Page 1
SENSOR ARRAY ASSISTED SPECTRUM SENSING AND PERFORMANCE
OPTIMIZATION IN COGNITIVE RADIO NETWORKS
KIRAN SULTAN
A Thesis Submitted
in Partial Fulfillment of the Requirement
for the Degree of
Doctor of Philosophy
DEPARTMENT OF ELECTRICAL ENGINEERING
AIR UNIVERSITY
2013
Page 2
SENSOR ARRAY ASSISTED SPECTRUM SENSING AND PERFORMANCE
OPTIMIZATION IN COGNITIVE RADIO NETWORKS
Ph.D. Dissertation
SUBMITTED BY
KIRAN SULTAN
REG. NO. Ph.D.-EE-091315
SUPERVISOR
PROF. DR. IJAZ MANSOOR QURESHI
DEPARTMENT OF ELECTRICAL ENGINEERING
AIR UNIVERSITY
ISLAMABAD
December, 2013
Page 3
CERTIFICATE
Department of Electrical Engineering
It is hereby certified that Kiran Sultan (Reg # Ph.D.-EE-091315) has successfully completed her
dissertation.
_____________________________
Dr. Ijaz Mansoor Qureshi Air University
Supervisor
____________________________ ____________________________ Dr. Fida Muhammad Khan Dr. Zafar Ali Shah
Internal Examiner 1 Internal Examiner 2 Guidance and Evaluation Committee Guidance and Evaluation Committee
____________________________ ____________________________
Dr. Abdul Jalil Dr. Noor Muhammad Khan External Examiner 1 External Examiner 2
Guidance and Evaluation Committee Guidance and Evaluation Committee
____________________________ ____________________________
Dr. Fida Muhammad Khan Dr. Zafarullah Koreshi Chair Department Dean
Page 4
SENSOR ARRAY ASSISTED SPECTRUM SENSING AND PERFORMANCE
OPTIMIZATION IN COGNITIVE RADIO NETWORKS
Ph.D. Dissertation
KIRAN SULTAN
REG. NO. Ph.D.-EE-091315
SUPERVISOR
PROF. DR. IJAZ MANSOOR QURESHI
FOREIGN RESEARCH EVALUATION EXPERTS
Prof. Dr. AJITH ABRAHAM, DIRECTOR, MIR LABS, USA
Prof. Dr. WEN HSIEN FANG, NTUST, Taiwan
DEPARTMENT OF ELECTRICAL ENGINEERING
AIR UNIVERSITY
ISLAMABAD
2013
Page 5
ABSTRACT
Cognitive Radio has gained worldwide attention from research communities and is expected to
be a revolutionary technology for the next generation (4G) wireless systems. In this dissertation,
Amplify-and-Forward (AF) based relay-assisted cognitive radio networks (RCRNs) are studied
in an underlay spectrum sharing environment. The primary issue faced by underlay networks is
the limited transmit power ability of the secondary users (SUs) due to the interference constraints
towards the primary users (PUs), which reduces secondary throughput and allows only short-
range communication. Thus, performance enhancement of secondary communication in the
frequency bands allocated to the PUs is a major design challenge faced by the underlay RCRNs.
It requires relay selection along with the fine tuning and adjustment of the transmit power of the
secondary relays.
In this thesis, we proposed advanced multiple relay selection schemes for secondary network in
the Rayleigh flat-fading scenario considering the availability of perfect instantaneous channel
state information (CSI). The effects of variations in the instantaneous CSI, transmit power of
source and relays, interference threshold of the primary network, signal-to-noise ratio (SNR)
threshold of the secondary network and size of potential relay network on multiple relay
selection in underlay RCRNs are the main issues that are analyzed in depth in this research.
Furthermore, the performance analysis of multiple relay selection has been carried out and closed
form expressions for the outage probability and average probability of error have been derived
through the cumulative distributive function (CDF) of the received SNR at secondary
Page 6
destination, which is a new contribution to the AF-based underlay RCRNs. The optimization tool
used in this study is Artificial Bee Colony (ABC) global optimizer.
Another novel idea proposed in this dissertation is Fuzzy Rule Based System (FRBS) for
multiple relay selection and transmit power allocation (RSTPA), which is a new contribution to
the underlay RCRNs. The proposed FRBS assisted RSTPA schemes aim to perform intelligent
multiple relay selection for performance enhancement of secondary communication in power
constrained RCRNs. It is proved through simulations that FRBS is an optimal choice to solve the
non-linear optimization problems of SNR maximization and transmit power minimization.
Another contribution of this research is in the field of spectrum sensing in CRNs. Spectrum
sensing faces a lot of challenges in terms of reliability and accuracy of information for detection
and estimation of primary transmissions in CRNs. The advantages and limitations of different
cooperative and non-cooperative spectrum sensing schemes have been studied in detail, and a
novel spectrum sensing scheme based on uniform linear array (ULA) of sensors is proposed,
which not only detects the number of sources, but also estimates their parameters such as
frequency, Direction-of-Arrival (DOA) and power strength. The effectiveness and reliability of
the proposed scheme is proved under low SNR conditions. Genetic Algorithm (GA) hybridized
with Pattern Search (PS) is used to optimize the results.
All the proposed algorithms have been investigated through simulations under different design
requirements, constraints and a well-defined range of different parameters to validate their
significance and effectiveness.
Page 7
Copyright by
KIRAN SULTAN
2013
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 permission from the
author.
Page 8
DEDICATED TO
My Parents,
Brother and Sisters
Page 9
i
CERTIFICATE OF APPROVAL It is certified that the research work contained in this Ph.D. dissertation has been carried out
under my supervision in the Department of Electrical Engineering, Air University, Islamabad. It
is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. It
has been set against the plagiarism and the report has been attached alongwith.
Signature: _____________________
Supervisor:
Prof. Dr. Ijaz Mansoor Qureshi Department of Electrical Engineering
Air University,
Islamabad.
Page 10
ii
LIST OF PUBLICATIONS
List of Published Papers
1. Kiran Sultan, Ijaz Mansoor Qureshi, Aqdas Naveed Malik, Muhammad Zubair,
“Performance Analysis of Relay Subset Selection for Amplify-and-Forward Cognitive Relay
Networks”, The Scientific World Journal by Hindawi Publishing Corporation, 2013, (ISI
Indexed Journal with IF: 1.732).
2. Kiran Sultan, Ijaz Mansoor Qureshi, Muhammad Zubair, Aqdas Naveed Malik, “Power
Minimization through Relay Subset Selection in Underlay Cognitive Radio Networks”,
World Applied Sciences Journal, Vol. 23(5), 2013, pp. 714-717, (ISI Indexed Journal with
IF: 0.234).
3. Kiran Sultan, Ijaz Mansoor Qureshi, Muhammad Zubair, “Detection and Estimation of
Multiple far-field Primary Users using Sensor Array in Cognitive Radio Networks”, Journal
of Computing, Vol. 5(2), 2013, pp. 7-14, (ISI Indexed Journal with IF: 0.21).
4. Kiran Sultan, Ijaz Mansoor Qureshi, Bahman Ramzan Ali Alyaei, Ali Azad, “Performance
Enhancement of Secondary Communication through Multiple Relay Selection and Power
Allocation in Non-Regenerative Cognitive Radio Networks”, J. Basic Appl. Sci. (JBASR),
Vol. 3(10), 2013, pp. 416-420, (ISI Indexed Journal).
5. Kiran Sultan, Ijaz Mansoor Qureshi, Waseem Khan, Atta-ur-Rahman, “Performance
Enhancement of Secondary Network using Fuzzy Rule based System in Cognitive Relay
Networks”, European Journal of Scientific Research (EJSR), 2013, (ISI Indexed Journal).
6. Kiran Sultan, Ijaz Mansoor Qureshi, Muhammad Zubair, “SNR maximization through
Relay Selection in Cognitive Radio Networks”, Research Journal of Applied Sciences,
Engineering and Technology, 6(7), 2013, pp. 2616-2620, (ISI Indexed Journal).
7. Kiran Sultan, Ijaz Mansoor Qureshi, Muhammad Zubair, Babar Sultan, “ SNR
Maximization through CSI based Relay-Subset Selection in Amplify-and-Forward Cognitive
Radio Networks”, Presented in ICACELT, Abu Dhabi UAE, 2013, pp. 61-66.
8. Kiran Sultan, Ijaz Mansoor Qureshi, Muhammad Zubair, “SNR maximization through
Relay Selection and Power Allocation for Non-Regenerative Cognitive Radio Networks”,
Presented in INMIC IEEE, 2012, pp. 361-364.
Page 11
iii
9. Kiran Sultan, Ijaz Mansoor Qureshi, Babar Sultan, “Performance Enhancement of
Secondary Communication in Underlay Cognitive Relay Networks”, accepted in INAAR,
IEEE, 2013.
10. Waseem Khan, I. M. Qureshi, Kiran Sultan, "Ambiguity Function of Phased-MIMO Radar
and its Properties", IEEE Geoscience and Remote Sensing Letters, Vol. PP(99), 2013, pp. 1-
5, (ISI Indexed Journal with IF: 1.823).
11. Ayesha Khaliq, Fawad Zaman, Kiran Sultan, Ijaz Mansoor Qurehsi, “3-D near field source
localization by using hybrid Genetic Algorithm”, Research Journal of Applied Sciences,
Engineering and Technology, Vol. 6(23),2013. (ISI Indexed Journal).
12. Shahid H. Abbassi, I. M. Qureshi, Bahman R. Alyaei, Hameer Abbasi, Kiran Sultan, “An
Efficient Spectrum Sensing Mechanism for CR-VANETs”, J. Basic Appl. Sci., (JBASR),Vol.
3(12), 2013, pp. 365-378, (ISI Indexed Journal).
13. Habibullah Jamal, Kiran Sultan, “Performance Analysis of Loss-Based High-Speed TCP
Congestion Control Algorithms”, WSEAS International Conference, Ningbo, China, 2008.
14. Habibullah Jamal, Kiran Sultan, “Performance Analysis of TCP Congestion Control
Algorithms”, International Journal of Computers and Communications, Vol. 2(1), 2008.
List of Submitted Papers
1. Kiran Sultan, Ijaz Mansoor Qureshi, Atta-ur-Rahman, and Shahid Hussain Abbassi,
“Transmit Power Minimization in Cognitive Relay Networks using Fuzzy Rule Based
System”, submitted in Journal of Intelligent and Fuzzy Systems.
2. Kiran Sultan, Ijaz Mansoor Qureshi, Atta-ur-Rahman, Waseem Khan, “SNR Maximization
using Fuzzy Rule Base System in Relay Assisted Cognitive Radio Networks”, submitted in
Journal of Multiple-valued Logic and Soft Computing.
3. Kiran Sultan, Ijaz Mansoor Qureshi, Muhammad Zubair, Aqdas Naveed Malik, “Artificial
Bee Colony Optimization for Relay Selection with SNR Maximization in Underlay
Cognitive Radio Networks”, submitted in Iranian Journal of Science and Technology,
Transactions of Electrical Engineering.
4. Waseem Khan, I. M. Qureshi, Kiran Sultan, "Ambiguity Function of Frequency-Diverse-
Array Radar and its Properties", submitted in WASJ (ISI Indexed Journal with IF 0.234).
The material presented in this dissertation is based on the published papers 1 to 8 and submitted
papers 1 to 3.
Page 12
iv
ACKNOWLEDGEMENTS
I am thankful to Almighty Allah for his guidance at each step of this work through the
blessings of eeman, health, supporting family, brilliant teachers, cooperative friends and
colleagues. Peace and prayer for marvelous human and the torch bearer of wisdom Muhammad
(Peace be upon him), who’s objective was enlightenment of the whole world.
I owe thanks to many people for where I have arrived today.
First and foremost, I offer my special and sincere thanks to my supervisor Dr. Ijaz
Mansoor Qureshi for his direction, motivational guidance and support that remained with me
throughout the research work. He is a man of inducing brain waves and initiating the sparks of
good ideas in my mind. He helped me with building up a strong foundation for my future
scholarly career. Despite his hyper dimensional commitments, he never let me relax. I would like
to thank him for giving me space and freedom to glide and explore the topics I felt more
comfortable with. I also thank him for listening to me with patience and tolerance. He is one of
the best samples of a good supervisor.
I am indebted to Dr. Kamal Athar for his encouragement and administrative support. He
always stands up for us compassionately, and looks for great opportunities for us. I owe to
express my thanks to Dr. Fida Muhammad Khan for his high cooperation and strong
administrative support from time to time. He always tries to make things smooth and manageable
for us. I am grateful to Dr. Atta-ur-Rahman for his fruitful advices and profound comments on
my research work. I found him eager to help me in my research. I also thank Dr. Zafar Ali Shah
for all I learned from him that I have to be organized, precise and smart to be a good academic
scholar.
I feel very fortunate to be given an opportunity to have the company of a collection of
awesome people in the Department of Electrical Engineering, AU. I like to thank my friends,
colleagues and students for being cooperative during my Ph.D. period. They are all the most
caring and smart people I have ever had the privilege to know. My deepest thanks to Mr.
Bahman Ramzan Ali Alyaei, whom one can never stop learning from. It is really hard to count
many scholars as smart and humble as him. I cannot thank him in words for his enormously
helping and directing comments on my research. I am really thankful to Ms. Sundas Amin from
Page 13
v
depth of my heart for being so nice, sincere and helpful. I would like to express my gratitude to
Mr. Shahid Abbasi for his encouragement and moral support.
At last I express my immense gratitude and respect to my lovely parents and family who
never stopped supporting, encouraging and standing by me during the ups and downs of my life,
and love to see me flourishing and progressing. I am thankful to my brother Capt. Babar for
providing me the educational and moral support during the whole Ph.D. and setting up goals for
me. I dedicate this thesis, which is the outcome of my life to my family to whom I owe every
single success in my life.
(Kiran Sultan)
Page 14
vi
TABLE OF CONTENTS
LIST OF TABLES x
LIST OF FIGURES xi
LIST OF ABBREVIATIONS xiii
Chapter 1 1
Introduction
1.1 Background 1
1.2 The System Model 4
1.3 Objectives and Contributions 7
1.4 Organization of the Thesis 8
Chapter 2 11
Cognitive Radios and Cognitive Relay Networks
2.1 History of Cognitive Radios 11
2.2 Spectrum Sharing in Cognitive Radios 15
2.2.1 Underlay Spectrum Sharing 16
2.2.2 Overlay Spectrum Sharing 16
2.2.3 Interweave Spectrum Sharing 16
2.3 Cooperative Communication 17
2.3.1 Relay Protocols 18
2.3.1.1 Amplify-and-Forward Relaying 19
2.3.1.2 Decode-and-Forward Relaying 19
2.4 Cognitive Relay Networks 20
2.4.1 Relay Selection in Underlay Cognitive Relay Networks 21
2.4.1.1 Best Relay Selection Schemes Proposed in Literature 22
2.4.1.2 Multiple Relay Selection Schemes Proposed in Literature 23
2.4.1.3 Observations in the Literature Review of Relay Selection 23
Schemes
2.5 Concluding Remarks 24
Chapter 3 25
Nature Inspired Algorithms and Fuzzy Logic
3.1 The Evolutionary Algorithms 25
3.2 Artificial Bee Colony (ABC) Optimization 27
3.2.1 Optimization Phases of ABC 28
3.2.1.1 Initialization Phase 28
3.2.1.2 Best Solution Search Phase 28
3.2.2 Flow Chart of ABC Optimization 29
3.3 Genetic Algorithm 30
Page 15
vii
3.3.1 Optimization Phases in GA 30
3.3.2 Flow Chart of GA Optimization 31
3.3.3 Applications of GA 32
3.4 Fuzzy Logic 32
3.4.1 History of Fuzzy Logic 32
3.4.2 Fuzzy Control System 33
3.4.3 Applications of Fuzzy Logic 34
3.4.4 Fuzzy Logic in CRNs 34
3.5 Concluding Remarks 35
Chapter 4 36
SNR Maximization in Underlay Networks
4.1 Problem Formulation 1 36
4.2 Proposed Algorithms 37
4.2.1 Proposed Algorithm 1 37
4.2.1.1 ABC Optimization 38
4.2.1.2 Simulation Results 40
4.2.2 Proposed Algorithm 2 41
4.2.2.1 Simulation Results 42
4.3 Comparison of the Proposed Algorithms 43
4.3.1 Problem Formulation II 43
4.4 Problem Formulation 2 45
4.5 Proposed Algorithms 46
4.5.1 Proposed Algorithm I 45
4.5.1.1 Simulation Results 47
4.5.1.2 Concluding Remarks 48
4.5.2 Proposed Algorithm II 48
4.5.2.1 Simulation Results 50
4.5.2.2 Concluding Remarks 51
4.5.3 Algorithm III 51
4.5.3.1 Simulation Results 54
4.5.3.2 Concluding Remarks 55
4.6 Comparison of the Proposed Algorithms 55
4.7 Concluding Remarks 57
Chapter 5 58
Outage Analysis of Multiple Relay Selection
5.1 Problem Formulation 58
5.2 The Proposed Algorithm 60
5.3 Performance Analysis 63
5.3.1 Multiple Relay Selection 63
5.3.2 Best Relay Selection 67
5.4 Simulation Results 70
5.5 Concluding Remarks 73
Page 16
viii
Chapter 6 74
Transmit Power Minimization in Underlay CRNs
6.1 Problem Formulation 74
6.2 The Proposed Algorithm 75
6.3 Simulation Results 77
6.4 Concluding Remarks 79
Chapter 7 80
Performance Enhancement of CRNs using Fuzzy Rule Based System
7.1 SNR Maximization 80
7.2 FRBS Assisted System Design 1 81
7.2.1 Mamdani Fuzzy Control 82
7.2.1.1 Fuzzificaion 82
7.2.1.2 Rule Based Decision 85
7.2.1.3 Defuzzifier 85
7.2.2 The Proposed Algorithm 87
7.2.3 Simulation Results 88
7.2.4 Concluding Remarks 91
7.3 FRBS Assisted System Design 2 91
7.3.1 Mamdani Fuzzy Control 93
7.3.1.1 Fuzzification 93
7.3.1.2 Rule Based Decision 96
7.3.1.3 Defuzzifier 97
7.3.2 The Proposed Algorithm 99
7.3.3 Simulation Results 100
7.4 Comparisons of the Proposed Algorithms 101
7.4.1 Concluding Remarks 102
7.5 Transmit Power Minimization 102
7.5.1 The Proposed FLS Design 103
7.5.2 Mamdani Fuzzy Control 103
7.5.2.1 Fuzzification 104
7.5.2.2 Rule Based Decision 105
7.5.2.3 Defuzzifier 105
7.5.3 The Proposed Algorithm 105
7.5.4 Simulation Results 106
7.5.5 Concluding Remarks 110
Chapter 8 111
Detection and Estimation of Multiple Far-Field Primary Users using Sensor Array
8.1 Background 111
8.2 Spectrum Sensing Methods 112
Page 17
ix
8.2.1 Non-Cooperative Spectrum Sensing 112
8.2.1.1 Energy Detector 112
8.2.1.2 Matched Filer Detection 112
8.2.1.3 Cyclostationary based Detection 113
8.2.2 Interference based Spectrum Sensing 113
8.2.3 Cooperative Spectrum Sensing 113
8.3 Source Localization 113
8.4 Contribution of Thesis 114
8.5 System Model and Problem Formulation 115
8.6 Proposed Algorithm for Detection of PUs 117
8.7 Simulation Results and Discussions 119
8.8 Conclusion 127
Chapter 9 128
Conclusions and Future Work
9.1 Conclusions 128
9.2 Future Work 129
References 131
Page 18
x
LIST OF TABLES
Table 4.1 Pseudocode of the Proposed Algorithm 38
Table 4.2 ABC Optimization to solve SNR Maximization Problems 39
Table 4.3 The Parameter Settings 40
Table 4.4 Pseudo Code For Proposed Algorithm 2 (Power Set Algorithm) 42
Table 4.5 Comparison Of The Proposed Algorithms 44
Table 4.6 Psuedocode for Algorithm I 46
Table 4.7 No. of Selected Relays Obtained From Proposed Algorithm 1 48
Table 4.8 Pseudo Code for Algorithm II 49
Table 4.9 No. Of Selected Relays From Proposed Algorithm 2 50
Table 4.10 The Pseudo Code For The Proposed Algorithm 53
Table 4.11 No. Of Selected Relays From Proposed Algorithm 3 55
Table 4.12 Comparison Of The Proposed Algorithms 56
Table 5.1 Pseudocode for the Proposed Relay Subset Selection Algorithm 61
Table 5.2 Performance Analysis Of Best Relay, Multiple Relay and All Relays
Participation Schemes 71
Table 6.1 The Proposed Algorithm 76
Table 6.2 Total Transmit Power required For Different Values of 78
Table 6.3 Transmit Power Allocation To Relay Network For 79
Table 7.1 Corresponding Number Of Selected Relays 89
Table 7.2 Total No. of Selected Relays 101
Table 7.3 Comparisons of Proposed Algorithms in Chp 4 and Chp 7 101
Table 8.1 Parameter Settings for GA-PS 119
Table 8.2 Pseudocode of the Proposed Algorithm for Detection of Number 120
of Sources
Table 8.3 Amplitude, DOA and frequency estimation for different SNR 123
levels with M = 2, L = 20
Table 8.4 Amplitude, DOA and frequency estimation for different SNR levels 125
with M = 4, L = 25
Table 8.5 Amplitude, DOA and frequency estimation for different SNR levels 126
and different number of sensors in the array with M = 2
MandI thth ,,dBdBI thth 0,0
Page 19
xi
LIST OF FIGURES
Fig. 1.1 Strict allocations of frequency bands 1
Fig. 1.2 The System Model 4
Fig. 2.1 FCC Spectrum Allocation Chart 12
Fig. 2.2 The concept of Spectrum Hole 14
Fig. 2.3 Multiple Input Multiple Output (MIMO) System 17
Fig. 2.4 AF and DF Protocols 20
Fig. 2.5 Conceptual Model for Relay(s) Selection 21
Fig. 3.1 Flow Chart for Artificial Bee Colony Optimization 29
Fig. 3.2 Flow Chart for GA Optimization 31
Fig. 3.3 Fuzzy Control System 33
Fig. 4.1 Performance Analysis of Algorithm 1 41
Fig. 4.2: Performance Analysis Of Proposed Algorithm 2 43
Fig. 4.3: Comparison Of The Proposed Algorithms 44
Fig. 4.4: Performance Analysis Of Proposed Algorithm 1 47
Fig. 4.5: Performance Analysis Of Proposed Algorithm 2 50
Fig. 4.6 Performance Analysis Of Proposed Algorithm 3 54
Fig. 4.7 Performance Analysis Of Proposed Algorithms 56
Fig. 5.1: Flowchart of the Proposed Algorithm 62
Fig 5.2: Performance Analysis Of Different Schemes 71
Fig. 5.3: Outage Behavior of Best and Multiple Relay Selection Schemes 72
Fig. 5.4: BER of Best and Proposed Multiple Relay Selection 73
Fig. 6.1: Transmit Power Allocation to Relay Network keeping dBI th 0 77
Fig. 6.2: Transmit Power Allocation to Relay Network for 79
Fig. 7.1 The Proposed System Design 1 81
Fig. 7.2 MFs Of The Antecedents And The Consequents Of FLS 1 And FLS 2 84
Fig 7.3(a) Rule Surface for FLS 1 86
Fig. 7.3(b) Rule Surface for FLS 2 86
Fig. 7.4 The Flow Chart of the Proposed Fuzzy Rule Based RSTPA Design 87
Fig. 7.5 SNR Performance of Proposed Scheme 88
Fig. 7.6 SNR Performance For Different Source Transmit Power Levels 89
Fig. 7.7 Comparison of the Proposed Scheme and the Greedy Scheme 90
Fig. 7.8 Proposed System Design 2 91
Fig. 7.9 Proposed FLS Modules 92
Fig. 7.10(a) MFs For Antecedents And Consequents of FLS1 94
Fig. 7.10(b) MFs For Antecedents And Consequents of FLS1 96
Fig. 7.11(a) Rule Surface for FLS 1 97
Fig. 7.11(b) Rule Surface for FLS 2 98
Fig. 7.11(c) Rule Surface for FLS 3 98
Fig. 7.12 Flow Chart of the Proposed Algorithm 99
Fig. 7.13 Performance Of The Proposed Scheme Vs Interference Threshold thI 100
Fig. 7.14 Proposed FRBS 103
dBdBI thth 0,0
Page 20
xii
Fig. 7.15 Fuzzy Sets for the Antecedents and the Consequent 104
Fig. 7.16 The Rule Surface 105
Fig. 7.17 The Flow Chart of the Proposed Algorithm 106
Fig. 7.18(a) Total Transmit Power Vs Interference Threshold thI for 1th 107
Fig. 7.18(b) Corresponding Number Of Selected Relays 108
Fig. 7.19 Total Transmit Power Vs SNR Threshold th 108
Fig. 7.20 Total Transmit Power Of For Different Source Transmit Power Levels 109
Keeping dBI th 10 And 1th
Fig.8.1. The System Model 115
Fig 8.2(a) Detection of M = 2 PUs 121
Fig. 8.2(b) Error in DOA vs SNR for M = 2, L = 20 122
Fig. 8.2(c) Error in frequency vs SNR for M = 2, L = 20 122
Fig. 8.3(a) Detection of M = 4 PUs 123
Fig. 8.3(b) Error in DOA vs SNR for M = 4, L = 25 124
Fig. 8.3(c) Error in frequency vs SNR for M = 4, L = 25 124
Fig. 8.4(a) Error in DOA estimation for different SNR levels and different number 125
of sensors in the array considering M = 2
Fig. 8.4(b) Error in frequency estimation for different SNR levels and different 126
number of sensors in the array considering M = 2
Page 21
xiii
LIST OF ABBREVIATIONS
Adaptive Fuzzy Logic AFL
Additive White Gaussian Noise AWGN
Amplify-and-Forward AF
Ant Colony Optimization ACP
Artificial Bee Colony ABC
Artificial Immune Optimization AIP
Artificial Intelligence AI
Artificial Neural Networks ANN
Binary Phase Shift Keying BPSK
Bit Error Rate BER
Channel State Information CSI
Cognitive Radio CR
Cognitive Radio Network CRN
Compress-and-Forward CF
Computer Added Design CAD
Consider for Selection CS
Cumulative Distributive Function CDF
Cyclostationary Detection CD
Decode-and-Forward DF
Differential Evolution DE
Direction of Arrival DOA
Dynamic Spectrum Access DSA
Employed Bee EB
Energy Detection ED
Evolutionary Algorithm EA
Expert System ES
Fuzzy Control system FCS
Fuzzy Inference Engine FIE
Page 22
xiv
Fuzzy logic FL
Fuzzy logic system FLS
Fuzzy Rule Based System FRBS
Genetic Algorithm GA
Gradient Search GS
Hybrid Evolutionary Algorithm HEA
Independent and Identically distributed IID
Interior Point Algorithm IPA
Invasive Weed Optimization IWO
Matched Filter MF
Maximum Eigenvalue Detection MED
Maximum Eigenvalue to Trace MET
Maximum Minimum Eigenvalue MME
Membership Function MF
Memetic Particle Swarm Optimization MPSO
Multiple-Input Multiple-output MIMO
Not Selected NS
Onlooker Bee OB
Particle Swarm Optimization PSO
Pattern Search PS
Primary User PU
Probability Distributive Function PDF
Quality of Service QOS
Relay Selection Factor RSF
Relay Selection and Transmit Power allocation RSTPA
Scout Bee SB
Secondary User SU
Selected S
Sequential Quadratic Programming SQP
Signal-to-Noise Ratio SNR
Simulated Annealing SA
Page 23
xv
Software Defined Radio SDR
Spectrum Sensing SS
Strong Consideration for Selection SCS
Time Division Multiple Access TDMA
Ultra-Wideband UWB
Uniform Linear Array ULA
Weak Consideration for Selection WCS
Page 24
1
Chapter 1
INTRODUCTION
1.1 BACKGROUND
Conventionally, spectrum regulatory bodies follow highly inflexible and authoritative approach
in specifying the services for a particular band of frequencies and the permitted technologies to
deliver those services. Such authorized or licensed users of the spectrum are known as primary
users (PUs) [1]. The strict policies for the use of spectrum are very effective to manage the
interference as depicted in Fig. 1.1. The guard bands in the figure ensure that the neighboring
services do not interfere each other’s transmissions [2],[3]. However, as a consequence of this
“Command and Control” strategy of spectrum management, some bands are heavily loaded in
vast temporal and geographical locations, whereas, a large number of frequency bands are highly
underutilized.
Fig. 1.1: Strict Allocations Of Frequency Bands
Page 25
2
Thus, broadly speaking, the spectrum bands are classified into black spaces (highly occupied
frequency bands), grey spaces (partially occupied frequency bands) and white spaces (vacant
frequency bands except for ambient noise) [3]. Joseph Mitola, pioneer of cognitive radio (CR)
technology, innovated the concept of CR, for the first time in the history of wireless
communication in 1999 [4]. Mitola defined CR as a smart radio which is “self-aware” and
“alert”. A CR is expected to operate with a clear understanding of its operating environment, the
communication requirements of other user(s), spectrum regulatory policies and its own
capabilities [2].
The cognitive users, commonly known as secondary users (SUs) in CR terminology, utilize the
licensed band using overlay, underlay and interweave approaches [5]. In the overlay spectrum
sharing, the SU uses specialized signal processing techniques for the performance enhancement
of primary transmissions, while transmitting concurrently in the frequency band of PU. In the
underlay mode, the SU is allowed to transmit in the frequency band assigned to the PU, as long
as the interference offered to the PU by secondary transmissions is below the interference
threshold of the PU. In the interweave approach, the SU looks for the spectrum opportunities
where the PU is currently absent and transmit with full power in the detected spectrum holes.
These spectrum sharing modes will be discussed in more detail in the next chapter.
CR technology has gained world-wide attention from the research communities as a potential
candidate for future wireless world. Being an emerging technology, CR faces a lot of challenges
to replace the currently deployed wireless communication systems. The hot areas of research in
CR are spectrum sensing algorithms, cooperative communication, CR architecture, dynamic
spectrum access, security issues, dynamic resource management, and development of adaptive
algorithms [6]-[7]. A lot of development has been done in designing the individual components
of CRs and building protocol architecture of cognitive radio networks (CRNs), but still the
integration of these parts for large scale deployment of CRNs is a major area of research.
Spectrum sensing (SS), aims to obtain information about the local spectrum, and is the key
enabling technology for the establishment of CR. CRs are built on the ability to sense the radio
spectrum and gain knowledge about the available frequency bands, transmit powers and
Page 26
3
direction-of-arrival (DOA) of the active users, available networks, spectrum management
policies regarding the use of detected spectrum and other operating restrictions. Various features
of spectrum sensing are shown in figure [8].
Earlier spectrum sensing techniques mainly focused on detecting the underutilized or vacant
bands of spectrum, however, with new challenges and dimensions in CRNs, sensing frequency
only may not be enough [9]. Thus it requires exploration of new dimensions of direction of
arrival (DOA), frequency, strength of signal, range and a critical parameter which is the number
of active PUs. In order to ensure secure, reliable and efficient communication keeping in view
the privilege of PUs, advanced SS algorithms capable of identifying occupancy in all of the
above dimensions of spectrum space to locate spectrum holes need to be developed, which have
not been considered simultaneously in CRNs yet according to the best of our knowledge.
The concept of cooperative diversity is based on the idea of introducing multiple nodes between
source-destination pair in such a way that each intermediate node listens to the signal transmitted
by the source. These partners are known as “relays” in wireless communication terminology
[10]. The relays cooperate with each other and behave like a virtual array of transmit antennas to
facilitate the source-destination pair in their communication even in worst-case scenarios when
direct communication is not possible between them due to deep fading, shadowing etc. The
whole study considers dual-hop one-way relay network, in which each relay is equipped with a
single antenna and one complete data transmission occurs in two time-slots, to be explained in
detail in the later chapters. Relaying efficiently improves system throughput, combats channel
fading effect, reduces power consumption, increases transmission reliability, and extends
coverage area [11].
Cooperative diversity techniques have been extensively utilized in the CRNs. Cooperation
between SUs [6]-[7], and cooperation between PUs and SUs [8] are two approaches followed for
this purpose. Cognitive relay networks, inspired by CR and relay networks, exist as a versatile
choice to assist SUs and extend their coverage area, but employing all relays in a power-
restricted environment may cause high interference to the concurrent primary transmissions.
Page 27
4
Relay selection has been observed as a fascinating technique to solve the aforementioned
complex interference-mitigation issue faced by multi-relay networks. In this context, several best
and multiple relay selection schemes have been proposed, with relatively less contributions
observed in the area of multiple relay selection. Furthermore, a lot of effort has been done to
investigate the performance of best relay selection in terms of outage probability and bit error
rate, but no one has carried out derivation for outage probability and bit error rate in closed form
for multiple relay selection in AF based underlay networks.
1.2 THE SYSTEM MODEL
Fig. 1.2. illustrates the cooperative dual-hop CRN comprising M randomly distributed cognitive
relays, a source S , and a destination D . The whole relay network operates in the vicinity of a PU
Q . Each node in the network is equipped with a single antenna, thus simultaneous transmission
and reception is not possible. Rayleigh flat fading is assumed for the whole scenario, in which
M
mmg1 , M
mmh1 and M
mmf 1 denote the independent and identically distributed (i.i.d.) channel
coefficients between source-relay, relay-destination and relay-PU respectively.
Fig. 1.2: The System Model
S
D
R1
R2
2
R3
RM
Q
gm
hm
fm
Source
Destination
PU
Relay Network
Page 28
5
Some assumptions made for the whole study are as follows. 1) Instantaneous channel state
information (CSI) is available at each secondary node. Moreover, the relays have perfect
knowledge of their forward and backward channels. 2) Binary Phase Shift Keying (BPSK)
modulation is used, hence, the symbol SS P,Ps , where SP represents the symbol power
transmitted by the source. 3) The relays operate in half-duplex mode, so they are unable to
transmit and receive on the same frequency simultaneously. 4) Amplify-and-Forward (AF)
relaying is assumed at the relay network. 4) Line-of-sight path suffers from deep fading, thus
making direct communication between source-destination pair impossible. 5) Underlay spectrum
sharing model is assumed at the relay network, thus the secondary communication can only take
place if the total interference offered to the PU by the potential relay network remains below a
predefined threshold thI , which is the maximum tolerable interference level for the PU. 6)
Additive white Gaussian noise (AWGN) with zero mean unit variance is assumed for each hop.
Based on the third assumption, one data transmission is completed in two time-slots. The source
transmits a symbol s in time slot 1 and the received signal my at the thm relay is given as:
mmsm sgPy 1 )1.1(
where, )1,0(~1 Nm is modeled as AWGN at the thm relay. In time slot 2, the destination
receives the scaled version of the received message from the relay network while the source is
silent. The signal Dy received at the destination D is expressed as:
DmmD hyy )2.1(
where, )1,0(~ ND represents AWGN with variance 0N , received at the destination. The signal
my amplified according to AF scheme is given as:
m
ms
mm y
NgP
Py 1
0
2|| )3.1(
Page 29
6
where, mP represents the transmit power of thm relay in the above eq. adjusted via AF protocol
and max0 PPm . maxP is the maximum transmit power each thm relay is allowed to transmit. The
transmission power mP of each relay is limited not only by the battery capacity due to regulations
specifying the maximum power that each node is allowed transmit, but also by the interference
threshold of the PU.
Substituting (1.1) and (1.3) in (1.2) and solving the resulting expression, end-to-end signal-to-
noise ratio (SNR) m of the thm relay link can be expressed as [12]-[14],
0
2
0
20
2
0
2
||||1
||||
N
hP
N
gP
N
hP
N
gP
mmms
mmms
m
Or in compact form,
mm
mmm
21
21
1
)4.1(
where, 0
2
1
||
N
gP msm and
0
2
2
||
N
hP mmm denote the instantaneous SNR achieved at the
source-relay and relay-destination links respectively.
The total instantaneous end-to-end SNR D at the secondary destination due to M relaying
links is then given by [15],
M
m mm
mmM
m
mD1 21
21
1 1
)5.1(
In underlay networks, sophisticated signal processing techniques are employed to mitigate the
interference offered to the PU. However, due to the inherent simplicity of AF protocol, such
Page 30
7
computationally complex techniques may not be supported at the AF relays. Thus, enabling the
secondary communication exploiting the services of all relays in the potential relay set may not
be a viable idea in terms of total interference offered by the relay network to the nearby PU.
Alternatively, suppressing the transmit power of candidate relays reduces the interference power
but at the same time makes it difficult to enable the secondary communication with minimum
QoS requirements. For simultaneous primary and secondary transmissions in such energy-
constrained environment, the total interference power experienced by the PU due to the
transmissions of relay network must satisfy the predefined interference threshold given as,
th
M
m
mm
M
m
m IfPII 1
2
1
|| )6.1(
where thI is the interference threshold set by the PU.
Relay selection stands as a fascinating solution to this problem. Therefore, relay selection is
performed to choose the best combination of relays that maximizes the SNR achieved at the
destination keeping in view the privilege of the PU. Thus eq. (1.5) takes the form,
SS
Dm mm
mm
m
m21
21
1
)7.1(
where, S denotes the selected subset of relays.
This dissertation proposed multiple relay selection and power allocation schemes in this
framework to select the best subset of relays S which meets the objectives and constraints.
1.3 OBJECTIVES AND CONTRIBUTIONS
This dissertation highlights several issues in CR technology. The objectives of this dissertation
are to study the SS issues in CRs, and the design issues in AF based cognitive relay networks
operating in an underlay mode of spectrum sharing, aiming to highlight the deficiencies found in
the literature. For this purpose, a detailed study of best and multiple relay selection schemes and
Page 31
8
the existing SS techniques has been carried out, paying special attention on the scenarios, in
which line-of-sight path between secondary source-destination pair undergoes deep fading, thus
making direct communication impossible. The main contributions of this dissertation are
summarized as follows:
Various multiple relay selection schemes have been proposed in the setups of Rayleigh
flat-fading scenario assuming availability of perfect instantaneous CSI. The effects of
variations in the instantaneous CSI, transmit powers of source and relays, interference
threshold of the primary network, SNR threshold of the secondary network and size of
potential relay network on multiple relay selection in underlay cognitive relay networks
are the main subjects that are studied in this research.
Outage behavior of secondary network for multiple relay selection is investigated and
closed-form expressions for outage probability and bit-error rate are derived through the
CDF of the SNR received at the destination.
FRBS assisted multiple relay selection and transmit power allocation schemes are
proposed aiming to enhance secondary performance, which is another new contribution
to underlay cognitive relay networks.
A novel idea of SS is proposed for CRNs. The proposed scheme not only detects the
number of active PUs, but also provides the estimates of their parameters such as
frequency, power strength and Direction-of-Arrival (DOA) upto high accuracy.
The performance of each proposed scheme has been evaluated under different design
requirements, assumptions and constraints.
1.4 ORGANIZATION OF DISSERTATION
The dissertation has been written in a manuscript style. The contributions in the form of
published and submitted manuscripts are included as the central body of the dissertation.
Footnotes are also added for clarification where necessary. The chapter wise distribution of the
dissertation is organized as follows:
In Chapter 2, current spectrum management policy is introduced and the fundamental knowledge
of CR starting from history to the current research challenges is discussed. After going through
Page 32
9
the comprehensive survey of this emerging technology, the discussion proceeds towards
cooperative communication and develop the understanding of the main concepts involved in
cognitive relay networks. Finally, some best and multiple relay selection schemes proposed for
the interference-constrained cognitive relay networks are surveyed and their findings are
highlighted. The system model and the basic assumptions made for this study are also explained.
Chapter 3 discusses the tools used in the performance evaluation of different schemes. A review
of the nature inspired evolutionary algorithms and artificial intelligence is carried out, aiming to
understand their advantages, properties, limitations, and applications. In this context, the phases
of Artificial Bee Colony optimization, Genetic Algorithm and Fuzzy Logic are further discussed
in depth with the aid of flow charts and block diagram where necessary, since these tools are
employed to solve relay selection and spectrum sensing problems in the dissertation.
Chapter 4 highlights the contributions in the area of multiple relay selection, aiming to maximize
the secondary performance under strict interference constraints imposed on the cognitive relay
network. The effect of individual relay’s transmit power constraint, entire relay network’s
transmit power constraint and interference threshold levels are studied for in-depth analysis.
In chapter 5, a problem of transmit power minimization is formulated for cognitive relay network
under same design assumptions. A multiple relay selection scheme is proposed in this regard,
aiming to minimize the total power consumed at the relay network, while satisfying minimum
quality-of service (QoS) requirements of both primary and secondary networks.
Chapter 6 highlights the deficiencies found in literature regarding the performance analysis of
multiple relay selection in underlay networks, and derives the closed form expressions of the
outage probability and bit error rate of the SNR received at the destination, which is another new
contribution to the cognitive relay networks operating in an underlay spectrum sharing
environment.
Chapter 7 presents a novel idea of using FRBS assisted intelligent relay selection schemes for
relay-assisted CRNs. The proposed schemes takes the CSI of each candidate relay in the
Page 33
10
potential relay network as an input and assigns relay selection factor (RSF) to each relay. The
RSF eventually sets the precedence in which the relays are selected, aiming to enable the
coexistence of the primary and the secondary networks.
The last topic that is investigated in chapter 8 is SS. In order to preserve the PUs’ rights of
interference-free operation, the SUs are required to sense the licensed bands at regular intervals,
and reliably detect the primary signals. A novel idea of uniform linear array (ULA) based SS is
proposed, which not only detects the number of active PUs, but also provides the estimates of
amplitude, frequency and DOA of the active users upto high accuracy.
In chapter 9, the whole research work is concluded and a comprehensive summary is provided.
We also put forward some future directions to extend the proposed techniques.
Page 34
11
Chapter 2
COGNITIVE RADIOS AND COGNITIVE
RELAY NETWORKS
Objective
This chapter discusses the current spectrum management policies and the motivation behind the
CR technology. After a comprehensive review of the history of the proliferating CR technology,
the research challenges and the deficiencies in designing the CRNs are discussed. Furthermore,
cognitive relay networks are included in the discussion and different single and multiple relay
selection schemes are surveyed and their findings are highlighted for power-constrained CRNs.
2.1 HISTORY OF COGNITIVE RADIOS
The wireless applications are proliferating very rapidly and large variety of communication
systems exist for different applications in the licensed and unlicensed frequency bands [16]. The
global wireless communication standards include personal area networks (IEEE 802.15), local
area networks (IEEE 802.11), metropolitan area networks (IEEE 802.16) and wide area networks
(IEEE 802.20). The flourishing wireless technology has urged the deployment of mesh networks.
A careful survey shows that up to 1 trillion wireless devices are expected to be operational by
2020. Moreover, due to non-line-of-sight propagation, radio propagation favors the use of
spectrum below 3GHz in the entire radio spectrum ranging from 3MHz to 300GHz. According to
current spectrum regulatory framework, the licensed frequency bands have been exclusively
allocated to the specific services [17]. The stringent spectrum assignment used by Federal
Communications Commission (FCC) is provided in Fig. 2.1.
Page 35
12
Fig. 2.1: FCC Spectrum Allocation Chart
This “command and control” strategy of spectrum allocation and management effectively
protects the authorized users from unwanted interference of other users of the radio spectrum on
one hand, and does not allow the users to switch between highly occupied and underutilized
bands of frequencies on the other hand [18]. This ever increasing growth of wireless services,
huge demand of internet access, evolution of smart phones and spectrum analysis demand more
and more spectrum resources, which creates a scenario of spectrum scarcity. The need of
continuous and fast technological evolution creates the demand of new dedicated spectrum
bands. Another strong observation revealed through the careful analysis of FCC in 2002 is the
underutilization of most of the licensed spectrum bands allocated to the current wireless
subscribers, in temporal and spatial domains [19]. FCC reports show that the variation in the
licensed spectrum occupancy ranges from 15% to 85% [20]. This underutilization stems from the
existing fixed spectrum allocation strategies for the valuable resource as mentioned above. For
example in US, a frequency band of 512-608 MHz is dedicated for television broadcasting for
channels 21-36, while the frequencies from 960-1215 MHz are reserved for radio navigation
[21]. Spectrum management is one of the major responsibilities of the communication regulators
and more efficient spectrum utilization can result by better spectrum management.
Page 36
13
Earlier in 1993, Joseph Mitola III proposed a novel idea of dynamic spectrum sharing of the
spectrum bands exclusively allocated to the licensed users using novel devices called Cognitive
Radios [22]. Mitola envisioned CR as a combination of existing wireless technology and
artificial intelligence. The need for CRs is primarily motivated by the complexity of radio
systems itself. Conventional radio design is targeted for single purpose and single environment.
The key enabling technology for cognitive wireless networks is the software defined radio (SDR)
which first emerged in 1990 [23]. SDR aims to bring radio electronics into the digital age, thus
adding new degrees of freedom in designing wireless networks by enabling radios to adapt to the
requirements at hand [24]. These radios perform signal processing in software, thus enabling the
devices that can be reconfigured via software after deployment. SDRs find their applications in
industry, academia, government and military organizations, communication research, data
acquisition and many more. However, the radios built on SDR technology are expensive, since
they support multiple interface technologies e.g. GSM, CDMA with a single modem by
reconfiguring it in software. Thus, a CR is an SDR, that is fully programmable to interact with its
operating environment and dynamically adapts its parameters i.e. carrier frequency, modulation
technique, transmit power, channel access method and networking protocols to deliver the best
application performance. However, the fundamental differences between CRs and SRs are as
follows [25]. First, contrary to CR technology, the SDRs do not have the ability to sense and
detect the unoccupied or partially occupied slots in the spectrum. Second, CRs are capable of
operating at any frequency in the entire radio spectrum, whereas, SDRs are designed for certain
standards and their assigned frequency bands. Third, SDRs are built on the availability of a priori
knowledge of the interfering channels, whereas, the CRs have the ability to tolerate interference
at any frequency in the bandwidth defined for cognitive wireless devices.
Based on the above facts and requirements, FCC took the initiative in 2004 by allowing the
unlicensed users to utilize television spectrum in the regions, where spectrum is not in use.
Broadly speaking, the term CR refers to various solutions of the spectrum underutilization
problem by enabling transmissions from the unlicensed wireless devices in the underutilized
licensed frequency bands, in such a way that the licensed users are as uninterrupted as possible.
Thus, the CR devices are capable of detecting the spectrum holes1 and dynamically and
1Spectrum hole is a band of frequencies primarily assigned to PU, but at a specific time and location, is not in use by the PU
Page 37
14
autonomously adapt themselves to the changing network conditions to deliver the best possible
performance. The networks of such nodes are recognized as CRNs. Fig. 2.2 illustrates the
concept of spectrum hole.
Fig. 2.2: The concept of spectrum hole
The licensed users are known as the primary users (PUs) in CR terminology, the unlicensed
users, who are given opportunistic access of the spectrum, are known as secondary users (SUs)
or cognitive users, the underutilized bands of frequencies are known as “white spaces” or
“spectrum holes”, and such opportunistic spectrum sharing in a noninterfering manner is known
as dynamic spectrum access (DSA).
Thus each transmission process in CRs is completed in two steps. First is spectrum sensing to
detect vacant or underutilized bands of spectrum and second is transmission of data of the source
towards the destination. Both the spectrum sensing [26]-[27] and data transmission phases have
been extensively studied.
Simon Haykin in 2005 identified three specific tasks that lie in the core of CR [3]. First, radio-
scene analysis to be performed by the SU receiver, which includes detection of spectrum holes
and estimation of interference temperature around a PU receiver. Second, channel identification
Page 38
15
which is required by the SU receiver to improve spectrum utilization efficiency and to perform
coherent detection of the PU signal. Third, dynamic spectrum management and transmit power
control to be performed by the SU transmitter. For this purpose, there is an obvious need to
connect the transmitter and the receiver via dedicated feedback channel to keep harmony
between both devices and to mitigate the interference offered to both the licensed and unlicensed
users.
CR is emerging as a promising technology for future wireless world. The salient features of
CRNs are spectrum sensing, spectrum management, spectrum sharing, mission-oriented
configuration, adaptive algorithms, distributed collaboration, routing and security [6]. CRs have
made rapid transition from an idea to reality during the last decade. The motivation behind this
remarkable progress to enable CR technology is the deployment of mature DSA systems. An
extensive research is being conducted by CR research community in the areas of spectrum
sensing and management, CR architecture, cooperative communication, dynamic spectrum
access algorithms, protocol architectures for CRNs, resource management, security issues and
finally large-scale deployment [6],[28]-[29].
Moving one step further, the large-scale deployment of CR technology heavily depends on the
interconnection of these intelligent devices, which work in collaboration to enhance the overall
system performance forming CRNs. In literature, infrastructure-based CRNs, cognitive ad-hoc
networks and hybrid networks have been proposed for CRNs [30]. Being wireless in nature,
CRNs are prone to all security threats inherent in conventional wireless networks [31]. The
common security threats are confidentiality, integrity, availability and access control.
2.2 SPECTRUM SHARING IN COGNITIVE RADIOS
A natural question is to explore the means by which the SUs can be accommodated in the
licensed spectrum bands without disrupting the PUs of the spectrum. Broadly speaking, three
spectrum sharing models proposed in CRs are underlay, overlay and interweave [5],[32].
Page 39
16
2.2.1 UNDERLAY SPECTRUM SHARING
The underlay spectrum sharing follows the ultra-wideband (UWB) system strategy to support
simultaneous primary and secondary transmissions over same frequency bands. Underlay CRNs
guard the primary transmissions by enforcing a spectral mask on the secondary signals so that
the PUs remains undisturbed of the interference caused by the secondary transmissions. To
compensate the spectral masking, secondary signals are spread and de-spread over a wide
bandwidth to provide desired SNR at the secondary receiver. The main problem faced by
underlay networks is the limited transmit power ability of the secondary users due to the
interference constraints, which reduces secondary throughput and allows only short-range
communication. Thus, enabling secondary communication with minimum QoS in the frequency
spectrum allocated to the PUs is hot area of underlay research, and it requires fine tuning and
adjustment of the transmit power of the SUs. In [33], the authors suggested a transmit power
allocation scheme for dual-hop CRNs operating in AF mode, under transmit power constraints
and interference constraints. First, the optimization problem was simplified by relaxing the
transmit power constraint to obtain sub-optimal solution, which was then further utilized to
propose a power allocation scheme in order to satisfy both constraints all the time.
2.2.2 OVERLAY SPECTRUM SHARING
The overlay spectrum sharing enables simultaneous primary and secondary transmissions. For
this purpose, the SUs use a part of their transmit power for secondary communication, and the
remaining power to relay primary signals. This versatile mode of communication offsets the
decrease in PU’s SNR due to the interference caused by the SU’s transmit power by exactly
increasing the PU’s SNR due to the relaying services provided by SU. Sophisticated signal
processing and coding techniques are also employed for interference mitigation depending on the
available side information.
2.2.3 INTERWEAVE SPECTRUM SHARING
The interweave spectrum sharing is primarily based on the idea of opportunistic spectrum access,
i.e. to exploit the spectrum holes, that are temporarily not in use by the primary users. These
Page 40
17
spectrum gaps highlight the spectrally underutilized regions in terms of time and space, and can
be used to accommodate the secondary users to enhance the spectral efficiency. The spectrum
holes vary with time and geographic location. Thus, they need to be periodically monitored to
intelligently detect unoccupied parts of the spectrum. The periodic scanning of the spectrum aims
to avoid the interference to the primary users once they are active again. Due to these reasons,
interweave mode is also known as interference avoidance mode of spectrum sharing.
2.3 COPERATIVE COMMUNICATION
Multiple-Input Multiple-Output (MIMO) systems [34], an extension of developments in antenna
array communication, employ multiple transmit and receive antennas as shown in Fig. 2.3, and
provide a number of advantages over single-antenna-to-single-antenna communication. These
advantages are: less sensitivity to fading effects of communication channel and increased
resistance to local interference due to existence of multiple spatial paths between transmit-
receive antennas, improved gain, reduced power requirements, capacity enhancement and many
more. However, these advantages are gained at the cost of implementation complexity and
increased size, which many wireless devices may not be able to support. Cooperative diversity
was proposed as an alternative potential solution.
Fig. 2.3: Multiple Input Multiple Output (MIMO) System
1
2
M
1
2
N
Transmitter Receiver
MIMO Channel
Page 41
18
Cooperative communication [35]-[36] is an effective means of increasing the spatial diversity of
a signal in wireless communication networks. It has been recorded in [37]-[38] that cooperative
diversity networks can achieve a diversity order equal to the number of end-to-end source-
destination routes, however, capacity enhancement is limited by the transmission of symbols in a
time-division multiple access (TDMA) manner. It lays down the foundation of adhoc networks
and holds the promise of supporting sensor networks, communication networks for providing
public safety, strategic networks, cellular networks which are hierarchical in nature and military
applications [39].
Such communication strategy efficiently improves system throughput, combats channel fading,
reduces power consumption, increases transmission reliability and coverage area [40]-[43].
Cooperative communication techniques follow such approaches as collaborative signal
processing, cooperative coding and relaying [44]. Relaying [45] is a powerful cooperative
diversity technique in which multiple spatially distributed terminals, commonly known as relays,
assist the source by relaying its information to the destination. In such communication strategy,
the probability that all links are simultaneously down is much smaller than that for a single link
[39]. Relay networks [46], introduced by Van Der Meulen [3], have attracted tremendous
research attention and particularly find their applications in the networks with transmit power
constraints and portable mobile terminals, where mounting multiple antennas is difficult. A relay
network operates in either full-duplex or half-duplex mode [46]. In full-duplex mode of
operation, a relay can simultaneously transmit and receive at the same frequency, whereas, in
half-duplex mode, simultaneous transmission and reception on the same frequency band is not
supported on the relay.
2.3.1 RELAYING PROTOCOLS
In order to implement a cooperative communication network, efficient relaying strategies and
received signal combining schemes have been developed. Depending on the signal processing
strategy employed at the relay, two primary and most widely employed relaying techniques are
Amplify-and-Forward (AF) and Decode-and-Forward (DF) [41],[46]-[47].
Page 42
19
2.3.1.1 Amplify-and-Forward Relaying
In AF relaying, also known as non-regenerative relaying, a relays simply adjusts the
amplification of the received signal and retransmits it, thus requiring less processing and low
power consumption at the relay [46]-[47]. AF relaying aims to overcome the power loss on the
source-relay link, however, a very important issue encountered by AF protocol is the
amplification factor required at the relay for scaling the received signal, which may result in an
unbounded power at the relay. In practice, peak transmit power constraints apply on the analog
circuitry involved in the communication devices. [48]. This unboundedness of the relay
amplification factor may result in peak power saturation and harmful interference to other
cochannel users, for example, CRs.
2.3.1.2 Decode-and-Forward Relaying
DF relays are more complex than the AF relays since they decode the received message, perform
error correction, encode the corrected message and retransmit it. Generally speaking, in DF
mode, source broadcasts its data which is heard by the destination and the relays. The set of
relays which are able to decode the received signal successfully constitute a set called the
decoding set [49]. DF protocol can be applied for both the coded sequences and the uncoded
signals. For coded sequences, coded DF is used in which error correction codes are added to the
symbols at the time of transmission [50]. The relays validate the reliability of the received signal
upon decoding via known error correction code. For uncoded symbols, a relay becomes a
member of the decoding set, if the received signal-to-noise (SNR) ratio at the relay exceeds a
predefined SNR threshold. Different variants of AF and DF protocols are already mature in
literature, proposed to enhance their accuracy and efficiency.
In addition, another relaying strategy, although less popular than AF and DF relaying is
Compress-and-Forward (CF), which is employed in the situations when a relay is unable to
decode the received signal and it sends an estimate of the source’s message to the destination
[51]. Fig. 2.4 shows the basic principle of AF and DF relaying.
Page 43
20
Fig. 2.4: AF and DF Protocols
2.4 COGNITIVE RELAY NETWORKS
The combination of CR and user cooperation emerged as cognitive relay networks [31],[52],
which considerably improve the bandwidth efficiency, tackle unfavorable effects of wireless
channels and improve performance tradeoffs for both the PUs and the SUs [53]-[55]. In CR
research, cooperative communication is being extensively applied in spectrum sensing and
sharing, dynamic resource allocation, interference management. Cognitive relay networks follow
one of three approaches [56]. The first approach involves mutual understanding of primary and
secondary users whereby and the SUs act as relays to assist the PUs in their transmission, which
in turn provides more transmission opportunities to the SUs. The second approach is based on
the collaboration among SUs and in this scenario, the SUs relay signals for each other. In third
approach, spectrum-rich SUs help spectrum-short SUs and such communication strategy is
known as cooperative relaying. In addition to the design challenges inherent in the single-hop
CRNs, multi-hop CRNs also face research challenges related to resource sharing among different
nodes and performance enhancement of secondary network keeping in view the privilege of PUs
[55].
In order to deliver the best performance in cognitive relay networks operating in an underlay
spectrum sharing environment, the SUs, including the source and the relays, are allowed to
transmit concurrently with the primary transmissions over the same frequency band as long as
Page 44
21
the interference offered by the SUs remain below a predefined threshold [52]. The interference
threshold can be defined by average or instantaneous interference power received at the primary
receiver [56]. The instantaneous or peak interference power requires the knowledge about
instantaneous channel gains of the interference channels and it is suitable for real-time traffic.
The average interference power applies to non-real time traffic where average SNR determines
the QoS. However, in practice, it is very hard to determine the interference threshold. In [57], the
instantaneous CSI between the secondary transmitter and the primary receiver along with the
instantaneous CSI of PU link determine the interference threshold.
2.4.1 RELAY SELECTION IN COGNITIVE RELAY NETWORKS
As mentioned above, the performance of underlay CRNs is significantly enhanced by
incorporating cognitive relays, which convey the message transmitted by the source to the
destination, but engaging the whole relay network may not be a feasible idea because the
interference produced by the relays to the concurrent primary communication may exceed the
threshold [58]. This practical limitation demands efficient alternatives of all-relays participation
in cognitive relay networks. Fortunately, cognitive relay networks offer a fascinating solution to
this problem in the form of relay selection. Relay selection aims to select the best combination of
relays or single best relay keeping in view the objectives and constraints of the system under
consideration. Owing to the half-duplex mode of communication [56], relay selection is
performed in two time-slots as shown in Fig. 2.5. In time slot 1, the message broadcast by the
source is heard by the potential relay network, whereas, in time slot 2, the selected relay(s)
retransmit the received message to the destination after necessary processing, while the source is
silent.
(a) (b)
Fig. 2.5: Conceptual Model for Relay(s) Selection, (a): Time Slot 1, (b): Time Slot 2
Source
broadcast
s Source Destination
Destination
Page 45
22
However, a critical parameter to be considered in relay selection is the selection speed. Owing to
the time-varying nature of communication links, relay selection must be performed no slower
than the channel coherence time which is defined as the time duration over which the channel
impulse response is considered to be flat. If relay selection scheme fails to satisfy this constraint,
this might degrade the overall system performance by selecting a wrong relay because in this
case, the selection entirely based on the old CSI, while the channel conditions are changed at the
time selection is performed.
In this dissertation, relay selection in the context of AF based cognitive relay networks operating
in underlay mode is investigated in detail. Several best and multiple relay selection schemes have
been studied in this regard based on different design requirements and assumptions.
Note: “Multiple Relay Selection” and “Relay Subset Selection” terms will be used
interchangeably in the whole dissertation, where Relay Subset Selection aims to select multiple
relays from a potential relay set.
2.4.1.1 Best Relay Selection Proposed In Literature
Few research contributions involving best relay selection in underlay CRNs under interference
constraints are highlighted as follows. In best relay selection, only the single “best” relay, which
satisfies an index of merit, is nominated as a selected relay to participate in the communication.
In [59], Fredj et. al presented a scenario in which a secondary transmitter used the services of
intermediate relays to communicate to its receiver. In this scenario, best relay was selected from
the potential relay set to enable secondary communication under interference constraints.
Furthermore, end-to-end SNR statistics were derived and BER was evaluated for different
modulation schemes. D. Li investigated best relay selection based on full and partial CSI in [60],
and compared the performance of both schemes by deriving the closed-form expressions for
outage probability. For this purpose, a cluster of cognitive relays assisting a single source-
destination pair was considered. It was proved that partial-CSI-based relay selection was
outperformed by the full-CSI based relay selection. In [61], Seyfi et. al proposed a best relay
selection scheme for dual-hop cognitive relay network under transmit power constraints and
interference constraints. Furthermore, the outage probability of the secondary network with relay
selection was derived while considering the effect of PU interference. The derived results were
Page 46
23
tested through simulations. In [62], Hussain et. al considered a cognitive source communicating
with its destination through direct transmission path and also with the help of intermediate fixed-
gain cognitive relays. In this scenario, best relay selection criteria was proposed aiming to
maximize the SNR achieved on relay-destination link. Furthermore, outage probability and BER
were derived in closed form for performance analysis. In [63], Bao et. al proposed best relay
selection and considered tight lower bound of the end-to-end SNR to derive the closed-form
expressions for CDF and probability density function (PDF) over non-identical Rayleigh fading
channels. The derived results were used to investigate the outage probability and average symbol
error probability of proposed system. The performance was evaluated against some key
parameters. The asymptotic analysis of the scenario showed that interference constraint does not
affect the diversity gain. In [64], the authors carried out derivations of the outage probability and
symbol error rate for cognitive relay networks under interference and transmit power constraints
over Rayleigh fading channels. It was shown through numerical results that the transmit power
constraint and interference power constraint cause the outage saturation phenomenon. The
analytical results were validated through Monte Carlo simulations.
2.4.1.2 Multiple Relay Selection Proposed In Literature
Research contributions in the area of multiple relay selection are, however, quite limited. In [65],
the authors proposed optimal and two suboptimal schemes of multiple relay selection in
cognitive relay networks with an objective to maximize the SNR received at the destination
under interference constraints. The comparative analysis of all schemes was carried out against
well-defined range of source transmit power for different interference threshold levels and
different number of candidate relays. Naeem et. al considered a dual-hop CRN, and proposed a
multiple relay selection scheme with interference awareness for underlay CR systems in [66]. It
was proved through simulations that the performance of the proposed scheme approached
exhaustive search technique, while having low implementation complexity.
2.4.1.3 Observations In The Literature Review Of Relay Selection Schemes
All the above mentioned research contributions for best and multiple relay selection are selected
for discussion, because they were built on some common assumptions which are as under. First,
underlay spectrum sharing model was assumed for each scenario. Second, all schemes, except
Page 47
24
[62], assumed severe shadowing on the line-of-sight path between source-destination pair, thus
making direct communication impossible. Third, all system models were built up using single
antenna terminals. Fourth, AF relaying was assumed at the cognitive relay network. Fifth, all the
highlighted contributions for best and multiple relay selection assume the availability of CSI of
the interference channels.
Each of the proposed scheme studied above has been analyzed with interference and transmit
power constraints. However, there is a strong observation that all relay selection schemes
performing best relay selection study the outage behavior of the secondary network in detail. On
the other hand, the effect of relay subset selection on the outage probability and bit error rate of
the secondary system operating in an underlay spectrum sharing environment is not presently
available in literature, to the best of our knowledge.
These prior works have significantly improved our understanding of relay-assisted CRNs.
Inspired by the contributions in the field of relay selection, we focus on the deficiencies
highlighted in the performance evaluation of multiple relay selection schemes.
2.5 CONCLUDING REMARKS
This chapter starts with a comprehensive introduction to CRs alongwith the salient features,
different spectrum sharing modes and research challenges faced by this emerging technology.
Then the discussion proceeds towards the cooperative communication techniques highlighting
the significance of relay networks. Furthermore, cognitive relay networks have been discussed as
a merging technology of CR and cooperative communication. Finally, several contributions in
the areas of multiple relay selection and best relay selection proposed in literature for underlay
cognitive relay networks have been highlighted.
Page 48
25
Chapter 3
NATURE INSPIRED ALGORITHMS AND
FUZZY LOGIC
Objective
This chapter discusses the history, advantages, optimization phases and applications of the tools
used in this disseration. After a comprehensive review of the history and applications of
Evolutionary Algorithms (EAs), two versatile EAs, namely, Artificial Bee Colony and Genetic
Algorithm are discussed in detail. Finally, the discussion moves towards Artificial Intelligence
(AI), and a well-known AI tool, the Fuzzy Logic is studied in depth.
3.1 THE EVOLUTIONARY ALGORITHMS
The difficulties associated with the mathematical modeling of large-scale engineering
optimization problems seek to develop alternative solutions. An optimization problem is the one
which aims to find out the best solution from all candidate solutions and requires specialized
problem solving techniques. When optimization is to be performed within complex domains of
available information, bio-inspired EAs based on the behavior of biological entities, have
emerged as a fascinating area in this framework [67]. The behavior of social insects, such as
finding the best food source, building of optimal nest structure, clustering etc. show intelligent
behavior on the swarm level. The swarm behavior heavily depends on the interactions among
individuals in addition to the behavior of individuals.
EAs are derivative-free methods, which belong to the class of probabilistic optimization method
[68] and perform very well to solve non-convex and non-differentiable problems [69]. The
classical and the most prominent EAs are Genetic Algorithm (GA), Artificial Bee Colony
Page 49
26
(ABC), Particle Swarm Optimization (PSO), Artificial Immune Optimization (AIO), Differential
Evolution (DE), Invasive Weed Optimization (IWO), Ant Colony Optimization (ACO), and
Simulated Annealing (SA) [69],[70]-[71]. The suitability of any optimizing technique [68] for a
particular application is a compromise between accuracy, speed, complexity, memory demand,
utilization of a priori information, and balancing of global search (exploration) and local
refinement (exploitation).
During the past decades, these nature-inspired algorithms have gained tremendous attention by
research community, and have been successfully adopted in many applications requiring
optimization finding the optimal routes, scheduling, image and data analysis etc. for finding
near-optimum solutions of the optimization problems [72]-[75]. Broadly speaking, EAs find their
applications in engineering, social sciences, arts, economics, robotics and all fields of real-world.
Furthermore, most of the EAs are able to intelligently solve epistatic problems, in which the
quality of one variable is highly dependent on the other. EAs are also found to be robust in
nature, i.e. similar results are obtained from multiple runs of EA to solve the same problem.
Moving one step further, multiobjective evolutionary algorithms (MOEAs) have already been
developed to solve optimization problems involving multiple conflicting objectives in science,
economics and engineering [67],[76]. In such problems, MOEAs aim to identify a set of all
possible solutions rather than one optimal solution obtained in the case of single-objective
evolutionary algorithms (SOEAs) [67]. The set of all feasible solution represent the best
compromise between the multiple objectives defining that particular problem. These algorithms
can be classified into aggregating function algorithms, population based algorithms and Pareto
based approaches.
However, a common problem associated with all EAs is the imbalance between exploration and
exploitation. High degree of exploitation results in premature convergence to local minima, and
on the other hand excessive exploration slows down the execution [77]. Thus, a severe limitation
of the population based global optimizers is the lack of ability to do fine-tuning of the obtained
results. The global optimization tools are found to be good in exploration of the search space but
less good in the exploitation [78]. On the other hand, the local search algorithms like pattern
Page 50
27
search (PS), Interior Point Algorithm (IPA) efficiently improve the accuracy of the results.
Hybridization of global optimizers with local optimizers came up as a fascinating solution to
tackle this issue, for instance, GA hybridized with PS [78],[79], Hybrid Evolutionary and
Gradient Search [41], Memetic PSO [80], DE hybridized with Sequential Quadratic
Programming (SQP) [81] are few prominent contributions in this regard.
Two famous global optimization algorithms employed to solve the optimization problems in this
research are Artificial Bee Colony and Genetic Algorithm. The history, applications and
optimization phases of these algorithms are discussed below.
3.2 ARTIFICIAL BEE COLONY OPTIMIZATION
Artificial Bee Colony (ABC) is a relatively new population based global optimization algorithm
proposed by Dervis Karaboga in 2005 [69], [74]-[75], [82]-[85] and it simulates the foraging
behavior of honey bees. In the initial phase, it was proposed to search the optimal solutions to
unconstrained problems [82]. Later, ABC and its extended versions have gained remarkable
attention of research groups due to ease of implementation, robustness, employing fewer control
parameters (mainly colony size and maximum iteration number), and good convergence
properties [85]-[87]. ABC performs smart handling of linear/non-linear constrained problems
and non-convex problems. Furthermore, it shows reduced computational overhead and does not
suffer from memory limitation problems since each candidate solution in the population is not
examined from the start to the end of optimization procedure [85].
ABC emulates honey bees intelligent behavior of searching for quality food source and sharing
that information with their fellows in the hive. A quality food source is the one, which contains
the highest amount of nector. Thus, the important decision making parameters in ABC execution
are: The amount of nector which corresponds to the fitness value associated with a particular
food source, and the number of iterations for which the algorithm is evaluated repeatedly [74].
ABC finds its applications in science, engineering and medicine [77], [88]-[89]. In order to
further enhance the accuracy of ABC and to overcome its limitations, certain modifications have
been carried out making it a versatile choice to solve multimodal and non-differentiable
Page 51
28
problems [90]. These enhancements enabled ABC to outperform other state-of-the-art
metaheuristic algorithms (GA, DE, PSO) in efficiency, effectiveness and speed of convergence
[69],[74]-[75],[83],[85],[88],[90].
3.2.1 OPTIMIZATION PHASES OF ABC
ABC algorithm is based on the notion of Artificial Bees and Greedy Search procedure. The
entire optimization procedure in ABC is divided into two phases:
3.2.1.1 Initialization Phase
In the Initialization Phase, potential solutions (food sources) are randomly generated.
3.2.1.2 Best Solution Search Phase
In this optimization phase, search processes of Artificial Bees are recursively operated until the
best solution is achieved or the maximum number of iterations is expired. The best solution is
memorized by means of Greedy Search approach in ABC.
Broadly speaking, the artificial bees are divided into two categories: Employed Bees (EBs) and
the Unemployed Bees (UBs). Unemployed Bees are further classified into Onlooker Bees
(OLBs) and Scout Bees (SB)s. Thus, ABC combines local (employed and onlookers) and global
(scouts) search methods to achieve global or near-global optimum solution [85]. The tasks
performed by these foraging bees are summarized as follows.
a) Employed Bees (EBs)
Employed Bees are the “search agents” which search for the neighborhood solutions in the
vicinity of the initialized solutions and update their memory using Greedy approach by the
best solution that improves the fitness function and satisfies the constraints. There is a
dedicated EB for each potential solution.
b) Unemployed Bees (UBs)
i. Onlooker Bees (OBs)
OLBs are the “selector agents” which rely on the information shared by EBs about the
discovered solutions and exploit only those solutions chosen according to the probability
Page 52
29
of their fitness function relative to the sum of all by Roulette wheel Mechanism. UBs
become EBs whenever they find a solution (food source) to act upon.
ii. Scout Bees (SBs)
SBs are the “replace agents” which carry out the random search in the whole search space
to replace the abandoned solutions by new ones. An abandoned solution is the one that
fails to improve the fitness function after several attempts w.r.t. the threshold level.
3.2.2 Flow Chart of ABC Optimization
For better understanding, main steps in ABC are summarized in the flowchart in Fig. 3.1.
Randomly initialize the population of SN number of solutions
Place EBs on the initialized solutions, kiYYYZ ikiiii ,10),(
SNi ,,2,1
Greedy Search between iY and iZ
Use Roulette Wheel to Spread Onlookers
Greedy Search to memorize the best solution
Discover new solutions via scouts
Memorize the best solution
Loop Expired or
criteria met? Return best solution
Yes No
Start
Fig. 3.1: Flow Chart of Artificial Bee Colony Optimization
Page 53
30
3.3 GENETIC ALGORITHM
Genetic Algorithm (GA), was developed by Holland in 1970 on the basis of Darwin’s Theory of
natural evolution [91]. Holland’s study of natural adaptation phenomena aimed to find out ways
to apply the principles of natural evolution to the optimization problems. Later in 1975, he
presented GA as an abstraction of natural evolution in his book “Adaptation in Natural and
Artificial Systems”, and provided a theoretical framework to build the first GA. The primary
search procedures in GA like other evolutionary models are natural selection and survival of the
fittest [92]. The population of candidate solutions or chromosomes in GA is updated in each run
by selecting the best chromosome and discarding the unhealthy one.
3.3.1 OPTIMIZATION PHASES IN GA
Three steps followed in GA to produce successive generations are selection, crossover and
mutation [92]-[93]. Selection aims to produce next generation individuals, whereas, crossover
and mutation are the methods for reproduction. We briefly explain each one as follows.
a) Selection
Selection procedure aims to generate intermediate population by choosing those
chromosomes for survival in the next generation which exhibit the finest fitness scores,
while the remaining ones are discarded. The commonly used methods of selection are
stochastic uniform, remainder, roulette and tournament, rank, and scaling.
b) Crossover
Crossover combines two parents from the intermediate population to create offsprings.
Different commonly used crossover criteria are single point, two point, intermediate,
heuristic, scattered and arithmetic.
c) Mutation
Mutation functions introduce genetic diversity by making small random changes in the
individuals within a population. The purpose of mutation is spreading the search to a
broader space, thus preventing the algorithm from being stuck in the local minima.
Page 54
31
Gaussian, uniform and adaptive feasible are the famous mutation function used for this
purpose.
3.3.2 FLOW CHART OF GA OPTIMIZATION
The sequence of steps followed by GA optimization can be better understood with the help of
flowchart shown in Fig. 3.2.
Fig. 3.2: Flow Chart Of GA Optimization
Create initial population
Parents’ Selection
Fitness evaluation of individuals
Fitness evaluation of children
Update population
Loop Expired
or fitness
achieved?
Return best solution
Yes No
Crossover to create children
Mutation
Start
End
Page 55
32
3.3.3 APPLICATIONS OF GA
GA finds its applications in different domains including multimodal optimization, power
electronics, bioinformatics, economics, scheduling applications, robotics, security (encryption
and decryption), communication networks design, adaptive modulation, all fields of engineering,
and many more. Moreover, GAs have been extensively employed to solve miscellaneous issues
in CR systems, such as cooperative spectrum sensing [92],[93]-[94] joint channel and power
allocation [95], relay selection and resource allocation issues [96]. Efficiency, accuracy and
reliability of GA can be considerably improved by hybridization with any other competent and
well-balanced computational technique such as Interior Point Algorithm (IPA), Pattern Search
(PS) etc.
3.4 FUZZY LOGIC
Broadly speaking, Artificial intelligence (AI) is the automation of activities that are linked with
human thinking. These activities involve decision making, problem solving, learning, perception,
and reasoning [97]. CRNs encourage the use of AI tools for reconfiguration to meet the
requirements of changing network conditions. The AI tools of interest extensively applied in
huge number of real-time applications include fuzzy logic (FL), adaptive fuzzy logic (AFL),
expert systems (ESs), rough set (RS) theory and artificial neural networks (ANNs) [72]. FL,
based on human perception and cognition, is a powerful variation of crisp logic, which closely
relates the knowledge representation to human thinking. FL holds the power of natural
knowledge representation as well as strong inference capabilities of expert systems. FL is being
utilized in this research to intelligently solve the relay selection problems and is explained in
more detail.
3.4.1 HISTORY OF FUZZY LOGIC
The notion of FL, was first introduced by Lotfi A. Zadeh, in 1965 [98]. In the initial phase, it
made a very slow development, but by early 1970’s, the world paid attention to this new theory
when Zadeh delineated the motivation behind fuzzy control in 1972 [99]. An important
breakthrough in this progress was made in 1973 [100], introducing the basic idea of a linguistic
Page 56
33
variable, defined as a variable whose values are linguistic terms rather than numbers. By the end
of 1970’s, FRBS captured the imagination of research groups and scientific laboratories started
implementation of fuzzy inference engines. After going through significant development over
five decades, FL systems are now used as a successful and dominant tool of AI.
3.4.2 FUZZY CONTROL SYSTEM
Block diagram of a fuzzy control system (FCS) is shown in Fig. 3.3 [101]-[103].
Fiure 3.3: Basic structure of FLC
Fig. 3.3: Fuzzy Control System
FCS is conceptually split into four components:
a) Knowledge Base
b) Fuzzifier
c) Fuzzy Inference Engine
d) Defuzzifer
The tasks performed by each component are summarized as follows.
a) Knowledge Base
The knowledge base contains all the knowledge required by the FCS and it comprises
fuzzy control rule base and a data base. The rule base illustrates the relations between the
input and the output variables through IF-THEN rules based on fuzzy reasoning. The rule
base must contain rules for every possible combination of the input space.
Crisp
Input
Crisp
Output
Fuzzy Input Sets Fuzzy Output Sets
Knowledge
Base
Fuzzification
Interface
Defuzzification
Interface
Inference
Page 57
34
b) Fuzzifier
The fuzzification interface fuzzifies each input linguistic variable to produce a set of
fuzzy numbers. This is done by comparing each input variable with the predefined MFs
to allocate the variable, a value between 0 and 1.
c) Fuzzy Inference Engine (FIE)
The heart of FCS is the inference engine, fed by the fuzzy numbers from the fuzzifier and
it derives reasonable control actions on the basis of predefined rule base.
d) Defuzzifier
The fuzzy variables produced by FIE are converted into the crisp values by the
defuzzifier to represent the actual output of the system.
3.4.3 APPLICATIONS OF FUZZY LOGIC
FL has been heavily employed in power systems to solve issue related to automatic power
restoration and control, power optimization, system diagnosis and stability, classifying PQ
disturbances, protection, fault diagnosis and load forecasting [72]. FL based computer aided
design (CAD) tools are used to address issues in analog and digital circuit design. In the industry,
FL has been successfully applied in the modeling of complex systems and smart handling of
design, manufacturing and control issues. One of the most demanding applications of fuzzy set
theory is pattern recognition [104]-[105]. In addition to the above highlighted applications, fuzzy
set theory has been applied in incredibly diverse real-world applications, including engineering
design, social science, robotics, economics, management, finance, web mining, heuristic control
and regression analysis. [73], [101],[106]-[109].
3.4.4 FUZZY LOGIC IN CRNS
FL adds more degrees of freedom to make sophisticated and reliable decisions, thus it has
already been employed in CRNs, especially in the areas of spectrum sensing, interference
management and power control [111]-[119]. Fuzzy logic based transmit power control schemes
result in simple, reliable and cost effective implementations. Some methods of FL based transmit
power allocation and power management in CRNs are highlighted as under. In [120], the authors
designed a fuzzy logic system (FLS) based transmit power controller to enable coexistence of
Page 58
35
primary and secondary users. It was proved through simulation results that FLS based power
control mechanism significantly reduced the average values of outage probability and transmit
power increase. In [121], Tabakovic et al. designed a simple and cost effective FL based transmit
power controller aiming to enable the secondary communication with the desired QoS
requirements, while ensuring that the interference offered to the PU is minimized and the mutual
interference of the SUs. Another fuzzy rule based opportunistic power allocation strategy
proposed in [122] for the efficient utilization of radio spectrum. However, the application of FL
to perform multiple relay selection for performance enhancement of secondary communication
in underlay cognitive radio networks has not been done so far to the best of our knowledge.
3.5 CONCLUDING REMARKS
In this chapter, the literature review of Evolutionary based algorithms has been carried out with
focus on Artificial Bee Colony and Hybrid Genetic Algorithm. Furthermore an efficient
Artificial Intelligence tool, the Fuzzy Logic has been studied in detail. For each algorithm, the
optimization phases are described in detail and the applications are highlighted.
Page 59
36
Chapter 4
SNR MAXIMIZATION IN UNDERLAY
CRNS
Objective
In this chapter, performance enhancement of secondary communication via AF based cognitive
relay network under Rayleigh flat-fading channel model is studied. Underlay spectrum sharing is
assumed at the intermediate relay network, and six algorithms for relay selection and power
allocation have been proposed based on the availability of perfect CSI. Each proposed algorithm
aims to maximize the SNR received at the destination under interference and transmit power
constraints. Impact of source and relay(s) transmit power, interference threshold levels, number
of candidate relays have been very well investigated for each algorithm keeping in view the
objectives and constraints.
4.1 PROBLEM FORMULATION 1
The first optimization problem is different from the other problems of SNR maximization in
defining the transmit power constraint for the relay network. No individual transmit power
constraint is imposed on any relay, rather the selected subset of relays S must constrain the total
transmit power Sm
mP below the predefined threshold maxP . The sum transmit power constraint
aims to give relaxation to individual relays to transmit at any power for secondary performance
enhancement, keeping in view the interference threshold of the PU. Hence, the mathematical
formulation of the multiple relay selection and power allocation problem can be expressed as,
Page 60
37
SS m mm
mm
m
mD21
21
1max
..ts
max:1 PPC
Sm
m
thm
m
m IfPC
S
2:2
)1.4(
where maxP is the maximum power that can be transmitted by cluster of M selected relays and
thI is the maximum tolerable interference for the PU. 1C defines the transmit power constraint,
whereas 2C is the interference constraint. Both constraints must be satisfied by the relay
network to enable the secondary communication. As obvious from the constraint 2C in 4.1, sum
transmit power of the relay network Sm
mP critically effects the total interference power towards the PU.
4.2 PROPOSED ALGORITHMS
Two algorithms are proposed to solve the highlighted problem, which are explained as under.
4.2.1 PROPOSED ALGORITHM 1
The relay selection and power allocation algorithm works as follows. Let },......,2,1{ Minitial be
the initial set of relays. The proposed algorithm is a two-phase algorithm. After initialization of
transmit power of potential relay set satisfying constraint C1, the first phase performs relay
selection to satisfy constraint 2C . For this purpose, the sum interference power I due to initial
towards the PU is computed and S is updated by excluding the relays in the descending order of
individual interference offered by each relay, aiming to satisfy the interference threshold thI .
After ensuring the security of the primary communication, the second phase works on improving
the SNR achieved at the destination. To achieve this goal, each selected relay gets an equal
increment in its transmit power to increase the corresponding relay-destination link SNR. The
Page 61
38
equal increment in the transmit power of each selected relay aims to reduce the computational
complexity, while satisfying sum transmit power threshold maxP . The pseudocode of the proposed
algorithm is provided in table 4.1.
Table 4.1: Pseudocode of the Proposed Algorithm 1
4.2.1.1 ABC Optimization
To solve the above mentioned non-linear constrained optimization problem, ABC is used for
optimization. ABC is preferred due to the advantages and applications highlighted in chapter 3.
Since the optimization phases have already been explained in chapter 3 with the help of flow
chart, the pseudo code of ABC algorithm to solve the SNR maximization problem is directly
presented in table 4.2.
The Proposed Algorithm
MNMmforfhgIPNPInputs SinitialmmmthS ,,,,,,,,: max0
);()( 0
22 NgPAP mSmm m
max
1
PPPM
m
msum
1// Csatisfying
thm
m
m
m
m IfPIwhile2
2// Ccheck
mm IforP max;0 ;1& MNS
endwhile
;_ m
mnewsum PP Smfor
;
_
N
PP newsumsum
; mm PP Smfor 2&1// CCsatisfying
;1 21
21
SS m mm
mm
m
mD
Page 62
39
Table 4.2: ABC Optimization to solve SNR Maximization Problems
Artificial Bee colony Optimization
orithmaproposedtoaccordingsolutionspotentialSNInitialize lg
iterationsLCfor :11
)(:1 EBsBeesEmployedPhase
SNnfor ,,2,12
ntsegergeneraterandomly ...,int
;* m
n
m
n
m
n
m
n
memp PPPP m 10 n
m
21&int CandCsatisfyPonphaseialiationofnscomputatioapplyn
memp
;
1 21
21
SS m
n
newmm
n
newmm
m
n
newm
n
newD
ocedureSearchGreedyifreplace n
D
n
newD
n
D Pr//
2forend
:2Phase Onlooker Bees (OBs)
;
1
SN
n
n
D
n
Dnp
SNn 1 fitnesscheck//
SNnfor ,,2,13
methodWheelRoulettebysolutionstheontosonlooPlace ker
3forend
Phase 3: Scout Bees (SBs)
if (rem(LC,2) = =0)
for4 n = 1,2,……, SN
Generate new food source for the abandoned ones
end for4
end if
end for1
Page 63
40
4.2.1.2 Simulation Results The performance of the proposed scheme for multiple relay selection is analyzed against
different interference threshold levels and different sizes of potential relay network. Furthermore,
the convergence behavior of ABC is strongly observed. The parameter settings for all
simulations of this chapter are listed in table 4.3.
Table 4.3: The Parameter Settings
Parameters Values
mg 9.03.0
mf 5.01.0
mh 9.03.0
0N 1
SP 10
maxP 10
For all simulations presented in this chapter, N denotes the number of selected relays. Fig. 4.1
evaluates the multiple relay selection algorithm at different interference threshold levels and for
different sizes of potential relay network. The interference thresholds are set to
)5,0,5( dBdBdBI th and M is set to )10,5(M . There are three strong observations First,
quick convergence of ABC, second, high SNR achieved at high interference threshold which
gives more freedom to the relays to transmit at high power, and finally SNR is further enhanced
by increasing the size of potential relay network, because large network gives more choices to
select the best possible combination of relays The NM relays offering relatively high
interference are not allowed to participate in communication.
Page 64
41
Fig. 4.1: Performance Analysis of Algorithm 1
4.2.2 PROPOSED ALGORITHM 2
The second algorithm proposed to solve the same problem is the “Power Set Algorithm”, which
works as follows. Let },,2,1{ Minitial denotes the initial set of potential relays. The
proposed algorithm initializes the transmit power of each relay while satisfying individual relay’s
transmit power constraint and creates all possible non-trivial subsets of transmit power vector
],,,[ 21 MPPPP , where the number of non-trivial subset in is given by,
M
n n
MJ
1
.
Given J subsets in , the relay subset selection algorithm selects JL subsets denoted by
Lll 1
, such that, each element in the thl subset l satisfies the condition mm hf . This relay
selection criteria aims to select those relays which are less harmful to the primary
communication. Next, the interference offered by each thl subset in is computed, where
interference offered by each thm relay is defined as2|| mmm fPI . Thus, the better is the channel
coefficient mf towards the PU, the higher is the interference offered by that relay. Finally, that
0 5 10 15 200
5
10
15
20
Iterations
D(d
B)
Ith
= -5dB, M = 5, N = 2
Ith
= 0dB, M = 5, N = 3
Ith
= 5dB, M = 5, N= 3
Ith
= -5dB, M = 10, N = 2
Ith
= 0dB, M = 10, N = 3
Ith
= 5dB, M = 10, N = 3
Page 65
42
subset is declared as the finally selected subset out of L selected subsets which maximizes the
SNR D received at the destination, keeping in view the constraints. The pseudo code of the
proposed algorithm is given in table 4.4 below.
Table 4.4: Pseudo Code For Proposed Algorithm 2 (Power Set Algorithm)
4.2.2.1 Simulation Results
For fair comparison, the SNR achieved by the power set algorithm is also analyzed against
different interference threshold levels and different sizes of potential relay network. The results
obtained are shown in Fig. 4.2. SNR is significantly improved because the power set algorithm
only allows those relay to participate in communication which strictly satisfy the condition
0 mm fh .
else
1//max
1
CsatisfyingPPPM
m
msum
mNgPAP mSmm )||( 0
22 1// Csatisfyingwhile
PJ
j 1][ PofsubsetstrivialnonAll //
L
ll 1
0..// mmm fhsatisfiesinelementeachts
l
th subsetleachforLlfor //:1
MNMmforfhgNPIPInputs SinitialmmmSth ,,,,,,,,: 0max
lbysatisfiedisconstraerferenceif intint
2&1 CCsatisfyingrelaysofpowertransmitadjust
lm
l
m
l
m
l
m
l
ml
D
21
21
1
ifend
forend
imumistoingcorrespondts sD
L
lls max..1
initialS N SofycardinalitthedenotesN //
SDOutputs ,:
Page 66
43
Fig. 4.2: Performance Analysis Of Proposed Algorithm 2
4.3 COMPARISON OF THE PROPSOSED ALGORITHMS
The comparison of the above proposed algorithms is shown in the Fig. 4.3 alongwith table 4.5,
showing the number of selected relays for both algorithms at different interference threshold
levels and different number of potential relays. The comparisons show that increase in the size of
potential relay network does not necessarily increase the number of selected relays but increases
the SNR achieved at the destination significantly because more choices are available to select the
best combination of relays. Algorithm 2 outsmarts algorithm 1 because it exclude the relays
which exhibit bad channel coefficients towards destination as compared to the corresponding
coefficients towards the PU.
4.3.1 CONCLUDING REMARKS
Performance enhancement of secondary communication in underlay CRNs has been focused, and
two multiple relay selection and power allocation scheme are proposed for relay-assisted CRNs.
It is proved that relay selection based cooperative diversity scheme increases the SNR at the
-5 -3 -1 1 3 5
10
12
14
16
18
I th(dB)
D(d
B)
M = 5
M = 10
Page 67
44
destination keeping the total interference offered by the selected relays to the PU below a certain
threshold level. Furthermore, the CSI based power set algorithm offers better results because it
makes decision about selecting a particular relay on the basis of each relay’s outgoing channel
coefficients.
Fig. 4.3: Comparison Of The Proposed Algorithms
Table 4.5: Comparison Of The Proposed Algorithms
Proposed Algorithm Total No. of
Relays “M”
No. of Selected
Relays “N”
Algorithm 1 5 -5 8.2 2
0 12.2 3
5 15.3 3
Algorithm 1 10 -5 8.7 2
0 12.9 3
5 16.1 3
Algorithm 2 5 -5 9.3 2
0 14.7 3
5 17.2 3
Algorithm 2 10 -5 10.1 2
0 15.3 4
5 17.6 4
)(dBIth )(dBD
-5 0 55
10
15
20
Ith
(dB)
D(d
B)
Algorithm 1, M = 5
Algorithm 1, M = 10
Algorithm 2, M = 5
Algorithm 2, M = 10
Page 68
45
4.4 PROBLEM FORMULATION II
The second problem of SNR maximization imposes individual transmit power constraint on each
individual potential relay, rather than defining sum transmit power constraint for the whole relay
network. The transmission power mP of each relay is limited not only by battery capacity due to
regulations specifying the maximum power that each node is allowed transmit, but also by the
interference threshold of the PU.
Thus, the mathematical formulation of the multiple relay selection and power allocation problem
can be expressed as,
SS m mm
mm
m
mD21
21
1max
..ts
:1C maxPPm
:2C thm
m
m IfP
S
2
)2.4(
where, maxP is the maximum power that can be transmitted by any selected relay and thI is the
maximum tolerable interference for the PU. 1C defines the individual relay’s transmit power
constraint, whereas 2C is the interference constraint. Both constraints must be satisfied by the
relay network to enable the secondary communication.
4.5 THE PROPOSED ALGORITHMS
The algorithms proposed to solve the SNR maximization problem formulated above assume the
availability of perfect CSI at each relay. The algorithms are explained one by one in this section.
4.5.1 PROPOSED ALGORITHM I
The optimization problem is again solved using ABC. The pseudocode of the proposed algorithm
is explained in table 4.6.
Page 69
46
Table 4.6: Psuedocode For Algorithm I
Let },......,2,1{ Minitial be the initial set of relays. The proposed multiple relay selection and
power allocation algorithm is a two-phase algorithm. In the first phase, total interference power
due to initial towards the PU is computed and S is updated by excluding those relays offering
The Proposed Algorithm
MNMmforfhgNPIPInputs SinitialmmmSth ,,,,,,,,: 0max
mNgPAP mSmm );||( 0
22
1// Csatisfying
m
th
m
mm IfPmIwhile )||)(( 2 // Constraint 2C
1&)max(;0 NIforP Smm
computeRe m
mI )(
];|[ mmm hff Smfor
{})( if
][ mf Smfor
endif
))(:1( sizeifor
;mmm PP );min(for
min// previousmm
;, max mm PPwhere
)2( satisfiednotCif
satisfiedisIuntiltakenPlastdecrement thm
exitfor
endif
endfor
;1 21
21
SS m mm
mm
m
mD
SDOutputs ,:
endwhile
Page 70
47
relatively high interference until thI is satisfied. In the second phase, transmit power of relays in
S is increased to improve SNR at secondary destination while satisfying the constraints. This is
achieved by increasing transmit power of those relays in S which exhibit good channel gains
towards the secondary destination as compared to the corresponding channel gains towards the
PU assuring that constraints C1 and C2 are still satisfied. If none of the relays is able to satisfy
mm hf , then transmit power of relays in S are increased by selecting relays in the ascending
order of mf .
4.5.1.1 Simulation Results
All parameter settings are done according to table 4.3 for fair comparison. Fig. 4.4 illustrates the
performance of the proposed algorithm for different levels of interference threshold thI and
different number of potential relays M . The number of selected relays are provided in table 4.7.
Hence the proposed algorithm exploits spatial diversity and multiple relays participate in
improving secondary network’s performance ensuring reliability of primary communication.
Fig. 4.4: Performance Analysis Of Proposed Algorithm 1
-5 -3 -1 1 3 58
10
12
14
16
18
Ith(dB)
D(d
B)
M = 5
M = 10
Page 71
48
Furthermore, high SNR is achieved at the destination as compared to the proposed algorithm
presented in section 4.2 above, since each selected relay gets an increment in its transmit power
depending on its corresponding channel coefficients towards the destination and the PU. Thus,
the relays which exhibit mm fh are preferably allowed to transmit at maxP . It is also observed
that an increase in the value of thI allows more relays to take part in communication to enhance
SNR D achieved at the destination, because the higher the value of thI , the more is the freedom
given to the relays to transmit at maximum available power maxP .
Table 4.7: No. of Selected Relays Obtained From Proposed Algorithm 1
4.5.1.2 Concluding Remarks
A CSI-based multiple relay selection with adjustable power allocation scheme is proposed for
underlay CRNs aiming to maximize the SNR achieved at the destination while adhering to
interference power constraint towards the PU. The proposed scheme achieved satisfactory results
at different interference threshold levels and outsmarts previously proposed algorithm in section
4.2 at all interference threshold levels.
4.5.2 PROPSOED ALGORITHM II
This algorithm is the same power set algorithm proposed in section 4.2.2 to solve the SNR
maximization problem I, and it works as follows. Let },,2,1{ Minitial denotes the initial set
of potential relays. The proposed algorithm initializes the transmit power of each relay while
satisfying individual relay’s transmit power constraint and creates all possible non-trivial subsets
M N
5
-5 11.3 2
0 13.7 3
5 18.2 3
10
-5 11.7 2
0 14.3 3
5 18.7 3
)(dBIth)(dBD
Page 72
49
of transmit power vector ],,,[ 21 MPPPP , where the number of non-trivial subset in is
given by,
M
n n
MJ
1
. Given J subsets in , the relay subset selection algorithm selects JL
subsets denoted by Lll 1
, such that, each element in the thl subset l satisfies the condition
mm hf . This relay selection criteria aims to select those relays which are less harmful to the
primary communication. Next, the interference offered by each thl subset in is computed,
where interference offered by each thm relay is defined as2|| mmm fPI . Thus, the better is the
channel coefficient mf towards the PU, the higher is the interference offered by that relay.
Finally, that subset is declared as the selected subset out of L selected subsets which maximizes
the SNR D received at the destination, keeping in view the constraints. The pseudo code of the
proposed algorithm is given in table 4.8 below.
Table 4.8: Pseudocode For Algorithm II
The Proposed Algorithm
MNMmforfhgNPIPInputs SinitialmmmSth ,,,,,,,,: 0max
)||( 0
22 NgPAP mSmm m
1// Csatisfyingwhile
PJ
j 1][ PofsubsetstrivialnonAll //
L
ll 1 0..// mmm fhsatisfiesinelementeachts
Llfor :1 l
th subsetleachfor //
if (interference constraint is satisfied by l )
l mm
mm
ml
DR
l
SR
l
DR
l
SRl
D
1
else
ifend
forend
imumistoingcorrespondts sD
L
lls max..1
initialS N SofycardinalitthedenotesN //
SDOutputs ,:
Page 73
50
4.5.2.1 Simulation Results
The effectiveness of the proposed CSI based relay subset selection scheme is proved in this
section. Fig. 4.5 shows the obtained results, where the number of potential relays and range of
interference thresholds is kept the same for fair comparison with other proposed algorithms. It is
strongly observed that the proposed “Power Set Algorithm” outperforms the proposed algorithm
1, due to the same reason highlighted with the simulation results of power set algorithm of
section 4.3.1. The improved performance is due to the freedom of selecting that subset of relays
from the whole relay network which exhibit good channel conditions towards the destination.
Fig. 4.5: Performance Analysis Of Proposed Algorithm 2
Table 4.9: No. Of Selected Relays From Proposed Algorithm 2
-5 -3 -1 1 3 510
12
14
16
18
20
Ith(dB)
D(d
B)
M = 5
M = 10
M N
5
-5 12.8 2
0 16.5 3
5 19.1 3
10
-5 13.2 2
0 16.8 4
5 19.5 4
)(dBIth)(dBD
Page 74
51
Whereas, if full diversity is achieved by allowing all relays to forward the source’s data, the
transmit power of the source and/or each relay need(s) to be suppressed keeping in view the
interference constraint of the PU, which in turn causes negative effect on the SNR received at the
destination. Again the larger relay network increases the probability of selecting those relays
which exhibit favorable channel conditions towards destination as compared to PU thus showing
higher improvement. Finally, as mentioned earlier, relaxing thI allows the relays to transmit at
high power thus enhancing secondary SNR for both cases.
4.5.2.2 Concluding Remarks
A power set algorithm is proposed to solve multiple relay selection problem in AF based
Underlay CRNs. The proposed scheme outperforms the previously proposed multiple relay
selection scheme due to the CSI-based exhaustive search involved in finding the best subset of
relays from all non-trivial subsets of the potential relay network which is able to achieve the
objective while satisfying the constraints.
4.5.3 PROPOSED ALGORITHM III
This algorithm is based on a strong observation about the behavior of m . In literature, tight upper
bound for end-to-end SNR m given in 1.4 for thm relay link, exist in for a comprehensive
performance analysis [123]. In terms of these bounds, m is expressed as,ub
mm
lb
m , where,
),min(2
121 mm
lb
m and, ),min( 21 mm
ub
m . The lower bound clearly indicates, the minimum
value of m occurs when mm 12 , and, for this case,2
1m
m
, and, if m2 is increased further by
performing high amplification at the corresponding relay, the upper bound is approached. Thus
any attempt to improve end-to-end SNR m of thm relay link through power allocation to those
relays with mm 21 will not produce any significant change in m , as m in this case is entirely
dependent on the SNR of first hop m1 , i.e. mm 1 .Thus, if m2 is increased by performing high
Page 75
52
amplification at the corresponding relay, the upper bound is approached. On the basis of this
strong observation, our relay selection and power allocation algorithm works as follows.
Let },,.........2,1{ Minitial be the initial set of potential relays. The proposed algorithm is a two-
phase algorithm and it performs decision making on the basis of the SNR m1 of first hop, the
SNR m2 of the second hop and the interference mI offered by each relay to the PU.
Phase-1 performs relay selection to satisfy interference constraint towards the PU by calculating
total interference power I offered by the relays in initial . In the underlay model, the relay with
good SNR may be destructive for primary network by generating high interference to the
neighboring PU and hence may not be chosen to participate in communication. Thus, if I
exceeds thI , the relay having best channel gain mf towards the PU is excluded from the set and
thI is recomputed after updating S . If S still fail to satisfy thI , then the algorithm builds up 1
considering those relays in S having mm 12 and updates S by deselecting relays in the
decreasing order of elements of mm 121 until interference constraint is satisfied.
Phase-2 works on S and SNR maximization is performed by creating 2 such that 2 is a
complement of 1 and increasing the transmit power of relays in 2 by selecting relays in the
increasing order of elements of mm 212 while satisfying constraints C1 and C2.
The pseudo code for the proposed algorithm is provided in table 4.10.
Page 76
53
Table 4.10: The Pseudo Code For The Proposed Algorithm
The Proposed Algorithm
mfhgNNMNMIPPInputs mmmSinitialths ,,,,,,,,,: 0max
)||( 0
22 NgPAP mSmm 1// Csatisfying m
1// Phase
))(( thIIsumif 2int// Cconstra
)(max;0 mm fforP 1 NS
endif
))(( thIIsumif
]|[ 12121 mmmm Sm
))(:1( 1 lengthjfor
;0mP )max( 1for
1 NS
)( Isumrecompute
)2( satisfiedisCif
;break
endif
endfor
endif
NS where MN
2// Phase
]|[ 21212 mmmm Sm
))(:1( 2 lengthkfor
;mmm PP for )min( 2
..,, max tsPPwhere mm 2min previousofanyformm
2int// CconstrasatisfytoassoPtostepssmallinfedis mm
endfor
m m mm
mmmD
21
21
1
Sm
SDOutputs ,:
Page 77
54
4.5.3.1 Simulation Results
In this section, performance analysis of the proposed algorithm is carried out through
simulations. The performance of cognitive relay network is again evaluated for different
interference threshold levels thI , and different number of potential relays M .
In Fig. 4.6, again it is observed that D increases with an increase in the value of thI as expected.
However, there is a strong observation that the proposed algorithm does not perform well under
low interference thresholds, because the relay selection criteria does not take into account the
interference offered to the PU, rather a selfish approach is adopted in which the decisions are
taken on the basis of SNRs achieved on the source-relay and relay-destination links. Thus, a
selected relay might not be allowed to transmit at high power because of being harmful to the
PU. On the other hand, when interference threshold is relaxed, the algorithm outsmarts the power
set algorithm, because it strictly considers the bounds on the SNR of selected relay link, and
increases the transmit power of those selected relays in the second phase that have SNR of relay-
destination m2 close to SNR of source-relay m1 .
Fig. 4.6: Performance Analysis Of Proposed Algorithm 3
-5 -3 -1 1 3 55
10
15
20
25
Ith(dB)
D(d
B)
M = 5
M = 10
Page 78
55
Table 4.11: No. Of Selected Relays From Proposed Algorithm 3
4.5.3.2 Concluding Remarks
A CSI based multiple relay selection algorithm is proposed for underlay CRNs which maximizes
SNR at the intended destination under PU’s interference constraint and individual relay’s
transmit power constraint. The performance of the proposed algorithm is very-well studied
against different levels of interference threshold and number of potential relays. The low
computational complexity of the proposed algorithm makes it a suitable candidate for designing
underlay systems.
4.6 COMPARISON OF THE PROPOSED ALGORITHMS
The comparison of the above proposed algorithms is shown in the Fig. 4.7 alongwith table 4.12,
showing the number of selected relays for the three algorithms at different interference threshold
levels and different number of potential relays. Similar trends are observed as with the case of
algorithms proposed to solve problem 1. However, Algorithm 2 (Power Set Algorithm)
outsmarts algorithm 1 and Algorithm 3 at all interference threshold levels, because it performs
exhaustive search and considers each and every possible subset of potential relay subset by
excluding the relays which exhibit bad channel coefficients towards destination as compared to
the corresponding coefficients towards the PU. However, at high interference threshold levels,
algorithm 3 gives satisfactory results.
M N
5
-5 8.9 1
0 10.9 2
5 20.3 3
10
-5 9.2 2
0 11.3 2
5 20.8 3
)(dBIth)(dBD
Page 79
56
Fig. 4.7: Comparison Of The Proposed Algorithms
Table 4.12: Comparison Of The Proposed Algorithms
M N
Algorithm 1
5
-5 11.3 2
0 13.7 3
5 18.2 3
10
-5 11.7 2
0 14.3 3
5 18.7 3
Algorithm 2
5
-5 12.8 2
0 16.5 3
5 19.1 3
10
-5 13.2 2
0 16.8 4
5 19.5 4
Algorithm 3
5
-5 8.9 1
0 10.9 2
5 20.3 3
10
-5 9.2 2
0 11.3 2
5 20.8 3
-5 -3 -1 1 3 58
12
16
20
24
Ith(dB)
D(d
B)
Algorithm 1, M = 5
Algorithm 1, M = 10
Algorithm 2, M = 5
Algorithm 2, M = 10
Algorithm 3, M = 5
Algorithm 3, M = 10
)(dBIth)(dBD
Page 80
57
4.7 CONCLUDING REMARKS
In this chapter, several multiple relay selection schemes are proposed in an underlay scenario
under interference constraints and transmit power constraints. For each algorithm, SNR achieved
at the destination is strictly observed. All the algorithms provide high SNR at high interference
threshold levels and for large potential relay network. However the power set algorithm
outsmarts all the algorithms in both cases of transmit power thresholds, i.e. for individual relay’s
transmit power constraint and for total transmit power constraint of the whole relay network
outsmarts a. Another important observation is that, increase in the number of available choices of
relays increases SNR at the destination, because a system which assigns more relays to the users
gives more opportunities to select those relays which effectively participate in enhancing
secondary communication performance without degrading the quality of primary communication
taking place in parallel. Furthermore, convergence behavior of ABC is also showed for one of
the proposed algorithms. ABC proved its ease of implementation, fast convergence and
effectiveness in solving multi-constrained optimization problem.
Page 81
58
Chapter 5
OUTAGE ANALYSIS OF MULTIPLE
RELAY SELECTION
Objective
In this chapter, we investigate the outage behavior of multiple relay selection over Rayleigh flat-
fading channels. In particular, we derive closed-form expressions for outage probability and bit-
error rate of underlay relay-assisted CRN. For this purpose, a dual-hop CRN operating in AF
mode is considered, and a multiple relay selection scheme is proposed, while ignoring the line-
of-sight path between source-destination pair. Finally, simulation results are presented to verify
the derived results.
5.1 PROBLEM FORMULATION
Refer to chapter 1 for the system model and basic assumptions that apply here also. The total
instantaneous end-to-end SNR D at the secondary destination due to M relaying links
expressed in 1.5 is again expressed here,
M
m mm
mmM
m
mD
1 21
21
11
)1.5(
In underlay networks, sophisticated signal processing techniques are employed to mitigate the
interference offered to the PU. However, due to the inherent simplicity of AF protocol, such
computationally complex techniques may not be supported at the AF relays. Thus, enabling the
secondary communication exploiting the services of all relays in the potential relay set may not
Page 82
59
be a viable idea in terms of total interference offered by the relay network to the nearby PU. For
simultaneous primary and secondary transmissions in such power-constrained environment, the
total interference power experienced by the PU by the potential relay set must satisfy the
predefined interference threshold given as,
th
M
m
mm
M
m
m IfPII 1
2
1
|| )2.5(
where thI is the interference threshold set by the PU.
Relay selection stands as a fascinating solution to this problem. The proposed relay subset
selection problem as follows. Let P represents the transmit power vector of the potential relays
in the network, i.e. ],......,,,[ 321 MPPPPP . The number of all non-trivial subsets P is
given by
M
k k
MS
1
. The thl subset is denoted as l where }).,,.........2,1{( Ml . The
cardinality of thl subset is lM . Next is to compute total interference power due to each
thl subset
of relays towards the PU, where interference offered by each thm relay in any subset is defined
as2|| mmm fPI .
Let J be the number of subsets out of S , denoted asJ
jjL 1}{ , which satisfy the interference
threshold thI towards the PU. The interference constraint forthj such subset can be given as,
th
Lm Lm
mimj IfPII
j j
2|| Jj .,,.........2,1 )3.5(
The mathematical formulation of this optimization problem is as follows:
jj
ij Lm mm
mm
Li
RDL 21
21
1max
th
j IIts .. )4.5(
Page 83
60
In order to investigate the performance of the overall system in terms of outage
probability and average probability of error, we need to know the distribution of D which is not
mathematically tractable. To overcome this problem, tight upper and lower bounds for end-to-
end SNR m , given in 5.1, exist in literature for a comprehensive performance analysis, as
explained in section 4.4.4.
Keeping the behavior of m under consideration, we aim to maximize m2 through controlled
transmit power allocation to each relay so that m of each relay link tends to approach its upper
bound causing an overall favorable impact on ,
j
DLm
m while keeping the sum interference
constraint satisfied. Thus, the relay subset selection algorithm aims to pick up that subset of
relays, which maximizes combined SNR of relay-links, j , where, for each thj subset jL , j is
defined as
jLm
mj 2 .
Thus, the mathematically tractable form of our optimization problem is given as,
jjLm
mj
L
2max
s.t. thj II )5.5(
5.2 THE PROPOSED ALGORITHM
Let },,2,1{ Minitial be the initial set of potential relays. The proposed algorithm works as
follows. Transmit power of each relay is initialized, followed by selecting all possible subsets of
relays which are able to satisfy sum interference power threshold thI set by the PU. For all such
subsets, combined SNR of relay-destination links is computed and finally that subset is declared
as the selected subset which maximizes the SNR. The pseudo code of the proposed algorithm is
Page 84
61
table 5.1 below. For more clarity, the flowchart of the proposed algorithm is also presented in
Fig. 5.1.
Table 5.1: Pseudocode For The Proposed Relay Subset Selection Algorithm
:INPUTS initialmmminitials mfhgNMP ,,,},,......,2,1{,, 0
Mmfor .,,.........2,1
)||()( 0
22 NgPAP msmm ionamplificatimumuse min//
2|| mmm fPI
forend
NP 1][ Pofsubsetstrivialnon//
j jLm
thmm
Lm
i
j
j
th IfPIILsubsetjfortsofoutsubsetsAllL 2||,..
Jjfor :1 Linsubsetsselectedofnumber//
th
j IIwhile
mm PP
m
m
f
hfor max
j
thth Lsubsetselectedjofelementmdenotesm//
endwhile
jj Lm
mm
Lm
mj hP 2
2 || Jj ,,1
forend
imumisLtoingcorrespondtsLL jjj max..
jjLOUTPUTS ,:
Page 85
62
Fig. 5.1: Flowchart Of The Proposed Algorithm
?Jj
Yes
No
Initialize transmit power of each potential relay
Obtain S non-trivial subsets of potential relay set
Extract J
jjL1
subsets
which satisfy interference
constraint, where, SJ
1j
mmm PP for
m
m
f
hmax , jLm
m is adjusted to ensure that interference
constraint remains satisfied
1 jj
Pick the subset jL that offers maximum SNR
on the relay-destination links
Outputs: jjL ,
Start
End
Page 86
63
5.3 PERFORMANCE ANALYSIS
In this section, the performance of the proposed multiple relay selection scheme is investigated
and has been compared with the best relay selection scheme. The criterion for relay selection is
kept the same for both schemes for fair comparison. Performance evaluation is carried out in
terms of outage probability and average probability of error. We consider both cases separately
as follows.
5.3.1 MULTIPLE RELAY SELECTION
Rayleigh distributed channel coefficients are assumed with their squared amplitudes being
exponential random variables. Therefore, the PDFs of m2 and mI being independent and
exponentially distributed are given by,
mm
epm
1
)(2
and mm
x
mI exp
1
)( J
jjLm 1][ )6.5( a
and the corresponding CDFs are given by,
mm
eP
1)(2
and mm
x
I exP
1)( J
jjLm 1][ )6.5( b
Where m denotes the average second hop SNR for thm relaying link, and m is the average
strength of interference channel from the thm relay and PU.
Given J subsets for selection, the conditional PDF of the SNR of the finally chosen subset sel
according to the proposed relay subset selection scheme where jsel L is given as,
]Pr[]Pr[)(]Pr[]Pr[)()|( 212121 21 JJ ppJpsel
]Pr[]Pr[)( 11 JJJJp )7.5(
Page 87
64
In order to simplify further analysis of above PDF, we assume that, during a hop transmission,
instantaneous SNRs have same average values for all relays. Hence, M21 .
Using this assumption, 5.7 is rewritten as,
1]Pr[)()|(
Jjjjsel
JpJp
1]Pr[)(
Jjjj
Jp jj )8.5(
In the above equation, first part )( jp
is the PDF of combined SNR of final selected subset
being evaluated at . Since each element in the selected subset jL is exponentially distributed,
thus the PDF of the combined SNR j being the sum of exponential random variables with same
mean will be Erlang distributed and is given by,
1/
)(
1)(
j
j
N
jN
ep
)9.5(
Where jN is the cardinality of selected set jL .
The second part of 5.8 i.e. ]Pr[ jj is the CDF of SNR of thj subset j being evaluated at j .
As mentioned above, SNR of each thj subset follows Erlang distribution, thus the CDF of j will
be expressed as,
nj
N
n
jjj
jj
j neP
1
0
/
!
11]Pr[)( )10.5(
Therefore, the PDF of selected relay subset in 5.8 will take the form as,
11
0
/1/
!
11
)()|(
Jn
jN
nj
N
N jj
j
j
sel ne
N
JeJp
)11.5(
Page 88
65
An important consideration is that the PDF given above is conditioned over J .i.e. the number of
subsets which are able to satisfy the interference constraints. The value of J may vary from 0 to
S . If ,0J communication between secondary source-destination pair is not possible. This
situation occurs if sum interference threshold thI imposed by PU is too low that no subset of
relays is able to meet the requirement without amplification, thus making secondary
communication impossible. But this is not the case in our scenario as the relay network is
assumed to be far away from the PU. Thus J takes the values between 1 to S . If ,1J there
would be no relay subset selection and if ,2J the destination will decide which relay subset is
the one satisfying the proposed criteria. The interference constraint thI can be satisfied by each
subset in with a probabilitythIP , where, )Pr( th
j
I IIPth
dictates the Erlang distribution
following the same assumption for interference strengths of relayed links i.e.
M21 . Thus, the PDF can be obtained in the same way as,
1/
)(
1)(
j
jN
j
I
II
N
eIp
)12.5(
And the corresponding CDF is written as,
n
thN
n
thj
I
I
neIIP
j
th
1
0
/
!
11]Pr[ )13.5(
Thus, the probability of availability of J subsets out of S subsets which satisfy interference
threshold follows binomial distribution,
JM
I
J
IIJ thththPP
J
SPSJp
)1()(),;( )14.5(
The unconditional PDF of SNR at the destination sel due to selected subset can be found by
using 5.13 and 5.14 in 5.11 as,
Page 89
66
1/
11
0
/
1 )(!
11)1()();(
j
j
jj
ththsel
N
j
N
Jn
jN
n
S
J
JSI
JIth
j
N
e
nePPJ
J
SIIp
)15.5(
The corresponding CDF can be obtained by integrating the above PDF w.r.t. and using [124,
Eq. (3.381.1)], thus,
,~
!
11)1()(
)(
1)(
11
0
/
1
j
Jn
jN
n
S
J
JSI
JI
j
Nn
ePPJJ
S
NP
jj
ththsel )16.5(
),(~ xa denotes the incomplete gamma function given in [124, Eq. (8.354.1)] as,
kN
k j
k j
kNkxa
0)(!
)1(),(~
The outage event occurs in a communication system if the SNR received at the destination falls
below a set threshold th . The probability of this event can be directly obtained from the CDF of
the received SNR given in 5.16 evaluated at th i.e. )( thout selPP .
Average bit error probability is usually evaluated using the probability of error conditioned over
a given SNR in AWGN. This conditional probability of error is defined in terms of standard Q
function and its average is taken over the PDF of received SNR. Therefore,
0
)()/( dpPPselselee )17.5(
where selsele QP )/( and is a constant and its selection depends on the modulation
scheme employed. Refer to the technique in [125], the above eq. takes the form,
0
22
2
2
1dte
tPP
t
e sel )18.5(
Page 90
67
Solving the above equation using [124, Eq. 3.461.2], we obtain,
0)(
11
0
/
1 )(
!)!1)(2(
!
)1(
!
11)1()(
)(2
1
kkN
j
j
kJ
nj
N
n
S
J
JSI
JI
je
j
jj
thth
kN
kNknePPJ
J
S
NP
)19.5(
5.3.2 BEST RELAY SELECTION
In order to verify the effectiveness of the proposed multiple relay selection scheme, similar
derivation has been carried out for best relay selection. Based on the same criteria for multiple
relay selection, the relay which is able to maximize the SNR of relay-destination link, while
satisfying the primary interference threshold, is declared the best relay by the destination. Thus,
the optimization problem formulated in 5.5 can be expressed as,
m
m
2max
s.t. thm II )20.5(
Where mdenotes the index of the best relay selected for communication.
In order to investigate the system performance for best relay selection, we follow the same
assumptions for channel conditions as stated in the above section. Thus, the PDFs and CDFs of
m2 and mI will be exponentially distributed as given in 5.6 for each candidate relay satisfying
interference threshold.
Given K relays for selection out of M potential relays, such that, the interference offered by
each thk relay is below the interference level thI set by the PU, the conditional PDF of sel i.e.
the SNR of the final selected relay, where Ksel , is given according to the proposed relay
subset selection scheme as,
]Pr[]Pr[)(]Pr[]Pr[)()|( 22221222212221 2221 KK ppKpsel
]Pr[]Pr[)( )1(222122 KKKKp )21.5(
Page 91
68
Assuming same average values of instantaneous SNRs for all relays to simplify further analysis,
5.21 can be rewritten as,
122 ]Pr[)()|(
2
Kkkksel
KpKp
1
22 ]Pr[)(2
Kkkk
Kp kk )22.5(
In the above equation, first part )(2
kp
is the PDF of SNR of best chosen relay being evaluated
at .
The second part of the equation i.e. ]Pr[ 22 kk is the CDF of SNR of thk relay being
evaluated at k 2 . Since the SNR of each thk relay follows exponential distribution as mentioned
above, thus the conditional PDF of selected relay using 5.6 will take the form as given by,
1// 21)|(
Kk
selee
KKp
)23.5(
An important consideration is that the PDF obtained in the above eq. is conditioned over K .i.e.
the number of relays which satisfy the interference constraints. The value of K may take any
value from 0 to M . If ,0K communication between secondary source-destination pair is not
possible. This situation occurs if relay network experiences too high interference threshold set by
the PU which is not satisfied by even a single relay, thus making secondary communication
impossible. But this is not the case in our scenario as the relay network is assumed to be far away
from the PU. Thus K takes the values between 1 to M . If ,1K there would be no relay
selection and if ,2K the destination picks up the best relay satisfying the proposed criteria.
Each member of the potential relay set can satisfy the interference constraint thI with a
probabilitythIP ,where, )Pr( thkI IIP
th dictates the exponential distribution and
/1 th
th
II eP
following the same assumption for interference strengths of relayed links i.e.
K21 .
Page 92
69
Thus, the probability of availability of K relays out of M relays which satisfy interference
threshold follows binomial distribution,
KMI
KIIK ththth
PPK
MPMKp
)1()(),;( )24.5(
The unconditional PDF of SNR sel due to the best selected relay can be found by using 5.6 and
5.24 in 5.23 as,
1/
1
/21)1()();(
KM
K
KNI
KIthk
kththsel
ePPKK
MeIIp
)25.5(
The corresponding CDF can be obtained by integrating the above PDF w.r.t. . The resulting
CDF is,
1/
1
/ 21)1()(1)(
KM
K
KMI
KI
kththsel
ePPKK
MeP
)26.5(
Outage probability can be directly obtained from the CDF of the received SNR given in 5.26
evaluated at th i.e. )( thout selPP .
Average bit error probability using the same technique as employed for multiple relay selection
and using [124, Eq. 3.321.3 ] is given by,
1/
1
21)1()(2/12
1 KM
K
KMI
KIe
kthth
ePPKK
MP
)27.5(
In the next section, the results derived for the single best relay selection and multiple relay
selection have been investigated for well-defined range of certain parameters for the primary and
secondary networks.
Page 93
70
5.4 SIMULATION RESULTS
This section verifies the effectiveness of the proposed scheme for selecting the subset of relays.
Zero mean unit variance AWGN is assumed for each link. Furthermore, for the relay subset
selection algorithm, M and jN represent the number of potential relays and selected relays
respectively. BPSK with 2 is the modulation scheme employed. The interfering channels
towards the PU are generated by setting 9.0 . Parameter settings are kept the same as
provided in table 4.3 for 0,,, Nfhg mmm and SP for fair analysis.
In Fig. 5.2, the performance of best relay selection, multiple relay selection and all relays has
been compared. SNR achieved at the relay-destination links against different levels of
interference threshold thI . The figure shows that the multiple relay selection algorithm
outperforms both best relay selection and all relay techniques due to freedom of selecting the
best subset of relays which can maximize secondary system performance through controlled
transmit power allocation to the relay network keeping in view the privilege of PUs. Whereas, in
order to allow all relays to participate in transmission, source transmit power needs to be
suppressed keeping to satisfy the interference constraint, which in turn produces negative effect
on the power received at the relay network, eventually decreasing the SNR received at the
destination. Furthermore, a single best relay is also unable to maximize the secondary
performance.
The corresponding total number of selected relays for best and multiple relay selection schemes
is provided in table 5.2. The greater is the number of candidate relays in the potential relay
network, the higher is the flexibility added to the system to allow more relays to participate in the
communication, which are favorable for secondary communication and not harmful for primary
communication at the same time. Thus, the multiple relay selection scheme is the optimal choice
as it provides the optimal combination of relays to maximize secondary performance as
compared to best relay selection and all relay participation.
Page 94
71
Fig. 5.2: Performance Analysis Of Different Schemes
Table 5.2: Performance Analysis Of Best Relay, Multiple Relay and All Relays Participation
Schemes
M N
Best Relay
Selection
5
-5 9.06 1
0 16.8 1
5 20.5 1
10
-5 9.8 1
0 18.2 1
5 23.3 1
Multiple Relay
Selection
5
-5 15.7 2
0 28.15 3
5 34.12 3
10
-5 17.76 3
0 29.54 4
5 36.23 4
All Relays
5
-5 12.8 5
0 21.5 5
5 28.23 5
10
-5 14.36 10
0 23.75 10
5 31.32 10
-10 -5 0 5 100
10
20
30
40
Ith
(dB)
D(d
B)
Best Relay Selection, M =5
Best Relay Selection, M = 10
All Relays, M = 10
All Relays, M = 5
Multiple Relay Selection, M =5
Multiple Relay Selection, M = 10
)(dBIth )(dBD
Page 95
72
In Fig. 5.3 and Fig. 5.4, outage probability and bit error rate of the best and multiple relay
selection schemes are investigated by varying the average SNR per hop for different number of
potential relays N , thI and th are both set to 1 respectively [62],[126].
Fig. 5.3: Outage Behavior Of Best And Multiple Relay Selection Schemes
As obvious from Fig. 5.3, the outage probability is maximum for the single best relay selection
and significantly decreases in the case of proposed relay subset selection due to the fact that
spatial diversity enhances system performance by improving SNR received at the destination. An
important observation is the improved system performance in the case of proposed multiple-relay
selection scheme because in order to design an underlay network with full cooperative diversity,
transmit power of the source needs to be suppressed even if the relays just forward the received
signal without any further amplification. On the other hand, in multiple relay selection,
increasing the number of potential relays generates more subsets which are able to satisfy
interference threshold set by the PU, thus giving more freedom to choose the optimal
combination of relays which exhibit good channel conditions towards the destination.
Furthermore, relay selection gives priority to those relays that exhibit good channel conditions
0 4 8 12 16 2010
-6
10-4
10-2
100
(dB)
Po
ut
Best Relay Selection, M = 5
Best Relay Selection, M = 10
Multiple Relay Selection, M = 10, Nj' = 5
Multiple Relay Selection, M = 5, Nj' = 3
Best Relay Selection
Relay Subset Selection
Page 96
73
towards secondary destination and allow them to transmit at high power to improve secondary
throughput. Similar trends are observed in Fig. 5.4 due to the same reasons.
Fig. 5.4: BER Of Best And Proposed Multiple Relay Selection
5.5 CONCLUDING REMARKS
The major contribution of this chapter is the derivation of the outage probability and bit error rate
for multiple relay selection scheme. For this purpose, a relay subset selection algorithm is
proposed for CRNs operating in an underlay environment near a PU. In this scenario, we select
the optimal combination of relays from the potential relay aiming to maximize the SNR received
at the destination, keeping in view the interference threshold of the primary network. The
proposed scheme proves the effectiveness of multiple relay selection in energy-constrained
CRNs. Finally, the outage probability and average probability of error have been derived in
closed forms through CDF of the received SNR at secondary destination, which has not been
done in literature so far for multiple relay selection. Performance evaluation shows that multiple
relay selection outperforms best relay and all relay techniques. Simulation results recommend
different operating points for the entire system under different levels of interference threshold
and number of potential relays.
0 4 10 12 16 2010
-6
10-4
10-2
100
(dB)
BE
R
Best Relay Selection, M = 10
Best Relay Selection, M = 5
Multiple Relay Selection, M = 10, Nj' = 5
Multiple Relay Selection, M = 5, Nj' = 3
Best Relay Selection
Relay Subset Selection
Page 97
74
Chapter 6
TRANSMIT POWER MINIMIZATION IN
UNDERLAY CRNS
Objective
In this chapter, a relay subset selection algorithm is proposed to select optimal combination of
relays from a potential relay set aiming to ensure minimum QoS requirements at primary and
secondary networks. The proposed scheme considers AF relaying and declares that subset as the
optimal choice after exhaustive search which minimizes the total transmit power at the relay
network while satisfying interference and SNR thresholds of the primary and secondary
networks. Simulation results are provided to prove the effectiveness of relay selection for
underlay CRNs.
6.1 PROBLEM FORMULATION
This chapter addresses the deficiency found in literature in the area of AF-based cognitive radio
networks. All the best and multiple relay selection algorithms found in literature based on
assumptions highlighted in section 2.4.1.3 solve the SNR maximization problem under transmit
power and interference constraints, but no one has solved the problem of relay selection aiming
to minimize the total transmit power at the relay network while guaranteeing to satisfy the
minimum QoS of both primary and secondary networks in underlay CRNs. Thus, the objective
behind the proposed multiple relay selection is to choose the optimal subset of relays which is
able to achieve secondary target SNR theshold th utilizing minimum sum transmit power at the
relay network, while satisfying primary interference threshold thI . Thus the relay subset selection
problem is mathematically expressed as,
Page 98
75
minimize
m
mP
s.t.
th
m mm
mmD
n
C
21
21
1:1
)1.6(
nm
thm IIIC :2
where, m denotes any relay in the thn subset n that satisfies constraints C1 and C2.
6.2 THE PROPOSED ALGORITHM
Let },,2,1{ Minitial be the initial set of potential relays and P denotes the transmit power
vector of the relay network, i.e. ],,,[ 21 MPPPP . The number of all non-trivial subsets
P is given by
M
m l
MR
1
. The thn subset is denoted as n , where })1.,,.........2,1{( Rn .
The cardinality of subset || n is denoted as M where MM .
Relay subset selection works as follows. First interference power I due to eachthn subset of
relays towards the PU is computed where interference power mI due to thm relay in any subset is
given by, 2|| mmm fPI . Let K be the number of subsets out of R, such that for every subset
Kkk 1
, interference threshold thI towards the PU is satisfied. The interference constraint forthk
such subset can be given as,
th
m m
mmik IfPII
k k
2|| Kk .,,.........2,1 )2.6(
In a similar way, P number of subsets, given by Ppp 1
are selected out of R which are
able to satisfy SNR threshold thI at the secondary destination. The SNR constraint for thp such
subset can be given as,
Page 99
76
th
mpm
pm
pm
pmp
D
p
21
21
1 Pp .,,.........2,1 )3.6(
After having two types of subsets, i.e. Pp 1 which achieves SNR threshold th and Kk 1 which
satisfies interference threshold thI , the intersection of PpK
k 11 is performed to extract all
matching subsets having same number of relays and same index position of each relay in the
matching subsets. Finally, that subset is declared as the selected subset from the matching
subsets which consumes minimum sum transmit power at the relay network according to the
proposed criterion.
Summarizing, the proposed algorithm initializes the transmit power of each relay in the potential
relay set followed by performing exhaustive search on
M
m l
M
1
combinations of relays. The aim
is to sorting out all possible subsets of relays which satisfy interference threshold thI and target
SNR threshold th at the same time. Finally, the selected subset consumes the minimum transmit
power out of all subsets. The pseudocode of the proposed algorithm is shown in table 6.1.
Table 6.1: The Proposed Algorithm
:INPUTS initialmmmsinitialththS mfhgNNMIP ,,,,,,,, 0
)||()( 0
22 NgPAP mSmm max0 PPm initialm
P PofsubsetstrivialnonAll //
th
m m
mmm
k IfPIIk k
2||
K
kkwhere1
,
th
mpm
pm
pm
pmp
D
p
21
21
1
P
ppwhere1
,
PpK
km 11 indicessameatPhavingsubsetsmatchingallcontain mm //
imumisinPts sel
m
mmsel
sel
min..
initials N relaysofsubsetselectedthedenotesS//
Selm
ms POUTPUTS ,:
Page 100
77
6.3 SIMULATION RESULTS
In this section, we will prove the effectiveness of our proposed multiple relay selection scheme
to achieve cooperative diversity while satisfying minimum QoS requirements of both the primary
and secondary networks. The channel coefficients are kept the same as provided in table 4.3.
Furthermore, M and N represent the number of potential relays and selected relays
respectively. In Fig. 6.1, the behavior of the relay network is evaluated in terms of total transmit
power required for different levels of SNR threshold and different number of available relays.
SP and thI are set to dB10 and dB0 respectively. The figure shows that total transmit power
required at the relay network increases significantly by increasing the SNR threshold th , since
more power is required to satisfy high SNR threshold. Furthermore, the higher is the number of
available relays, the lower is the total transmit power required by the selected relay subset. This
is due to the fact that the total number of non-trivial subsets R increases by increasing the
number of potential relays M , which in turn increases the probability of selecting those relays
which exhibit good channel conditions towards the destination than the PU, thus reducing the
total transmit power required at the relay network to meet th while satisfying .thI
Fig. 6.1: Transmit Power Allocation to Relay Network keeping dBI th 0
0 2 4 62
3
4
5
6
th(dB)
Pt(d
B)
M = 10
M =5
Page 101
78
Table 6.2 provides total transmit power required at different interference threshold levels. The
relaxation in the value of thI not only generates more matching subsets from PpK
k 11 but
also gives more freedom to the relays to communicate at high transmit power.
Table 6.2: Total Transmit Power required For Different Values of
M
N N
0 5 2 3.06 2 3.56
10 2 2.98 2 3.87
2 5 2 3.89 2 4.25
10 2 3.64 2 4.92
4 5 2 4.79 2 5.39
10 2 4.54 3 5.12
6 5 3 5.95 3 6.65
10 2 5.72 3 6.95
Fig. 6.2 demonstrates the behavior of the proposed algorithm under different levels of source
transmit power with corresponding no. of selected relays provided in table 6.3. At low values of
source transmit power, higher amplification is required at the relay network to satisfy the SNR
threshold of the destination. As the transmit power of source increases, transmit power of the
selected subset of relays decreases, since less amplification is required at the relay to satisfy the
SNR threshold. However, at very high values of source transmit power and low interference
threshold level, relay selection problem becomes critical, as the higher is the power received at
the potential relay subset, the higher is the power transmitted by the relays with even less
amplification performed, due to which the interference power experienced by the PU increases
significantly.
dBI th 5)(dBPt
dBI th 0)(dBth )(dBPt
MandI thth ,,
Page 102
79
Fig. 6.2: Transmit Power Allocation To Relay Network Considering
Table 6.3: Transmit Power Allocation To Relay Network For
M N
5
0 3.77 3
5 3.28 2
10 3.06 2
15 2.65 1
10
0 3.69 2
5 3.14 2
10 2.98 2
15 2.54 1
6.4 CONCLUDING REMARKS
A multiple relay selection algorithm is proposed for CRNs operating in underlay environment in
the vicinity of a PU, aiming to select an optimal combination of relays from the potential relay
set, which consumes minimum transmit power while satisfying interference threshold and SNR
threshold of the primary and secondary network respectively.
0 5 10 152.5
3
3.5
4
PS(dB)
Pt (
dB
)
M = 5
M = 10
dBdBI thth 0,0
dBdBI thth 0,0
dBdBI thth 0,0
)(dBPS )(dBPt
Page 103
80
Chapter 7
PERFORMANCE ENHANCEMENT OF
CRNS USING FUZZY RULE BASED
SYSTEM
Objective
In this chapter, Fuzzy Logic has been used to solve the problem of performance enhancement of
secondary communication in underlay spectrum sharing environment. FRBS assisted relay
selection and transmit power allocation (RSTPA) schemes exploit the degrees of freedom
involved in fuzzy logic to solve the highlighted problems of performance enhancement in CRNs.
Standard Mamdani fuzzy control is used for this purpose. The proposed schemes for SNR
maximization and transmit power minimization assume the availability of perfect CSI and select
the best combination of relays keeping in view the objectives and constraints. Furthermore, the
proposed schemes have been investigated for a well-defined range of certain parameters.
7.1 SNR MAXIMIZATION
In this section, two algorithms based on FL have been proposed to solve the modified problem of
SNR maximization formulated in chapter 5 for underlay cognitive relay networks. For the sake
of convenience, the multiple relay selection and power allocation problem formulated in section
5.1 is again expressed here as,
Sm
m22max )(1.7 a
s.t. th
m
m III
S
)(1.7 b
Page 104
81
where, S denotes the selected set of relays.
The algorithms proposed to solve this constrained problem are explained as follows.
7.2 FRBS ASSISTED SYSTEM DESIGN 1
A novel idea of FRBS-assisted RSTPA system in AF-based cognitive relay networks over
Rayleigh flat-fading channels has been proposed in this section. The proposed scheme comprises
two FLSs as shown in Fig. 7.1.
Fig. 7.1: The Proposed System Design 1
Page 105
82
As shown in the Fig. 7.1, FLS 1 takes mm hP , and mf corresponding to each thm relay as inputs
and the fuzzy inference engine computes two outputs, one is the SNR m2 achieved on the link
from thm relay to destination, and the other is the interference mI offered to the PU at the thm
relay link corresponding to each rule. The consequents of FLS 1 are then fed to FLS 2 to
determine the relay selection factor RSF for each thm relay.
7.2.1 MAMDANI FUZZY CONTROL
The basic structure of FLS is already explained in chapter 3. The fuzzy rule based system to
solve the SNR maximization problem is explained below.
7.2.1.1 Fuzzification
Fuzzification represents the entire range of each antecedent (input variable) and the consequent
(output variable) through distinguishing fuzzy sets, with each fuzzy set assisted by a MF. The
MFs of the antecedents and consequents of FLS 1 and FLS 2 are given in Fig. 7.2.
Page 107
84
Fig. 7.2: MFs Of The Antecedents And The Consequents Of FLS 1 And FLS 2
Triangular MFs are used to represent the antecedent mf , whereas, for all other cases of
antecedents and consequents, trapezoidal MFs are chosen for minimum and maximum levels and
triangular MFs are used to represent medium levels. The fuzzy sets for the final consequent, the
RSF, taken from FLS 3 are explained as follows. The output variables are NS (Not Selected),
WCS (Weak Consideration for Selection), CS (Consider for Selection), SCS (Strong
Consideration for Selection), and, S (Selected). Same MFs for RSF will be used in all algorithms
proposed in this chapter.
Page 108
85
7.2.1.2 Rule Based Decision
Referring to the MFs shown in Fig. 7.2, three fuzzy sets are used for the first input variable mf ,
four fuzzy sets for the second input variable mh , and six fuzzy sets for the third input variable mP .
Thus, there are 72643 “IF-THEN” rules for FLS 1 of the form,
NSisISisTHENfisfhishPisPIFR llllll
121314161 ,,,,,: Ll ,,2,1
where L denotes the total number of rules for FLS 1. The outputs DRm and mI for each thm
relay from FLS 1 are then fed to FLS 2 to determine the consequent Relay Selection Factor
(RSF) of each thm relay. In FLS 2, there are 2555 “IF-THEN” rules of the form,
SisRSFTHENSisISCSisIFR kkkk
1121 ,,,: Kk ,,2,1
where K represents the total number of rules for FLS 2.
The rule base contains rules for every possible combination of the input space. Standard
Mamdani Inference Engine (MIE) serves the purpose.
7.2.1.3 Defuzzifier
The fuzzy variables produced by FIE are converted into the crisp values by the Defuzzifier to
represent the actual output of the system. Center Average Defuzzifier (CAD) is preferred
because of its computationally simplicity. After applying CAD method, the relationship of the
consequents m2 and mI with the antecedents mP , mh and mI in FLS1 are shown by the rule
surfaces in figure 7.3(a), whereas, figure 7.3(b) shows the rule surface for RSF according to rule-
based decisions taken in FLS2.
Page 109
86
Fig. 7.3 (a): Rule Surface For FLS 1
Fig.7.3 (b): Rule Surface For FLS 2
Fig. 7.3: The Rule Surface
Page 110
87
7.2.2 THE PROPOSED ALGORITHM
According to the proposed algorithm, FLS 1 computes m2 and mI based on the CSI available at
each relay, followed by FLS 2 which assigns RSF to each relay on the basis of consequents of
FLS 1. The relays are picked up in the descending order of RSF while ensuring to satisfy the
interference threshold thI . Finally, the fine tuning of transmit power of the relay with highest RSF
is carried out as it exhibits the best channel condition towards the destination and offers the
minimum interference to the nearby PU. The flowchart of the proposed algorithm is shown in
Fig. 7.4.
m = 1
Initialize transmit
power of mth
relay
FLS1 computes
m2 and mI
FLS2 computes RSF of mth
relay based on m2 and
mI
m = m + 1
m=M?
k = 1
Pick the relay with highest
RSF (excluding previous
selection)
interference
constraint
satisfied?
k = k + 1
k = M?
Fine tuning of transmit
power of selected relays
Outputs:
sm
ms 2,
Yes No
Yes
No
No
Start
Yes
Fig. 7.4: Flow Chart of the Proposed Fuzzy
Rule Based RSTPA Design
Page 111
88
7.2.3 SIMULATION RESULTS
The performance of the proposed schemes is presented in this section. The channel gains mm hg ,
and mf are taken according to table 4.3. Noise variance is assumed to be 1.0 . Fig. 7.5 illustrates
the performance of the proposed scheme in terms of total SNR of relay-destination link 2 given
in 7.1 achieved for different interference thresholds levels. Two different cases of total number
of potential relays are considered which are taken to be }10,5{M .
Fig. 7.5: SNR Performance Of The Proposed Scheme
From Fig. 7.5, it is observed that SNR 2 increases significantly by increasing the interfernce
threshold thI due to the fact that relaxing thI allows the relays to transmit at high power thus
enhancing secondary network’s performance. Furthermore, another strong observation is that, as
the relay network grows, there is considerable improvement in the SNR, as large relay network
adds more flexibility to choose those relays according to FRBS, which exhibit favorable channel
conditions towards the destination and at the same time are not harmful for the primary
communication. Table 7.1 shows the corresponding number of selected relays for different cases
considered in figure 7.5(a). Similar trends are observed here by relaxing interference threshold,
-5 -3 -1 1 3 512
14
16
18
20
22
24
26
Ith
(dB)
2(d
B)
M = 10
M = 5
Page 112
89
but higher SNR is achieved, due to FRBS assisted relay selection which intelligently assigns a
RSF to each individual relay to make reliable decision about relay selection.
Table 7.1: Corresponding Number Of Selected Relays
In Fig. 7.6, the proposed FRBS is extended to evaluate 2 given in 7.1 against different levels of
source transmit power. Interference threshold thI is set to 0dB. We consider the system’s
response for different number of potential relays }15,10,5{M . The obtained results show the
improvement in received SNR for low and medium levels of source transmit power, but SNR
decreases at high transmit power levels. This is due to the reason that the higher is the power
received at the relay network, the higher is the total interference offered by the relay network to
the PU, which makes the relay selection problem difficult and reduces the number of selected
relays, thus resulting in performance degradation.
Fig. 7.6: SNR Performance For Different Source Transmit Power Levels
M N
5
-5 11.3 2
0 13.7 3
5 18.2 3
10
-5 11.7 2
0 14.3 3
5 18.7 4
)(dBIth )(dBD
0 5 10 15 204
6
8
10
12
14
PS (dB)
2 (dB
)
M = 5, Ith
= -5dB
M = 10,Ith
= -5dB
M = 15, Ith
= -5dB
Page 113
90
Thus, in energy-constrained relay-assisted networks, where the interference threshold of the
primary network limits the transmit power of the relays, increasing the transmit power of the
source does not enhance the communication performance.
In Fig. 7.7, the performance of the proposed scheme is validated by comparison with the greedy
scheme [65] for multiple relay selection. The comparison has been carried out in terms of total
end-to-end SNR D . All parameter settings are done according to [65]. The figure shows
obvious performance difference and prove that the proposed scheme is able to achieve high total
end-to-end SNR as compared to the greedy scheme. This improvement is achieved due to two
strong factors. First, using FRBS assisted RSTPA design of the secondary system which checks
every possible combination of relay-destination and relay-PU channel conditions and picks the
best combination of relays to enhance secondary performance while operating in underlay mode.
Second, the flexibility in the transmit power allocation to the selected relays with the priority
given to the relay with the highest RSF obtained from FRBS keeping in view the interference
threshold. On the other hand, in the greedy scheme, either the relay does not participate in the
communication or transmits at fixed powerSm PP , thus, there is no flexible transmit power
allocation for the selected relays to for enhancement of secondary communication. However, in
high power areas, the SNR decreases in both cases due to the same reason mentioned in the
comments given for Fig.7.6.
Fig. 7.7: Comparison of the Proposed Scheme and the Greedy Scheme
0 5 10 15 20 252
4
6
8
10
12
PS (dB)
D (
dB
)
Proposed Scheme, M = 5, Ith
= 10
Greedy Scheme, M = 5, Ith
= 10
Page 114
91
7.2.4 CONCLUDING REMARKS
A novel fuzzy rule based RSTPA scheme is proposed to enable the secondary communication in
the frequency band of PUs. Once the CSI based knowledge is pumped into the FRBS, it provides
the optimum RSF upon offering the three parameters that are power received at each relay, relay
to PU interference channel gain and relay to destination channel gain. The RSF obtained from
the FRBS, is used as a priority factor that highlights the relays having ability to enhance
secondary system performance and are less harmful to the PUs. It is proved that the proposed FL
based relay selection scheme outsmarts an existing scheme in literature for multiple relay
selection.
7.3 FRBS ASSISTED SYSTEM DESIGN 2
The second FRBS-assisted RSTPA technique to solve the SNR maximization problem works as
follows. The proposed scheme uses three-stage FL system as shown in Fig. 7.8. This FRBS
assisted design is more complex than the design 1, since it contains three FLS module in cascade
and makes decisions not only on the bases of m2 and mI , but also includes the effect of m1 to
assign relay selection factor.
Fig. 7.8: Proposed System Design 2
As the figure shows, the proposed design, being a three-stage FLS is a bit computationally
complex, however, decisions taken in FLS1 are based on single antecedent (input), thus reducing
the overall computational cost involved. The three Mamdani fuzzy control modules FLS1, FLS2
and FLS3 are shown in Fig. 7.9.
FLS 3 hm
fm
Pm
γ1m
γ2m
Im
FLS 1 FLS 2 gm
RSFm
Page 115
92
Fig. 7.9: Proposed FLS Modules
Page 116
93
FLS1 takes the thm source-relay channel coefficient mg as input, and computes power mP
received at the corresponding relay and SNR of the first hop (source-relay link) m1 . FLS 2
follows FLS1 and takes three inputs for each thm relay; mP obtained from FLS1, relay-
destination channel coefficient mh and relay-PU channel coefficient mf . The outcomes of FLS2
are SNR of second hop (relay-destination link) m2 and the interference power mI of each thm
relay. Finally, FLS3 assigns RSF to each thm relay based on the consequents of FLS1 and FLS2.
The RSF is assigned to each relay to make a decision about its selection, aiming to give the
highest priority to those relays in the relay selection procedure, which exhibit the ability to
maximize SNR on the corresponding link with minimum amplification, and at the same time are
less harmful to the PU.
Thus, the selection of a particular thm relay depends on the following factors:
i. SNR achieved at the source-relay link m1
ii. SNR achieved at the relay-destination link m2
iii. Interference offered to the PU mI
7.3.1 MAMDANI FUZZY CONTROL
FRBS to solve the SNR maximization problem is explained below.
7.3.1.1 Fuzzification
Fuzzification represents the entire range of each antecedent (input variable) and the consequent
(output variable) through distinguishing fuzzy sets, with each fuzzy set assisted by a MF. The
entire range of each linguistic variable is covered with sufficient number of fuzzy sets for
accuracy and reliability of the system as shown in Fig. 7.10. Triangular MFs represent each
fuzzy set.
For FLS1, twenty three fuzzy sets are used for the antecedent mg and eight fuzzy sets each are
used for the consequents mP and m1 as shown in Fig. 7.10(a).
Page 117
94
Fig. 7.10(a): MFs For Antecedents And Consequents of FLS1
Page 118
95
The fuzzy sets for the inputs to FLS 2 are shown in Fig. 7.10(b). Fuzzy sets for the input mP are
shown earlier, whereas, six fuzzy sets cover the entire range of the input mh and three fuzzy sets
represent the input mf . For the outputs, eight fuzzy sets are used for mI , and seven fuzzy sets are
used for m2 .
Page 119
96
Fig. 7.10(b): MFs For Antecedents And Consequents of FLS2
Since FLS 3 is fed from the consequents of FLS 1 and FLS 2, the fuzzy sets for the antecedents
mmI 1, and m2 of FLS 3 are shown in Fig. 7.10(a) and Fig. 7.10(b), where the fuzzy sets for the
final output RSF are the same as shown in Fig. 7.2.
7.3.1.2 Rule Based Decision
Since there is a single antecedent for FLS1, there are 23 “IF-THEN” rules with each thl rule of
the form,
1111 ,,,,: BisandPisPTHENGisgIFR l
m
l
m
l
m
l Ll ,,2,1
where, L denotes the total number of rules for FLS 1.
There are 144386 “IF-THEN” rules for FLS2, with eachthn rule of the form,
424624 ,,,,,,: SisandIisITHENPisPFisfHishIFR n
m
n
m
n
m
n
m
n
m
n Nn ,,2,1
where, N represents the total number of rules for FLS 2.
For FLS 3, there are 448788 “IF-THEN” rules for FLS3, with each thp rule of the form,
SisRSFTHENIisISisBisIFR p
m
p
m
p
m
p
m
p ,,,,: 17261 Pp ,,2,1
where, P represents the total number of rules for FLS3.
Page 120
97
7.3.1.3 Defuzzifier
After applying CAD method, the relationship of the consequents of each FLS with corresponding
antecedent is shown by the rule surfaces in Fig. 7.11(a) and 7.11(b) respectively, whereas, Fig.
7.11(c) shows the rule surface for RSF according to rule-based decisions taken in FLS2.
Fig. 7.11(a): Rule Surface For FLS 1
Page 121
98
Fig. 7.11(b): Rule Surface For FLS 2
Fig. 7.11(c): Rule Surface For FLS3
Fig. 7.11: The Rule Surface
Fig. 7.11 (c) clearly indicates that, the selection factor of that relay is the highest, which exhibit
mm 12 , thus requiring less amplification to approach upper bound and so less harmful for the
Page 122
99
primary communication. RSF goes on decreasing for other values of m1 and m2 , and RSF for a
particular relay goes to zero for mm 21 , since very high amplification is required in this case
to achieve even lower bound of SNR, thus resulting in high interference towards the PU.
7.3.2 THE PROPOSED ALGORITHM
The algorithm proposed for secondary performance enhancement based on FL is explained with
the help of flow chart in Fig. 7.12 below.
Fig. 7.12: Flow Chart Of The Proposed Algorithm
m = 1
FLS1 computes Pm
and γ1m based on gm
FLS3 computes RSFm
based on γ1m, γ2m and Im
FLS2 computes γ2m and Im
based on Pm, hm and fm
m = m + 1
m = M ?
j= 1
Pick the relay with highest
RSF (excluding previous
selection)
interference
constraint
satisfied?
j = j + 1
j= M ?
Fine tuning of transmit
power of the selected relays
Outputs:
selj
msel 2,
Yes No
Yes
No
No
Yes
Start
Page 123
100
7.3.3 SIMULATION RESULTS
The performance of the proposed FRBS assisted relay selection technique is evaluated to select
the best possible combination of relays, which maximizes secondary communication
performance, in such a way that minimum amplification is required at the intermediate relay
network and at the same time, the primary interference constraint is not violated. Same parameter
settings provided in section 7.2.3 are maintained for the purpose of comparison of both schemes.
Fig. 7.13 illustrates the SNR performance of the fuzzy rule based relay selection algorithm
against different levels of interference threshold thI . The system response is observed for three
different sizes of the intermediate relay network, i.e. 10,5M . It is observed from that figure
that the total SNR 2 achieved at the selected relay-destination links significantly increases as the
interference threshold is relaxed, thus allowing the relays to transmit at high power. Furthermore,
adding more relays to the network considerably increases 2 , because more options are available
for FRBS to select those relays which can efficiently increase SNR by increasing signal
diversity, thus enhancing the overall system performance.
Fig. 7.13: Performance Of The Proposed Scheme Vs Interference Threshold thI
-5 -3 -1 1 3 512
14
16
18
20
22
24
26
28
Ith
(dB)
2(d
B)
M = 10
M = 5
Page 124
101
Table 7.2: Total No. of Selected Relays
Table 7.2 shows the corresponding number of selected relays. The nu1mber of selected relays
not only increases by relaxing the interference threshold, but also by increasing the number of
candidate relays, since more candidates present more choices for FRBS to pick the best possible
combination. Algorithm 2 outsmarts algorithm 1 because it takes into account SNR of source-
relay and relay-destination links alongwith interference offered to the PU by each relay and
assigns RSF to each individual relay.
7.4 COMPARISONS OF THE PROPOSED ALGORITHMS
Now we compare the performance of the multiple relay selection problems proposed in chapter 4
and in this chapter. For this purpose we compare the SNR achieved by the relay selection
schemes proposed in sec. 4.5.1 and 4.5.3 with the algorithms proposed in sec. 7.2 and 7.3
respectively. The observed results are provided in table 7.3.
Table 7.3: Comparisons of Proposed Algorithms in Ch. 4 and Ch. 7
M N
5
-5 11.3 2
0 13.7 3
5 18.2 3
10
-5 11.7 2
0 14.3 3
5 18.7 4
M = 10
Interference
Threshold
No. of Selected Relays
Algorithm 1 Algorithm 2 Algorithm 1 Algorithm 2
without
FL
with FL without
FL
with FL without
FL
with FL without
FL
with FL
11.7 12.1 9.2 12.8 2 2 2 2
14.3 15.1 11.3 16.2 3 3 2 3
18.7 20.3 20.8 24.1 3 4 3 4
)(dBIth )(dBD
dBIth 5
dBIth 0
)(dBD
dBIth 5
Page 125
102
As the table shows, Algorithm 1 solved with fuzzy logic (sec. 7.2) outsmarts the algorithm
proposed in sec 4.5.1. Similarly, algorithm 2 proposed in sec. 7.3 outsmarts the algorithm
proposed in sec. 4.5.3.
7.4.1 CONCLUDING REMARKS
The proposed FRBS assisted design 1 and 2 takes the information about incoming and outgoing
channel coefficients for each relay into account, and allocates a priority factor or RSF to each
relay. The RSF indicates the preference in which the relays are selected, aiming to maximize the
secondary performance in an underlay spectrum sharing environment. The effectiveness of the
proposed scheme is highlighted through the simulation results, which shows that the design 2
achieves high SNR as compared to design 1, although more computational complexity is
involved in design 2. The better results are obtained because the priority factor is assigned to
each thm relay on the bases of SNR of corresponding source-relay and relay-destination links and
the interference offered by the thm relay, which results in more sophisticated decision making.
Furthermore, it is proved that FL based relay selection outsmarts ABC optimization.
7.5 TRANSMIT POWER MINIMIZATION
This section solves the problem of transmit power minimization in underlay cognitive relay
networks using FL. For this purpose, a FRBS is proposed for intelligent relay selection such that
sum transmit power at the relay network is minimized, while achieving the desired SNR at the
destination and keeping the primary communication undisturbed. For the sake of inconvenience,
the mathematical formulation of the relay subset selection problem is again expressed as,
..ts
thm
m
m hP
S
22 ||
thm
m
m IfPIS
2|| 2.7
Sm
mPmin
Page 126
103
where, th and thI denotes the SNR threshold and the interference threshold respectively. s
denotes the selected set of relays, and the cardinality of the selected subset S is MN .
7.5.1 THE PROPOSED FLS DESIGN
The FLS proposed to solve the constrained problem is explained as under. FRBS is employed to
choose the best combination of relays to enable the secondary communication in an underlay
environment while consuming minimum overall transmit power. The proposed Mamdani-based
FLS is shown in Fig. 7.14.
Fig. 7.14: Proposed FRBS assisted Design
As Fig. 7.14 shows, the power mP transmitted by the thm relay in the relay network and the ratio
of relay-destination channel coefficient relative to the relay-PU channel coefficient act as the
antecedents and the fuzzy inference engine computes a RSF for each candidate relay based on
the fuzzy rules.
7.5.2 MAMDANI FUZZY CONTROL
The phases involved in the Mamdani fuzzy control are explained as under.
Page 127
104
7.5.2.1 Fuzzification
The fuzzy sets for each input and output variable is shown in Fig. 7.15. To make the whole
design sophisticated and accurate, the whole range of the inputs and the output are covered with
sufficient number of fuzzy sets. Fifteen fuzzy sets are used for the both antecedents mP and m
m
f
h
and five fuzzy sets for the consequent mRSF , with trapezoidal MFs for the minimum and
maximum levels and triangular MFs for the intermediate levels. Refer to Fig. 7.2 for the MFs of
RSF.
Fig. 7.15: Fuzzy Sets for the Antecedents and the Consequent
Page 128
105
7.5.2.2 Rule Based Decision
There are 2251515 “IF-THEN” rules with each thk rule of the form,
SisRSFTHENGisf
hPisPIFR k
k
m
k
mkk
115131 ,,,: Kk ,,2,1
where, K represents the total number of rules.
7.5.2.3 Defuzzifier
The rule surface specifying the relationship between each input and the output is shown in Fig.
7.16 below.
Fig. 7.16: The Rule Surface
7.5.3 THE PROPOSED ALGORITHM
FRBS assisted relay selection scheme works as follows. The inputs to the FRBS are the transmit
power mP of each relay without amplification, and the ratio of coefficients mh and mf , and RSF is
assigned to each thm candidate relay. Let ],,,[ 21 MRSFRSFRSFR be the RSF vector of the
potential relay network obtained from FRBS. The relays are then picked up in the descending
order of the RSF in R . The flowchart of the proposed technique is shown in Fig. 7.17.
Page 129
106
Fig. 7.17: The Flow Chart of the Proposed Algorithm
7.5.4 SIMULATION RESULTS
Fig. 7.18 illustrates the behavior of the proposed scheme for different levels of interference
threshold thI . SNR threshold th is set to 1. Source transmit power is set to 10. We consider three
1m
Initialize transmit
power of thm relay
FRBS computes RSF of thm
relay based on
m
m
f
hand mP
1mm
?Mm
Outputs:
s
m
m
DRs ,
Yes No
Start
Pick the relays in the descending
order of RSF and perform fine
tuning of transmit power of each
selected relay to satisfy
constraints
Page 130
107
different sizes of the potential relay network, i.e. 15,10,5M . As Fig. 7.18(a) shows, overall
transmit power of the relay network increases by relaxing the interference threshold thI , because
high levels of thI enable the relays to freely transmit at high power to meet the QoS requirements
of the secondary network without interfering the primary signals. However there is another
strong observation that as the potential relay network grows in size, the total transmit power
decreases. This is due to the reason that a large relay network provides more opportunities to
FRBS to intelligently pick those relays, which are capable of enabling the secondary
communication performing minimum amplification, at the same time not harmful for the primary
communication. Fig. 7.18(b) shows the total number of selected relays corresponding to different
cases of Fig. 7.18(a). As the figure shows, full cooperative diversity is observed for the
interference threshold level dBI th 10 .On the other hand, setting the interference threshold too
low makes the relay selection problem very difficult and a single best relay is selected to
participate in the communication. Thus, relay selection will become impossible below this level
of interference threshold.
Fig. 7.18(a): Total Transmit Power vs Interference Threshold thI for 1th
-10 0 10 200
5
10
15
20
Ith
(dB)
To
tal T
ran
sm
it P
ow
er
(W)
M = 15
M = 10
M = 5
Page 131
108
Fig. 7.18(b): Corresponding Number Of Selected Relays
In Fig. 7.19, the proposed scheme is analyzed by setting different SNR thresholds for the
secondary network while the interference threshold is set to dB0 . Transmit power of the source
is kept the same as in the case of Fig. 7.18. Again the performance is evaluated considering
three different sizes of potential relay networks, i.e. 15,10,5M . As the figure shows, the
transmit power of the relay network increases when SNR threshold is increased, because in order
to satisfy high SNR threshold, high transmit power is required at the intermediate relay network.
Fig. 7.19: Total Transmit Power Vs SNR Threshold th
-10 0 10 200
4
8
12
16
Ith
(dB)
Nu
mb
er
of S
ele
cte
d R
ela
ys
M = 5
M = 10
M = 15
0.5 1 1.5 27
8
9
10
th
To
tal T
ran
sm
it P
ow
er
(W)
M = 5
M = 10
M = 15
Page 132
109
Furthermore, only slight variations in the SNR threshold were possible in order to study the
system response, owing to the fact that underlay environments does not allow the SUs to transmit
at high power keeping in view the privilege of the PUs. Thus, setting too high SNR threshold
eventually makes the secondary communication impossible. Finally, for large potential relay
networks, total transmit power significantly decreases, because a large network provides more
opportunities to the FRBS to intelligently pick the best combination of relays to satisfy the QoS
requirements of both primary and secondary networks while saving the resources available at
each node.
Fig. 7.20 extends the proposed FRBS assisted relay selection system design to study the behavior
of the relay network for different cases of source transmit power. thI and th are set to dB10 and1
respectively. As observed from the figure, the total power transmitted by the relay network
significantly decreases by increasing the transmit power of the source because the higher is the
power received at the intermediate relay, the lower is the amplification required at each selected
relay to satisfy the SNR threshold.
Fig. 7.20: Total Transmit Power Of For Different Source Transmit Power Levels Keeping
dBI th 10 And 1th
0 5 10 15 20 2510
12
14
16
18
20
PS (dB)
To
tal T
ran
sm
it P
ow
er
(W)
M = 15
M = 10
M = 5
Page 133
110
7.5.5 CONCLUDING REMARKS
A FRBS is proposed to solve multiple relay selection problem for underlay cognitive relay
networks. Proposed FRBS takes the CSI of each candidate relay in the potential relay network as
an input and assigns RSF to each relay, aiming to enable the coexistence of the primary and the
secondary networks with minimum QoS requirements, while utilizing the minimum transmit
power at the relay network. The simulation results confirm the effectiveness of the FRBS-
assisted relay assignment scheme.
Page 134
111
Chapter 8
DETECTION AND ESTIMATION OF
MULTIPLE FAR-FIELD PRIMARY
USERS USING SENSOR ARRAY
Objective
In this chapter, an efficient, reliable and low-complexity spectrum sensing scheme is proposed
for CRNs which not only detects the number of active PUs but also estimates their parameters
such as frequency, power strength and DOA. The proposed scheme is based on GA as global
optimizer hybridized with PS as local optimizer. The system model used for this purpose
constitutes a uniform linear array of sensors. Fitness function is derived from Maximum
Likelihood (ML) principle and defines the MSE between actual and estimated signals. The
effectiveness of the proposed scheme is proved under low SNR conditions. Far-field
approximation is assumed and the signals are detected in the frequency band of 80MHz-
108MHz. The snapshots are available to us after 10-15 seconds.
8.1 BACKGROUND
As explained earlier, spectrum sensing [127] is a process conducted to become aware of the
status of the spectrum usage which involves detection of active signals then estimation of the
signal parameters, followed by decision but it has revamped as a very active area of research
with the advent of cognitive radio technology [128]. In CR, spectrum sensing is a decision
making technique in which SUs are required to dynamically detect spectrum holes to become
aware of the presence of the PUs which have high priority being the licensed users. Being the
core component of CRN, spectrum sensing faces many challenges [129] in terms of hardware
Page 135
112
requirements, hidden terminal problem, detection of spread spectrum primary users,
data/decision fusion in scenarios of cooperative sensing, multipath fading, noise power
uncertainty, implementation complexity, security etc. In order to meet these challenges
efficiently, spectrum sensing requires innovative techniques for not only detecting the number of
PUs but also estimating their amplitudes, DOAs and frequencies to avoid interference between
primary and secondary transmissions.
8.2 SPECTRUM SENSING METHODS
A number of spectrum sensing methods to detect spectrum holes in CRs have been proposed in
literature which have been broadly categorized into three main classes: Non-cooperative
spectrum sensing [130], cooperative spectrum sensing [131]-[132] and interference based
spectrum sensing [133].
8.2.1 NON-COOPRATIVE SPECTRUM SENSING
Non-cooperative spectrum sensing also known as transmitter detection is further classified into
Energy Detection (ED), Matched Filter Detection (MFD) and Cyclostationary based Detection
(CBD).
8.2.1.1 Energy Detector
Energy Detector [134] is the most widely studied spectrum sensing technique because of its less
complexity and no requirement of prior knowledge of PU signal, but it is accompanied by a
number of shortcomings which include noise power uncertainty, poor performance under low
SNR and inefficient to detect spread spectrum signals.
8.2.1.2 Matched Filter Detector
MFD [135] is considered as the optimum method of signal detection when perfect knowledge of
PU is available otherwise it performs poorly. Implementation complexity of MF is impractically
large because it demands CR to have dedicated receivers for all signal types.
Page 136
113
8.2.1.3 Cyclostationary Detector
CBD [136] relies on the prior knowledge of PU signals and exploits cyclostationary features of
the received signals, hence it is capable of differentiating PU signals and noise. Its
implementation complexity lies between energy detector and matched filter.
8.2.2 INTERFERENCE BASED SPECTRUM SENSING
The focus of interference-based spectrum sensing is to design the CRNs to operate in underlay
spectrum sharing environment. In this method, SUs do not perform spectrum sensing to find
spectrum opportunities rather they identify spectrum occupancy status of PUs and an interference
power threshold is set up for SUs towards PUs for a particular frequency band and location.
8.2.3 COOPERATIVE SPECTRUM SENSING
In cooperative spectrum sensing, SUs collaborate and share sensing information to solve
problems like hidden terminal problem, receiver uncertainty and multipath fading at the cost of
increased detection delay and high implementation complexity due to requirement of control
channels efficient information sharing algorithms.
8.3 SOURCE LOCALIZATION
Source localization by means of sensor arrays has been one of the fundamental and effective
ways to estimate amplitude, frequency, DOA and range estimation of both far and near field
sources upto high accuracy in many systems including radar, navigation and wireless
communication systems. In order to achieve optimum performance of a sensor array [137], array
geometry, the number of sensors and the physical separation between the sensors are critical
design parameters in addition to the number of other factors including signal-to-noise ratio.
Many algorithms have already been proposed in array signal processing for source localization
which can be categorized into far-field source localization and near-field source localization on
the basis of range between the radiating source and the array of sensors.
Page 137
114
Far-field source localization algorithms make assumption that sources are located in the far-field
region of the array. Thus each signal arriving at the array has planar wavefront. ESPRIT
algorithm [138] and MUSIC algorithm [139] are among the widely studied far-field source
localization algorithms. However, the far-field assumption is no longer valid when the sources
are located close to the array and are described by spherical wavefront assumption, thus range
parameter is also included in addition to amplitude, DOA and frequency to characterize the
sources. A number of techniques have been proposed in this area such as 2-D MUSIC [140],
Linear Prediction method [141], higher order ESPRIT-method [142] etc but most of these
algorithms are computationally complex.
8.4 CONTIBUTION OF THE CHAPTER
This dissertation addresses the problem of detecting the number of active PU signals and then
estimating their signal parameters to ensure interference free communication in CRNs. Most of
the existing techniques to determine the number of sources are based on the Singular Value
Decomposition SVD of the covariance matrix of the snapshots which yields M distinct
eigenvalues, where M is the number of signals present and the remaining eigenvalues are either
zero or non-zero repeated eigenvalues [143] or non-zero eigenvalues less than threshold.
However, SVD has high uncertainty in terms of decisions about setting of the threshold and so
different schemes [144] have been proposed for threshold setting to detect the presence of
signals. These include Maximum Eigenvalue Detection (MED), Maximum Minimum Eigenvalue
(MME), Maximum Eigenvalue to Trace (MET) etc. Unfortunately, most of the existing methods
are either problem specific or computationally complex due to exhaustive comparisons of test
hypothesis involved to achieve high accuracy. In [145], a technique is proposed to detect number
of signals in order to solve problem of DOA.
In this dissertation, a generalized spectrum sensing method is proposed to first detect the number
of possibly active PUs located in the far field region of the array and then estimate their
amplitudes, DOAs and frequencies. The proposed spectrum sensing scheme is not application
specific. It can be used for cooperative as well as non-cooperative spectrum sensing. Mean
Square Error (MSE) is used as a fitness function which defines an error between actual and
Page 138
115
estimated signals at different sensors of the ULA and is derived from ML Principle. MSE is one
of the easy and optimum fitness functions to be minimized using array of sensors and fairly good
results are obtained even in the scenario of low SNR. We employed heuristic optimization
techniques to minimize the error in which GA being one of the most popular evolutionary
algorithm because of its reliability, efficiency and robustness is used as global optimizer
hybridized with PS as local optimizer. This simple and elegant technique simply demands a
passive sensor array whose snapshots should be readily available to us for calculation after every
10-20 seconds.
8.5 SYSTEM MODEL AND PROBLEM FORMULATION
We have an array of sensors that is sensing the signals from different base stations of primary
users. If the array is almost at the same height as that of the base station transmitters, we do not
have to detect the elevation angle . So, consider a uniform linear array as shown in Fig. 8.1
consisting of L omnidirectional sensors observing M far-field primary signals radiating with
different unknown carrier frequencies. The distance d between two consecutive elements is kept
one-quarter of the minimum wavelength of received signals i.e. 4
min.
Fig. 8.1: The System Model
Page 139
116
The composite signal ix received by the thi sensor is expressed as,
i
M
m
idjk
mi zeax mm
1
sin)1( Li 1 )1.8(
where ma and m represent the amplitude and DOA of the thm source impinging on the array, mk
is the propagation constant and cfk mmm /2/2 with mf representing the frequency of
the thm signal incident on the array and iz is the AWGN added to the output of thi sensor. Thus
the parameters to be estimated for M incident sources are expresses in a vector as,
where M is the number of active PUs and is also unknown and has to be detected first.
The received signal vector X at the L-element ULA is expressed as,
T
LLi xxxxxX ],.,,..................,,[ 121
where superscript T denotes the transpose.
Thus the problem in hand is to develop a novel technique for two purposes, first detecting the
number of active PU signals impinging on ULA and second, performing joint estimation of
amplitude, DOA and frequency of the detected sources considering the sensor array as reference.
We also consider the effect of variation in SNR on the detection and estimation results. The
fitness function can be expressed mathematically as,
2
,,,
ˆmin XXfaM g
)2.8(
where X defines the estimated signal vector at the sensor array and is given as,
T
LLi xxxxxX ]ˆ,ˆ.,,.........ˆ.........,ˆ,ˆ[ˆ121
ix is the estimated output at the ith
sensor and is expressed as,
g
mm
M
m
idkj
mi eax1
ˆsin)1(ˆ
'
''
'ˆˆ
Li 1 )3.8(
],......,,,......,,,......,[ 111 MMM ffaa
Page 140
117
where gM is the number of sources randomly selected to detect the possibly active PUs.
Thus the elements of the estimated vector obtained through optimization algorithm are given
by,
8.6 PROPOSED ALGORITHM FOR DETECTION OF PUS
In this section, we give an overview of the procedural steps carried out in GA optimization,
parameter settings for GA and hybrid scheme PS, and pseudo code for the proposed algorithm.
We solve the problem of detection first. To achieve this purpose, we randomly select gM
number of sources in the estimated signal vector and calculate mean square error MSE given
in 8.2. The value of gM is then increased or decreased aiming to decrease the MSE in each
selection. This process is repeated until minimum mean square error MMSE is obtained with
gM corresponding to MMSE indicating the number of active PUs. After detection of the number
of sources is done, we perform joint estimation of amplitude, DOA and frequency of the detected
signals by further refining the MMSE. The optimization problem given in 8.2 is solved through
GA hybridized with PS.
GA has been widely used to solve optimization problems in communication and array signal
processing because of being simple in concept, reliable, ease in implementation and with less
probability of getting stuck in local minima [146]. Efficiency, accuracy and reliability of GA can
be considerably improved by hybridization with any other competent computational technique
such as Interior Point Algorithm (IPA), PS etc. In [147], performance of GA, PS and Simulated
Annealing (SA) is compared with GA-PS and SA-PS in the joint estimation of amplitude and
DOA of multiple far-field sources incident on L-type array considering MSE as fitness function.
The flowchart for GA optimization has been provided in chapter 3. The steps followed in GA-PS
optimization are summarized below.
]ˆ,......,ˆ,ˆ,......,ˆ,ˆ,......,ˆ[ˆ 111 ggg MMM ffaa
Page 141
118
_________________________________________________________________________________________
Algorithm: GA hybridized with PS _______________________________________________________________________________________________
Step (i): Initialization
Randomly generate P number of chromosomes (potential solutions). Lower and upper
bounds are specified for the genes.
Step (ii): Fitness Function Evaluation
Fitness of each chromosome in the population is computed using mean square error MSE
derived from ML Principle as fitness function and is given in 8.2. The chromosomes are
sorted on the basis of their fitness values.
Step (iii): Termination Criteria
The algorithm terminates if any of the following two criteria are met, i.e. reaching the
maximum number of cycles or achieving the predefined fitness value.
Step (iv): Create New Generations
Select the best chromosome depending on the value of its fitness and create next
generation by employing mutation and crossover.
Step (v): Fine-Tuning via Local Search
The PS algorithm takes the best chromosome obtained from GA as a starting point for
further improvement and refinement of results.
Step (vi): Storage:
Store the global best and to achieve better results repeat the steps 2 to 5 for sufficient
numbers of iterations for better statistical analysis.
_____________________________________________________________________________
MATLAB optimization toolbox is used for this purpose and parameter settings for GA and PS
are shown in table 8.1. Pseudo code of the proposed algorithm to solve the detection and
estimation problem is provided in table 8.2.
Page 142
119
Table 8.1: Parameter Settings For GA-PS
8.7 SIMULATION RESULTS AND DISCUSSIONS
In this section, the performance of the proposed technique is evaluated in terms of accuracy for
two purposes, first, to detect of number of far-field sources impinging on ULA, and second, for
joint estimation of amplitude, DOA and frequency of the detected signals. Inter-element spacing
in the array is kept4
min. We perform spectrum sensing in the frequency band of 81MHz –
108MHz. The signals received at the array were polluted by AWGN. Different cases are
discussed on the basis of different number of sources M impinging on ULA, different number of
sensors L, and for different SNR levels, with SNR to be as high as 35dB and as low as 15dB. All
the values of DOA and SNR are taken in degrees and dB respectively.
GA PS
Parameters Settings Parameters Settings
Population size 300 Start point Optimal values
from GA
No. of generations 2000 Poll method GPS positive
basis 2N
Selection Stochastic uniform Polling order Consecutive
Mutation function Adaptive feasible Maximum
iterations
1000
Crossover function Heuristic Maximum
function
evaluation
10000
Crossover Fraction 0.2 Function
Tolerance
1e-18
Hybridization PS
No. of generations 3000 Expansion
Factor
2.0
Function Tolerance 1e-15 Contraction
Factor
2.5
Migration Direction Both Way
Penalty Factor
100 Scaling Function Rank
Elite Count 8
Page 143
120
Table 8.2: Pseudocode Of The Proposed Algorithm For Detection Of Number Of Sources
.1,,,: MmwherefadInputs mmm
MforguessaasMChoose g
.1
M
m
idjk
mimmeax
1
sin)1(
Li 1
.2
g
mm
M
m
idkj
mi eax1
ˆsin)1(ˆ
'
''
'ˆˆ
Li 1
.3 compute2
0ˆ
gMM XXE
ErrorSquareMean//
.4 let 1 g
new
g MM
// sourceoneadd
.5 compute
newg
mm
M
m
idkj
mi eax1
ˆsin)1(ˆ
'
''
'ˆˆ
Li 1
.6 compute
2'0
ˆnewgMM XXE
.7 if )( 0'0 EE
.i MofvaluepossibleaasMsaveandEE new
g
'
00
.ii newgMupdate updatelastportingMM new
gnewg sup1//
.iii newgMlastofrecordkeepingwhiletostepsrepeat 75
gincreastartsMSEuntilacquired sin
else
.i 1 g
new
g MM
.ii 75 tostepsrepeat
ifend
directionsbothinEaroundEofvaluesthreeatleastObservemin00.8
new
gg MasEinincreaseensuretoMthatgconsiderin 00
min0EtocorrespondwhichMarounddecreasesorincreases new
g
min0..Re: EeiMMSEtocorrespondthatMturnOutput new
g
Fig. 8.2 illustrates the performance of GA for two incoming sources i.e. M = 2 under different
SNR conditions. A ULA with L = 20 sensors is employed for this purpose. The amplitude A ,
DOA and frequency f of the incoming signals are taken as ,5.4,0.3 21 aa
Page 144
121
,145,75 21
oo MHzfMHzf 100,85 21 where 111 ,, fa correspond to the first PU and
222 ,, fa correspond to the second PU. The obtained results are averaged over 20 snapshots.
Fig. 8.2(a) illustrates the detection of two sources with gM ranging from 1 to 6. Minimum
Mean square error (MMSE) is plotted against the number of sources gM in the estimated signal
vector which clearly gives the minimum value when gM coincides with M. The figure also
indicates that increase in error is less significant in the case when gM > M as compared to the
case when gM < M which represents an under-determined system i.e. number of solutions are
less than the number of unknowns. After the detection of active sources, table 8.3 provides the
estimates of amplitudes, DOAs and frequencies of both PUs for different values of SNR. Fig.
8.2(b) and Fig. 8.2(c) plot error in DOA and frequency of the incident signals versus SNR
respectively and it is obvious from the figures that estimation accuracy increases to 99.87% in
DOA and 99.77% in frequency as the SNR increases from 15dB to 35dB.
Fig. 8.2(a): Detection of M = 2 PUs
1 2 3 4 5 610
-4
10-3
10-2
10-1
100
101
102
Mg
Me
an
Sq
ua
re E
rro
r
SNR = 30dB
SNR = 25dB
SNR = 20dB
Page 145
122
Fig. 8.2(b): Error in DOA vs SNR for M = 2, L = 20
Fig. 8.2(c): Error in frequency vs SNR for M = 2, L = 20
15 20 25 30 350.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0.11
0.12
0.13
SNR (dB)
Err
or
in
(D
eg
ree
s)
delta
1
delta 2
15 20 25 30 350.1
0.12
0.14
0.16
0.18
0.2
0.22
SNR (dB)
Err
or
in f(
MH
z)
delta f
1
delta f2
Page 146
123
Table 8.3: Amplitude, DOA And Frequency Estimation For Different SNR Levels With M = 2,
L = 20
SNR
35dB 3.00 4.50 75.04 144.97 84.89 100.10
30dB 3.00 4.50 74.95 144.96 85.13 100.12
25dB 2.99 4.51 75.07 145.06 84.84 99.85
20dB 2.98 4.52 74.91 145.08 84.82 100.19
15dB 3.02 4.48 74.90 144.89 85.19 100.21
In Fig. 8.3 illustrates the performance of GA-PS with M = 4 primary users. ULA with L = 25
sensors is used for this purpose. The values of amplitude, DOA and frequency of the sources are
taken as },81,60,2{ MHzo },88,90,5.2{ MHzo },95,135,3{ MHzo
and }.105,160,5.3{ MHzo
Figure 8.3(a)
plots MMSE versus gM to detect the number of active sources by setting gM in the range of 1
to 7 and it is obvious from the figure that error is minimum when MM g giving a clear
indication of 4 active PUs. Fig. 8.3(b) and 8.3(c) plot error in DOA and frequency estimates of
the detected users versus different SNR levels with SNR raised from 15dB to 35dB. The values
estimated by GA are tabulated in table 8.4. The results are averaged over 20 snapshots. Table 8.4
provides the amplitude, DOA and frequency estimates obtained. Fig. 8.3 proves the validity of
the proposed technique when the number of signals incident on the array increases and it can still
simultaneously estimate amplitudes, DOAs and frequencies with high estimation accuracy.
Figure 8.3(a): Detection of M = 4 PUs
1 2 3 4 5 6 710
-4
10-3
10-2
10-1
100
101
102
Mg
Me
an
Sq
ua
re E
rro
r
SNR = 30dB
SNR = 25dB
SNR = 20dB
1a 2a1 2 )(ˆ
1 MHzf )(ˆ2 MHzf
Page 147
124
Fig. 8.3(b): Error in DOA vs SNR for M = 4, L = 25
Fig. 8.3(c): Error in frequency vs SNR for M = 4, L = 25
15 20 25 30 35
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
SNR (dB)
Err
or
in
(Degre
es)
delta fi
1
delta fi2
delta fi3
delta fi4
15 20 25 30 350.2
0.25
0.3
0.35
0.4
0.45
SNR (dB)
Err
or
in f (
MH
z)
delta f
1
delta f2
delta f3
delta f4
Page 148
125
Table 8.4: Amplitude, DOA and frequency estimation for different SNR levels with M = 4, L =
25
In Fig. 8.4 we evaluate the performance of our proposed scheme with different number of
sensors L in the array as the SNR is raised from 15dB to 35dB. Number of PUs and the values
of amplitudes, DOAs and frequencies of the PUs are kept the same as in the case of figure 8.2.
The values estimated by GA-PS are tabulated in table 8.5. The results are averaged over 20
snapshots. It is obvious from Fig. 8.4(a) and Fig. 8.4(b) that the greater the number of sensors in
the array, the higher is the accuracy in the estimated values with further improvement achieved at
high SNR levels.
Fig. 8.4(a): Error in DOA estimation for different SNR levels and different number of sensors in
the array considering M = 2
SNR
35dB 2.00 2.50 3.00 3.50 60.20 90.18 135.19 159.84 81.23 88.24 94.79 104.77
30dB 1.99 2.51 3.00 3.48 59.78 90.24 134.75 159.79 80.71 88.28 94.74 104.72
25dB 1.98 2.41 3.01 3.47 59.72 90.31 134.73 160.28 80.68 87.67 95.33 104.69
20dB 2.02 2.52 2.99 3.53 59.62 89.66 135.37 160.34 80.63 88.35 95.39 105.38
15dB 2.03 2.47 2.98 3.54 60.40 90.34 134.59 159.63 81.45 87.39 95.43 105.42
6 8 10 12 14 16 18 20 220.06
0.08
0.1
0.12
0.14
0.16
0.18
L
Err
or
in
(De
gre
es)
SNR = 30dB
SNR = 25dB
SNR = 20dB
SNR = 15dB
1a 2a 3a14a 2 3 4 )(ˆ
1 MHzf )(ˆ2 MHzf )(ˆ
3 MHzf )(ˆ4 MHzf
Page 149
126
Fig. 8.4(b): Error in frequency estimation for different SNR levels and different number of
sensors in the array considering M = 2
Table 8.5: Amplitude, DOA and frequency estimation for different SNR levels and different
number of sensors in the array with M = 2
SNR
L
30dB
6 3.04 4.54 75.12 144.88 85.22 100.23
10 3.03 4.53 75.11 144.90 85.21 100.22
14 3.02 4.53 74.91 144.93 84.83 100.19
18 2.98 4.49 75.07 145.05 85.16 99.82
22 3.00 4.50 74.94 144.96 85.15 100.16
25dB
6 2.96 4.55 75.13 145.12 85.23 99.77
10 2.95 4.46 75.12 145.11 85.22 100.22
14 2.95 4.48 74.91 145.09 84.81 100.18
18 3.02 4.48 74.91 145.07 84.82 100.17
22 2.99 4.51 74.92 144.93 85.17 99.83
20dB
6 3.06 4.57 75.15 144.84 85.25 99.76
10 3.05 4.56 75.13 145.15 84.77 99.75
14 3.03 4.53 75.12 145.12 84.80 100.23
18 2.98 4.51 74.90 145.09 84.82 99.80
22 2.99 4.49 74.91 144.92 85.17 100.18
15dB
6 3.07 4.42 75.17 144.82 85.27 100.26
10 2.96 4.57 74.83 144.84 84.78 99.75
14 3.03 4.45 74.86 145.12 85.79 100.23
18 2.98 4.46 74.88 145.08 85.81 99.79
22 2.98 4.47 75.10 145.09 84.82 99.80
6 8 10 12 14 16 18 20 22
0.16
0.18
0.2
0.22
0.24
0.26
0.28
L
Err
or
in f
1(M
Hz)
SNR = 30dB
SNR = 25dB
SNR = 20dB
SNR = 15dB
1a 2a 1 2 )(ˆ1 MHzf )(ˆ
2 MHzf
Page 150
127
8.8 CONCLUDING REMARKS
In this chapter, a novel idea based on GA hybridized with PS is presented for detecting the
number of active PUs and estimation of joint amplitudes, DOAs and frequencies of the detected
users for CRNs. The proposed method is not application specific, the signal parameters are
paired automatically and estimated with high accuracy. Moreover, the proposed algorithm has
less computation burden and offers satisfactory results even when number of users increases. The
simulation results verify the effectiveness of the proposed algorithm in AWGN environment.
Page 151
128
Chapter 9
CONCLUSIONS AND FUTURE WORK
9.1 CONCLUSIONS
In this dissertation, we studied AF based relay-assisted cognitive radio networks in an underlay
spectrum sharing environment. The underlay networks impose strict interference constraints
towards the secondary users which limit their transmit power and allows only short-range
communication. Thus, performance enhancement of secondary communication in the frequency
bands allocated to the PUs is a major design challenge faced by the underlay RCRNs. It requires
relay selection along with the fine tuning and adjustment of the transmit power of the secondary
relays. The main contributions of the dissertation in this area are summarized as follows.
Various “Multiple Relay Selection” schemes are proposed to enable secondary communication
in an underlay spectrum sharing scenario. Rayleigh flat-fading is considered for this purpose,
assuming the availability of perfect instantaneous channel state information (CSI). Relay
selection aims to select the best combination of relays aiming to maximize the signal-to-noise
ratio (SNR) achieved at the destination while adhering to the interference constraint of the
primary network.
Another critical issue in the underlay networks is to optimize the transmit power consumption at
the cognitive relay network while ensuring the minimum quality-of-service requirements of the
primary and secondary networks. This dissertation formulates a transmit power minimization
problem and proposes a relay subset selection algorithm aiming to select the optimal
combination of relays that minimizes the total transmit power utilized at the relay network while
satisfying the minimum interference threshold of the primary network and the SNR threshold of
the secondary network simultaneously. The problem of transmit power minimization of the
Page 152
129
cognitive relay network operating in an underlay scenario has not been solved by relay selection
so far to the best of our knowledge.
This dissertation also highlights the deficiency in the performance analysis of multiple relay
selection in AF-based underlay RCRNs. For this purpose, a relay subset selection algorithm is
proposed aiming to maximize the SNR received at the destination, keeping in view the
interference threshold of the primary network. The closed form expressions for the outage
probability and average probability of error have been derived through the CDF of the received
SNR at secondary destination, which has not been done in the literature so far, for multiple relay
selection in AF-based underlay RCRNs.
Moving one step further and acknowledging the increasing popularity of the famous Artificial
Intelligence tool, the fuzzy logic, FRBS assisted relay selection and transmit power allocation
(RSTPA) schemes are proposed for intelligent relay selection to solve the highlighted problems
of SNR maximization and transmit power minimization in CRNs.
The effects of variations in the instantaneous CSI, transmit power of source and relays,
interference threshold of the primary network, SNR threshold of the secondary network and size
of the potential relay network on multiple relay selection in underlay RCRNs are the main issues
that are analyzed in depth for all the schemes proposed in this research.
The last topic that is studied in this dissertation is spectrum sensing in Cognitive Radios. In order
to preserve the PUs’ rights of interference-free operation, the SUs are required to sense the
licensed bands at regular intervals, and reliably detect the primary signals. For this purpose, a
novel idea of uniform linear array (ULA) based spectrum sensing is proposed alongwith a hybrid
GA based algorithm. It not only detects the number of active PUs, but also provides the estimates
of amplitude, frequency and DOA of the active users upto high accuracy. The effectiveness and
reliability of the proposed scheme is proved under low SNR conditions.
9.2 FUTURE WORK
The work presented in this dissertation can be further extended in many directions.
Page 153
130
It is assumed in all schemes that the perfect knowledge of CSI is available to perform
relay selection and power allocation to the selected relays. However, the accurate
information about the incoming and outgoing channels might not always be available,
which may distract the relay selection procedure. In future, the relay selection procedures
in the scenarios of imperfect CSI will be studied.
The line-of-sight path between source-destination pair and the phenomena of frequency
selectivity are neglected throughout the study, which significantly affect relay selection.
In future, these dominant factors will be incorporated in formulating relay selection
problem and performing relay selection and power allocation.
The spectrum sensing scheme proposed in this dissertation is based on the novel idea of
employing uniform linear array of sensors. Since, the relay network behaves like a virtual
array of distributed antennas in space, the proposed spectrum sensing method will be
extended to the relay network. The idea is to perform cooperative spectrum sensing
through potential relay network to detect the number of primary signals and the signal
parameters. Furthermore, relay selection and/or beamforming techniques will be applied
to enhance secondary performance, while guaranteeing to satisfy the interference
threshold of primary network.
The focus of the whole research is to enable secondary communication under interference
constraints, i.e. underlay mode of spectrum sharing. The problem of performance
enhancement of secondary communication in overlay and underlay mode when primary
signal is present and the interweave mode when primary signal is absent can be jointly
formulated to make secondary users smart enough to respond to the changing network
conditions and adapt themselves according to the operating environment in which they
operate.
Page 154
131
REFERENCES
[1]. P. Lin, J. Jia, Q. Zhang, M. Hamdi, “Dynamic Spectrum Sharing With Multiple Primary
and Secondary Users”, IEEE Transactions on Vehicular Technology, vol. 60, no. 4, pp. 1756–
1765, 2011.
[2]. L. E. Doyle, “Essentials of Cognitive Radios”, Cambridge Wireless Essentials Series, Cambridge
University, Press, 2009.
[3]. Cognitive Wireless Communication Networks, Springer Science + Business Media, LLC, 2007.
[4]. J. Mitola, G. Q. Maguire, “Cognitive Radio: Making Software Radios ore Personal”, IEEE
Personal Communications, vol. 6, no. 4, pp. 13-18, 1999.
[5]. J.Lee,H. Wang, J.G.Andrews, and D. Hong, “Outage Probability of Cognitive Relay Networks with
Interference Constraints”, IEEE Trans. Wireless Commun., vol. 10, no. 2, pp. 390-395, 2011.
[6]. P. Steenkiste, D. Sicker, G. Minden, and D. Raychaudhuri, “Future Directions in Cognitive Radio
Network Research”, NSF Workshop Report, 2009.
[7]. J. Mietzner, L. Lampe, and R. Schober, “Distributed Transmit Power Allocation for Multihop
Cognitive-Radio Systems," IEEE Trans. Wireless Commun., vol. 8, no. 10, pp. 5187-5201, 2009.
[8]. O. Simeone, U. Spagnolini, and Y. Bar-Ness, “Stable Throughput of Cognitive Radios with and
without Relaying Capability," IEEE Trans. Commun., vol. 55, no. 12, pp. 2351-2360, 2007.
[9]. C. Stevenson, G. Chouinard, Z. Lei, W. Hu, S. Shellhammer, and W. Caldwell, “IEEE 802.22: The
First Cognitive Radio Wireless Regional Area Networks (WRANs) Standards”, IEEE
Communications Magazine, vol. 47, no. 1, pp. 130–138, 2009.
[10]. A.H. Bastami, A. Olfat, “Optimal SNR-based Selection Relaying Scheme with M
Ary modulations in Cooperative Wireless Networks”, IET Commun., vol. 5, no. 6, pp. 797-810,
2011.
[11]. D. Chen, H. Ji, X. Li, “Optimal Distributed Relay Selection in Underlay Cognitive Radio
Networks: An Energy-Efficient Design Approach”, IEEE WCNC, 2011, pp. 1203-1207.
[12]. F. GABRY, N. LI, N. Schrammar, M. Girnyk, E. Karipidis, R. Thobaben, L. K. Rasmussen, E.
G. Larsson and M. Skoglundi, “Secure Broadcasting in Cooperative Cognitive Radio Networks”,
Future Network & Mobile Summit (FutureNetw), 2012, pp. 1-8.
[13]. M. Chen, Ted C.-K. Liu, and X. Dong, “Opportunistic Multiple Relay Selection with Outdated
Channel State Information”, IEEE Trans. Vehicular Technology, vol. 61, no. 3, pp. 1333-1345,
2012.
[14]. T. Nechiporenko, P. Kalansuriya, and C. Tellambura, “Performance of Optimum Switching
Adaptive M-QAM for Amplify-and-Forward Relays”, IEEE Trans. Vehicular Technology, vol. 58,
no. 5, pp. 2258-2268, 2009.
Page 155
132
[15]. M. Naeem, U. Pareek, and D. C. Lee, “Power Allocation for Non-Regenerative Relaying in
Cognitive Radio System”, WiMob IEEE, 2010, pp. 720-725.
[16]. K-C Chen, R. Prasad, “Cognitive Radio Networks”, Willey Press, 2009.
[17]. E. Hossain, D. Niyato, Z. Han, “Dynamic Spectrum Access and Management in Cognitive Radio
Networks”, Cambridge University Press, 2009.
[18]. C.Bazelon,“Licensed or Unlicensed: The Economic Considerations
in Incremental Spectrum locations”, IEEE Commun. Magazine, vol. 47, no. 3, pp. 110-116, 2009.
[19]. Y. Xiao, G. Bi, and D. Niyato, “Game Theoretic Analysis for Spectrum Sharing with Multi-Hop
Relaying” , IEEE Trans. Wireless Commun., vol. 10, no. 5, pp. 1527-1537, 2011.
[20]. M. Gandetto, C. Regazzoni, “Spectrum Sensing: A Distributed Approach for Cognitive
Terminals”, IEEE J. Selected Areas in Communications, vol. 25, no. 3, pp. 546-557, 2007.
[21]. A. G. Fragkiadakis, E. Z. Tragos, I. G. Askoxylakis, “A Survey on Security Threats and
Detection Techniques in Cognitive Radio Networks Communications”, IEEE Surveys & Tutorials,
vol. 15, no. 1, pp. 428-445, 2013.
[22]. C. Ma, G. Yu, J. Zhang, “A Node Grouping Algorithm for Joint Relay Selection and Resource
Allocation in Cooperative Cognitive Radio Networks” , Int. Conf. Wireless Commun. and Signal
Processing (WCSP), 2011, pp. 1-5.
[23]. A. M. Wyglinski, M. Nekovee, T. Hou, “Cognitive Radio Communications and Networks,
Principles and Practice”, Elsevier, 2010.
[24]. G. J. Minden, J. B. Evans, L. S. Searl, D. DePardo, R. Rajbanshi, J. Guffey, Q. Chen, T. R.
Newman, V. R. Petty, F. Weidling, M. Peck, B. Cordill, D. Datla, B. Barker, and A. Agah,
“Cognitive Radios for Dynamic Spectrum Access- An Agile Radio for Wireless Innovation”, IEEE
Commun. Magazine, vol. 45, no. 5, pp. 113-121, 2007.
[25]. B. Razavi, “Challenges in the Design of Cognitive Radios”, IEEE 1 Custom Integrated Circuits
Conference, 2009, pp. 391-398.
[26]. T. Yucek and H. Arslan, “A Survey of Spectrum Sensing Algorithms for Cognitive Radio
Applications”, IEEE Commun. Surveys & Tutorials, vol. 11, no. 1, pp. 116-130, 2009.
[27]. H. Sadeghi, P. Azmi and H. Arezumand, “Cyclostationarity-based Soft Cooperative Spectrum
Sensing for Cognitive Radio Networks”, IET Commun., vol. 6, no. 1, pp. 29-38., 2012.
[28]. A. A. Khan, S. M. I. Rahman, M. Ahmed, “Research Challenges of Cognitive Radios”, Int. J.
Engg. Research Tech, vol 1, no. 3, pp. 1-4., 2012.
[29]. J. Marinho, E. Monteiro, “Cognitive Radio: Technology Survey and Future Research Directions”,
ANACOM 2013, pp. 1-9.
[30]. K. C. Chen, Y. J. Peng, N. Prasad, Y. C. Liang, and S. Sun, “Cognitive Radio Network
Architecture: Part I - General Structure”, ACM Int. Conf. Ubiquitous Information Management
Commun. (ICUIMC), 2008, pp. 114-119.
Page 156
133
[31]. A. G. Fragkiadakis, E. Z. Tragos, I. G. Askoxylakis, “A Survey on Security Threats and
Detection Techniques in Cognitive Radio Networks”, IEEE Coomun. Survey and Tutorials, vol. 15,
no. 1, pp. 428-445, 2013.
[32]. S. Srinivasa and S. A. Jafar, “The Throughput Potential of Cognitive Radio: A
Theoretical Perspective”, IEEE Commun. Magazine, vol. 45, no. 5, pp. 73-79., 2007.
[33]. M. Choi, J. Park, and S. Choi, “Simplified Power Allocation Scheme for Cognitive Multi-Node
Relay Networks”, IEEE Trans. Wireless Commun., vol. 11, no. 6, pp. 2008-2012, 2012.
[34]. D. W. Bliss, K.W. Forsythe, A. M. Chan, “MIMO Wireless Communication”, Lincoln Laboratory
Journal, vol. 15, no. 1, pp. 97-126, 2005.
[35]. T. Y. Wang, and J. Y. Wu, “Cooperative Communications Using Reliability-Forwarding
Relays”, IEEE Trans. Commun., vol. 61, no. 5, pp. 1776-1785, 2013.
[36]. D. Liu W. Wang, and W. Guo, “ ‘Green’ Cooperative Spectrum Sharing Communication”, IEEE
Commun. Letters, vol. 17, no. 3, pp. 459-462, 2013.
[37]. P. Anghel and M. Kaveh, “Exact Symbol Error Probability of a Cooperative Network in a
Rayleigh-Fading Environment,” IEEE Trans. Wireless Commun., vol. 3 (5), 2004, pp. 1416–1422.
[38]. S. Ikki and M. Ahmed, “Performance Analysis of Cooperative Diversity Wireless Networks over
Nakagami Fading Channel,” IEEE Commun. Letters, vol. 11, no. 4, pp. 334–336, 2007.
[39]. J. Mietzner, R. Schober, L. Lampe, W. H. Gerstacker, and P. A. Hoeher, “Multiple-Antenna
Techniques for Wireless Communications – A Comprehensive Literature Survey”, IEEE
Commun. Surveys & Tutorials, vol. 11, no. 2, pp. 87-105, 2009.
[40]. B. Nazari, A. Jamalipour, “Cooperative Communication with Asymmetric Channel State
Information: A Contract Theoretic Modeling Approach” IEEE China Commun., vol. 10, no. 1, pp.
31-43, 2013.
[41]. H. A. Suraweera, T. A. Tsiftsis, G. K. Karagiannidis, and A. Nallanathan, “Effect of Feedback
Delay on Amplify-and-Forward Relay Networks With Beamforming”, IEEE Trans. Vehicular
Tech., vol. 60, no. 3, pp. 1265-1271, 2011.
[42]. Y. Zou, J. Zhu, B. Zheng, and Y. D. Yao, “An Adaptive Cooperation Diversity Scheme With
Best-Relay Selection in Cognitive Radio Networks”, IEEE Trans. Signal Processing, vol. 58, no.
10, pp. 5438-5445, 2010.
[43]. D. Chen, H. Ji, and V.C.M. Leung, “Distributed Best-Relay Selection for Improving TCP
Performance Over Cognitive Radio Networks: A Cross-Layer Design Approach”, IEEE J. Selected
Areas Commun., vol. 30, no. 2, pp. 315-322, 2012.
[44]. D. Raychaudhuri, and N. B. Mandayam, “Frontiers of Wireless and Mobile Communications”,
Proceedings of the IEEE, vol. 100, no. 4, pp. 824-840, 2012.
[45]. Q. Li, R. Q. HU, Y. Qian, and G. Wu, “Cooperative Communications for Wireless Networks:
Techniques and Applications in LTE-Advanced Systems”, IEEE Wireless Commun., vol. 19, no. 2,
pp. 22-29, 2012.
Page 157
134
[46]. R. Mo, Y. H. Chew, and C. Yuen, “Information Rate and Relay Precoder Design for Amplify-
and-Forward MIMO Relay Networks With Imperfect Channel State Information”, IEEE Trans.
Vehicular Technology, vol. 61, no. 9, pp. 3958-3968, 2012.
[47]. Q. Liu, W. Zhang, X. Ma, G. T. Zhou, “Designing Peak Power Constrained Amplify-and-
Forward Relay Networks with Cooperative Diversity” IEEE Trans. Wireless Communications, vol.
11,no. 5, pp. 1733-1743, 2012.
[48]. A. N. D’Andrea, V. Lottici, and R. Reggiannini, “RF Power Amplifier Linearization Through
Amplitude and Phase Predistortion,” IEEE Trans. Commun., vol. 44 (11), 1996, pp. 1477–1484.
[49]. A. Riberio, X. Cai, and G. Giannakis, “Symbol error probabilities for general cooperative links,”
IEEE Trans. Wireless Commun., vol. 4, no. 3, pp. 1264–1273, 2005.
[50]. P. Merkey and E. C. Posner, “Optimum Cyclic Redundancy Codes for Noise Channels,” IEEE
Trans. Inf. Theory, vol. 30, no. 6, pp. 865–867, 1984.
[51]. A. Chakrabarti, A. Sabharwal, B. Aazhang, “Cooperative Communications: Fundamental Limits
and Practical Implementation”, Rice University.
[52]. J. Si, Z. Li, X. Chen, B. Hao, and Z. Liu, “On the Performance of Cognitive Relay Networks
Under Primary User’s Outage Constraint”, IEEE Commun. Letters, vol. 15, no. 4, pp. 422-424,
2011.
[53]. I. Krikidis, J. N. Laneman, J. S. Thompson, and S. McLaughlin, “Protocol Design and
Throughput Analysis for Multi-User Cognitive Cooperative Systems”, IEEE Trans. Wireless
Commun., vol. 8, no. 9, pp. 4740-4751, 2009.
[54]. L. Li, X. Zhou, H. Xu, G. Y. Li, D. Wang, and A. Soong, “Simplified Relay Selection and Power
Allocation in Cooperative Cognitive Radio Systems”, IEEE Trans. Wireless Communications, vol.
10, no. 1, pp. 33-36, 2011.
[55]. S. Sengupta, K. P. Subbalakshmi, “Open Research Issues in Multi-Hop Cognitive Radio
Networks, IEEE Commun. Magazine, vol. 51, no. 4, pp. 168-176, 2013.
[56]. M. Xia and S. Aissa, “Cooperative AF Relaying in Spectrum-Sharing Systems: Performance
Analysis under Average Interference Power Constraints and Nakagami-m Fading”, IEEE Trans.
Commun., vol. 60, no. 6, pp. 1523-1533, 2012.
[57]. Y. Chen, G. Yu, Z. Zhang, H. Chen, and P. Qiu, “On Cognitive Radio Networks with
Opportunistic Power Control Strategies in Fading Channels”, IEEE Trans. Wireless Commun., vol.
7, no. 7, pp. 2752-2761, 2008.
[58]. M. Naeem, D.C. Lee, U. Pareek, “An Efficient Multiple Relay Selection Scheme for Cognitive
Radio Systems”, IEEE ICC 2010, pp. 1-5.
[59]. K. B. Fredj, and S. Aissa, “Performance of Amplify-and-Forward Systems with Partial Relay
Selection under Spectrum-Sharing Constraints”, IEEE Trans.
Wireless Communications, vol. 11, no. 2, pp. 500 –504, 2012.
Page 158
135
[60]. D. Li, “Outage Probability of Cognitive Radio Networks with Relay Selection”, IET Commun.,
vol. 5, no. 18, pp. 2730-2735, 2011.
[61]. M. Seyfi, S. Muhaidat, J. Liang, “Relay Selection in Cognitive Radio Networks with Interference
Constraints”, IET Commun., vol. 7, no. 10, pp. 922-930, 2013.
[62]. S. I. Hussain, M. S. Alouini, K. Qaraqe and M. Hasna, “Reactive Relay Selection in Underlay
Cognitive Networks with Fixed Gain Relays”, IEEE Int. Conf. Commun., 2012, pp. 1784-1788.
[63]. V. N. Q. Bao, T. Q. Duong, D. B. Costa, G. C. Alexandropoulos, and A. Nallanathan, “Cognitive
Amplify-and-Forward Relaying with Best Relay Selection in Non-Identical Rayleigh Fading, IEEE
Commun. Letters, vol. 17, no. 3, pp. 475-478, 2013.
[64]. H. Yu, W. Tang, and S. Li, “Outage Probability and SER of Amplify-and-Forward Cognitive
Relay Networks, IEEE Wireless Commun. Letters, vol. 2, no.2, pp. 219-222, 2013.
[65]. J. Xu, H. Zhang, D. Yuan, Q. Jin, and C. X. Wang, “Novel Multiple Relay Selection Schemes in
Two-Hop Cognitive Relay Networks”, IEEE INt. Conf. Commun. and Mobile Computing, 2011,
pp. 307-310.
[66]. M. Naeem, D. C. Lee, and U. Pareek, “An Efficient Multiple Relay Selection Scheme For
Cognitive Radio Systems”, IEEE Int. Conf. Commun., 2010, pp.1-5.
[67]. A. Ponsich, A. L. Jaimes, and C. A. Coello, “A Survey on Multiobjective Evolutionary
Algorithms for the Solution of the Portfolio Optimization Problem and Other Finance and
Economics Applications”, IEEE Trans Evolutionary Computation, vol. 17, no. 3, pp. 321-344,
2013.
[68]. T. Blickle, “Theory of Evolutionary Algorithms and Application to System Synthesis”, Ph.D.
Dissertation, 1967.
[69]. W. F. Gao, S. Y. Liu, and L. L. Huang, “A Novel Artificial Bee Colony Algorithm Based on
Modified Search Equation and Orthogonal Learning”, IEEE Trans. Cybernetics, vol. 43, no. 3, pp.
1011-1024, 2013.
[70]. W. Peng, Y. Zeng, H. Gong, Y. Q. Leng, Y. H. Yan, and W. Hu, “Evolutionary Algorithm and
Parameters Extraction for Dye-Sensitised Solar Cells One-Diode Equivalent Circuit Model”, Micro
and Nano Letters, vol 8, no. 2, pp. 86-89, 2013.
[71]. A. Sharaqa, N. Dib, “Position-Only Side Lobe Reduction of a Uniformly Excited Elliptical
Antenna Array using Evolutionary Algorithms”, IET Microw. Antennas Propag., vol. 7, no. 6, pp.
452-457, 2013.
[72]. W. R. A. Ibrahim and M. M. Morcos, “Artificial Intelligence and Advanced Mathematical Tools
for Power Quality Applications: A Survey”, IEEE Trans. Power Delivery, vol. 17, no. 2, pp. 668 –
673, 2002.
[73]. S. K. Pal, V. Talwar, and P. Mitra, “Web Mining in Soft Computing Framework: Relevance,
State of the Art and Future Directions”, IEEE Trams. Neural Networks, vol. 13, no. 5, pp. 1163-
1177, 2002.
Page 159
136
[74]. R. A. Vural, T. Yildirim, T. Kadioglu, and A. Basargan, “Performance Evaluation of
Evolutionary Algorithms for Optimal Filter Design”, IEEE Trans. Evolutionary Computation”, vol.
16, no.1, pp. 135-147, 2012.
[75]. D. Karaboga and B. Basturk, “On the performance of Artificial Bee Colony (ABC) Algorithm,”
Appl. Soft Comput., vol. 8, no.1, pp. 687–697, 2008.
[76]. R. Wang, R. C. Purshouse, and P. J. Fleming, “Preference-Inspired Coevolutionary Algorithms
for Many-Objective Optimization”, IEEE Trans Evolutionary Computation, vol. 17, no. 4, pp. 474-
494, 2013.
[77]. L. D. S. Coelho and P. Alotto, “Gaussian Artificial Bee Colony Algorithm Approach Applied to
Loney’s Solenoid Benchmark Problem”, IEEE Conf. Electromagnetic Field Computation, 2010,
pp.1.
[78]. L. Costa, I.A.C.P.E. Santo, R. Denysiuk and E.M.G.P. Fernandes, “Hybridization of a Genetic
Algorithm with a Pattern Search Augmented Lagrangian Method”, Int. Conf. Engineering
Optimization, 2010, pp.1-10.
[79]. Z. He, P.F. Zhu, “A Hybrid Genetic Algorithm for Multiresponse Parameter Optimization within
Desirability Function Framework”, Int. Conf Industrial Engineering and Engineering Management,
2009, pp. 612-617.
[80]. Y.G. Petalas, K.E. Parsopoulos, M.N. Vrahatis, “Memetic Particle Swarm Optimization”, Ann
Oper Res, 156, 2007, pp. 99-127.
[81]. L. D. S. Coelho and V. C. Mariani, "Combining of Chaotic Differential Evolution and Quadratic
Programming for Economic Dispatch Optimization with Valve-Point Effect," IEEE Trans. Power
Systems, vol. 21, pp. 989-996, 2006.
[82]. F. S. Abu-Mouti, and M. E. El-Hawary, “Optimal Distributed Generation Allocation and Sizing
in Distribution Systems via Artificial Bee Colony Algorithm”, IEEE Trans. Power Delivery, vol.
26, no. 4, pp. 2090-2101, 2011.
[83]. D. Karaboga and B. Basturk, “A Powerful and Efficient Algorithm for Numerical Function
Optimization: Artificial Bee colony (ABC) Algorithm”, J. Global Optimiz., vol. 39, pp. 459–471,
2007.
[84]. D. Karaboga, B. Akay, “A Comparative Study of Artificial Bee Colony Algorithm”, Appl. Math.
Comput., vol. 214, no. 1, pp. 108–132, 2009.
[85]. D. Jeya Mala, V. Mohan, M. Kamalapriya, “Automated Software Test Optimization Framework –
an Artificial Bee Colony Optimization-Based Approach”, IET Software, 2010.
[86]. X. Li, M. Yin, “Hybrid Differential Evolution with Artificial Bee Colony and its Application for
Design of a Reconfigurable Antenna Array with Discrete Phase Shifters”, IET Microwaves,
Antenna and Propagation, vol. 6, no. 14, pp. 1573-1582, 2012.
Page 160
137
[87]. N. Taspinar, D. Karaboga, M. Yildirim, and B. Akay, “Partial Transmit Sequences based on
Artificial Bee Colony Algorithm for Peak-to-Average Power Ratio Reduction in Multicarrier Code
Division Multiple Access Systems”, IET Commun., vol. 5, no. 8, pp. 1155-1162, 2011.
[88]. Y. Wang, W. Chen, and C. Tellambura, “A PAPR Reduction Method Based on Artificial Bee
Colony Algorithm for OFDM Signals”, IEEE Trans. Wireless Commun., vol. 9, no. 10, pp. 2994-
2999, 2010.
[89]. N. Todorovic and S. Petrovic, “Bee Colony Optimization Algorithm for Nurse Rostering”, IEEE
Trans. Systems, Man, and Cybernetics, vol. 43, no. 2, pp. 467-473, 2013.
[90]. X. Zhang , X. Zhang , S. Yin Yuen , S. L. Ho , and W. N. Fu, “An Improved Artificial Bee
Colony Algorithm for Optimal Design of Electromagnetic Devices”, IEEE Trans. On Magnetics,
vol. 49, no. 8, pp. 4811-4816, 2013.
[91]. D. S. Weile and E. Michielssen, “Genetic Algorithm Optimization Applied to Electromagnetics:
A Review”, IEEE Trans. Antenna and Propagation, vol. 45, no. 3, pp. 343-353, 1997.
[92]. M. Sanna and M. Murroni, “Optimization of Non-Convex Multiband Cooperative Sensing With
Genetic Algorithms”, IEEE J. Selected Topics Signal Process., vol. 5, no. 1, pp. 87-96, 2011.
[93]. L. Khalid and A. Anpalagan, “Cooperative Sensing With Correlated Local Decisions in Cognitive
Radio Networks”, IEEE Trans. Vehicular Tech., vol. 61, no. 2, pp. 843-849, 2012.
[94]. A. Balieiro, P. Yoshioka, K. Dias, C. Cordeiro and D. Cavalcanti, “Adaptive Spectrum Sensing
for Cognitive Radio based on Multi-Objective Genetic Optimization”, IEEE Electronics Letters,
vol. 49, no. 17, pp. 1099-1101, 2013.
[95]. M. Canales and J.R. Gallego, “Potential Game for Joint Channel and Power Allocation in
Cognitive Radio Networks”, IEEE Electronics Letters, vol. 46, no. 24, pp. 1632-1634, 2010.
[96]. A. Alsharoa, F. Bader, and M.S. Alouini, “Relay Selection and Resource Allocation for Two-
Way DF-AF Cognitive Radio Networks”, IEEE Wireless Commun. Letters, vol. 2, no. 4, pp. 427-
430, 2013.
[97]. S. L. Blond, R. Aggarwal, “A Review of Artificial Intelligence Techniques as Applied to
Adaptive Autoreclosure, with Particular Reference to Deployment with Wind generation”,
Universities Power Engg. Conference, 2009, pp. 1-5.
[98]. D. G. Schwartz, G. J. Klir, H. W. Lewis, and Y. Ezawa, “Applications of Fuzzy Sets and
Approximate Reasoning”, Proceedings of IEEE, vol. 82, no. 4, pp. 482-498, 1994.
[99]. L.A. Zadeh. “A rationale for fuzzy control,” J. Dynamic Syst., Meas., Contr., vol. 94, no. 1, pp.
3-4, 1972.
[100]. -, “Outline of a new approach to the analysis of complex systems and decision processes,” IEEE
Trans. Syst., Man, Cyhern., vol. SMC-3, no. 1, pp. 28-44, 1973.
[101]. K. P. Philip, E.L. Dove, D.D. McPherson, N.L. Gotteiner,W. Stanford, K.B. Chandran,
“The Fuzzy Hough Transform-Feature Extraction in Medical Images”, IEEE Trans. Medical
Imaging, vol. 13, no. 2, pp. 235-240, 1994.
Page 161
138
[102]. J. M. W. Kickert, H. Koppelaar, “Application of Fuzzy Set Theory to Syntactic Pattern
Recognition of Handwritten Capitals”, IEEE Trans. Systems, Man and Cybernetics, vol. SMC-6,
no. 2, pp. 148-151, 1976.
[103]. G. Fang, “A Survey on Analysis and Design of Model-Based Fuzzy Control Systems”, IEEE
Trans. Fuzzy Sys., vol. 14, no. 5, pp. 676-697, 2006.
[104]. A.B. Ajiboye, R.F. Weir, “A Heuristic Fuzzy Logic Approach to EMG Pattern Recognition for
Multifunctional Prosthesis Control”, IEEE Trans. Neural Systems Rehabilitation Engg., vol. 13, no.
3, pp. 280-291, 2005.
[105]. Cui-Yan, J. Y. Tao, W.J. Tao, “Pattern Recognition Based on Fuzzy Three-Valued Logic
Reasoning”, Int. Conf. Information Engg. Computer Science, 2009, pp. 1-4.
[106]. K. L. Wood, “Computations with Imprecise Parameters in Engineering Design: Background and
Theory,” ASME J. Mechanisms, Transmissions, and Automation in Design, vol. 111, no. 4, pp.
616-625, 1989.
[107]. M. Smithson, “Fuzzy Sets Analysis for Behavioral and Social Sciences”, New York Springer-
Verlag, 1987.
[108]. A. Billot, “Economic Theory of Fuzzy Equilibria: An Axiomatic Analysis”, Berlin-New York:
Springer-Verlag, 1992.
[109]. Elkan, C. ; Berenji, H.R. ; Chandrasekaran, B. ;de Silva, C.J.S. ; Attikiouzel, Y. ; Dubois,
D. ; Prade, H. ;Smets, P. ; Freksa, C. ; Garcia, O.N. ; Klir, G.J. ; Bo Yuan ;Mamdani,
E.H. ; Pelletier, F.J. ; Ruspini, E.H. ; Turksen, B. ;Vadiee, N. ; Jamshidi, M. ; Pei-Zhuang
Wang ; Sie-Keng Tan ;Shaohua Tan ; Yager, R.R. ; Zadeh, L.A. , “The Paradoxical Success of
Fuzzy Logic”, IEEE Expert, vol. 9, no. 4, pp. 3-49, 1994.
[110]. J. Kacprzyk and M. Fedrizzy, “Fuzzy Regression Analysis”, Warsaw, Poland: Omnitech Press,
1992.
[111]. N. Baldo, and M. Zorzi, “Cognitive Network Access using Fuzzy Decision Making”, IEEE Trans.
Wireless Commun, vol. 8, no. 7, pp. 3523-3535, 2009.
[112]. K. Ahmed, F. Bashir, Najum-ul-Hassan, M. Ehsan ul Haq, “Comparative study of Centralized
Cooperative Spectrum Sensing in Cognitive Radio Networks”, IEEE ICSPS, 2010, pp. 246-249.
[113]. N. Baldo, M. Zorzi, “Fuzzy Logic for Cross-layer Optimization in Cognitive Radio Networks”,
IEEE Commun. Magazine, vol. 46, no. 4, pp. 64-71, 2008.
[114]. W. Liu, T. Lv, L. Gao, W. Wang, B. Liu, “A Novel Cooperative Spectrum Sensing Scheme
Based on Fuzzy Integral Theory in Cognitive Radio Networks”, IEEE WiCom, 2009, pp. 1-4.
[115]. W. Liu, T. Lv, L. Gao, W. Wang, B. Liu, “A Novel Cooperative Spectrum Sensing Scheme
Based on Fuzzy Integral Theory in Cognitive Radio Networks”, WiCom IEEE, 2009, pp. 1-4.
[116]. H. Zhang, X. Wang, “A Fuzzy Decision Scheme for Cooperative Spectrum Sensing in Cognitive
Radio”, VTC IEEE, 2011, pp. 1-4.
Page 162
139
[117]. W. Ejaz, N. ul Hasan, S. Aslam, H. S. Kim, “Fuzzy Logic Based Spectrum Sensing for Cognitive
Radio Networks”, 2011, NGMAST IEEE, pp. 185-189.
[118]. S. R. Aryal, H. Dhungana, K. Paudyal, “Novel Approach for Interference Management in
Cognitive Radio”, 2012, AH-ICI IEEE. pp. 1-5.
[119]. W. Mustafa, J.S. Yu, E. Rakus-Andersson, A. Mohammed, W.J. Kulesza, “Fuzzy-based
Opportunistic Power Control Strategy in Cognitive Radio Networks”, 2010, ISABEL IEEE. pp. 1-
9.
[120]. H.S.T. Le and Q. Liang, “An Efficient Power Control Scheme for Cognitive Radios”, WCNC
IEEE, 2007, pp. 2559-2563.
[121]. Z. Tabakovic, S. Grgic, M. Grgic, “Fuzzy Logic Power Control in Cognitive Radio”, IWSSIP
IEEE, 2009, pp. 1-5.
[122]. A. Dey, S. Biswas, S. Panda, “A New Fuzzy Rule Based Power Management Scheme for
Spectrum Sharing in Cognitive Radio”, ICCIA IEEE, 2011, pp. 1-4.
[123]. G. Amarasuriya, M. Ardakani, and C. Tellambura, “Output-Threshold Multiple-Relay-Selection
Scheme for Cooperative Wireless Networks”, IEEE Trans. Vehicular Tech., vol. 59, no. 6, pp.
3091-3097, 2010.
[124]. I. S. Gradshteyn, I. M. Ryzhik. “Table of Integrals, Series and Products”, 5th
Ed. New
York, USA: Academic Press, 1994.
[125]. Y Zhao, R Adve T. J., Lim, “Symbol Error Rate of Selection Amplify-and-Forward Relay
Systems”, IEEE Commun. Letters, vol. 10, no. 11, pp.757-759, 2006.
[126]. S. I. Hussain, M. M. Abdallah, M. S. Alouini, M. Hasna, K. Qaraqe, “Best Relay Selection using
SNR and Interference Quotient for Underlay Cognitive Networks”, IEEE ICC, 2012, pp.4176-4180.
[127]. S. Haykin, "Cognitive Radio: Brain-Empowered Wireless communications", IEEE J. Sel. Areas
Comm., vol. 23, no. 2, pp. 201-220, 2005.
[128]. J. Mitola, “Cognitive Radio Architecture Evolution,” 2009, Proceedings of the IEEE, pp. 626–
641.
[129]. A. Ghasemi, E. S. Sousa,” Spectrum Sensing in Cognitive Radio Networks: Requirements,
Challenges and Design Trade-offs”, IEEE Comm. Magazine, vol. 46, pp. 32-39, 2008.
[130]. S. Kapoor, G. Singh, “Cooperative Spectrum Sensing: A Hybrid Model Approach”, ICDECom
IEEE, 2011, pp. 1-5.
[131]. S. Hussain, X. Fernando, “Spectrum Sensing in Cognitive Radio Networks: Up-to-date
Techniques and Future Challenges”, TIC-STH IEEE, 2009, pp. 736-741.
[132]. N. Pratas, N. R. Prasad, A. Rodrigues and R. Prasad, “Cooperative Spectrum Sensing: State of the
Art Review”, Wireless VITAE IEEE, 2011, pp.1-6.
Page 163
140
[133]. M. Subhedar and G. Birajdar, “Spectrum Sensing Techniques in Cognitive Radio Networks: A
Survey”, vol. 3, IJNGN.
[134]. A. Chandran, A. Karthik, A. Kumar, R. Naidu, S. Siva, U. S. Iyer, R. Ramanathan, “Evaluation of
Energy Detector based Spectrum Sensing for OFDM Based Cognitive Radio”, INCOCCI, IEEE,
2010, pp. 163-167.
[135]. S. M. Kay, “Fundamentals of Statistical Signal Processing: Detection Theory”, vol. 2, Prentice
Hall, Upper Saddle River, NJ, USA, 1998.
[136]. W. A. Gardner, “Exploitation of Spectral Redundancy in Cyclostationary Signals,” IEEE Signal
Processing Magazine, vol. 8, pp. 14–36,1991.
[137]. L. Gupta, R. P. Singh, “Array Signal Processing: DOA Estimation for Missing Sensors”,
ICPCES IEEE, 2010, pp. 1-4.
[138]. R. Roy, T. Kailath, “ESPRIT-Estimation of Signal Parameters via Rotational Invariance
Techniques,” IEEE Trans. Acoust., Speech, Signal Process., vol. 37, no. 7, pp. 984–995, 1989.
[139]. R. Schmidt, “Multiple Emitter Location and Signal Parameter Estimation,” IEEE Trans. Antennas
Propagation, vol. 34, no. 3, pp. 276–280, 1986.
[140]. Y. D. Huang and M. Barkat, “Near-Field Multiple Source Localization by Passive Sensor Array”.
IEEE Trans. Antennas Propag., vol 39, no. 7, pp. 968–975, 1991.
[141]. K. A. Meraim, Y. Hua and A. Belouchrani, “Second-Order Near-Field Source Localization:
Method and Performance Analysis”, Conference Record of the Thirtieth Asilomar Conference on
Signals, Systems and Computers, vol. 1, pp. 723–727, 1996.
[142]. N. Yuen and B. Friedlander, “Performance Analysis of Higher Order ESPRIT for Localization of
Near-Field Sources,” IEEE Trans. Signal Process., vol. 46, no. 3, pp. 709–719, 1998.
[143]. Q.T. Zhang and K.M. Wong, “A New Information-Theoretic Criterion for Detection of the
Number of Signals in Spatially Correlated Noise with Covariance Matrix”, ICASSP IEEE, 1992,
vol. 5, pp. 381-384.
[144]. U. Konda, Y. Cheng, T. Singh, P. D. Scott, “Source Identification of Puff-Based Dispersion
Models Using Convex Optimization”, IEEE Conf. Computing & Processing
(Hardware/Software) ;Engineered Materials, Dielectrics & Plasmas ; Nuclear Engineering, 2010,
pp. 1-6.
[145]. C. Candan and S. Koc, “Beamspace Approach for Detection of the Number of Coherent
Sources”, Radar Conference IEEE, 2012, pp. 913-918.
[146]. J. Zhang, H. S. Chung, and W. L. Lo, “Clustering-based Adaptive Crossover and Mutation
Probabilities for Genetic Algorithms," IEEE Trans. on Evol. Comput., vol. 11, no. 3, pp. 326-335,
2007.
[147]. F. Zaman, I. M. Qureshi, A. Naveed, J. A. Khan, and M. A. Z. Raja, “Amplitude and Direction
of Arrival Estimation: Comparison between Different Techniques”, Progress In Electromagnetics
Research B, vol. 39, pp. 319-335, 2012.