Design of spectrum sensing and mac in cognitive radio networks

average received SNR dB of ED andABD for 8PSK modulated primary signals over AWGN channelsand Rayleigh fading channels in the high SNR regime.. average received SNR dB of ED andABD for 8

DESIGN OF SPECTRUM SENSING AND MAC IN

COGNITIVE RADIO NETWORKS

ZHENG SHOUKANG

NATIONAL UNIVERSITY OF SINGAPORE

2012

COGNITIVE RADIO NETWORKS

ZHENG SHOUKANG(M Eng., National University of Singapore)

A THESIS SUBMITTEDFOR THE DEGREE OF DOCTOR OF PHILOSOPHY

DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING

NATIONAL UNIVERSITY OF SINGAPORE

2012

To my family.

my co-supervisor, for their enlightened supervision, advice, time and tremendousefforts spent in teaching me, discussing with me, refining all my writings andmost of all, their overall genuine concern for me.

I want to thank the Department of Electrical and Computer Engineering,National University of Singapore I also thank the Institute for Infocomm Re-search for providing me with all the necessary computing and communicationsfacilities, and a congenial environment to carry out my research

I must also thank my family for their constant love, care, concern and supportduring my embarkment and completion of my pursuit of a doctoral degree Lastbut not least, I would like to express my heartfelt gratitude to my colleaguesand friends at the Institute for Infocomm Research for providing hearty help andhappy hours together In particular, thanks to my buddies, Dr David Wong and

Dr Zeng Yonghong, who have helped to proofread the thesis for me

1.1 Background 1

1.2 Motivation 3

1.3 Contributions of Thesis 6

1.3.1 Bayesian Detector for MPSK Modulated Signals 6

1.3.2 Cross-layered Design of Spectrum Sensing and MAC Protocol 8 1.3.3 MAC Protocol Design for Cooperative Spectrum Sensing 8 1.4 Thesis Organization 9

Chapter 2 Literature Review: Spectrum Sensing and Cognitive Radio MAC 11 2.1 Spectrum Sensing 11

2.1.1 Binary Hypothesis Testing 13

2.1.2 Detection Performance and Threshold 15

2.1.3 LRT-based Detection Sensing 16

2.1.4 Energy Detection Based Sensing 17

Contents iv

2.1.5 Cyclostationary Detection Based Sensing 18

2.1.6 Matched-filter Based Sensing 20

2.1.7 Covariance-based Sensing 21

2.1.8 Wavelet-based Sensing 25

2.1.9 Cooperative Sensing 27

2.2 Cognitive Radio MAC 28

2.2.1 MAC for Centralized CRNs 28

2.2.2 MAC for Ad Hoc CRNs 30

2.2.3 Joint Design of Spectrum Sensing and MAC 38

Chapter 3 Bayesian Spectrum Sensing for BPSK Modulated Pri-mary Signals 39 3.1 Introduction 39

3.2 System Model and Optimal Detector Structure 40

3.3 Suboptimal Detector Structure 43

3.3.1 Approximation in the Low SNR Regime 44

3.3.2 Approximation in the High SNR Regime 45

3.4 Detection and False Alarm Probabilities 45

3.4.1 Detection Probability 46

3.4.2 False Alarm Probability 48

3.4.3 Analysis of ABD1 in the low SNR Regime 48

3.4.4 Analysis of ABD2 in the Low SNR Regime 51

3.4.5 Analysis in the High SNR Regime 52

3.5 Detection Threshold and Number of Samples 53

3.6 Simulation and Numerical Results 55

3.6.1 Low SNR Regime 56

3.6.2 High SNR Regime 61

3.6.3 Impact of Incorrect Prior 64

3.7 Summary 66

Chapter 4 Bayesian Detector for MPSK Modulated Primary Sig-nals 67 4.1 System Model and Optimal Detector Structure 67

4.2 Suboptimal Detector Structure 71

4.2.1 Approximation in the Low SNR Regime 71

4.2.2 Approximation in the High SNR Regime 74

4.3 Detection Threshold and Number of Samples 75

4.4 Simulation and Numerical Results 76

4.4.1 AWGN Channels 77

4.4.2 Rayleigh Fading Channels 81

4.4.3 AWGN Channels versus Rayleigh Fading Channels 90

4.4.4 Performance Comparison of ABDs for B/Q/8PSK Signals 97 4.5 Summary 102

Chapter 5 Bayesian Detector for Unknown Order MPSK Modu-lated Primary Signals 106 5.1 System Model and Optimal Detector Structure 106

5.2 Suboptimal Detector Structure 108

5.2.1 Low SNR Regime 108

5.2.2 High SNR Regime 109

5.3 Simulation and Numerical Results 109

5.3.1 AWGN Channels 110

5.3.2 Rayleigh Fading Channels 115

5.3.3 AWGN Channels versus Rayleigh Fading Channels 122

5.3.4 Performance Comparison of ABDs for Unknown and Known Order PSK Signals 128

5.4 Summary 133

Chapter 6 Joint Design of Spectrum Sensing and MAC Protocol for Opportunistic Spectrum Access 136 6.1 Introduction 136

6.2 System Model 138

6.2.1 Spectrum Sensing 139

6.2.2 MAC Random Access 140

6.2.3 Constrained Optimization 143

6.3 Cross-layered and Layered Design Approaches 144

6.4 Numerical Results 150

6.4.1 Simulation Model 150

Contents vi

6.4.2 Mean Number of Backlogged SUs 150

6.4.3 PU Idle Time 154

6.4.4 Interference Constraint 157

6.5 Summary 158

Chapter 7 Design of MAC Protocol for Cooperative Spectrum Sensing in Ad Hoc Cognitive Radio Networks 161 7.1 Introduction 161

7.2 Random Access for Cooperative Sensing 163

7.2.1 System Model 163

7.2.2 Cooperative Spectrum Sensing 165

7.2.3 Sequential Detection 167

7.2.4 Random Access in Control Channel 167

7.3 Upper Bound for Overall Throughput 171

7.3.1 Average Service Time for Sensing Decision 171

7.3.2 Saturation Problem 176

7.3.3 Upper Bound for Overall Throughput 176

7.3.4 Cooperative Sensing-Throughput Tradeoff 177

7.4 Sequential Detection with Prioritized Reporting 178

7.4.1 OR-rule Decision Fusion 180

7.4.2 AND-rule Decision Fusion 180

7.4.3 MAJORITY-rule Decision Fusion 181

7.5 Numerical Results 182

7.5.1 Impact of Frame Size 184

7.5.2 Impact of SNR 187

7.5.3 Impact of Sequential Detection 187

7.5.4 Comparison among Different Decision Rules 189

7.6 Summary 190

Appendix A Computation of µY (k) and σ2

Y (k) in (3.50) and (3.52) 215 Appendix B Computation of µY (k) and σ2

Y (k) in (4.17) and (4.18) 219

B.1 Computation of µY (k) and σ2

Y (k) under H0 219B.2 Computation of µY (k) and σ2

Y (k) under H1 222

D.1 Proposition 230D.2 Proof of Theorem 3 231D.3 Proof of Theorem 4 231

Summary

Recently, in order to improve spectrum usage more efficiently, many researchershave been actively exploring a few important issues in cognitive radio networks, ofwhich one is spectrum sensing technique on how to detect the primary signals andanother is how to make spectrum access among the contention by secondary usersafter identifying the spectrum opportunity The protection to the primary usersmotivates the research in spectrum sensing and cognitive radio MAC (MediumAccess Control) protocol design to provide efficient manner of detecting the pri-mary signals over the channel so as to determine whether the frequency band isfree or not, and sharing the available spectrum among the secondary users

An optimal Bayesian detector, based on the prior information on the highprobability that primary user is idle in cognitive radio networks, is proposed forspectrum sensing, assuming that primary signals are digitally PSK (Phase ShiftKeying) modulated but the sequence of bit transmission is not known to thesecondary users The proposed scheme considers not only BPSK (Binary PSK)modulated primary signals but also MPSK (M-ary PSK) modulated primary sig-nals, over both AWGN (Additive White Gaussian Noise) channels and fadingchannels The structure of the optimal Bayesian detector can approximately bereduced to that of an energy detector in the lower SNR (Signal-to-Noise Ratio)

regime, and can be approximated to that of a detector employing the sum of thereceived signal magnitudes in the high SNR regime, to detect BPSK modulatedprimary signals The energy detector structure is applicable to MPSK signals

in the low SNR regime as well The analyses for the optimal Bayesian detectorand its corresponding suboptimal detector structure in both low and high SNRregimes are given for the case of BPSK modulated primary signals, and the anal-ysis for a suboptimal detector structure in the low SNR regime is also conductedfor the case of MPSK modulated primary signals The detection performance

of the optimal/suboptimal detector is compared with those of energy detectorand Neyman-Pearson detector (the optimal detector given by Neyman-Pearsontheorem that maximizes the detection probability for a given false alarm prob-ability) with its analysis being verified with the simulation results A detectorwithout knowing the exact order of MPSK modulated primary signals is also pro-posed and studied as a further generalization Compared with energy detectorand Neyman-Pearson detector, the newly proposed detector can achieve higheroverall spectrum utilization and secondary users’ throughput and at the sametime the primary user is well-protected from the secondary users’ interference

In a distributed OSA (Opportunistic Spectrum Access) network where thesecondary users sense the channel independently and contend for channel ac-cess on a frame-by-frame basis, two design approaches are studied Differentfrom the layered design approach, the cross-layered design considers the randommedium access control protocol in conjunction with the spectrum sensing proto-col design In particular, physical layer parameters (frame duration and sensingtime) and random access probability in the MAC layer are considered jointly tomaximize the secondary network throughput performance while protecting theprimary users from the interference caused by secondary users’ operations Fromthe described system model, nonlinear constrained optimization problems are for-

Summary x

mulated for both the cross-layered and layered approaches Through numericalresults, the cross-layered approach is shown to perform much better than thelayered approach

To mitigate the degradation of the channels between active primary users andsecondary users in the wireless ad hoc cognitive radio networks, secondary usersare required to employ cooperative spectrum sensing A MAC protocol design,based on random access MAC protocols of IEEE 802.11 DCF (Distributed Co-ordination Function) and IEEE 802.11e EDCA (Enhanced Distributed ChannelAccess), is proposed to support cooperative sensing, allowing multiple transmis-sions of sensing reports and fused decisions over the control channel The tradeoffbetween cooperative sensing gain and channel reuse efficiency is exploited to im-prove the overall achievable throughput for all the channels The sequential de-tection approach has been proposed to reduce the average service time for sensingdecisions on top of the random access MAC scheme and numerical results haveshown the advantage of this enhancement

List of Tables

2.1 MAC Protocols for Ad Hoc Cognitive Radio Networks 31

7.1 Parameters for Numerical Results of MAC Protocols with erative Sensing 183

List of Figures

3.1 Receiver structure for a secondary user 423.2 Detection probabilities of ED, NPD, BD and ABD vs Es/N0 (dB)for BPSK modulated primary signals over AWGN channels in thelow ES/N0 regime 573.3 False alarm probabilities of ED, NPD, BD and ABD vs Es/N0

(dB) for BPSK modulated primary signals over AWGN channels

in the low ES/N0 regime 573.4 Spectrum Utilization of ED, NPD, BD and ABD vs Es/N0 (dB)for BPSK modulated primary signals over AWGN channels in thelow ES/N0 regime 583.5 Normalized SU throughput of ED, NPD, BD and ABD vs Es/N0

(dB) for BPSK modulated primary signals over AWGN channels

in the low ES/N0 regime 583.6 Simulation and numerical results of detection probability of ABD

vs Es/N0 (dB) for BPSK modulated primary signals over AWGNchannels in the low ES/N0 regime 593.7 Simulation and numerical results of false alarm probability of ABD

vs Es/N0 (dB) for BPSK modulated primary signals over AWGNchannels in the low ES/N0 regime 593.8 Numerical results of detection probability of ABD1/ABD2 based

on (3.16) and (3.20) vs Es/N0 (dB) for BPSK modulated primarysignals over AWGN channels in the low ES/N0 regime 60

3.9 Numerical results of false alarm probability of ABD1/ABD2 based

on (3.16) and (3.20) vs Es/N0 (dB) for BPSK modulated primarysignals over AWGN channels in the low ES/N0 regime 613.10 Simulation and numerical results of detection probability of ED,NPD, BD and ABD vs Es/N0 (dB) for BPSK modulated primarysignals over AWGN channels in the high ES/N0 regime 623.11 Simulation and numerical results of false alarm probability of ED,NPD, BD and ABD vs Es/N0 (dB) for BPSK modulated primarysignals over AWGN channels in the high ES/N0 regime 623.12 Simulation and numerical results of spectrum utilization of ED,NPD, BD and ABD vs Es/N0 (dB) for BPSK modulated primarysignals over AWGN channels in the high ES/N0 regime 633.13 Simulation and numerical results of normalized SU throughput of

ED, NPD, BD and ABD vs Es/N0 (dB) for BPSK modulatedprimary signals over AWGN channels in the high ES/N0 regime 633.14 Numerical results of normalized SU throughput vs Es/N0 (dB)for BPSK modulated primary signals over AWGN channels in thelow ES/N0 regime with incorrect prior 643.15 Numerical results of spectrum utilization vs Es/N0(dB) for BPSKmodulated primary signals over AWGN channels in the low ES/N0

regime with incorrect prior 653.16 Numerical results of normalized SU throughput vs Es/N0 (dB)for BPSK modulated primary signals over AWGN channels in thehigh ES/N0 regime with incorrect prior 653.17 Numerical results of spectrum utilization vs Es/N0(dB) for BPSKmodulated primary signals over AWGN channels in the high ES/N0regime with incorrect prior 66

List of Figures xiv

4.1 Detection probabilities of ED, NPD, BD and ABD vs Es/N0 (dB)for 8PSK modulated primary signals over AWGN channels in thelow ES/N0 regime 784.2 False alarm probabilities of ED, NPD, BD and ABD vs Es/N0

(dB) for 8PSK modulated primary signals over AWGN channels

in the low ES/N0 regime 784.3 Normalized SU throughput of ED, NPD, BD and ABD vs Es/N0

(dB) for 8PSK modulated primary signals over AWGN channels

in the low ES/N0 regime 794.4 Spectrum utilization of ED, NPD, BD and ABD vs Es/N0 (dB)for 8PSK modulated primary signals over AWGN channels in thelow ES/N0 regime 794.5 Simulation and numerical results of detection probability of ABD

vs Es/N0 (dB) for 8PSK modulated primary signals over AWGNchannels in the low ES/N0 regime 804.6 Simulation and numerical results of false alarm probability of ABD

vs Es/N0 (dB) for 8PSK modulated primary signals over AWGNchannels in the low ES/N0 regime 804.7 Detection probabilities of ED, NPD, BD and ABD vs Es/N0 (dB)for 8PSK modulated primary signals over AWGN channels in thehigh ES/N0 regime 814.8 False alarm probabilities of ED, NPD, BD and ABD vs Es/N0

(dB) for 8PSK modulated primary signals over AWGN channels

in the high ES/N0 regime 824.9 Detection probabilities of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the low SNR regime with unknown average re-ceived SNR 83

4.10 False alarm probabilities of ED, NPD, BD and ABD vs age received SNR (dB) for 8PSK modulated primary signals overRayleigh fading channels in the low SNR regime with unknownaverage received SNR 844.11 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the low SNR regime with unknown average re-ceived SNR 844.12 Detection probabilities of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the low SNR regime with known average re-ceived SNR 854.13 False alarm probabilities of ED, NPD, BD and ABD vs aver-age received SNR (dB) for 8PSK modulated primary signals overRayleigh fading channels in the low SNR regime with known aver-age received SNR 864.14 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the low SNR regime with known average re-ceived SNR 864.15 Detection probabilities of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the high SNR regime with unknown average re-ceived SNR 884.16 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the high SNR regime with unknown average re-ceived SNR 88

aver-List of Figures xvi

4.17 Detection probabilities of ED, NPD, BD and ABD vs average ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the high SNR regime with known average re-ceived SNR 894.18 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for 8PSK modulated primary signals over Rayleighfading channels in the high SNR regime with known average re-ceived SNR 904.19 Simulation and numerical results of detection probability of ABD

re-vs average received SNR (dB) for 8PSK modulated primary nals over Rayleigh fading channels in the low SNR regime withunknown average received SNR 914.20 Simulation and numerical results of false alarm probability of ABD

vs average received SNR (dB) for 8PSK modulated primary nals over Rayleigh fading channels in the low SNR regime withunknown average received SNR 914.21 Simulation and numerical results of detection probability of ABD

vs average received SNR (dB) for 8PSK modulated primary nals over Rayleigh fading channels in the low SNR regime withknown average received SNR 924.22 Simulation and numerical results of false alarm probability of ABD

vs average received SNR (dB) for 8PSK modulated primary nals over Rayleigh fading channels in the low SNR regime withknown average received SNR 924.23 Detection probabilities vs average received SNR (dB) of ED andABD for 8PSK modulated primary signals over AWGN channelsand Rayleigh fading channels in the low SNR regime 94

sig-4.24 False alarm probabilities vs average received SNR (dB) of ED andABD for 8PSK modulated primary signals over AWGN channelsand Rayleigh fading channels in the low SNR regime 954.25 Spectrum utilization vs average received SNR (dB) of ED andABD for 8PSK modulated primary signals over AWGN channelsand Rayleigh fading channels in the low SNR regime 954.26 Normalized SU throughput vs average received SNR (dB) of EDand ABD for 8PSK modulated primary signals over AWGN chan-nels and Rayleigh fading channels in the low SNR regime 964.27 Detection probabilities vs average received SNR (dB) of ED andABD for 8PSK modulated primary signals over AWGN channelsand Rayleigh fading channels in the high SNR regime 984.28 False alarm probabilities vs average received SNR (dB) of ED andABD for 8PSK modulated primary signals over AWGN channelsand Rayleigh fading channels in the high SNR regime 984.29 Normalized SU throughput vs average received SNR (dB) of EDand ABD for 8PSK modulated primary signals over AWGN chan-nels and Rayleigh fading channels in the high SNR regime 994.30 Spectrum utilization vs average received SNR (dB) of ED andABD for 8PSK modulated primary signals over AWGN channelsand Rayleigh fading channels in the high SNR regime 994.31 Detection probabilities vs Es/N0 (dB) of ABD for B/Q/8PSKmodulated primary signals over AWGN channels in the low SNRregime 1004.32 False alarm probabilities vs Es/N0 (dB) of ABD for B/Q/8PSKmodulated primary signals over AWGN channels in the low SNRregime 101

List of Figures xviii

4.33 Spectrum utilization vs Es/N0(dB) of ABD for B/Q/8PSK ulated primary signals over AWGN channels in the low SNR regime.101

mod-4.34 Normalized SU throughput vs Es/N0 (dB) of ABD for B/Q/8PSKmodulated primary signals over AWGN channels in the low SNRregime 1024.35 Detection probabilities vs Es/N0 (dB) of ABD for B/Q/8PSKmodulated primary signals over AWGN channels in the high SNRregime 1034.36 False alarm probabilities vs Es/N0 (dB) of ABD for B/Q/8PSKmodulated primary signals over AWGN channels in the high SNRregime 1034.37 Spectrum utilization vs Es/N0(dB) of ABD for B/Q/8PSK mod-ulated primary signals over AWGN channels in the high SNRregime 1044.38 Normalized SU throughput vs Es/N0 (dB) of ABD for B/Q/8PSKmodulated primary signals over AWGN channels in the high SNRregime 104

5.1 Detection probabilities of ED, NPD, BD and ABD vs Es/N0 (dB)for unknown order MPSK modulated primary signals over AWGNchannels in the low Es/N0 regime 1115.2 False alarm probabilities of ED, NPD, BD and ABD vs Es/N0

(dB) for unknown order MPSK modulated primary signals overAWGN channels in the low Es/N0 regime 1115.3 Spectrum utilization of ED, NPD, BD and ABD vs Es/N0 (dB)for unknown order MPSK modulated primary signals over AWGNchannels in the low Es/N0 regime 112

5.4 Normalized SU throughput of ED, NPD, BD and ABD vs Es/N0

(dB) for unknown order MPSK modulated primary signals overAWGN channels in the low Es/N0 regime 1125.5 Detection probabilities of ED, NPD, BD and ABD vs Es/N0 (dB)for unknown order MPSK modulated primary signals over AWGNchannels in the high Es/N0 regime 1135.6 False alarm probabilities of ED, NPD, BD and ABD vs Es/N0

(dB) for unknown order MPSK modulated primary signals overAWGN channels in the high Es/N0 regime 1145.7 Spectrum utilization of ED, NPD, BD and ABD vs Es/N0 (dB)for unknown order MPSK modulated primary signals over AWGNchannels in the high Es/N0 regime 1145.8 Normalized SU throughput of ED, NPD, BD and ABD vs Es/N0(dB) for unknown order MPSK modulated primary signals overAWGN channels in the high Es/N0 regime 1155.9 Detection probabilities of ED, NPD, BD and ABD vs averagereceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the low SNR regime withunknown average received SNR 1165.10 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the low SNR regime withunknown average received SNR 1165.11 Detection probabilities of ED, NPD, BD and ABD vs averagereceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the low SNR regime withknown average received SNR 118

List of Figures xx

5.12 False alarm probabilities of ED, NPD, BD and ABD vs averagereceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the low SNR regime withknown average received SNR 1195.13 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the low SNR regime withknown average received SNR 1195.14 Detection probabilities of ED, NPD, BD and ABD vs averagereceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the high SNR regime withunknown average received SNR 1205.15 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the high SNR regime withunknown average received SNR 1205.16 Detection probabilities of ED, NPD, BD and ABD vs averagereceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the high SNR regime withknown average received SNR 1225.17 Spectrum utilization of ED, NPD, BD and ABD vs average re-ceived SNR (dB) for unknown order MPSK modulated primarysignals over Rayleigh fading channels in the high SNR regime withknown average received SNR 1235.18 Detection probabilities vs average received SNR (dB) of ED andABD for unknown order MPSK modulated primary signals overAWGN channels and Rayleigh fading channels in the low SNRregime 125

5.19 False alarm probabilities vs average received SNR (dB) of EDand ABD for unknown order MPSK modulated primary signalsover AWGN channels and Rayleigh fading channels in the low SNRregime 1255.20 Spectrum utilization vs average received SNR (dB) of ED andABD for unknown order MPSK modulated primary signals overAWGN channels and Rayleigh fading channels in the low SNRregime 1265.21 Normalized SU throughput vs average received SNR (dB) of EDand ABD for unknown order MPSK modulated primary signalsover AWGN channels and Rayleigh fading channels in the low SNRregime 1265.22 Detection probabilities vs average received SNR (dB) of ED andABD for unknown order MPSK modulated primary signals overAWGN channels and Rayleigh fading channels in the high SNRregime 1285.23 False alarm probabilities vs average received SNR (dB) of EDand ABD for unknown order MPSK modulated primary signalsover AWGN channels and Rayleigh fading channels in the highSNR regime 1295.24 Normalized SU throughput vs average received SNR (dB) of EDand ABD for unknown order MPSK modulated primary signalsover AWGN channels and Rayleigh fading channels in the highSNR regime 1295.25 Spectrum utilization vs average received SNR (dB) of ED andABD for unknown order MPSK modulated primary signals overAWGN channels and Rayleigh fading channels in the high SNRregime 130

List of Figures xxii

5.26 Detection probabilities vs Es/N0 (dB) of ABD for unknown andknown order MPSK modulated primary signals over AWGN chan-nels in the low SNR regime 1315.27 False alarm probabilities vs Es/N0 (dB) of ABD for unknownand known order MPSK modulated primary signals over AWGNchannels in the low SNR regime 1315.28 Spectrum utilization vs Es/N0 (dB) of ABD for unknown andknown order MPSK modulated primary signals over AWGN chan-nels in the low SNR regime 1325.29 Normalized SU throughput vs Es/N0 (dB) of ABD for unknownand known order MPSK modulated primary signals over AWGNchannels in the low SNR regime 1325.30 Detection probabilities vs Es/N0 (dB) of ABD for unknown andknown order MPSK modulated primary signals over AWGN chan-nels in the high SNR regime 1335.31 False alarm probabilities vs Es/N0 (dB) of ABD for unknownand known order MPSK modulated primary signals over AWGNchannels in the high SNR regime 1345.32 Spectrum utilization vs Es/N0 (dB) of ABD for unknown andknown order MPSK modulated primary signals over AWGN chan-nels in the high SNR regime 1345.33 Normalized SU throughput vs Es/N0 (dB) of ABD for unknownand known order MPSK modulated primary signals over AWGNchannels in the high SNR regime 135

6.1 Markovian model of PU’s idle and active states 1396.2 SU frame structure with PU’s idle and active periods 1406.3 Maximum average achievable throughput vs mean number of SUsfor layered and cross-layered approaches SNR = -18 dB 151

6.4 False alarm probability vs mean number of SUs for layered andcross-layered approaches SNR = -18 dB 1526.5 Sensing time vs mean number of SUs for layered and cross-layeredapproaches SNR = -18 dB 1526.6 p-persistent probability vs mean number of SUs for layered andcross-layered approaches SNR = -18 dB 1536.7 Ratio of effective transmission time to PU’s mean idle time vs.mean number of SUs for layered and cross-layered approaches.SNR = -18 dB 1536.8 Frame duration vs mean number of SUs for layered and cross-layered approaches SNR = -18 dB 1546.9 Maximum average achievable throughput vs PU’s mean idle timefor layered and cross-layered approaches m = 0.5 and 3 1556.10 Ratio of effective transmission time to frame duration vs PU’smean idle time for layered and cross-layered approaches m = 0.5and 3 1566.11 Frame duration vs PU’s mean idle time for layered and cross-layered approaches m = 0.5 and 3 1566.12 Maximum average achievable throughput vs interference con-straint for layered and cross-layered approaches m = 0.5 and

3 1576.13 False alarm probability vs interference constraint for layered andcross-layered approaches m = 0.5 and 3 1586.14 p-persistent probability vs interference constraint for layered andcross-layered approaches m = 0.5 and 3 1596.15 Ratio of sensing time to frame duration vs interference constraintfor layered and cross-layered approaches m = 0.5 and 3 159

7.1 Frame structure of data channels with one control channel 164

List of Figures xxiv

7.2 Overall throughput gain vs number of data channels for OR-rule.SNR = -20 dB and tb = 5 ms 1857.3 Number of cooperative users vs number of data channels for OR-rule SNR = -20 dB and tb = 5 ms 1857.4 Sensing time vs number of data channels for OR-rule SNR = -20

dB and tb = 5 ms 1867.5 Average service time of sensing decision vs number of data chan-nels for OR-rule SNR = -20 dB and tb = 5 ms 1867.6 Number of cooperative users vs number of data channels forMAJORITY-rule T = 40 ms and tb = 5 ms 1877.7 Sensing time vs number of data channels for MAJORITY-rule T

= 40 ms and tb = 5 ms 1887.8 Average service time of sensing decision vs number of data chan-nels for sequential detection T = 30 ms and tb = 5 ms 1897.9 Overall throughput gain vs number of data channels for sequentialdetection T = 30 ms and tb = 5 ms 1907.10 Overall throughput gain vs number of data channels for 3 decisionrules SNR = -20 dB and tb = 5 ms 1917.11 Number of cooperative users vs number of data channels for 3decision rules SNR = -20 dB and tb = 5 ms 1917.12 Sensing time vs number of data channels for 3 decision rules SNR

= -20 dB and tb = 5 ms 1927.13 Average service time of sensing decision vs number of data chan-nels for 3 decision rules SNR = -20 dB and tb = 5 ms 192

ABD Approximate Bayesian Detector

AHCRN Ad Hoc Cognitive Radio Network

AIFS Different Arbitration Interframe Space

AWGN Additive White Gaussian Noise

BD Bayesian Detector

BP Beacon Period

BPSK Binary Phase Shift Keying

CAF Cyclic Autocorrelation Function

CCC Common Control Channel

CLT Central Limit Theorem

CTMC Continuous-time Markov Chain

CR Cognitive Radio

CRN Cognitive Radio Network

CSD Cyclic Spectrum Density

CSMA/CA Carrier Sense Multiple Access with Collision Avoidance

CTS Clear-To-Send

CWT Continuous Wavelet Transform

DCF Distributed Coordination Function

DIFS DCF Interframe Space

LRT Likelihood Ratio Test

MAC Medium Access Control

MPSK M-ary Phase Shift Keying

NPD Neyman-Pearson Detector

OFDM Orthogonal Frequency Division Multiplexing

OSA Opportunistic Spectrum Access

PDF Probability Density Function

POMDP Partially Observable Markov Decision Process

PU Primary User

QP Quiet Period

RP Report Period

RTS Request-To-Send

SCF Spectral Correlation Function

SIFS Short Interframe Space

UCA Uniform Cost Assignment

WSS Wide Sense Stationary

∗ Convolution operation or complex conjugation operation

F[·] Fourier transform function

x Second order derivative of function f with respect to x

[·]T Matrix transpose operation

P (·) Probability of a random variable

pX(·) Probability density function of random variable X

E[·] Expectation of Random variable X

µX Mean of random variable X

σ2

X Variance of random variable X

ℜ[·] Operation of taking real part

N Number of samples for spectrum sensing

Q(·) Complementary distribution function of a standard Gaussian variable

Q−1(·) Inverse function of Q(·)

PF False alarm probability

TLRT LRT based detection value

τ Spectrum sensing time

as a result of technical evolutions For example, the switchover to digital vision evacuates the bands at about 50 MHz Therefore, a large portion of theassigned spectrum is under-utilized as shown in [1, 5, 7–10], which illustrate thatthe fixed spectrum allocation rules result in inefficient spectrum usage.

tele-Cognitive radio (CR), as first proposed in [3], is a promising key technologythat enables dynamic spectrum access (DSA) networks (next generation commu-nication networks [7]), to utilize the spectrum more efficiently Recent researchefforts on cognitive radio have opened the door to more efficient spectrum utiliza-

tion [3, 11, 12, 17] It is a paradigm for wireless communication in which the radiocan change its transmission or reception parameters to communicate efficientlyavoiding interference with licensed or unlicensed users, based on the active moni-toring of the information in the external and internal radio environment, such asradio transmission frequency, bandwidth, power, modulation, user behavior andnetwork state.

There are also a number of industrial applications on cognitive radio Newworldwide standards to coordinate the applications of CR, like IEEE 802.22 [4]and IEEE 802.11af, as well as those from the White Spaces Coalition [13], have ad-vocated using white spaces to provide wireless broadband Internet access, thoughthese efforts could affect wireless microphones, medical telemetry, and other tech-nologies that operate on these open frequencies European Conference of Postaland Telecommunications Administrations (CEPT), UK Office of Communications(OFCOM) and other countries have ridden on similar trends and made regula-tion rules for the secondary users (SUs) in the TV band white spaces for theirrespective domains [9]

Worldwide trials on cognitive radio are also carried out in addition to latory standardization activities For instance, Spectrum Bridge, in partnershipwith Google and another company, has launched the first Smart Grid wirelessnetwork trial utilizing TV white spaces spectrum in Plumas-Sierra County, CA,USA [14] The recent experiment in UK allows trials of a new breed of super WiFithat uses the white space between TV channels are set to begin in Cambridge [15].The involving companies will investigate on how the gaps in frequencies between

regu-TV broadcasts can be used for broadband transmission This technology is pected to create super WiFi networks that can support bandwidth-hungry mobileInternet devices such as smartphones and tablet computers

ex-With the requirement on more flexible use of the spectrum resources, the

1.2 Motivation 3

interest in developing efficient spectrum allocation schemes and spectrum accessprotocols to allow the SUs for exploring the spectrum opportunities in space, timeand frequency domains is increasing

In CR networks, the secondary users are allowed to use the spectrum nally allocated to primary users as long as the PUs are not using it temporarily.This operation is called opportunistic spectrum access (OSA) To avoid interfer-ence to the PUs, the SUs have to perform spectrum sensing before their attempts

origi-to transmit over the spectrum Upon detecting that the PU is idle, the SUs canmake use of the spectrum for transmission so that the overall utilization efficiency

of the spectrum is enhanced The protection to the PUs motivates the research

in spectrum sensing and cognitive radio medium access control (MAC) design

to provide efficient manner of detecting the primary signals over the channel todetermine whether the frequency band is free, and sharing the available spectrumamong the SUs

It is generally understood that certain kinds of spectrum users have icant variability in their spectrum use and much of their allocated spectrum isunder-utilized during non-peak periods [1] In [2], it reports the temporal and ge-ographical variations in the utilization of the assigned spectrum range from 15%

signif-to 85% The measurement results in Singapore for the frequency bands rangingfrom 80 MHz to 5850 MHz suggest that except for the frequency bands allocatedfor broadcasting and cell phones, most of the allocated frequencies are heavilyunderutilized [5] The similar observation in [6] also shows that there is a highprobability that the primary users are likely idle for most of the time This re-sulting global spectrum usage in the frequency bands allows the secondary users

to make use of the spectrum holes.

One of the important techniques in CR is spectrum sensing that determinesthe signal presence or absence of a primary user (PU) at the receiver of a sec-ondary user (SU) It is required that the SUs should frequently sense the spec-trum before they gain access to the free channel, to make sure they can reuse thechannel with a high probability, at the same time without causing severe inter-ference to the PUs For example, in order to utilize available free TV channels,90% probability of detection and 10% probability of false alarm is required atsignal-to-noise ratio (SNR) level as low as -20 dB [4]

Many detection methods have been proposed and studied [62, 75], for ample, energy detector [18–20], covariance based detection algorithm [21, 22],cyclostationarity based detection algorithms [23–26], matched filter based detec-tion [27, 28] and wavelet-based sensing [51, 112, 113] However, we find that thedetection methods such as energy detector and covariance based detection as-sume only random signals for primary signals Although cyclostationarity baseddetection exploits the feature of primary signals, it does not make full use of thecharacteristics of modulated signals Matched-filter based detection, however,requires the complete knowledge of primary signals The study has impressed usthat there is little work done for spectrum sensing of digitally modulated primarysignals

ex-Furthermore, current design methods for spectrum sensing try to ensure agood detection performance with a high probability of detection and a low proba-bility of false alarm Although it is advisable to optimize the spectrum efficiencythrough the secondary users making use of spectrum holes while at the same timeprotecting the primary user from unfavorable interference by secondary transmis-sions, it is worthwhile to study on the design principle that can maximize theoverall spectrum utilization, by considering the prior information that a primary

When the secondary users contend for the primary channels, the performance

of secondary network is dependent on both its detection performance and thespectrum sharing efficiency To achieve a better detection performance, a longspectrum sensing duration is generally required to generate a sufficiently largenumber of samples to reduce the probabilities of false alarm and misdetection.However, the sensing time consumes partly the transmission time, which degradesthe spectrum reuse efficiency This design issue is a fundamental tradeoff problem[17] Therefore, how to design a scheme for medium access control among thecontending secondary users to achieve high performance, at the same time with

a good protection to primary users, is challenging The exemplary research work[17, 81–85] considered the above difficulties

As the design of cognitive radio MAC is no longer an independent task forMAC layer, the researchers have been learning and investigating the approach onjoint design of spectrum sensing and MAC protocol [84, 93, 96] The joint designapproach does have the advantage over the layered design approach in the context

of cognitive radio networks as shown in the literature If a random access protocol

is applied in the secondary network, its MAC design should not only study thecontention among the competing SUs, but also consider imperfect sensing effect,especially the case when the distributed spectrum sensing is applied This willprevent a SU from accessing a primary channel if false alarm occurs, or interferewith the primary network if misdetection happens

When detection performance is severely compromised for the sake of theSUs experiencing fading effects and shadowing on the channels, and/or the re-

quirement on the sensitivity is stringent in the lower SNR regime to achieve thetarget of detection performance, it could be improved by cooperative sensing.The increasing number of cooperative sensing reports alleviates the degradingperformance caused by fading, but it also intensifies the contention on the re-porting channel and prolongs the reporting duration, which directly reduces thetransmission time or spectrum efficiency Such tradeoff design problem requires

an efficient MAC design, especially in the control channel assuming a number ofdata channels Sequential detection has been proposed and studied to achievethe cooperative sensing gain more efficiently [56, 57, 59, 99]

The thesis has a number of distinctive contributions

1.3.1 Bayesian Detector for MPSK Modulated SignalsFirstly, we propose an optimal Bayesian detector for spectrum sensing incognitive radio networks, assuming that the prior information of a primary chan-nel with high idle probability is known and the primary signals are digitally PSKmodulated Compared to energy detector and the optimal detector by Neyman-Pearson theorem that maximizes the detection probability for a given false alarmprobability, the proposed detector achieves higher overall spectrum utilizationand SU throughput and at the same time the primary user is well-protected fromthe secondary users’ interference

The sequence of the primary signals are not known to the SUs Although weinitially consider a coherent detector for BPSK signals, the proposed detector isextended to a non-coherent detector for MPSK signals other than BPSK signalsand can be also extended to a coherent or non-coherent detector with channel

1.3 Contributions of Thesis 7

estimation based on the pilots of the primary signals After approximation, thedetector is identical in structure to Neyman-Pearson detector, which is a likeli-hood ratio test (LRT) detector However, it is convenient for a Bayesian detector

to determine the detection threshold, which is dependent on the ratio of theprobabilities of two hypotheses By contrast, the detection threshold of energydetector or Neyman-Pearson detector hinges on the noise variance for a givenprobability of false alarm Similar to energy detector, the proposed Bayesiandetector is also vulnerable to noise uncertainty

Our proposed scheme considers not only BPSK modulated primary nals but also MPSK modulated primary signals, over both AWGN channels andRayleigh fading channels We find that the optimal Bayesian detector can ap-proximately be reduced to an energy detector in the lower SNR regime, and itcan be approximated to a detector, employing the sum of received signal magni-tudes in the high SNR regime, to detect BPSK modulated primary signals Wealso give the analyses for an optimal Bayesian detector and its correspondingsuboptimal detector structure in both low and high SNR regimes for the case ofBPSK modulated primary signals, and also give the analysis for the suboptimaldetector structure in the low SNR regime for the case of MPSK modulated pri-mary signals The analysis is verified to match well with the simulation results.The detection performance of the optimal/suboptimal detector is compared withthose of energy detector and Neyman-Pearson detector via the simulation results

sig-We further propose a detector when there is no prior knowledge on theexact order of MPSK modulated primary signals The scheme assumes that theprimary signals are equally likely modulated with different order MPSK Similarly

we derive the detector structure in both low and high SNR regimes

1.3.2 Cross-layered Design of Spectrum Sensing and MAC

Protocol

Secondly, we consider a distributed OSA network where the SUs sense thedata channels independently and contend for channel access on a frame-by-framebasis once a data channel is available The random medium access control pro-tocol in conjunction with the spectrum sensing protocol design has been studied

In particular, we are interested in the design of frame duration, sensing time andMAC random access to maximize the secondary network throughput performancewhile protecting the PUs from the interference of the secondary users’ operations

We formulate nonlinear constrained optimization problems for the described tem model with the cross-layered and layered approaches Through numericalresults, the cross-layered approach is shown to perform much better than that ofthe layered approach

sys-1.3.3 MAC Protocol Design for Cooperative Spectrum

Sensing

Thirdly, we propose a MAC protocol for wireless ad hoc cognitive radio works where secondary users employ cooperative spectrum sensing to mitigatethe degradation of the channels between the primary users and the secondaryusers The sensing reports and fused decisions are transmitted based on randomaccess protocols using IEEE 802.11 DCF and IEEE 802.11e EDCA on the con-trol channel, whose access schemes determine the overall achievable throughputamong all the data channels We propose several schemes and derive the upperbound of overall throughput (steady state) The saturation problem is also stud-ied to address the tradeoff between cooperative sensing gain and channel reuseefficiency Sequential detection has been further proposed to reduce the latency