Phy layer and mac layer protocol design and performance analysis for opportunistic spectrum access in cognitive radio networks

To improve detection performance, two cooperative spectrum sensing strategies called amplify-and-relay AR and detect-and-relay DR using data fusion policy are proposed to work at the PHY

LAYER PROTOCOL DESIGN IN COGNITIVE RADIO

NETWORKS

CHEN QIAN

NATIONAL UNIVERSITY OF SINGAPORE

2011

LAYER PROTOCOL DESIGN IN COGNITIVE RADIO

NETWORKS

CHEN QIAN

(M Eng., Xi’an Jiaotong University, China)

A THESIS SUBMITTEDFOR THE DEGREE OF DOCTOR OF PHILOSOPHY

DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING

NATIONAL UNIVERSITY OF SINGAPORE

2011

I would like to express my sincere gratitude and appreciation to my supervisorsProfessor Lawrence Wai-Choong Wong and Assoc Professor Mehul Motani for theirvaluable guidance and helpful technical support throughout my Ph.D course Had itnot been for their advices, direction, patience and encouragement, this thesis wouldcertainly not be possible

I would like to thank Dr Ying-Chang Liang at Institute for Infocomm Research(I2R), Agency for Science, Technology and Research (A-STAR), Dr ArumugamNallanathan at King’s College London, and Dr Yan Xin at NEC Laboratories America,for many helpful discussions on my research work

My thanks go to my colleagues in the ECE-I2R Wireless CommunicationsLaboratory at the Department of Electrical and Computer Engineering and AmbientIntelligence Lab at the Interactive and Digital Media Institute, and also go to theresearch group at I2R A-STAR for their generous help and warm friendship duringthese years

Last, I would like to thank my family, especially, my wife Xue Tian and my sonChen Xuesen Delmar, for their love and encouragement

Acknowledgement i

1.1 Spectrum Sensing Techniques 2

1.2 Spectrum Access Mechanisms 4

1.2.1 Opportunistic Spectrum Access Model 4

1.2.2 Spectrum Sharing Model 6

1.2.3 Overlay Model 7

1.3 Motivations and Challenges 8

1.4 Contributions and Organization of the Thesis 10

2 Existing Techniques and Literature Review 13

2.1 Spectrum Sensing Strategies 13

2.1.1 Non-cooperative Spectrum Sensing 14

2.1.2 Cooperative Spectrum Sensing 15

2.2 Opportunistic Spectrum Access 16

2.2.1 Spatial OSA 17

2.2.2 Temporal OSA 17

2.3 Packet Scheduling 18

2.3.1 Guaranteed Access Model 18

2.3.2 Random Access Model 19

2.4 IEEE 802.11 MAC Protocol in WLAN and Multi-hop Networks 20

3 Cooperative Spectrum Sensing Strategies for Cognitive Radio Mesh Networks 24 3.1 Introduction 25

3.2 System Model 26

3.3 Cooperative Spectrum Sensing Strategies for Single-Relay Model 29

3.3.1 Overview of Non-Cooperative Spectrum Sensing Method 29

3.3.2 Performance of AR 31

3.3.3 Performance of DR 35

3.4 Cooperative Spectrum Sensing Strategies for Multi-Relay Model 36

3.4.1 AR for Multi-Relay Model 37

3.4.2 DR for Multi-Relay Model 39

3.5 Cooperative Spectrum Sensing Strategies with Known-CSI Condition 42 3.6 Simulation Results 44

3.6.1 Performance of Single-Relay Model 44

3.6.2 Performance of Multi-relay Model 47

3.6.3 Effects of Known-CSI and Unknown-CSI Cases 49

3.7 Conclusions 50

4 A Two-Level MAC Protocol Strategy for Opportunistic Spectrum Access in Cognitive Radio Networks 51 4.1 Introduction 52

4.2 System Model 54

4.2.1 System Model 54

4.2.2 Spectrum Sensing method 55

4.2.3 Traffic Model and Assumptions 56

4.2.4 Spectrum Access Scheme 57

4.2.5 PU’s Activities and Performance Parameters 58

4.3 Slotted CR-ALOHA and its Performance 60

4.3.1 Slotted CR-ALOHA 60

4.3.2 Throughput Analysis 62

4.3.3 Delay Analysis 65

4.3.4 Optimal Spectrum Sensing Time 68

4.4 CR-CSMA and its Performance 68

4.4.1 CR-CSMA Protocol 68

4.4.2 Throughput Analysis 70

4.4.3 Delay Analysis 75

4.5 Simulation Results 76

4.5.1 Performance of slotted CR-ALOHA 76

4.5.2 Performance of CR-CSMA 81

4.5.3 Tradeoff between Performance and Interference 83

4.5.4 Tradeoff between Performance and Agility 84

4.5.5 Optimal Frame Length 84

4.5.6 Effects of PU’s Activities 89

4.6 Conclusions 92

5 MAC Protocol Design and Performance Analysis for Multi-hop Cognitive Radio Networks 93 5.1 Introduction 94

5.2 System Model 96

5.2.1 System Model 96

5.2.2 Spectrum Sensing and PU Protection 97

5.2.3 PTS/RTS/CTS Access Mechanism 98

5.2.4 Exponential Backoff and Blocking Mechanisms 101

5.3 Queueing Model for CR-CSMA/CA 102

5.3.1 The Packet Transmission Process 102

5.3.2 Markov Chain Model for Packet Transmission Probability τ 110 5.3.3 Packet Service Time T s 113

5.3.4 Queue Empty Probability P0 122

5.3.5 Performance Metrics for G/G/1 Queueing Model 123

5.3.6 Performance Metrics for M/G/1 Queueing Model 126

5.4 Simulation Results 126

5.4.1 Performance of CR-CSMA/CA for multi-hop CRN 128

5.4.2 Performance of CR-CSMA/CA for WLAN 135

5.4.3 Performance Comparisons 135

5.4.4 Effects of Spectrum Utilization of Primary Network 140

5.5 Conclusions 142

6 Conclusions and Future Work 143 6.1 Conclusions 143

6.2 Future Work 145

6.2.1 Spectrum Sensing Technique in Transmission Process 145

6.2.2 Multi-Channel Access Mechanism 145

6.2.3 Security Problem under CRN 146

6.2.4 Cross-Layer Protocol Design for CRN 146

A.1 Proof of Theorem 3.2 147

A.2 Proof of Theorem 3.3 147

A.3 Proof of Theorem 3.4 148

Cognitive radio (CR) communication technique is proposed to relieve the spectrumscarcity problem, which allows the unlicensed secondary users (SUs) to use thespectrum bands originally allocated to the licensed primary users (PUs) Generally,SUs have lower access priority and must operate in a transparent manner withoutinterfering the PUs’ work Thus, the coexistence of SUs and PUs at the same frequencybands brings the challenges for the protocol design of the secondary network on bothphysical (PHY) layer and medium access control (MAC) layer In this thesis, wefocus on cognitive radio networks (CRN) and aim to improve the spectrum sensingperformance at the PHY layer and to solve the access contention problem at the MAClayer Thus, the performance of the secondary network is optimized, while the primarynetwork also can be adequately protected

To improve detection performance, two cooperative spectrum sensing strategies

called amplify-and-relay (AR) and detect-and-relay (DR) using data fusion policy

are proposed to work at the PHY layer of CR mesh networks Considering thesingle-relay and multi-relay models with the conditions of known and unknownchannel state information (CSI), closed-form expressions of the performance metrics,e.g., false alarm probability and detection probability, are derived for each strategy.Then, the comparisons between our proposed strategies and an exiting method or thenon-cooperative spectrum sensing method are provided In addition, the effect of thenumber of relay users on detection performance is also investigated

To solve the channel access contention problem at the MAC layer, a two-level

OSA strategy is proposed, and two MAC protocols called Slotted CR-ALOHA and

CR-CSMA, are developed accordingly Moreover, closed-form expressions of networkmetrics -normalized throughput and average packet delay are derived, respectively.For various frame lengths and different number of SUs, the optimal performances areanalyzed Meanwhile, using the interference factor and the agility factor, the tradeoffsbetween the achievable performance of the secondary network and the protectioneffects on the primary network are studied, and the optimal frame length problem is

also addressed accordingly In addition, the performance comparisons between slotted

CR-ALOHA and CR-CSMA are provided, and the effects of the spectrum utilization ofthe primary network on the performance of the secondary network are also considered

To address the problem of MAC protocol design for a multi-hop network, weconsider the hidden terminal problem and the spectrum sensing technique, and thenextend the traditional RTS/CTS mechanism to the PTS/RTS/CTS mechanism which

involves an asynchronous spectrum sensing technique to perform detection during the

process of the transmission link establishment Based on this mechanism, a new MACprotocol, namely, CR-CSMA/CA, is proposed to coordinate the channel access for thesecondary network and avoid interference to the primary network Using the discretetime G/G/1 queuing model with the assumption of unsaturated network condition,the performance of CR-CSMA/CA is analyzed, and closed-form expressions of theperformance metrics are also derived, viz., packet successful transmission probability,normalized throughput, average packet service time, average buffer queue size, etc Inaddition, the achievable performance of the secondary network are studied for bothmulti-hop CRN and WLAN, in consideration with the number of neighboring SUs, theoffered traffic load, and the spectrum utilization of the primary network

List of Figures

1.1 The OSA model: The shadowed areas with solid line denote the spectra occupied by PUs and the white areas with dash line denote

the spectra occupied by SUs 5

1.2 The spectrum sharing model: SUs share the same spectrum with PUs while the interference power at each PU receiver is lower than a threshold 6

1.3 The overlay model: SU transmitter has a priori knowledge of PU transmitter’s message 7

2.1 Hidden/exposed terminal problem 22

3.1 Cooperative spectrum sensing model 27

3.2 The designed MAC frame structure for our proposed strategies 27

3.3 Detection performances for single-relay model (E0= 0 dB) 45

3.4 Detection performances for single-relay model (E0= 4 dB) 45

3.5 Detection performances for single-relay model (E0= 7.8 dB) 46

3.6 Detection performances for multi-relay model (E0= 0 dB) 48

3.7 Detection performances for multi-relay model (E0= 4 dB) 48

3.8 Detection performances for multi-relay model (E0= 7.8 dB) 49

4.1 The system model of CRN 54

4.2 The MAC frame structure for two-level OSA strategy: The blue arrowsdenote the packet arrivals 55

4.3 The operation scheme of slotted CR-ALOHA: Solid box indicates the

inactive Ptin the current frame and dotted box indicates the active case 61

4.4 The operation scheme of CR-CSMA: Solid box indicates the inactive

Ptin the current frame and dotted box indicates the active case 69

4.5 Normalized throughput S versus t for slotted CR-ALOHA . 77

4.6 Average packet delay D versus t for slotted CR-ALOHA . 77

4.7 Maximum normalized throughput S max versus N for optimal t and maximum t (slotted CR-ALOHA) . 79

4.8 Minimum average packet delay D min versus N for optimal t and maximum t (slotted CR-ALOHA) . 79

4.9 Normalized throughput S versus t for CR-CSMA . 80

4.10 Average packet delay D versus t for CR-CSMA . 80

4.11 Maximum normalized throughput S max versus N for optimal t and maximum t (CR-CSMA) . 82

4.12 Minimum average packet delay D min versus N for optimal t and maximum t (CR-CSMA) . 82

4.13 Tradeoff between performance and interference for slotted CR-ALOHA. 85

4.14 Tradeoff between performance and interference for CR-CSMA 86

4.15 Tradeoff between performance and agility for slotted CR-ALOHA . 87

4.16 Tradeoff between performance and agility for CR-CSMA 88

4.17 Effects of P H0 on the performance of slotted CR-ALOHA . 90

4.18 Effects of P H0 on the performance of CR-CSMA 91

5.1 System model of a multi-hop scenario: The user with black colordenotes the SU, and the user with grey color denotes the PU 96

5.2 PTS/RTS/CTS access mechanism 99

5.3 Packet transmission process 103

5.4 Network topology of a multi-hop CRN: The circle with solid line denotes the area X t or X i, and the circle with dash line denotes the area X c 104

5.5 The vulnerable period of SIFS prior to DATA 108

5.6 The vulnerable period of SIFS prior to ACK 109

5.7 Markov chain model for backoff steps 111

5.8 Generalized state transition diagram of packet service process 119

5.9 Normalized throughput of CR-CSMA/CA 129

5.10 Performance metrics for multi-hop CRN 133

5.11 Performance metrics for WLAN 139

5.12 Effects of the spectrum utilization of the primary network on the performance of the secondary network 141

5.1 System configurations used to obtain the simulation results 127

List of Notations

a lowercase letters are used to denote scalars

a boldface lowercase letters are used to denote column vectors

A boldface uppercase letters are used to denote matrices

(·) T the transpose of a vector or a matrix

(·) H the conjugate transpose of a vector or a matrix

E[·] or E{·} the statistical expectation operator

|a| or kak the length or magnitude or norm of a vector

Pr{·} the occurrence probability of an event

b·c the floor function of a real number

d·e the ceiling function of a real number

C k

S or¡S k¢ the set of all k-combinations of a set S

dom the domain of an element or a function

f 0 or df the first order derivative of a function f

f 00 or d2f the second order derivative of a function f

CCC Common Control Channel

CDF/cdf Cumulative Distribution Function

CGRV Complex Gaussian Random Variable

CRN Cognitive Radio Networks

CSCG Circularly Symmetric Complex Gaussian

CSI Channel State Information

CSMA Carrier Sense Multiple Access

CSMA/CA Carrier Sense Multiple Access with Collision AvoidanceCSR Carrier Sensing Range

DARPA Defense Advanced Research Projects AgencyDCF Distributed Coordination Function

DIFS Distributed Inter-Frame Space

DR Detect-and-Relay

EIFS Extended Inter-Frame Space

FCC Federal Communications Commission

FFT Fast Fourier Transform

MAC Medium Access Control

MBS Mesh Base Station

MSS Mesh Subscriber Station

NAV Network Allocation Vector

OSA Opportunistic Spectrum Access

PDF/pdf Probability Density Function

PGF/pgf Probability Generating Function

PMF/pmf Probability Mass Function

POMDP Partially Observable Markov Decision ProcessPSK Phase Shift Keying

PTI Packet Type Identifier

PTS Prepare-To-Sense

of opportunities for access at the slot time level This contradiction motivates thedevelopment of cognitive radio networks (CRN) [3–5], where unlicensed secondaryusers (SUs) are approved to be organized by using the spectrum bands originally

belonging to licensed primary users (PUs).

Obviously, the coexistence of PUs and SUs in the same frequency bands bringsthe challenges for both physical (PHY) and medium access control (MAC) protocoldesign in CRN As compared with the traditional networks, PUs and SUs in a CRenvironment are usually unknown to each other, and the PUs’ information (e.g.,modulation technique and location) seems to be a “black box” for SUs Obviously,

if SUs are located outside the carrier sensing range (CSR) of PUs, it is impossiblefor them to know PUs’ ON/OFF states only by the carrier sense technique adopted

in traditional collision detection In this case, we must consider using the spectrumsensing technique to perform the primary user detection at the PHY layer Moreover,

we must consider the MAC protocol design for CRN which solves the channel accesscontention problem between PUs and SUs To avoid interference and to protectthe PUs’ operations, SUs are required to agilely vacate the channel or transparentlytransmit their packets when PUs are being active

In this chapter, we will briefly introduce the background and implementationissues at the PHY layer and the MAC layer under CRN At the end of this chapter,

we present the objectives and contributions of this thesis

1.1 Spectrum Sensing Techniques

Spectrum sensing [6] plays an important role in the realization of CRN, whichenables SUs to exploit the unused spectrum bands adaptively to the changing radioenvironment Since SUs are assumed to have no real-time interaction with PUs, they

do not know the exact information of the ongoing transmissions within the primarynetwork Thus, SUs rely only on the local radio observations in the secondarynetwork to detect PUs’ ON/OFF states Therefore, the performance of a spectrumsensing method, determined by two parameters called detection probability and false

alarm probability in signal detection theory, not only deeply affects the achievableperformance of the secondary network, but also has a profound influence on theoperation of the primary network Here, the detection probability is defined as theprobability that a SU can correctly detect the active states of PUs when PUs are active,and false alarm probability refers to the occurrence probability that a SU claims theexistence of active PUs while the truth is that no PU is active at that time According

to these definitions, we can easily conclude that the lower the false alarm probability,the more access opportunities for SUs, and vice versa Similarly, higher detectionprobability decreases the possibility that the secondary network interferes with theprimary network Therefore, to achieve better performance for the secondary networkand to properly protect the operation of the primary network in a CRN, SUs mustmaintain a lower false alarm probability and a higher detection probability However,the performances of these two parameters cannot always be satisfied at the same time,thus a compromise between performance and protection arises In most of applications,the secondary network is required to operate in a transparent mode, i.e., the primarynetwork could work as usual and “feel” that no SU exists Therefore, the main problem

in the implementation of a CRN is to improve the spectrum utilization efficiency andmaximize the system performance of the secondary network while the constraint onprotecting the primary network is satisfied

Generally, three spectrum sensing techniques are widely used for differentapplications [7–12] The first technique called matched filter is a linear optimal filterused for coherent signal detection to maximize the signal-to-noise ratio (SNR) in thepresence of additive stochastic noise The second one called energy detector is optimal

in detecting the unknown signal if the noise power is known The third one calledcyclostationary feature detection determines the presence of PU signals by extractingtheir specific features such as pilot signals, cyclic prefixes, symbol rate, spreadingcodes, or modulation types In addition, the advantages and disadvantages of these

three techniques are summarized in [10] We will detail these techniques in the nextchapter.

1.2 Spectrum Access Mechanisms

According to the definition in [5], CR is an intelligent wireless communicationsystem that is aware of its surrounding environment, adapts its transmission to theelectromagnetic environment, and improves the utilization efficiency of the radiospectrum Based on the properties of the different access methods, spectrum accessmechanisms proposed to address the access contention problem can be classifiedinto three categories: opportunistic spectrum access (OSA) model, spectrum sharingmodel, and overlay model

1.2.1 Opportunistic Spectrum Access Model

OSA envisioned by the DARPA XG program [13] is a feasible and key approach

to implement the coexistence of SUs and PUs over the same bands in CRN, whichallows SUs access the unused channels only when PUs are detected to be inactive,

as seen in Fig 1.1 This mechanism is also called listening-before-transmission, where the listening function is fulfilled by spectrum sensing at the PHY layer, and the transmission function refers to the packet scheduling at the MAC layer Based on

the OSA model, when PUs are detected, SUs must defer their transmissions, vacate

the channel, and then try again later after a predefined duration called blocking time.

Conversely, if PUs are undetected, SUs are allowed to access immediately

To find more spectrum access opportunities without interfering with the primary

networks, Iran et al [11] considered two issues: (1) how long and frequently SUs

should sense the spectrum to achieve sufficient sensing accuracy in in-band sensing,

Figure 1.1: The OSA model: The shadowed areas with solid line denote the spectraoccupied by PUs and the white areas with dash line denote the spectra occupied bySUs.

and (2) how quickly SUs can find the available spectrum band in out-of-band sensing.The first issue is related to the sensing-throughput tradeoff problem studied in [14].Generally, in-band sensing aims at the performance optimization of CRN on a specificchannel, which adopts the periodic spectrum sensing policy during the whole channelaccess time In this case, longer sensing time leads to higher sensing accuracy, andhence results in less interference However, as the sensing time becomes longer, thetransmission time will decrease accordingly Therefore, how to decide the optimalsensing and transmission times over a single channel is an important issue in OSAmodel

The second issue touches upon the channel discovery and selection problemsamong multiple channels Since the spectrum environment changes over time, SUsmust find the new available spectrum bands in real time (out-of-band sensing)

As a result, spectrum discovery time and channel selection time arise Therefore,out-of-band sensing must not only discover as many spectrum opportunities as possible

Tx PU

Rx PU

1.2.2 Spectrum Sharing Model

Unlike the OSA model, SUs in the spectrum sharing model are allowed to transmitsimultaneously with PUs, which is shown in Fig 1.2 However, the resultinginterference from secondary network should not cause performance loss of the primarynetwork or at least remain below an acceptable level Therefore, the transmissionpower of SUs must be controlled lower than a threshold which is named as interferencepower constraint [15–17] To satisfy this constraint, SUs are assumed to have thechannel state information (CSI) from SUs to PUs

In either a single-antenna or a multi-antennas scenario, dynamic resource

Actually, many existing studies focus on performance optimization issuesregarding resource allocation in a spectrum sharing model, e.g [18–20].

have perfect a priori knowledge of PUs’ messages As illustrated in Fig 1.3, SUs

act as the role of relays to help PUs transmit If PUs have packets to transmit, SUsmust allocate part of their power for primary transmission, and the remaining powerstill can be used for secondary transmission On the contrary, if PUs have no packet,SUs can fully make use of their power on their own packet transmissions Obviously,

the principle that the primary network has higher priority to occupy the spectrum bandstill holds in this model.

Most of works regarding overlay model focus on the complex coding schemes[22–27], including cooperative coding, collaborative coding, and dirty paper coding,etc In addition, recent studies on overlay model have been summarized in [28]

1.3 Motivations and Challenges

In traditional networks, the spectrum bands are always licensed and users competefor access opportunities only within their own network However, in CRN, SUs sharethe same spectrum bands with PUs, but have lower access priority Therefore, SUsmust first detect and avoid interference to the primary network at the PHY layerand then compete to use the channel within the secondary network at the MAClayer Many existing works focus on this coexistence problem and aim to optimizesystem performance of the secondary network limited by the interference constraint ofprotecting the primary network However, these literatures are either limited to somerestrictions or based on ideal assumptions, which may not be applicable in the realworld For the sake of better illustration, we summarize the research gaps as follows:

• The detection performance of non-cooperative spectrum sensing technique islimited by the strength of the received signal from PUs Therefore, all thethree spectrum sensing techniques introduced in the Section 1.1 would have poorperformance in the presence of multi-path fading and shadowing

• Although the decision fusion based cooperative spectrum sensing technique canimprove the overall detection performance, this majority logic based decisionmechanism actually cannot improve the individual detection probability

• Most of the MAC protocols assume perfect spectrum sensing and continuous

channel access time which are actually idealistic conditions for CRN and therelated influence has not yet been addressed.

• The guaranteed access model employs an exclusive common control channel(CCC) or central coordinator to coordinate the packet scheduling in a sequentialmanner If PUs are undetected, SUs first compete for the right of access in thecontrol channel, and then only the successful competitor can fully or partiallyoccupy the available transmission channels in the following time slots However,the problem is that this control channel may be unavailable in the application

Moreover, it also easily suffers from control channel saturation problem, thus

we must consider the starvation and fairness problems on this model

• The random access model does not rely on the control channel to solve thepacket scheduling problem, and also can be easily implemented in a CRN.Based on this model, each SU can attempt to transmit whenever it haspackets awaiting transmission However, the retransmission mechanism must

be carefully considered when packet transmission failure occurs To the best ofour knowledge, fewer works concentrate on this research area now

• The existing MAC protocols are mainly proposed for a wireless local accessnetwork (WLAN), but seldom consider a multi-hop scenario In a WLAN, SUscan be easily synchronized to detect the PUs’ states at the same time, thusthey would not interfere with each other However, in a multi-hop scenario,

it is impossible to make SUs sense simultaneously, and hence the inferencefrom more than one-hop neighboring SUs would affect the current spectrum

sensing result In this case, we must consider an asynchronous spectrum sensing

mechanism to avoid interference from the secondary network to the primarynetwork, and analyze the achievable performance of the MAC protocol in amulti-hop CRN

1.4 Contributions and Organization of the Thesis

The main contributions of this thesis are to improve the detection performance at thePHY layer and to solve the channel access contention problem at the MAC layer.First, we adopt an energy detection technique and a data fusion policy to develop thecooperative spectrum sensing strategies under a CR mesh network as described in IEEE802.16 [29, 30] Second, considering the realistic conditions of imperfect spectrumsensing and discontinuous channel access time, we design a random access basedMAC protocols for the secondary network operating in a WLAN Last, we consider

a multi-hop CRN and develop the MAC protocol based on IEEE 802.11 DCF [31].The rest of this thesis is organized as follows

Chapter 2 outlines the non-cooperative and cooperative spectrum sensingstrategies, the spatial and temporal OSA mechanisms, the guaranteed access andrandom access models, and the performance analysis of IEEE 802.11 MAC protocol

in WLAN and multi-hop networks, respectively Meanwhile, we provide a literaturereview of existing works on both the PHY layer and the MAC layer

In Chapter 3, we consider the cooperative spectrum sensing problem for a

CR mesh network and propose two new cooperative spectrum sensing strategies,

called amplify-and-relay (AR) and detect-and-relay (DR), which improve the detection

performance of SUs with the help of other eligible relay SUs so as to agilely vacate thechannel when the neighboring PUs switch to the active state Based on the AR and DRstrategies, we derive closed-form expressions of false alarm probability and detectionprobability, considering the single-relay and multi-relay models with known-CSI andunknown-CSI conditions In addition, numerical results are also provided to show theadvantages of our proposed strategies

In Chapter 4, we focus on the MAC protocol design based on randomaccess model for CRN A two-level OSA strategy is proposed to optimize system

performance of the secondary network and to protect the operation of the primary

network Accordingly, two MAC protocols called Slotted cognitive radio ALOHA

(CR-ALOHA) and cognitive radio based carrier sensing multiple access (CR-CSMA)are developed to deal with packet scheduling of the secondary network Based onthis strategy, we employ normalized throughput and average packet delay as networkmetrics, and derive their closed-form expressions to evaluate the performances ofthe secondary network Moreover, we use interference factor and agility factor asperformance parameters to measure the protection effects on the primary network Forvarious frame lengths and number of SUs, we analyze the optimal performances ofthroughput and delay, respectively, and also study the tradeoff between the achievableperformance of the secondary network and the effects of protection on the primary

network In addition, we compare the performance between slotted CR-ALOHA and

CR-CSMA, and consider the effects of the spectrum utilization of the primary network

on the performance of the secondary network

Chapter 5 develops the MAC protocol for a multi-hop CRN Considering thehidden/exposed terminal problem and the spectrum sensing technique, we extend thetraditional RTS/CTS mechanism to the PTS/RTS/CTS mechanism which employs an

asynchronous spectrum sensing strategy to detect PUs, and establishes the transmission

link through RTS/CTS interaction Based on this mechanism, we propose a newMAC protocol called CR-CSMA/CA to coordinate channel access for the secondarynetwork and avoid interference to the primary network Furthermore, using thediscrete time G/G/1 queuing model with the assumption of unsaturated networkcondition, we evaluate the proposed MAC protocol and derive closed-form expressions

of the performance metrics for the secondary network including packet successfultransmission probability, normalized throughput, average packet service time, averagebuffer queue size, etc In addition, simulation results are provided to show theperformance of CR-CSMA/CA for various number of SUs, traffic load and spectrum

utilization of the primary network, respectively.

At the end of this thesis, we summarize the works that we have done and pointout the further research directions in Chapter 6

2.1 Spectrum Sensing Strategies

Spectrum sensing aims to search the available access opportunities for SUs bydetecting the PUs’ states According to the properties of different detecting methods,spectrum sensing strategies are divided into two classes: non-cooperative spectrumsensing and cooperative spectrum sensing The difference is whether the SUs need todetect the PUs with the help of other neighboring SUs or not Then, we will introducethem below

2.1.1 Non-cooperative Spectrum Sensing

As mentioned in Chapter 1, three spectrum sensing techniques are widely used fordifferent applications: matched filter, energy detection, and cyclostationary featuredetection [7–11]

The first technique, namely, matched filter, is a linear optimal filter, which can

be used for coherent signal detection to maximize the SNR in the presence of additivestochastic noise The main advantage of this technique is that it requires less time to

achieve higher processing gain since only O(1/SNR) samples are needed to meet a

given detection performance However, the implementation of this technique requires

the SU to have a priori knowledge of the PU signal at both the PHY and the MAC

layers (e.g modulation technology, pulse shaping, and packet format) Therefore, each

SU must configure a dedicated receiver for every PU class Obviously, this techniquewould not be smart enough to adapt to the dynamic spectrum environment

The second technique called energy detection is optimal in detecting the unknownsignal if the noise power is known, which can be easily implemented by means similar

to spectrum analysis using the Fast Fourier Transform (FFT) Due to the property of

non-coherent detection, only O(1/SNR2) samples are required to meet the detectionrequirement However, the threshold set for making a decision at each SU is highlysusceptible to unknown noise or interference level Moreover, energy detection isunlikely to differentiate between the PU signals and interference, therefore it cannotbenefit from interference canceling in the adaptive signal processing Besides, energydetection is not suitable in a multi-hop network scenario since transmissions from theneighboring SUs would interfere with the current detection on PUs Even if the energydetection works in a WLAN, the interference avoidance still requires that all the SUsmust be synchronized to perform the spectrum sensing at the same time

The third technique, cyclostationary feature detection, determines the presence of

PU signals by extracting their specific features such as pilot signals, cyclic prefixes,symbol rate, spreading codes, or modulation types As compared with energydetection, this technique can detect weak PU signals from the background of strongnoise or interference, even if PUs and SUs are working simultaneously However, theweakness of cyclostationary feature detection is that it suffers from the sampling clockoffset problem.

In general, the detection performance of non-cooperative spectrum sensingtechniques is limited by the strength of the received signal from the PU Therefore, all

of the three techniques described above would have poor performance in the presence

of multi-path fading and shadowing

2.1.2 Cooperative Spectrum Sensing

Non-cooperative spectrum sensing, indeed, provides an effective means to detect PUs

at SUs However, it is not easy to improve individual detection performance due to theinadequacy of the required detection information

To improve overall detection performance, some collaboration techniques havebeen proposed using either a centralized or a distributed approach, which are based

on a decision fusion (or hard fusion) policy [10, 14, 32–37] In a centralized approach,each SU receives the signals from the PUs, independently makes its local decision, andthen sends the decision result to an anchor node or a base station (BS) Consequently,the BS makes the global decision and immediately broadcasts to every SU once thePUs have been detected However, since the BS may be located far away from SUs,

it is inappropriate to implement this centralized fusion mechanism in a distributedscenario Moreover, the centralized fusion approach would be easily compromised byattackers If an attacker intends to invade the secondary network, it would imitate thePU’s working scheme and sends some fictitious signals in a periodic or an intermittent

manner to obstruct the normal transmission of SUs In this case, since SUs cannotdistinguish these attackers from the real PUs, they must keep quiet to avoid interferenceall the time Obviously, in a centralized approach, the BS becomes the most vulnerablepart of the whole system.

In a distributed approach, each SU collects the decisions from the neighboringSUs similar to the centralized approach The difference is that the SU makes a localdecision by itself instead of the global decision at the BS Thus, this collaborativeapproach can be used in a distributed environment and the risk of being attackedwould reduce accordingly Generally, both the centralized approach and the distributedapproach can be implemented by some classical fusion algorithms, such as ”OR”,

”AND” and ”k out of N” rules, etc According to the property of this type of fusionalgorithms, we know that the majority logic based decision fusion policy actuallycannot improve the individual detection probability

To overcome the weaknesses mentioned above, Ganesan et al in [38, 39]considered a relay model and proposed a cooperative spectrum sensing strategy using

a data fusion (or soft fusion) policy to improve the individual detection performance.According to this strategy, each SU receives signals instead of the decision results fromthe neighboring relay SUs, and then makes the decision based on the received signalswhich is a compound signal containing the signals originally received from PUs byeach relay SU and by itself Therefore, the detection performance can be improved ifthe relay SUs are eligible to be helpers We will detail the data fusion policy in Chapter

3 and propose two new cooperative spectrum sensing strategies accordingly

2.2 Opportunistic Spectrum Access

Opportunistic spectrum access (OSA) mechanism enables SUs to sense in-band orout-of-band to exploit the access opportunities on both space and time dimensions

Therefore, OSA can be divided into two classes: spatial OSA and temporal OSA.

2.2.1 Spatial OSA

The main issue addressed by spatial OSA is to coordinate the channel allocation orspectrum reuse in some particular areas or locations (e.g cellular-based networks),while PUs’ states are considered to be static or slowly varying in time Many existingworks, such as [18, 19, 40–42], propose their own channel allocation algorithms toimprove the spectrum efficiency through dynamic spectrum assignment

To address this issue, the most effective way is to formulate the channel allocationproblem as a graph coloring problem, thus the solution can be obtained by list coloringalgorithms

2.2.2 Temporal OSA

Temporal OSA exploits the temporal spectrum opportunities [14,43,44], where unused

time slots of PUs can be accessed by SUs in real time Therefore, sensing the rapidlyvarying spectrum opportunities becomes a critical technique, and the corresponding

expense also should be taken into account In [14], Liang et al studied the performance

tradeoff between sensing time and achieved throughput of SUs, and demonstratedthe existence of an optimal spectrum sensing time which yields maximum achievablethroughput for SUs and meanwhile protects PUs under certain interference constraints.However, although this policy can guarantee the maximum throughput of secondarylink pair, it aims only for a simple point-to-point transmission model, and is notsuitable to be used in a network environment In [43,44], a MAC protocol based on theframework of partially observable Markov decision processes (POMDPs) is developed

to exploit the optimal sensing and access strategy for CRN However, its complexityexponentially grows with the number of channels, and the assumption that PUs’ usage

statistics remain unchanged actually simplifies the MAC protocol design.

2.3.1 Guaranteed Access Model

Guaranteed access model arranges each SU to access the channel in an orderlymanner or following some rules, which ensures that no access contention occurs inthe transmission process

In traditional networks, round-robin (RR) is one of the simplest packet schedulingalgorithms based on the guaranteed access model, which assigns the time slots to eachuser in equal proportions and in a circular order without priority If a user has packets

to transmit, it just waits for the access assignment reserved for itself Obviously, we seethat RR is a contention free algorithm However, this type of scheduling algorithm isinappropriate to CRN, and the major disadvantage is that it cannot adapt to the dynamicspectrum environment Therefore, the starvation problem arises and SUs would neverget the opportunity to transmit since the secondary network has lower priority to usethe channel

To overcome this drawback, many existing works, e.g [45–54], apply theguaranteed access model into CRN using an exclusive common control channel (CCC)

or central coordinator to schedule SUs’ packets in a sequential manner According tothis policy, all the SUs must compete in the control channel, and only the successful

competitors have the right to use the channels In [49], Su et al considered this kind

of access model, where each frame of the control channel is divided into the reportphase and the negotiation phase In the report phase, two different channel spectrumsensing policies called random sensing policy and negotiation-based sensing policy areproposed to detect the available sub-channels and report the obtained information Atthe end of this phase, all the SUs have knowledge of channel states by listening to thebeacons from the control channel Then, in the negotiation phase, the SUs exchange

data following the p-persistent CSMA protocol to compete for the channel and get the

permission to utilize all the available sub-channels in the next frame

In fact, this centralized or local control channel may not be always available, and

it also easily suffers from control channel saturation problem [53] Since all the SUs

must compete for access opportunities in the CCC, this becomes a bottleneck of thewhole system Once the maximum throughput of the CCC is achieved, the systemperformance will not improve any more even if the number of the available channelsincreases

2.3.2 Random Access Model

A random access model does not rely on a CCC and enables each user to usethe channel whenever it has a packet ready for transmission Many classic MAC

protocols belong to this kind of access model, such as ALOHA, slotted ALOHA, 1-persistent CSMA, p-persistent CSMA, non-persistent CSMA, etc However, without

the coordination of a CCC, two or more SUs could transmit at the same time, resulting

in collisions To overcome the channel access contention problem, efforts can be madefrom two aspects The first one is to constraint the transmitter’s behavior, e.g users

working under CSMA are prohibited to transmit if any other users are transmitting.The second one is to design the remedy mechanism to avoid continuous collisions, e.g.the backoff mechanism.

To the best of our knowledge, fewer studies on packet scheduling consideredthe random access model under CRN The difficult is that in a CR environment,SUs not only compete for the channel with other SUs, but also need to vacate the

channel to avoid interference to PUs Huang et al [55] proposed three random access

schemes with different sensing, transmission, backoff mechanisms for SUs, namely,

VX, VAC, and KS Considering the case of one PU band and one SU, they investigatedthe capacity of SUs and derived the closed-form expressions of performance metricsfor each scheme However, since they assumed that PUs’ packet arrival process andspectrum sensing at SUs are independent of each other, spectrum sensing is actuallyunhelpful to increase the SUs’ access opportunities Therefore, the relevant achievableperformance is pessimistic

2.4 IEEE 802.11 MAC Protocol in WLAN and

Multi-hop Networks

IEEE 802.11 [31] is the standard protocol providing the wireless connectivity forWLAN, which has been widely extended into the wireless multi-hop ad hoc or meshnetworks [56, 57] The fundamental access method of the IEEE 802.11 MAC is adistributed coordination function (DCF) known as carrier sense multiple access withcollision avoidance (CSMA/CA), which belongs to the random access scheme

Since the performance of the IEEE 802.11 MAC protocol addresses thequality-of-service (QoS), many research interests focus on the performance analysis

of the considered network metrics like normalized throughput, average packet delay,

queuing empty probability, average queueing size, packet loss rate, etc Bianchi[58] proposed the first framework to compute the IEEE 802.11 DCF throughput forWLAN which employs a 2-dimensional Markov chain for binary exponential backoffmechanism, under the assumption of saturated network condition which means thatevery user always has packets awaiting in its buffer queue Based on this framework,

Foh et al [59] analyzed the average packet delay at different throughput values.

Furthermore, considering the unsaturated condition instead of the saturated condition,several research groups [60–62] proposed their own frameworks to analyze thedistribution of the packet service time, and then employed the discrete time queueingmodel to study the performance of IEEE 802.11 MAC protocol under WLAN

For multi-hop ad hoc or mesh networks, the performance analysis becomes morecomplicated due to spatial spectrum reuse, where nodes located far apart can transmitconcurrently, resulting in the hidden/exposed terminal problem [63] As seen in Fig

2.1(a), the hidden terminal problem occurs when node A and node C (out of the range

of each other) send packets to node B simultaneously, which definitely results in a collision at node B Moreover, the exposed terminal problem occurs when node

C intends to send to node D, but the ongoing transmission is prevented since its neighboring node A is currently transmitting to node B, as seen in Fig 2.1(b) In

a WLAN, the hidden/exposed terminal problem has been solved by IEEE 802.11DCF A four-way handshaking technique is adopted to establish the transmissionlink, which is known as Request-To-Send/Clear-To-Send (RTS/CTS) mechanism [31].However, in multi-hop scenarios, the RTS/CTS mechanism fixes the hidden terminalproblem well, but it cannot solve the exposed terminal problem due to the collision

avoid mechanism, which actually reduces access opportunities In [64], Xie et al.

extended the existing framework to evaluate the performance of a multi-hop ad hocnetwork, considering the situations at both the transmitter and receiver, respectively.However, the hidden terminal problem is totally neglected in the analysis, which

Figure 2.1: Hidden/exposed terminal problem.

would degrade the possibility of a link establishment between the transmitter and the

receiver using the RTS/CTS interaction In [65], Zhai et al focused on the carrier

sensing and spatial reuse problem and investigated the impacts of different factors

on determining the optimal carrier sensing range for multi-hop ad hoc networks.Furthermore, in [66, 67], the analysis concluded that when the carrier sensing range(CSR) is larger than twice the transmission range (TR), the hidden terminal problemcan be eliminated between the transmitter and the receiver However, we must note that