The TDD Bluetooth system is depicted in Figure A.1. Only about half a
duration of the slot can be used for transmission. After the system transmits the bits, it hops to
a free (unoccupied) frequency band and it needs time to stabilize itself in that band. The frame
duration must not exceed 366 µs.
According to [34] in order to fulfil 2 pJ(pico joule) per conversion step Figure of Merit of the
analog to digital converter (ADC), the signal bandwidth must not exceed 5 MHz. Since the
sampling frequency is twice the signal bandwidth, the sampling time must not be less than 0.1
µs.
As the spectrum scanning module indicates a 5 MHz bandwidth (available in a certain carrier
frequency (fc) e.g. between 800 -900 MHz), the front end module will hop to that frequency,
stabilizes itself and then transmits data. Depending on the cognitive radio device capability,
and assuming that the scanning process and transmitting data are independent of each other,
the scanning can be conducted in parallel to the front end’s oscillator stabilization and data
transmission process. After data transmission, the information on the free 5 MHz bandwidth
(provided by the spectrum scanning module) will be available to each CR transceiver through
a common control channel [11]. Upon receiving the fc information, the front end will set the
oscillator to the corresponding fc. This algorithm can obviously only be applied to the device
which has parallel processors. In this way, the scanning and transceiver are done separately
and parallel to each other. In this thesis, the device configuration is realized by combining the
USRP as spectrum scanning module with the P25M development board as transceiver
module. USRP can also be used as transceiver module. More about these two platforms and
demonstrator are presented in chapter 6.
Time required for the transmitter’s front end to transmit the data and stabilizing itself to the
new carrier frequency is about 625 µs; during this period also the spectrum scanner module
senses the frequencies. The process flow of the frequency hopping including spectrum
scanning and data transmission is described in Figure A.2.
188
For the devices without having the capability to scan and transmit data in parallel, a certain
time slot is required for dedicating the processor to only scan the spectrum and then proceed
the transmission process at the cost of a bit rate degradation.
Figure A.1: TX/RX cycle of the Bluetooth system
Figure A.2: Process flow of hopping and data transmission for a CR system, OFDM based
with a Bluetooth TDD frame design
The advantage of having the capability to scan and transmit in parallel is the following. In the
time a cognitive radio device does its transmission in a certain frequency band and an LU in
between that period also transmits its signal in that band, the spectrum scanning module of
189
that cognitive radio device will detect the LU presence and commands the transmission
module to stop its transmission. Then it scans for a new frequency in a different band. An
extended capability will be attained if the cognitive radio has two scanning module that can
work in parallel. One scans in the own band where it transmits data, and the other one scans a
different band to find an alternative if a sudden LU access appears in the band where the
cognitive radio system transmits data.
With respect to this mechanism, the spectrum sensing module of that cognitive radio device
should have knowledge about the data transmitted by the transmission module , e.g. in terms
of a pilot or training symbol, then should subtract this known data from the received signal at
that frequency band, and further apply LU signal detection. In the cooperative sensing
mechanism (where the logical OR operation is applied), the decision by one device in
claiming one spectrum region is occupied will be enough to notch the spectrum of all
cognitive radio devices within its network in that occupied spectrum region. In this work, we
assume that this sensing mechanism is available.
190
Appendix B
Training Symbol Repetition Period
Mathematical Proof by Zeros Insertion
To validate the zero insertion on the odd index of subcarriers, we can observe the following
equations.
The first one is if the zeros are inserted on the odd index of subcarriers.
N FFT − 2 N FFT − 2
N FFT −1 j 2π kn j 2π (2 k ) n j 2π ((2 k ) −1) n
1 1 2
1 2
xn =
N FFT
∑
k =0
Xke N FFT
= ∑ X 2k e
N FFT k =0
+ ∑ X (2k )−1e
N FFT k =0
N FFT N FFT

even index of subcarriers 0

odd index of subcarriers,
where values of the component X 2k-1 are zeros
N FFT − 2
j 2π (2 k ) n
1 2
=
N FFT
∑k =0
X 2k e N FFT
⎛N ⎞ N FFT − 2 ⎛N ⎞
j 2π k ⎜ FFT + n ⎟ j 2π (2 k )⎜ FFT + n ⎟
N FFT − 2 ⎝ 2 ⎠ ⎝ 2 ⎠
1 1 2
x N FFT
+n
=
N FFT
∑
k =0
Xke N FFT
= ∑ X 2k e
N FFT k =0
+ N FFT
2
even index of subcarriers
N FFT − 2 ⎛N ⎞
j 2π ((2 k ) −1)⎜ FFT + n ⎟
⎝ 2 ⎠
1 2
∑ X (2k )−1e
N FFT k =0
N FFT
0
odd index of subcarriers,
where values of the component X 2k-1 are zeros
N FFT − 2 N FFT − 2
j 2π (2 k ) n j 2π (2 k ) n
1 2
1 2
=
N FFT
∑k =0
X 2k e N FFT
e j 2π k
=
N FFT
∑
k =0
X 2k e N FFT
(B.1)
Since ej2πk is always 1 for every integer value of k, then if zeros are inserted on the component
of xk for every odd values of k, xn = x N FFT .
+n
2
191
The following equation depicts if zeros are inserted on the even index of subcarriers.
NFFT −2 NFFT −2
NFFT −1 j 2π kn j 2π (2 k ) n j 2π ((2k )−1) n
1 1 2
1 2
xn =
NFFT
∑
k =0
Xke NFFT
= ∑
NFFT k =0
X 2k e + ∑
NFFT k =0
NFFT
X (2k )−1e NFFT

0 odd index of subcarriers,

even index of subcarriers,
where values of the component X2k are zeros
NFFT −2
j 2π ((2k )−1) n
1 2
=
NFFT
∑k =0
X (2k )−1e NFFT
⎛N ⎞ NFFT −2 ⎛N ⎞
j 2π k ⎜ FFT +n ⎟ j 2π (2k )⎜ FFT +n ⎟
NFFT −2 ⎝ 2 ⎠ ⎝ 2 ⎠
1 1 2
xNFFT =
+n NFFT
∑
k =0
Xke NFFT
= ∑ X 2k e
NFFT k =0
+ NFFT
2
0
even index of subcarriers,
where values of the component X2k are zeros
NFFT −2 ⎛N ⎞
j 2π ((2k )−1)⎜ FFT +n ⎟
⎝ 2 ⎠
1 2
∑ X(2k )−1e
NFFT k =0
NFFT
odd index of subcarriers
NFFT −2 ⎛N ⎞ NFFT −2
j 2π ((2k )−1)⎜ FFT +n ⎟ j 2π ((2k )−1) n
⎝ 2 ⎠ jπ ((2k )−1)
1 2
1 2 NFFT
=
NFFT
∑
k =0
X (2k )−1e NFFT
=
NFFT
∑
k =0
X (2k )−1e e
NFFT −2 j 2π ((2 k )−1) n

1 2 NFFT
=-
NFFT
∑
k =0
X (2k )−1e
(B.2)
Since ejπ((2k-1)) is always -1 for every integer value of k, then if zeros are inserted on the
component of xk for every even values of k, x n = − x N FFT .
+n
2
192
Appendix C
Single User CCSK Detection in TDCS with

an Embedded Symbol
The CCSK detection by cross correlating the received signal after fading removal with the
reference basis signal according to (5.5) is as follows:
N FFT − 1 1 ⎡ N FFT − 1 ⎛ − 2 πτ r
⎞ j N2 πir ⎤
Ri ≈ ∑ ⎢ ∑ ⎜ A ' r Se e
2 j ϕ
j
N FFT
+ A' r* N r ⎟e FFT ⎥
n =0
2
( N FFT ) ⎢ r = 0 ⎝ ⎜ H r ⎟⎠ ⎥⎦
⎣
N FFT − 1 ⎛ ⎞ j N2 πir
− 2 πτ r
1 j
N
∑ ⎜⎜ A 'r Se e ⎟e FFT
2 ϕ
= j N FFT
+ A' r
* r
( N FFT ) r = 0 ⎝ H r ⎟⎠
⎛ ⎞
⎜ ⎟
⎜ −2πτ r
⎟ 2 π ( ir )
N F F T −1 ⎛ ⎞
+ N r ⎟ A '* r
1 ⎜ j ⎟ j N FFT
≈ ∑ jϕ
⎜ ⎜⎜ A ' r S e e
N FFT
⎟e (C.1)
( N FFT ) r=0 ⎜ ⎝
H
r ⎟⎠ ⎟
⎜ Yr ⎟
⎜
⎟
⎝ U r ⎠
2 π ( i −τ ) r 2 π ( ir )
1 ⎡ jϕ N FFT −1 2 j N
N FFT −1
N j ⎤
≈ ⎢ Se ∑ A 'r e FFT + ∑ A ' r e N FFT ⎥
* r
( N FFT ) ⎢⎣ r =0 r =0 Hr ⎥⎦
⎡ ⎤
1 ⎢ jϕ N FFT −1 j N FFT ⎥
2 π ( i −τ ) r 2 π ir
* N r
N FFT −1 j
= ⎢ Se ∑ e + ∑ A' r e N FFT
⎥
( N FFT ) ⎢
r =0

r =0 H r
⎥
⎥⎦
(C.2)
By observing (C.1) we can come to the vector of cross-correlation value:
R = IFFT ⎡⎣ FFT ( y ).FFT * (c ) ⎤⎦ (C.3)

193
Appendix D
The Power to Mean Sidelobe Power Ratio

(PMR) of TDCS basis function
In conventional TDCS-CCSK if we assume the transmitted CCSK data is 0, and by observing
(C.2) the peak of the correlation detection using the reference basis function will occur when
the received signal is cross-correlated with the unshifted reference basis function (0 shift).
The peak term in this case is defined as :
⎡ ⎤
⎢ N FFT −1 j 2 π 0 r 2π 0 r ⎥
⎡ jϕ N FFT −1 * ⎤
⎢ Se jϕ ∑ e N FFT + ∑ A '* r N r e N FFT ⎥ = N
N FFT −1
1 j 1
R0 = ⎢ Se + ∑ A ' r
r
⎥
N FFT ⎢ r =0

r =0
H r
⎥ N FFT ⎢⎣ r =0 H r⎥
⎦
⎥⎦
(D.1)
Accordingly based on the derivation for eq. (5.7), the peak power to mean sidelobe power
ratio (PMR) of the TDCS basis function with extra embedded symbol is derived as :
⎡ ⎛ jϕ ⎞
2
⎤
E⎢
1 N FFT −1
∑ ⎜⎜ Se + A ' r
* N r
⎟⎟ ⎥
E ⎡⎣ R02 ⎤⎦ ⎢ N FFT
⎣
r =0 ⎝ H r ⎠
⎥
⎦
PMR = =
E ⎡⎣ Ri2,i ≠ 0 ⎤⎦ ⎡ N FFT −1 ⎛ jϕ ⎞ j 2NπFFT
( ir ) 2
⎤
E⎢ ∑
1 N FFT −1
∑ ⎜⎜ Se + A ' r
* N r
⎟⎟e ⎥
⎢ i =1 N FFT
⎣
r =0 ⎝ H r ⎠
⎥
⎦
⎡⎛ 1 2
⎤ ⎡⎛ 2
⎤
E ⎢⎜
N FFT −1
∑
⎞
( Se jϕ ) ⎟ ⎥ + E ⎢⎜
1 N FFT −1
∑
⎛ *
⎜⎜ A ' r
N r ⎟⎞ ⎟⎞ ⎥
⎢⎣⎝ N FFT r =0 ⎠ ⎥⎦ ⎢⎝⎜ N FFT
⎣
r =0 ⎝ H r ⎠⎟ ⎠⎟ ⎥
⎦
=
⎡⎛ ⎞ ⎤
2
⎢⎜ N FFT −1 1 N FFT −1 2 π ( ir )
⎟ ⎥ ⎡⎛ N FFT −1 2π ( ir )
⎞
2
⎤
E ⎢⎜ ∑ ∑ jϕ
Se e
j
N FFT
⎟ ⎥ + E ⎢⎜⎜ ∑
1 N FFT −1
∑ *
A' r N r
e
j
N FFT
⎟ ⎥
⎟
⎢⎝
i =1 N FFT
⎢⎜
r =0 ⎟ ⎥
⎥
⎢⎝ i =1 N FFT
⎣
r =0 H r ⎠ ⎥
⎦
⎣ =0 ⎠ ⎦
(D.2)
194
N
A 'r = Ar e jθk with A ∈ {0,1} . |A’r| will then be either 0 or 1. It is assumed that H has a
r
r
Gaussian distribution with variance σN/H2. Hence (D.2) becomes [103]:
⎡⎛ 1 ⎞
2
⎤
2
S e j 2ϕ
+ E ⎢⎜
N FFT −1
∑ *
A' N r
⎟⎟ ⎥
⎢⎜⎝ N FFT
r
⎣ r =0 H r⎠ ⎥
⎦
PMR =
⎡ NFFT −1 ⎛ 2π ( ir )
⎞
2
⎤
E⎢ ∑ ⎜
1 N FFT −1
∑ *
A' r N r
e
j
N FFT
⎟ ⎥
⎢ i =1 ⎜⎝ N FFT r =0 H ⎟
⎠ ⎥ (D.3)
⎣ r
⎦
1
S 2e j 2ϕ + N FFT σ N2 / H
N FFT S 2e j 2ϕ + σ N2 / H
≈ =
1
N FFT σ N2 / H σ N2 / H
N FFT
195
Appendix E
Multi User CCSK Detection in TDCS with

an Embedded Symbol
In a multi-user environment CCSK detection, by cross correlating the received signal after
fading removal with the reference basis function at user 1 according to eq. (5.13), we derive
N FFT −1 ⎡NFFT −1⎛ −2πτ1r u −2πτu r 1

⎞ j N2πir ⎤
Ri,1 ≈ ∑
1
⎢ ∑ ⎜ A'r,1 S1e e
2 jϕ1
j
NFFT
U
* Hr
+ ∑ A'r,u A' r,1 1 Sue e
jϕu
j
NFFT N
+ A' r,1 1 ⎟e FFT ⎥
* r
(NFFT )2 ⎢ r =0 ⎜ Hr H r ⎟⎠ ⎥⎦
n=0
⎣ ⎝ u=2
1 NFFT −1⎛ ⎞ j N2πir

−2πτ1r u −2πτu r 1
≈ ∑ ⎜ A '
2
S e jϕ1
e
j
NFFT
+ ∑
U
A ' A '* Hr
S e jϕu
e
j
NFFT
+ A '* N r
⎟e FFT
⎜
H r ⎟⎠
r ,1 1 r ,u r ,1 1 u r ,1 1
(NFFT ) r=0
⎝ u=2 Hr
1 N FFT −1 ⎛ ⎛ ⎞ * ⎞ j 2Nπ ( ir )
−2πτ 1r u −2 πτ u r u
≈ ∑ ⎜ A 'r ,1 S1e e
⎜ jϕ1
j
N FFT
U
+ ∑ A 'r , u H r
Se e jϕu
j
N FFT
+ H r
⎟ A ' r ,1 ⎟e FFT
( N FFT ) r = 0 ⎜ ⎜⎝ Hr
1 u
H r ⎟⎠
1
⎟
⎝ u =2
⎠
(E.1)
⎡ jϕ NFFT −1 2π ( i −τ1 ) r u 2π ( i −τ u ) r
⎤
⎢ S1e ∑ A 'r ,1 e
1
2 j N
FFT
U N FFT −1
+ ∑ Su e ∑ A 'r ,u A ' r ,1
jϕu * H r
1
e
j
N FFT
⎥
≈
1 ⎢
⎢
r = 0 u = 2 r = 0 H r ⎥
⎥
( N FFT ) ⎢ NFFT −1 1 2π ir
N r e NFFT
j ⎥
⎢ + ∑ A ' r ,1
*
1 ⎥
⎣ r =0 Hr ⎦
⎡ jϕ NFFT −1 j 2πN(i −τ1 ) r U u 2π ( i −τ u ) r
⎤
⎢ S1e ∑ e
1 FFT
N FFT −1
+ ∑ Su e ∑ A 'r ,u A ' r ,1
jϕu * H r
1
e
j
N FFT
⎥
⎢
⎢
r = 0

u = 2 r = 0
Hr
⎥
⎥
1 signal term
multiple access interference term
≈ ⎢ ⎥
( N FFT ) ⎢ NFFT −1 1 2π ir
⎥
* Nr
j
⎢+ ∑ A ' r
N FFT
1
e ⎥
⎢ r = 0 Hr
⎥
⎢⎣ noise
term ⎥⎦
(E.2)
196
The inner bracket represents the received signal on user 1 after removal of its own fading. By
observing (E.1) we can come to the vector of cross-correlation value :
R1 = IFFT ⎡⎣ FFT (y1 ).FFT * (c1 )⎤⎦ (E.3)
where R1, y1 and c1 are the cross-correlation vector , the received signal vector, and the
reference basis signal vector for user 1 respectively.
197
Appendix F
V-BLAST Algorithm
The V-BLAST algorithm is implemented as [105] :
1. Build a Moore pseudo inverse matrix of H

G= H+= (H*H)-1H*
2. Find the row of G where its Euclidean norm is the smallest one
k= arg min j || Gj ||
and j is the column of matrix G. In TDCS system
the sum is applied for NFFT samples since one
transmitted symbol is represented by NFFT samples
3. Take the row k of G as the nulling vector w in TDCS w will be in the form of a matrix
since we want to estimate the whole NFFT samples
w = (G )k
4. Obtain the strongest transmit signal

rk = w . Y
in the TDCS system rk will be a vector of size NFFT.
5. The vector rk is the output from the V-BLAST signal processing module. It is in
frequency-domain. The vector rk will be transformed into time domain by IFFT, then
decorrelated by the reference signal to obtain the estimated symbol d2’ from the
source data 2. The vector rk will be used also to detect the transmitted data on source
data 1 by multiplying it with the inverse of the frequency-domain form of the shifted
reference signal. The shift depends on the result derived from the detection of data
from source data2. Later after averaging the signal over NFFT, symbol demapping is
proceed to have the estimated symbol d1’ on source data 1.
6. After detection of the strongest transmitted signal, its effect must be cancelled from
the received signal vector to reduce the detection complexity of the remaining transmit
signals. The symbol d2’ is mapped again into its TDCS with CCSK representation in
frequency-domain, and then multiplied with d1’. Later the results of this multiplication
are fed back to the V-BLAST signal processing module. The cancellation is applied as
Y = Y – (H)k* d1’* FFT(Mapping (d2’))

198
where k is the column index. The k-th column of channel matrix H is then zeroed for the
purpose of detection of the strongest transmitted signal on the next layer. The algorithm
returns to step 1 untill the transmitted symbols on all layers have been detected.
199
Appendix G
Future Work in Wavelet Packet
Multicarrier Modulation Based Cognitive
Radio System
The application of the tree extension on the subcarrier with good channel condition to
compensate the loss of data source due to de-activated carriers is depicted in Figure G.1. In
the figure an example of de-activated carrier is signed by the cross, and implemented by
allocating zero symbol to the carrier.
At this stage the design of the wavelet basis functions used in this thesis cannot accommodate
the notching part of its spectrum on a LU band by itself. Notching can only be applied by de-
activating carriers located at the LU band.
H
2↑ H’
G 2↑ H’
2↑ G’
2↑ H’
2↑ H’
2↑ G’
2↑ G’
0 2↑ H’
2↑ H'
2↑ G’
2↑ G’
2↑ H’
2↑ G‘
2↑ G’
Figure G.1: Proposed future work on WPMCM to compensate the loss of data source due to
de-activated carriers by extending the tree on a subcarrier with good channel condition
200
201
Bibliography
[1] Joseph Mitola III, Cognitive Radio An Integrated Agent Architecture for Software
Defined Radio, PhD thesis, Dept. of Teleinformatics, KTH Royal Institute of
Technology, Stockholm, Sweden, 2000.
[2] S. Haykin, ”Cognitive Radio: Brain Empowered Wireless Communications”, in IEEE

Journal on Selected Areas In Communications, vol. 23, no.2, pp. 201-220, Feb. 2005.
[3] B. Fette, ”Technical Challenges and Opportunities”, presented at the Conference

Cognitive Radio, Las Vegas, NV, Mar. 2004.
[4] M. Dillinger, K. Madani and N. Alonistioti, Eds., Software Defined Radio :

Architectures, Systems and Functions, Wiley, New York, 2003.
[5] M. A. Fischler and O. Firschein, Intelligence: The Brain, and the Computer, ser. MA.
Reading: Addison-Wesley, pp. 81, 1987.
[6] I. Budiarjo, M. K. Lakshmanan, and H. Nikookar, “Cognitive Radio Dynamic Access

Techniques”, in International Journal on Wireless Personal Communications-
Springer, vol. 45, no. 3, pp. 293-324, Feb. 2008.
[7] R. W. Brodersen, A. Wolisz, D. Cabric, S. M. Mishra and D. Wilkomm, “CORVUS : A

Cognitive Radio Approach for Usage of Virtual Unlicensed Spectrum, White Paper,
Dept of electrical Engineering and Computer Science University of Berkeley, July,
2004. http://bwrc.eecs.berkeley.edu/Research/ MCMA/CR_White_paper_final1.pdf.
[8] M.G. Di Benedetto and G. Giancola, Understanding Ultra Wide Band, Upper Saddle
River, NJ, Prentice Hall, 2004.
[9] V.D. Chakravarthy, A. K. Shaw, M. A. Temple, and J. P. Stephens, “Cognitive Radio –

An Adaptive Waveform with Spectral Sharing Capability”, in Proc. IEEE Wireless
Communications and Networking Conference, vol 2, pp. 724- 729, Mar. 2005.
[10] R. van Nee and R. Prasad, OFDM for Wireless Multimedia Communications, Artech
House Publishers, Boston, London, UK, 2000.
[11] T. Weiss and F. Jondral, “Spectrum Pooling - An Innovative Strategy for the
Enhancement of Spectrum Efficiency”, in IEEE Communications Magazine, vol. 42,
no. 3, pp. S8-14, March 2004.
202
[12] I. Budiarjo et al., “D3.31 Radio System : First Phase Verification and advances in
architecture and algorithms”, Adaptive Ad-hoc Freeband (AAF) wireless
communications, Netherland, Sci. Rep. P1.0.06, 2006.
[13] T. Weiss, J. Hillenbrand, A. Krohn, and F. K. Jondral, ”Mutual Interference in OFDM-

based Spectrum Pooling Systems”, in Proc. IEEE Vehicular Technology Conference,
Milan, Italy, pp. 1873- 1877, 17-19 May 2004.
[14] H. Nikookar and R. Prasad. “Wave Shaping of Multicarrier Signal for Data
Transmission Over Wireless Channels”, in Proc. 6 th IEEE International Conference
on Universal Personal Communications Record (ICUPC), pp. 173–177, 1997.
[15] P. Tan and N. C. Beaulieu. “Reduced ICI in OFDM Systems Using the Better than
Raised – Cosine Pulse”. in IEEE Communications Letter, vol. 8, no. 3, pp. 135 – 137,
March 2004.
[16] A. Assalini and A. M. Tonello,”Improved Nyquist Pulses”, in IEEE Communications

Letters, vol. 8, no.2, pp. 87-89, 2004.
[17] C. Muschalik , “Improving An OFDM Reception Using An Adaptive Nyquist

Windowing”, in IEEE Transactions on Consumer Electronics, vol. 42, no.3, pp. 259-
269, 1996.
[18] K. Witrisal, “OFDM Air Interface Design for Multimedia Communications”, Ph.D.
Thesis, Dept. Electrical Eng. Mathematics and Computer Science, Delft University of
Technology, Apr. 2002.
[19] H.Nikookar and R.Prasad," Optimal Waveform Design for Multicarrier Transmission
Over a Multipath Channel", in Proc. 47th IEEE Vehicular Technology Conference,
Phoenix, Arizona, USA, May 4-7, pp. 1812-1816, 1997.
[20] F. Harris, E. Kjeldsen, “A Novel Interpolated Tree Orthogonal Multiplexing (ITOM)

Scheme with Compact Time-Frequency Localization: an Introduction and Comparison
to Wavelet Filter Banks and Polyphase Filter Banks”, in Proc. Software Defined
RadioTechnical Conference and Product Exposition, Orlando, Florida, USA, Nov.
2006.
[21] H. Yamaguchi, “Active Interference Cancellation Technique for MB-OFDM Cognitive

Radio”, Proc. 34 th European Microwave Conference, Amsterdam, The Netherlands,
vol. 2, pp. 1105-1108, Oct. 2004.
[22] S. Brandes, I. Cosovic, and M. Schnell, “Sidelobe Suppression in OFDM Systems by

Insertion of Cancellation Carriers”, Proc. 62nd IEEE Vehicular Technology
Conference (VTC), Fall, Dallas, USA, Sept, 2005.
203
[23] S. Brandes, I. Cosovic, and M. Schnell, “Reduction of Out of Band Radiation in OFDM
Systems by Insertion of Cancellation Carriers”, in IEEE Comunications Letters, vol. 10,
no. 6, pp. 420-422, June 2006.
[24] S. Brandes, I. Cosovic, and M. Schnell, “Reduction of Out-of-Band Radiation in OFDM

Based Overlay Systems”, in Proc. First IEEE International Symposium New Frontiers
in Dynamic Spectrum Access Networks (DySPAN), Baltimore, Maryland, USA, pp.
662 – 665, Nov. 2005.
[25] U. Berthold, F. K. Jondral, S. Brandes and M. Schnell, “OFDM Based Overlay Systems:
A Promising Approach for Enhancing Spectral Efficiency” in IEEE Communications
Magazine, vol. 45, no. 12, pp. 52-58, Dec. 2007.
[26] I. Cosovic, S. Brandes, and M. Schnell, “A Technique for Sidelobe Suppression in

OFDM System”, in Proc. IEEE GLOBECOM, vol. 1, pp. 204-208, Nov. 2005.
[27] I. Cosovic, S. Brandes, and M. Schnell, “Subcarrier Weighting : A Method for Sidelobe
Suppression in OFDM Systems”, in IEEE Communications Letters, vol. 10, no. 6, pp.
444-446, June 2006.
[28] H.Nikookar and R.Prasad,” Weighted OFDM for Wireless Multipath Channels”, in
IEICE Transactions on Communications, vol. E83-B, no. 8, pp. 1864-1872, Aug. 2000.
[29] I. Cosovic and T. Mazzoni, ”Suppression of Sidelobes in OFDM systems by multiple

choice sequences”, in European Transactions on Telecommunications Special Issue on
Multi carrier – Spread Spectrum Wiley InterScience, pp. 623-630, June 2006.
[30] H. A. Mahmoud and H. Arslan, “Sidelobe Suppression in OFDM –Based Spectrum

Sharing Systems by Using Adaptive Symbol Transition”, in IEEE Communications
Letters, vol. 12, no. 2, pp. 133-135, Feb. 2008.
[31] I. Cosovic and T. Mazzoni, “Sidelobe Suppression in OFDM Spectrum Sharing Sys-
tems via Additive Signal Method”, in Proc. 65th IEEE Vehicular Technology
Conference (VTC), Dublin, Ireland, pp. 2692-2696, Apr. 2007.
[32] S. Pagadarai, R. Rajbanshi, A. M. Wyglinski, and G. J. Minden, “Sidelobe Suppression

for OFDM-Based Cognitive Radios Using Constellation Expansion”, in Proc. IEEE
Wireless Communicaitons and Networking Conference (WCNC), Las Vegas, Nevada,
USA, pp. 888-893, Apr. 2008.
[33] S. Pagadarai, A. M. Wyglinski, R. Rajbanshi, ”A Novel Sidelobe Suppression

Technique for OFDM-Based Cognitive Radio Transmission”, in Proc. IEEE Symposia
204
on New Frontiers Dynamic Spectrum Access Networks (DySPAN), Chicago, USA, Oct.
2008.
[34] F.W. Hoeksema, M. Heskamp, R. Schiphorst, and C.H. Slump. A Node Architecture for
Disaster Relief Networking, in Proc. 1st IEEE Int. Symposium on new Frontiers in
Dynamic Spectrum Access Networks (DySPAN’05), pp. 577-584, Nov 2005.
[35] I. Budiarjo, H. Nikookar, and L.P. Ligthart, “Performance Evaluation of OFDM Based
Cognitive Radio System With Wiener Filter Channel Estimation Using Frequency
Hopping GSM Channel Model at 900 MHz”, in Proc. 10th European Conference on
Wireless Technology (ECWT), Munich, Germany, pp. 74-77, Oct. 8-12, 2007.
[36] ETSI. Digital cellular telecommunications system (phase 2+). Technical Specification
ETS 300 910 GSM 05.05 version5.3.1, 1997.
[37] T. Keller and L. Hanzo, “Adaptive Multicarrier Modulation: A Convenient Framework

for Time-Frequency Processing in Wireless Communications”, in Proceeding of the
IEEE, vol . 88, no. 5, May 2000, pp 611–640, May 2000.
[38] L. Hanzo, C. H. Wong, and M. S. Yee, Adaptive Wireless Transceivers, John Wiley and
Sons Ltd, Baffins, Chichester, 2002.
[39] T. Keller and L. Hanzo, “Blind-Detection Assisted Sub-band Adaptive Turbo-Coded

OFDM Scemes”, in Proc. IEEE Vehicular Technology Conference (VTC), vol. 1,
Houston, TX, pp. 489-493, July 1999.
[40] S. B. Reddy, T. Yucek, and H. Arslan, “An Efficient Blind Modulation Detection
Algorithm for Adaptive OFDM Systems”, in Proc. IEEE Vehicular Technology
Conference (VTC), vol. 3, Orlando, USA, Oct. 2003,.
[41] P. S. Chow, J. M. Cioffi, and J. A. C Bingham, “A Practical Discrete Multitone

Transveiver Loading Algorithm for Data Transmission over Spectrally Shaped
Channels”, in IEEE Transactions On Communications, no. 2, vol. 43, pp. 773–775,
Feb. 1995.
[42] R. F. H. Fischer and J. B. Huber, “A New Loading Algorithm For Discrete Multitone
Transmission”, in Proc. IEEE GLOBECOM ’96, pp. 724–728, Nov. 1996.
[43] A. N. Barreto and S. Furrer, “Adaptive Bit Loading for Wireless OFDM Systems”, in
Proc. IEEE Int. Sym. Personal, Indoor, Mobile Radio Commun., vol. 2, pp. 88–92,
2001.
205
[44] R. Gruenheid, E. Bolinth, and H. Rohling, “A Blockwise Loading Algorithm for The
Adaptive Modulation Technique in OFDM Systems”, in Proc. Veh. Technol. conf., vol.
2, pp. 948–951, Fall 2001.
[45] R. Gruenheid, E. Bolinth, H. Rohling, and K. Aretz, “Adaptive Modulation for the
HIPERLAN/2 Air Interface”, in Proc. 5th Int. OFDM Workshop Hamburg,Germany,
2000.
[46] M. Lei and P. Zhang,”Subband Bit and Power Loading for Adaptive OFDM”, in 58th
IEEE Vehicular Technology Conference, vol. 3, pp. 1482-86, Oct. 200.
[47] I. Budiarjo, H. Nikookar and L. P. Ligthart, “Performance Evaluation of Adaptive Bit

Loading for OFDM With Channel Estimation Using 2x1-D Wiener Filter”, Proc. 9th
International Symposium on Wireless Personal Multimedia Communications (WPMC),
San Diego, USA, Sep. 2006.
[48] I. Budiarjo, H. Nikookar and L. P. Ligthart, “Combined Spectrum Pooling and Adaptive
Bit Loading for Cognitive Radio OFDM Based System”, Proc. 13th IEEE Symposium
on Communications and Vehicular Technology (SCVT), Liege, Belgium, Nov. 2006.
[49] R. Hassun, M. Flaherty, R. Matreci and M. Taylor, “Effective Evaluation of Link

Quality Using Error Vector magnitude Techniques”, Proc. Wireless Communications
Conf., Boulder, CO, USA, pp. 89-94, Aug. 1997.
[50] A. Jamin and P. Mahonen, “Wavelet Packet Modulation for Wireless Communica-
tions”, in Wireless Communications & Mobile Computing Journal, vol. 5, Issue 2, pp.
123-137, John Wiley and Sons Ltd, March 2005.
[51] B. G. Negash and H. Nikookar, “Wavelet-based Multicarrier Transmission Over Multi-

path Wireless Channels”, in IEE Electronics Letters, Vol. 36 No. 21, pp. 1787–1788,
October 2000.
[52] C.V. Bouwel, J. Potemans, S. Schepers, B. Nauwelaers and A.V. Capelle, "Wavelet
Packet Based Multicarrier Modulation", in Proc. IEEE Benelux Symposium on
Communications and Vehicular Technology, Leuven, Belgium, October 2000.
[53] P. P. Vaidyanathan, Multirate Systems and Filter Banks, Englewood Cliffs, NJ,
Prentice-Hall, Inc., 1993.
[54] I. Daubechies, Ten Lectures on Wavelets, SIAM, Philadelphia, 1992.
[55] M. Vetterli and I. Kovacevic, Wavelets and Subband Coding, Prentice Hall PTR,
Englewood Cliffs, New Jersey, 1995.
206
[56] S. Mallat, “A Theory for Multiresolution Signal Decomposition: The Wavelet

Representation”, in IEEE Trans. Pattern Anal. Machine Intell., vol. 11, pp. 674-693,
July 1989.
[57] C. S. Burrus, R. A. Gopinath, H. Guo, Introduction to Wavelets and Wavelet

Transforms, a Primer, Upper Saddle River, NJ (USA): Prentice Hall, 1998.
[58] B.Burke, The World According to Wavelets: The Story of a Mathematical Technique in
the Making, Second Edition, A K Peters, May 1998.
[59] O. Rioul and P. Duhamel, "A Remez Exchange Algorithm for Orthonormal Wavelets",
in IEEE Trans. Circuits Systems - II, vol. 41, no. 8, pp. 550-560, August 1994.
[60] O. Rioul and P. Duhamel, "Note on ‘A Remez Exchange Algorithm for Orthonormal
Wavelets’," Authors note available on the internet at
http://www.comelec.enst.fr/~rioul/publis/199302rioulduhamelnote.pdf.
[61] A. V. Oppenheim and R. W. Schafer, Discrete-Time Signal Processing, Englewood

Cliffs, NJ, Prentice Hall, 1989.
[62] A. Jensen and A.L. Cour-Harbo, Ripples in Mathematics: The Discrete Wavelet
Transform, Springer, 2001.
[63] M. K. Lakshmanan, I. Budiarjo, and H. Nikookar, “Maximally Frequency Selective

Wavelet Packets Based Multi Carrier Modulation Scheme for Cognitive Radio
Systems”, in Proc. IEEE GLOBECOM, Washington D.C., USA, pp. 4185-4189, Nov.
2007.
[64] E.R. de Lima, S.J. Flores, V. Almenar, M.J. Canet, “Performance Enhancements in
OFDM-WLAN Systems Using MIMO Access Techniques”, in Proc. International
Workshop on Telecommunications, Santa Rita do Sapucai, MG, Brazil, August, 2004.
[65] P.W. Wolniansky, G.J. Foschini, G. D. Golden, and R. A. Valenzuela, “V-BLAST: An

Architecture for Realizing Very High Data Rates Over the Rich Scattering Wireless
Channel”, in Proc. International Symposium on Signals Systems and Electronics
(ISSE), Pisa, Italy, pp 295-300, Oct 1998.
[66] M. K. Lakshmanan, I. Budiarjo, and H. Nikookar, “Wavelet Packet Modulation MIMO

Based Cognitive Radio System with V-BLAST Receiver Architecture”, in Proc. IEEE
Wireless Communications and Networking Conference, Las Vegas, USA, pp. 705-710,
March 2008.
207
[67] H. Che, Adaptive OFDM and CDMA Algorithms for SISO and MIMO Channels, Ph.D.
Thesis, Dept. Electrical Eng. Mathematics and Computer Science, Delft University of
Technology, May. 2005.
[68] B. Muquet, Z. Wang, G. B. Giannakis, M. de Courville, and P. Duhamel, “Cyclic

Prefixing or Zero Padding for Wireless Multicarrier Transmissions?”, in IEEE
Transactions on Communications, vol. 50, no. 12, pp. 2136-2148, Dec. 2002.
[69] N. Wang and S. D. Blostein, “Adaptive Zero-Padding OFDM over Frequency-Selective

Multipath Channels”, in EURASIP Journal on Applied Signal Processing, no. 10, pp.
1478-1488, 2004.
[70] N. Chotikakamthorn and H. Suzuki, “On Identifiability of OFDM Blind Channel

Estimation”, in Proc. 50 th IEEE Vehicular Technology Conference (VTC),
Amsterdam, The Netherlands, vol. 4, pp. 2358-2361, Sept. 1999.
[71] K. Amleh, H. Li and R. Wang, “Blind Channel Estimation, Equalisation and CRB for
OFDM with Unmodelled Interference”, in IET Communications , vol. 1, no. 3, pp. 489-
495, June 2007.
[72] M. C. Necker and G. L. Stuber, “ Totally Blind Channel Estimation for OFDM over
Fast Varying Mobile Channels”, in Proc. IEEE International Conference on
Communications (ICC), vol. 1, New York, USA, pp. 421-425, Apr. 2002.
[73] P. Hoeher, S. Kaiser and P. Robertson, “Pilot-symbol aided channel estimation in time
and frequency,” in Proceedings IEEE GLOBECOM, Phoenix, USA, pp. 90–96, Apr.
1997.
[74] P. Hoeher, S. Kaiser and P. Robertson, “Two-Dimensional Pilot Symbol Aided Chan-
nel Estimation by Wiener Filtering”, in Proc. IEEE International Conference on
Accoustics, Speech, and Signal Processing (ICASSP), vol. 3, Munich, Germany, pp.
1845-1848, Apr. 1997.
[75] S. Kaiser, Multi-Carrier CDMA Mobile Radio Systems – Analysis and Optimization of
Detection, Decoding and Channel Estimation, VDI Verlag Gmbh, Dusseldorf,
Germany, 1998.
[76] I. Budiarjo, OFDM Channel Estimation by Adaptive Filtering, MSc Thesis Report,
Munich University of Technology, 2003.
[77] P. Hoeher, “A Statistical Discrete-Time Model for The WSSUS Multipath Channel”,
IEEE Transactions on Vehicular Technology, vol.41, No.4, pp. 461-468, November,
1992.
208
[78] D.E. Dudgeon, and R.M. Mersereau, Multidimensional Signal Processing, Prentice-
Hall, 1983.
[79] M. J. Fernandez - Gettino Garcia, J.M Paez-Borallo and S. Zazo, “ Efficient Pilot
Patterns For Channel Estimation in OFDM Systems Over HF Channels,” in IEEE
Vehicular Technology Conference Fall 99, Amsterdam, The Netherlands, September
1999.
[80] M. J. Fernandez - Gettino Garcia, J.M Paez-Borallo and S. Zazo, “Pilot Patterns For
Channel Estimation in OFDM”, in Electronic Letters, vol. 36, no. 12, June 2000.
[81] J. L. Choi and Y. H. Lee, “Optimum Pilot Pattern for Channel Estimation in OFDM
Systems”, in IEEE Transactions on Wireless Communications, vol. 4, no. 5, Sept.
2005.
[82] J. Zhu and W. Lee, "Channel Estimation with Power-Controlled Pilot Symbols and
Decision Directed Reference Symbols", in Proc. IEEE 58th Vehicular Technology
Conference (VTC), vol. 2, pp. 1268-1272, Oct. 2003.
[83] I. Rashad, I. Budiarjo, and H. Nikookar, “Efficient Pilot Pattern for OFDM based
Cognitive Radio Channel Estimation – Part 1”, in Proc. 14 th IEEE Symposium on
Communications and Vehicular Technology (SCVT), Delft, The Netherlands, Nov.
2007.
[84] I. Budiarjo, I. Rashad, and H. Nikookar, “Efficient Pilot Pattern for OFDM based
Cognitive Radio Channel Estimation – Part 2”, in Proc. 14 th IEEE Symposium on
Communications and Vehicular Technology (SCVT), Delft, The Netherlands, Nov.
2007.
[85] S. Sand, A. Dammann, and G. Auer, “Adaptive Pilot Symbol Aided Channel Estima-
tion for OFDM Systems”, in Proc. 4 th International Workshop on Multi Carrier
Spread Spectrum & Related Topics (MC-SS 2003), Oberpfaffenhofen, Germany, pp.
227-234, Sept. 2003.
[86] G. Auer, “Modelling of OFDM Channel Estimation Errors”, in Proc. 10 th Inter-

national OFDM Workshop , Hamburg, Germany, Aug. 2005.
[87] I. Budiarjo, I. Rashad, and H. Nikookar, “On the Use of Virtual Pilots with Decision
Directed Method in OFDM Based Cognitive Radio Channel Estimation Using 2x1-D
Wiener Filter ”, in Proc. IEEE International Conference on Communications (ICC),
Delft, The Netherlands, pp. 703-707, Nov. 2007.
[88] I. Budiarjo, H. A. Sulaiman, and H. Nikookar, “On The Use of Virtual Pilots for 2x2
MIMO V-BLAST OFDM Channel Estimation with Wiener Filter in Hexagonal Pilot
209
Pattern”, in Proc. IEEE 15th Symposium on Communications and Vehicular

Technology (SCVT), Antwerp, Belgium, Nov. 2008.
[89] M. Sandell, J. J. V. D. Beek, and P. O. Boerjesson, “Timing and Frequency

Synchronization in OFDM Systems Using the Cyclic Prefix”, in Proc. International
Symposium on Synchronization, Essen, Germany, pp. 16-19, Dec. 1995.
[90] J. J. V. D. Beek, M. Sandell, and P. O. Boerjesson, “ML Estimation of Time and

Frequency Offset in OFDM Systems”, in IEEE Transactions on Signal Processing,
vol. 45, pp. 1800-1805, July 1997.
[91] B. Y. Prasetyo, F. Said, A. H. Aghvami, “On The Guard Band-Based Coarse Frequency
Offset Estimation Technique for Burst OFDM Systems”, in Proc. IEEE 51st Vehicular
Technology Conference, Tokyo, Japan, pp. 220-224, May 2000.
[92] T. Lu, J. Chen and H. Li, “Low –Complexity Blind Symbol Timing Offset Estimation
in OFDM Systems”, in EURASIP Journal on Applied Signal Processing, pp. 532-540,
2005.
[93] J. J. van de Beek et. al.,”Low Complex Frame Synchronization in OFDM Systems”, in
Proc. International Conference on Universal Personal Communications (ICUCC),
Tokyo, Japan, pp. 982-986, Nov. 1995.
[94] T. Weiss, A. Krohn, and F. Jondral, “Synchronization Algorithms and Preamble Con-
cepts in Spectrum Pooling Systems”, in Proc. IST Mobile and Wireless Telecommu-
nications Summit, June 2003.
[95] M. Sliskovic, “Carrier and Sampling Frequency Offset Estimation and Correction in
Multicarrier Systems”, in Proc. IEEE GLOBECOM, San Antonio, Texas, USA, pp.
285-289, Nov. 2001.
[96] K. B. Png et. al., “Two-Dimensional Iterative Sampling Frequency Offset Estimation
for MB-OFDM System”, in Proc. IEEE 63rd VTC, Melbourne, Australia, pp. 1344-
1348, May 2006.
[97] M. J. Fernandez-Getino Garcia, O. Edfors, and J. M. Paez-Borallo, “Peak Power Re-

duction for OFDM Systems with Orthogonal Pilot Sequences”, in IEEE Transactions
on Wireless Communications, vol. 5, no. 1, pp. 47-51, Jan. 2006.
[98] R. Rajbanshi, A. M. Wyglinski, and G. J. Minden, “Peak to Average Power Ratio Ana-
lysis for NC-OFDM Transmissions”, in Proc. IEEE 66 th Vehicular Technology
Conference, Baltimore, USA, pp. 1351-1355, Oct. 2007.
210
[99] V. Chakravarthy, A. S. Nunez, J. P. Stephens, A. K. Shaw and M. A. Temple, “TDCS,

OFDM, and MC-CDMA: A Brief Tutorial”, in IEEE Communications. Magazine, vol
43, no. 9, pp S11 –S16, Sept. 2005.
[100] M. J. Lee, Wavelet Domain Communication System (WDCS): Packet-Based Wavelet

Spectral Estimation And M-ary Signaling, Master thesis, Air Force Institute of
Technology, Ohio, USA, Mar. 2002.
[101] M. J. Lee, M. A. Temple, R. L. Claypoole Jr., R. A. Raines, J. P. Stephens, “Wavelet

Domain Communication System: Bit Error Sensitivity Characterization for
Geographically Separated Transceivers”, in Proc. MILCOM, Anaheim, CA, vol. 2, pp.
1378–82, Oct. 2002.
[102] M. J. Lee, “Wavelet Domain Communication System (WDCS): Packet- Based Wavelet
Spectral Estimation and M-ARY Signaling”, Master thesis, AFIT/GE/ENG/02M-14,
Graduate School of Engineering and Management Air Force Institute of Technology
Wright – Patterson Air Force Base, Ohio, USA, Mar. 2001, DTIC:ADA401433,
approved for public release.
[103] G.M. Dillard, M. Reuter, J. Zeidler and B. Zeidler, “Cyclic Code Shift Keying: A Low
Probability of Intercept Communication Technique”, in IEEE Trans. on Aerospace and
Electronic Systems, vol.39, no.3, pp. 786-798, July 2003.
[104] A. S. Nunez, Interference Suppression in Multiple Access Communications Using M-

Ary Phase Shift Keying Generated Via Spectral Encoding, M.Sc Thesis Report, Air
Force Institute of Technology, Ohio, USA, 2004.
[105] I. Budiarjo, H. Nikookar, and L.P. Ligthart,”Cognitive Radio with Single Carrier TDCS
and Multicarrier OFDM Approach with V-BLAST Receiver in Rayleigh Fading
Channel”, in Proc. 2nd International Conference on Cognitive Radio Oriented Wireless
Communications (CrownComm), Orlando,USA, Aug. 2007.
[106] J. G. Proakis and M. Salehi, Communication Systems Engineering, Upper Saddle

River, NJ, Prentice Hall, 2002.
[107] M. v. d. Oever, A Cognitive Radio Demonstrator, Thesis Report, Technische

Hochschule Rijswijk, Rijswijk, The Netherlands , June, 2007.
[108] S. B. Raghunathan, M. v. d. Oever, R. D. Mohammady, P. Pawełczak, I. Budiarjo, M.

Heskamp, Q. Zhang, A. Kokkeler, H. Nikookar, Z. Qin, R. Hekmat, and L. P. Ligthart,
“Dynamic Spectrum Access AAF Platform”, in IEEE Symposia on New Frontiers in
Dynamic Spectrum Access Networks (DySPAN), Oct 14-17, 2008, Chicago, Illinois,
USA.
211
[109] I. Budiarjo, M. Heskamp, Q. Zhang, and F.W. Hoeksema, ”D.3.31Radio System: First
Phase Verification and Advances in Architecture and Algorithms”, Adaptive Ad-hoc
Freeband (AAF), August, 2006, The Netherlands.
212
List of Abbreviations
AAF Adaptive Adhoc Freeband
ADC Analog to Digital Converter
AWGN Additive White Gaussian Noise
BER Bit Error Rate
BPSK Binary Phase Shift Keying Modulation
BTRC Better Than Raised Cosine
CCDF Complementary Cumulative Distribution Function
CCSK Cyclic Code Shift Keying
CIR Channel Impulse Response
CSI Channel State Information
CTF Channel Transfer Function
CR Cognitive Radio
DAC Digital to Analog Converter
ETSI European Telecommunications Standard Institute
FDD Frequency Division Duplex
FFT Fast Fourier Transform
FoM Figure of Merit
GSM Global System for Mobile Communications
ICI Inter Carrier Interference
IDWPT Inverse Discrete Wavelet Packet Transform
IFFT Inverse Fast Fourier Transform
ISI Inter Symbol Interference
LS Licensed System
LU Licensed User
MCM Multicarrier Modulation
MCS Multiple Choice Sequence
MIMO Multiple Input Multiple Output
MMSE Minimum Mean Square Error
MSE Mean Square Error
OFDM Orthogonal Frequency Division Multiplexing
PAM Pulse Amplitude Modulation
PAPR Peak to Average Power Ratio
PDF Probability Density Function
PMR Peak Power to Mean Sidelobe
PSD Power Spectral Density
QAM Quadrature Amplitude Modulation
QoS Quality of Service
QPSK Quadrature Phase Shift Keying
QMF Quadrature Mirror Filter
RDS Root mean square Delay Spread
213
rms Root mean square

RS Rental System
RU Rental User
SBLA Simple Blockwise Loading Algorithm
SISO Single Input Single Output
SINR Signal to Interference plus Noise Ratio
SNR Signal to Noise Ratio
TDCS Transform Domain Communication System
TDD Time Division Duplex
V-BLAST Vertical-Bell Laboratories -Layered -Space-Time
WDCS Wavelet Domain Communication System
WP-3 Work Package 3
WPMCM Wavelet Packet Multicarrier Modulation
WPT Wavelet Packet Transform
WSSUS Wide Sense Stationary Uncorrelated Scattering
ZF Zero Forcing
214
List of Symbols
α Window roll off factor
ε n,k Channel estimate mismatch error on subcarrier n of OFDM symbol
k
γmargin System performance margin (in dB) that defines the maximum
allowable noise quantity on the system
Γ SNR gap (in dB) which is a constant which estimates the difference
between channel capacity and the actual capacity usage by the
transmission scheme
φm[k] Complex basis wavelet function
П(l) Permutation matrix that will do the interleaving
σI2 Interference variance
σn2 Noise variance
τfilter Estimated maximum delay of the channel for filter design
τmax Maximum delay of the channel
θn Random phase on subband n of TDCS
θn−n'',k −k '' Channel time frequency correlation function
θ Random phase vector
φn'−n'',k '−k '' Autocorrelation function on pilot positions
ω Wiener filter coefficients matrix
ccN Total number of sidelobes cancellation carrier
df Distance between pilots in frequency direction
dt Distance between pilot in time direction
em Decision directed error energy between the ideal constellation
phasor positions and the received phasors
fD Doppler frequency
fDfilter Estimated Doppler frequency for filter design
fn Frequency on carrier n
fLU Frequency in the range of LU band,
FLU Fourier transformation matrix for the signal in LU band
g(t) Window function
Hn,l Actual channel state on subcarrier n of OFDM symbol l

H n ,k Final estimated value of the frequency-domain channel transfer
function (CTF) at subcarrier n of OFDM symbol k
Ĥs,n Estimated value of signaling subcarriers
In,l Interference term on subcarrier n of OFDM symbol l
I Signal part of the cognitive radio signal in LU band
Iˆ n ,l
Estimated interference on subcarrier n of OFDM symbol l
215
NFFT Number of FFT points which refers to the possible number of total
OFDM/TDCS subcarriers,
Nn,l Noise term on subcarrier n of OFDM symbol l
Nˆ n ,l Estimated noise on subcarrier n of OFDM symbol l
Pbudget Total allocated power per OFDM symbol
pn Allocated power on carrier n
ps(γ) Probability of a signaling error given SNR γ
PThreshold Limit of interference power that is still tolerable by the LUs
Q(x) Area covered by the normal distribution function from x to infinity
Rn Number of allocated bits in subcarrier n
Rs,n Received symbols in the signaling subcarriers
Rtarget Target total bits per OFDM symbol
TGI OFDM guard interval duration
TS OFDM symbol duration including the guard interval
Tu Useful OFDM symbol duration
Xn Modulated data on subcarrier n of OFDM symbol l
XPn,l Modulated pilot on subcarrier n of OFDM symbol l
X Symbol vector
ym The mth time domain OFDM symbol including the guard interval
216
Publications by the Author

Book Chapters
1. I. Budiarjo, M. K. Lakshmanan, and H. Nikookar, "Cognitive Radio Dynamic

Access Techniques for Mutual Interference Reduction and Efficient Spectrum
Utilization”, submitted for a chapter in book "Cognitive Radio Systems", IN-
TECH, ISBN 978-953-7619-25-1.
2. M. K. Lakshmanan, I. Budiarjo, and H. Nikookar, “Robust Bases for Spectrum

Pooling Cognitive Radio Systems on Wavelet Packet Multi-carrier Modulation
MIMO Architecture”, submitted for a chapter in book "Cognitive Radio
Systems", IN-TECH, ISBN 978-953-7619- 25-1.
Journal Papers
3. I. Budiarjo, M.K. Lakshmanan, and H. Nikookar, “Cognitive Radio Dynamic

Access Techniques”, in Springer International Journal on Wireless Personal
Communications special issue on Cognitive Radio Technologies, vol. 45, no. 3,
pp. 293-324, Feb. 2008.
4. I. Budiarjo, H. Nikookar and L.P. Ligthart, “Cognitive Radio with Single Carrier
TDCS and Multicarrier OFDM Approach with V-BLAST Receiver in Rayleigh
Fading Channel”, in ACM / Springer Mobile Networks and Application
(MONET) Journal special issue on Cognitive Radio Oriented Wireless Networks
and Communications, vol. 13, no. 5, pp. 416-423, July 2008.
5. I. Budiarjo, H. Nikookar and L. P. Ligthart, “Cognitive Radio Modulation

Techniques Reducing Mutual Interference”, in IEEE Signal Processing
Magazine special issue on Signal Processing for Cognitive Radio Networks, vol.
25, issue 6, pp. 24-34, Nov. 2008.
6. I. Budiarjo, H. Nikookar and L. P. Ligthart, “Combined Spectrum Pooling,

Adaptive Bit Loading and Sidelobes Cancellation Carriers : An Advanced
Technique for OFDM-Based Cognitive Radio in Overlay Spectrum Sharing
Access”, to be submitted to IEEE Wireless Communications.
7. I. Budiarjo, H. Nikookar, and L.P. Ligthart, “MIMO TDCS with Extra Embedded
Symbol and V-BLAST Receiver Architecture for Higher Data Rate Dynamic
Spectrum Access”, submitted to IEEE Transactions on Mobile Computing.
217
Conference Papers
8. I. Budiarjo, H.Nikookar, and L.P. Ligthart,“Overview of Adaptive OFDM in the

Context of Cognitive Radio”, in Proc. 12th Annual Symposium of IEEE on
Communications and Vehicular Technology (SCVT) in the Benelux, Enschede,
The Netherlands, pp. 1-5, Nov. 3, 2005.
9. I. Budiarjo, H. Nikookar, and L.P. Ligthart,“Performance Evaluation of Adaptive

Bit Loading For OFDM With Channel Estimation Using 2x 1-D Wiener Filter”,
in Proc. 9th Annual International Symposium on Wireless Personal and
Multimedia Communications (WPMC), San Diego, USA, pp. 1138-1142, Sept.
17-20, 2006.
10. I. Budiarjo, H. Nikookar, and L.P. Ligthart, “Combined Spectrum Pooling and
Adaptive Bit Loading for Cognitive Radio OFDM Based System”, in Proc. 13th
Annual Symposium of IEEE on Communications and Vehicular Technology
(SCVT) in the Benelux, Liege, Belgium, pp. 73-76, Nov. 23, 2006.
11. I. Budiarjo, H. Nikookar, and L.P. Ligthart, “Techniques for Cognitive Radio”,
invited paper on the 2nd Center for TeleInfrastruktur (CTIF) Workshop,
Aalborg, Denmark, May 22-23, 2007.
12. I. Budiarjo, H.Nikookar, L.P. Ligthart,”Cognitive Radio with Single Carrier

TDCS and Multicarrier OFDM Approach with V-BLAST Receiver in Rayleigh
Fading Channel”, in Proc. 2nd International Conference on Cognitive Radio
Oriented Wireless Networks and Communications (CROWNCOM), Orlando,
Florida, USA, pp. 140-146, August 1-3, 2007. (Note : This paper has been
selected by the CrownCom 2007 TPC Co-Chairs for publication in a special
issue of ACM / Springer Mobile Networks and Applications (MONET) journal )
13. I. Budiarjo, H. Nikookar, and L.P. Ligthart, “Performance Evaluation of OFDM

Based Cognitive Radio System With Wiener Filter Channel Estimation Using
Frequency Hopping GSM Channel Model at 900 MHz”, in Proc. 10th European
Conference on Wireless Technology (ECWT), Munich, Germany, pp. 74-77,
Oct. 8-12, 2007.
14. M. K. Lakshmanan, I. Budiarjo and H. Nikookar, “Maximally Frequency

Selective Wavelet Packets Based Multicarrier Modulation Scheme for Cognitive
Radio Systems”, in Proc. 50th IEEE Global Communications (GLOBECOM)
Conference, Washington, DC, USA, pp. 4185-4189, Nov. 26-30, 2007.
15. I. Rashad, I. Budiarjo and H. Nikookar, “Efficient Pilot Pattern for OFDM based
Cognitive Radio Part I”, in Proc. 14th IEEE Symposium on Communications
218
and Vehicular Technology (SCVT), Delft, The Netherlands , pp. 1-5, Nov. 15,
2007.
16. I. Budiarjo, I. Rashad and H. Nikookar, “Efficient Pilot Pattern for OFDM based
Cognitive Radio Part II”, in Proc. 14th IEEE Symposium on Communications
and Vehicular Technology, Delft, The Netherlands, pp. 1-5, Nov. 15, 2007.
17. M. K. Lakshmanan, I. Budiarjo and H. Nikookar, “Wavelet Packet Multi-carrier

Modulation MIMO Based Cognitive Radio Systems with VBLAST Receiver
Architecture”, in Proc. IEEE Wireless Communications and Networking
Conference (WCNC), Las Vegas, USA , pp. 705-710, March 31 – April 3, 2008.
18. I. Budiarjo, I. Rashad and H. Nikookar, “On The Use of Virtual Pilots with
Decision Directed Method in OFDM Based Cognitive Radio Channel
Estimation Using 2x1-D Wiener Filter”, in Proc. IEEE International Conference
on Communications (ICC), Beijing, China , pp. 703-707, May 19-23, 2008.
19. I. Budiarjo, H. Nikookar, and L.P. Ligthart,” On the Utilization of Embedded

Symbol for CCSK BER Improvement in TDCS Dynamic Spectrum Access”, in
Proc. IEEE European Wireless Technology Conference, Amsterdam, The
Netherlands, pp. 123-126, Oct. 28-29 2008.
20. I. Budiarjo, H. Abdulrahman Sulaiman, and H. Nikookar, “On the Use of Virtual
Pilots for 2x2 MIMO V-BLAST OFDM Channel Estimation with Wiener Filter
in Hexagonal Pilot Pattern”, in Proc. 15th IEEE Symposium on Communications
and Vehicular Technology (SCVT), Antwerp, Belgium, Nov. 2008.
21. I. Budiarjo and H. Nikookar, “MIMO TDCS with Extra Embedded Symbol for
Higher Data Rates in Overlay Spectrum Sharing System”, in Proc. IEEE
Wireless Communications and Networking Conference (WCNC), Budhapest,
Hungary, pp. 1-6, April 5-8, 2009.
Demonstrator Paper
22. S. B. Raghunathan, M. v. d. Oever, R. D. Mohammady, P. Pawełczak, I.
Budiarjo, M. Heskamp, Q. Zhang, A. Kokkeler, H. Nikookar, Z. Qin, R.
Hekmat, and L. P. Ligthart, “Dynamic Spectrum Access AAF Platform”, in
IEEE Symposia on New Frontiers in Dynamic Spectrum Access Networks
(DySPAN), Oct 14-17, 2008, Chicago, Illinois, USA.
219
Reports
23. I. Budiarjo, H. Nikookar,“D.3.12 Adaptive Baseband Processing : State of the

Art”, Adaptive Ad-hoc Freeband (AAF), May, 2005, The Netherlands.
24. M. Heskamp, I. Budiarjo, Q. Zhang and A. Kokkeler, “D.3.22 First Phase

Radio System Architecture, Algorithms and Mappings”, Adaptive Ad-hoc
Freeband (AAF), October, 2005, The Netherlands.
25. I. Budiarjo, M. Heskamp, Q. Zhang, and F.W. Hoeksema, ”D.3.31Radio

System: First Phase Verification and Advances in Architecture and
Algorithms”, Adaptive Ad-hoc Freeband (AAF), August, 2006, The
Netherlands.
26. I. Budiarjo, M. Heskamp, Q. Zhang, H. Nikookar , A. J. Kokkeler, “D.3.41

Radio system: Second phase verification and advances in architecture and
algorithms”, Adaptive Ad-hoc Freeband (AAF), July 2007, The Netherlands.
220
Publication Cross-reference Table

The thesis is based, in part, on the publications listed in the previous sections. The following
table indicates the relevance of the publications to the chapters within the thesis.
Publication Ch. 1 Ch. 2 Ch. 3 Ch. 4 Ch. 5 Ch. 6

1 ◙ ◙ ◙ ◙
2 ◙ ◙
3 ◙ ◙ ◙ ◙
4 ◙ ◙
5 ◙ ◙ ◙
6 ◙
7 ◙
8 ◙ ◙ ◙
9 ◙ ◙
10 ◙ ◙
11 ◙ ◙ ◙ ◙ ◙
12 ◙ ◙
13 ◙ ◙
14 ◙ ◙
15 ◙
16 ◙
17 ◙ ◙
18 ◙
19 ◙
20 ◙
21 ◙
22 ◙
23 ◙ ◙ ◙ ◙ ◙
24 ◙ ◙ ◙ ◙ ◙
25 ◙ ◙ ◙ ◙ ◙
26 ◙ ◙ ◙ ◙ ◙ ◙
221
Samenvatting
Adaptive Baseband Verwerkingstechnieken

voor Cognitieve Radio Systemen
Cognitieve Radio is een nieuw paradigma in de draadloze communicatie. Het voegt

intelligentie en een dimensie van ‘bewustzijn’ toe aan het radiocommunicatiesysteem. Een
cognitief systeem is zich bewust van veranderingen in zijn omgeving. Het ‘bewustzijn’ dat in
dit proefschrift beschreven wordt heeft betrekking op het spectrum en de kanaalkeuze. Door
op een intelligente manier te leren en de omgeving te begrijpen kan een cognitief systeem zijn
transmissieparameters aanpassen aan de veranderende omgeving. Het doel van een cognitief
systeem is betrouwbare communicatie en efficiënt spectrumgebruik.
Alhoewel het grootste deel van het radiospectrum al gelicentieerd is, hebben studies en
metingen aangetoond dat het spectrum niet druk bezet is. Sommige banden worden zelfs
nauwelijks gebruikt. Congestie treedt op als gevolg van een slechte
spectrumtoegangstechniek. Hierdoor ontstaat de mogelijkheid om spectrum te ‘huren’ of te
gebruiken als de gelicentieerde gebruiker niet actief is, danwel in combinatie met een
gelicentieerd systeem een extra gebruiker toe te laten door middel van een cognitief systeem
waarbij gebruik gemaakt wordt van de gaten in het spectrum. Op deze manier wordt het
spectrum efficiënter gebruikt. De uitdaging is om een win-win co-existentie te bereiken tussen
het cognitieve radiosysteem (het huursysteem) en het gelicentieerde systeem. In de literatuur
is Spectrum pooling voorgesteld als een techniek waarmee het spectrum gedeeld kan worden
met het gelicentieerde systeem, door gebruik te maken van Orthogonal Frequency Division
Multiplexing (OFDM) als modulatietechniek. Sommige OFDM-carriers die zich in de
gelicentieerde band bevinden worden ge-deactiveerd zodat deze draaggolven niet interfereren
met het gelicentieerde systeem. Door deze flexibiliteit wordt OFDM beschouwd als een
geschikte techniek voor cognitieve radio. Interferentie op het signaal van de gelicentieerde
gebruiker kan verder beperkt worden door meer carriers te de-activeren, door windowing en
door verschillende signaal-bewerkingstechnieken. Deze technieken worden in dit proefschrift
bestudeerd en de resultaten worden gepresenteerd. De techniek om spectrum te delen kan
alleen werken als er nauwkeurige informatie over de bezetting van het spectrum beschikbaar
is. Deze informatie is afkomstig uit een spectrum sensing module. Voor spectrum sensing is
een geavanceerde module nodig, welke veel aandacht vraagt. Dit proefschrift richt zich niet
op de spectrum sensing module, maar gaat er vanuit dat de informatie over de bezetting van
222
het spectrum beschikbaar en nauwkeurig is. Onze onderzoekspartner bij de Universiteit

Twente† behandelt het onderwerp van spectrum sensing.
Om het doel van cognitieve radio, namelijk betrouwbare communicatie, te realiseren, is het
belangrijk om niet alleen stil te staan bij de techniek aan de kant van de gelicentieerde
gebruiker, maar ook bij die van de huurgebruiker. Naast het veiligstellen van betrouwbare
communicatie op het gelicentieerde systeem moet de gewenste kwaliteit en Quality of Service
(QoS) van het cognitieve systeem worden gehaald. De praktische parameters waarnaar
gekeken moet worden zijn de Bit Error Rate (BER) en de bitrate van het systeem. Het
toepassen van OFDM met spectrum shaping met als doel de bijdrage van interferentie in het
gelicentieerde systeem te beperken, kan alleen gedaan worden ten koste van de ‘eigen’ QoS.
Onderzoeksresultaten in de literatuur hebben aangetoond dat door toepassing van Adaptive
Bit Loading de BER van een OFDM-systeem kan worden verbeterd. Bij Adaptive Bit
Loading worden afhankelijk van de kanaalkwaliteit de bits op een intelligentie manier
toegewezen aan elke afzonderlijke OFDM-carrier en worden de als doelgestelde BER en
bitrate van het hele systeem aangepast. In de thesis stellen we voor om Adaptive Bit Loading
toe te passen in combinatie met Spectrum Shaping, met als doel de QoS van het cognitieve
systeem te behouden. De impact van deze combinatie op de toename van de OFDM Peak to
Average Power Ratio (PAPR) wordt geëvalueerd door simulaties.
Er bestaat een groeiende interesse om de Fourier-transformatie in OFDM te vervangen door

Wavelet Basis functions. Deze techniek wordt ook wel Wavelet Packet Multi Carrier
Modulation (WPMCM) genoemd. In de literatuur zijn onderzoeksresultaten gepresenteerd
waarin WPMCM geëvalueerd wordt en de resultaten worden vergeleken met OFDM. Op
basis van succesvolle toepassing van een Frequency Selective Wavelet in een Ultra Wide
Band systeem evalueren we in dit proefschrift de geschiktheid van de Frequency Selective
Wavelet in WPMCM, gecombineerd met het concept van Spectrum Pooling, om te worden
toegepast in een cognitief radiosysteem. Omdat efficiënt spectrumgebruik één van de
hoofddoelen is van cognitieve radio, is er voldoende reden om Multiple Input Multiple-Output
(MIMO) op te nemen in het cognitieve radio systeem. In dit proefschrift wordt dit onderwerp
bestudeerd en worden de prestaties van MIMO in een OFDM- en WPMCM-gebaseerd
cognitief radiosysteem geëvalueerd.
Channel estimation (kanaal schatten) is de cruciale module in elk OFDM-systeem. Wij stellen
een effectief channel estimation schema voor, gebaseerd op optimale piloot patronen
aanwezig in conventionele OFDM en gebruikmakend van virtuele ‘piloot-tonen’ om te
worden toegepast in cognitieve radiosystemen. De virtuele piloottonen worden afgeleid uit de
combinatie van lineaire interpolatie/extrapolatie tussen twee echte piloottonen via de
zogenoemde decission directed method. Zonder beperking in algemene toepasbaarheid
†
Dit proefschrift is onderdeel van het Werkpakket 3 (WP3) in het Nederlandse Adaptive Adhoc Freeband
(AAF) project. In WP3 zullen 3 proefschriften worden voltooid, één over Spectrum Sensing, één op Baseband
Processing (dit proefschrift) en het derde proefschrift handelt over het in kaart brengen van Spectrum Sensing en
Baseband Verwerking tot een heterogene architectuur.
223
gebruiken we het Wiener-filter als channel estimation techniek, vanwege zijn efficiëntie en
recht-toe-recht-aan-manier waarmee de kanaalcorrelatie-eigenschappen van de afstand tussen
de piloottonen en de data worden uitgenut. We werken het Single Input Single Output (SISO)
op OFDM gebaseerde cognitieve radiosysteem uit en breiden het daarna uit met de
MIMO-applicatie.
Naast OFDM- en WPMCM-gebaseerde cognitieve radiosystemen zijn recentelijk ook het

Transform Domain Communication System (TDCS) en het Wavelet Domain Communication
System (WDCS) geïntroduceerd als veelbelovende modulatietechnieken voor cognitieve
radiotoepassingen. TDCS en WDCS hebben beperkingen in bitrate. In een poging om de
bitrate van TDCS en WDCS te verbeteren, stellen we voor om een extra embedded symbool
toe te voegen aan TDCS en WDCS. We analyseren de impact van het embedded symbool op
de datadetectie in een multi-useromgeving. Deze aanpak wordt al ondersteund in
conventionele TDCS en WDCS. Bovendien evalueren we de prestaties van TDCS met een
embedded symbool in een MIMO-systeem.
Als een soort verificatieplatform voor cognitieve radio hebben we een praktische
Demonstrator voorgesteld, die bestaat uit een spectrum scanning module en een transceiver
module werkend in basisband. De spectrum sensing wordt gebruikt door de Universal
Software Defined Radio Peripheral (USRP), terwijl de baseband processing transceiver
module is toegepast in een FPGA development board. De huidige versie van het cognitieve
radiosysteem verificatieplatform is simpel, maar heeft beperkte mogelijkheden. Verdere
ontwikkelingen op dit gebied zijn nodig en details voor gerelateerd toekomstig werk zijn ook
in dit proefschrift opgenomen.
224
225
Acknowledgement
This thesis is the result of a four year PhD journey with hard work. This thesis has come to its
present shape due to support of many people. Therefore, I would like to acknowledge the
people who have provided me with the support over the past years to complete this thesis.
First of all, I would like to thank my supervisor Dr. Homayoun Nikookar, for his guidance,
discussions, reviews and strong support during my research work. Also I would like to thank
Prof. Alex Yarovoy for introducing my resume to Dr Nikookar so I had the chance to pursue
my PhD in Cognitive Radio.
I would like to express my deepest gratitude to my promotor Prof. Leo Ligthart for the
opportunity given to me to do research in the field of Cognitive Radio, for his valuable
comments on shaping the manuscript, and for guiding me with the propositions and thesis
summary in Dutch.
I would like to thank the Freeband AAF project for funding my research, and also the team of
researchers in a fruitful cooperation: Ir. Fokke Hoeksema, Dr. Andre Kokkeler, Dr. Qiwei
Zhang, and Ir. Marnix Heskamp, all from Twente University; Dr. Ramin Hekmat and Dr.
Przemyslaw Pawelczak, both from WMC TU Delft.
I am thankfull to my room-mate colleague, Ir. Madan K. Lakshmanan for the collaboration

and discussion especially on the subject of wavelet based transmission and his words of
wisdom. I would like to acknowledge my master students at IRCTR TU Delft : Ir. Ikhsan
Rashad, Ir. Haval Abdulrahman Sulaiman, and Ir. Iman Ghaseminezhad Marandi for their
contribution to this thesis. My gratitude to student assistants Ir. Shreyas Raghunathan, Ir.
Maarten van de Oevers, and Ir. Rahman Doost for their effort in developing the IRCTR
Cognitive Radio demonstrator platform.
I would like to thank Mrs Mia van der Voort-Kester, Laura Bauman, Dominique Meijer,
Marjon Verkaik, Patricia Thoms and Rowena for their constant assistance with the
administrative matters. My acknowledgement go also to all my colleagues in IRCTR, I enjoy
the time in the International Research Centre for Pubs and Restaurants (IRCPR) gatherings.
My gratitudes to all of my Indonesian countrymen in Delft, the Delft Christian Fellowship
community, fellows in International Church Levend Water Delft for their support, and Delft
Student Chaplancy as well. My gratitude to my colleague in KPN Selmar Sneider and my
226
friend Dirk Goedbloed from Levend Water Church in Delft for helping me shaping the
Samenvatting.
I would like to dedicate this thesis to my parents Baik Ukurta Purba and Bedya Haris. I thank
them for their wisdom, patience and care in shaping me the way that I am. I am really
greatfull to have them, Yesus menyertai dan memberkati mu Ayah dan Ibu. My gratitude to
my beloved brother and friend Zakaria Budiarjo for his support while I am abroad.
Finally, I would like to thank Winny Vyedrametha Sihombing for being patience in
supporting and accompanying me in finishing this dissertation while she is in a different
continent with 6 hours time difference, the wonderfull love and care that she brings to my life,
and her wonderfull cover design for this thesis.
227
Curriculum Vitae
Ibrahim Budiarjo was born in 1978 in Bandung, West Java province, Indonesia. After
finishing his Bachelor Study in 2001 in Bandung Institute of Technology, Indonesia with
Sarjana Teknik (Bachelor of Engineering) degree in Electronics, he continued his study in
Communications Engineering in Munich University of Technology, Germany and received
his M.Sc degree in 2003. His M.Sc thesis was about OFDM channel estimation by adaptive
filtering. The M.Sc thesis was performed at Deutsche Luft und Raumfahrt (DLR)
Oberpfaffenhofen Germany and DoCoMo Eurolabs Muenchen Germany. He did a voluntary
internship in Knorr Bremse AG Germany working on CAN bus system for a year (till end of
2004).
In February 2005 he became a PhD candidate at the International Research Centre for
Telecommunications and Radar (IRCTR) TU Delft, the Netherlands within the Dutch
Adaptive Adhoc Freeband (AAF) Project which resulted in this dissertation. The focus of his
research is on the adaptive baseband processing for Cognitive Radio systems.

Bud I Ar Jo Dissertation 2 ND Edition

Uploaded by

Copyright:

Available Formats

Bud I Ar Jo Dissertation 2 ND Edition

Uploaded by

Document Information

Original Title

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

Bud I Ar Jo Dissertation 2 ND Edition

Uploaded by

Copyright:

Available Formats

Adaptive Baseband Processing Techniques

for Cognitive Radio Systems

Ter verkrijging van de graad van doctor

Master of Science in Communications Engineering

geboren te Bandung, Jawa Barat, Indonesien.

Prof. dr. ir. L. P. Ligthart

Rector Magnificus voorzitter

Copyright © 2009 by Ibrahim Budiarjo

Printed in The Netherlands

Adaptive Baseband Processing Techniques

As a sort of verification platform for Cognitive Radio we proposed a practical Demonstrator

3.3.2 Generation of Wavelet Packet Dual Bases 53

6.3 Spectrum Scanner Module 169

1.1 Research Background

In order to utilize the unused spectrum efficiently in dynamically changing environments, a

Figure 1.1: Cognitive Radio signal transmission cycle [6]

1.2 Research Motivation

In the RU perspective reliable communications is not connected to an interference avoidance

1.3 Scope and Novelties of this Dissertation

1.4 Outline of the Thesis

The organization of this thesis is as follows :

OFDM Based Cognitive Radio

In the overlay mode Orthogonal Frequency Division Multiplexing (OFDM) is considered as

Orthogonal frequency division multiplexing (OFDM) is a multicarrier modulation technique

Figure 2.1 : OFDM system model

Xn Serial to Add guard Digital

Figure 2.2 : OFDM subsystem description with spectrum sensing module

2.2 Spectrum Pooling with OFDM

Figure 2.3 : Illustration of spectrum pooling block diagram

0.6 RAISED COSINE α = 0.0625

Other available forms are Bartlett [14],

Better than Raised Cosine (BTRC) [15],

and Flipped Inverse Hyperbolic Secant (farchsech) [16] window,

Time domain Frequency domain

2.3 OFDM Sidelobes Reduction Techniques

2.3.1 Sidelobe Suppression by Cancellation Carriers Insertion

1 NFFT (1+α)Tu ccN

We introduce an alternative optimization approach by simplifying it to a set of ccN linear

Original OFDM without Deact.and Sidelobes Canc. Carriers

2.3.2 Sidelobe Suppression by Subcarrier Weighting

2.3.3 Sidelobe Suppression by using Multiple Choice Sequence

X’(l) = П(l)X, (2.8)

2.3.4 Sidelobe Suppression by Adaptive Symbol Transition

................. y(m) a(m) y(m+1) a(m+1) y(m+2) a(m+2) .................

Figure 2.10: Adaptive symbol transition scheme

2.3.5 Sidelobe Suppression by Frequency Domain Additive Signal

2.3.6 Sidelobe Suppression by Constellation Expansion

2.4 Frequency Hopping

2. 4. 1 Evaluation Using GSM 900 MHz Channel Model

Tap Relative Time (μs) Average relative

Tap Relative Time Average relative

Table 2.3: OFDM frame design according to TDD Bluetooth frame.

Baseband Processing for OFDM-Based

3.2 Combined Spectrum Pooling, Windowing, Adaptive Bit Loading and

transfer the transfer the

3.2.1 Modulation Detection by Signaling and Blind Detection

If some flexibility in choosing the transmission parameters is sacrificed within an adaptation

where Q(.) is the probability Gaussian Function (x ) =

where pi is the ith true probability outcome, π1 to πL correspond to elements of the

ψ i [k ] = w(k )* w(k / 2).... w(k / 2J −2 )* w(k / 2J −1 );