RESOURCE ALLOCATION FOR OFDM-BASED COGNITIVE RADIO SYSTEMS

Transcription

1 RESOURCE ALLOCATION FOR OFDM-BASED COGNITIVE RADIO SYSTEMS by YONGHONG ZHANG B.Eng., Xi an Jiaotong University, China, 1994 M.A.Sc., University of British Columbia, 2006 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Electrical and Computer Engineering) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) December 2008 c Yonghong Zhang, 2008

2 Abstract Cognitive radio (CR) is a novel wireless communication approach that may alleviate the looming spectrum-shortage crisis. Orthogonal frequency division multiplexing (OFDM) is an attractive modulation candidate for CR systems. In this thesis, we study resource allocation (RA) for OFDM-based CR systems using both aggressive and protective sharing. In aggressive sharing, cognitive radio users (CRUs) can share both non-active and active primary user (PU) bands. We develop a model that describes aggressive sharing, and formulate a corresponding multidimensional napsac problem (MDKP). Low-complexity suboptimal RA algorithms are proposed for both single and multiple CRU systems. A simplified model is proposed which provides a faster suboptimal solution. Simulation results show that the proposed suboptimal solutions are close to optimal, and that aggressive sharing of the whole band can provide a substantial performance improvement over protective sharing, which maes use of only the non-active PU bands. Although aggressive sharing generally yields a higher spectrum-utilization efficiency than protective sharing, aggressive sharing may not be feasible in some situations. In such cases, sharing only non-active PU bands is more appropriate. When there are no fairness or quality of service (QoS) considerations among CRUs, both theoretical analysis and simulation results show that plain equal power allocation (PEPA) yields similar performance as optimal power allocation in a multiuser OFDM-based CR system. We propose a low-complexity discrete bit PEPA algorithm. To improve spectrum-utilization efficiency, while considering the time-varying nature of the available spectrum as well as the fading characteristics of wireless communication channels and providing QoS provisioning and fairness among users, ii

3 this thesis introduces the following novel algorithms: (1) a distributed RA algorithm that provides both fairness and efficient spectrum usage for ad hoc systems; (2) a RA algorithm for non-real-time (NRT) services that maintains average user rates proportionally on the downlin of multiuser OFDM-based CR systems; and (3) cross-layer RA algorithms for the downlin of multiuser OFDM-based CR systems for both real-time (RT) services and mixed (RT and NRT) services. Simulation results show that the proposed algorithms provide satisfactory QoS to all supported services and perform better than existing algorithms designed for multiuser OFDM systems. iii

4 Table of Contents Abstract ii Table of Contents vii List of Tables viii List of Figures List of Abbreviations List of Symbols Acnowledgments Dedication ix xii xiv xviii xix Co-authorship Statement xx Chapter 1 Introduction Bacground Scope Motivation Related Previous Wor Resource Allocation Algorithms for OFDM-based CR Systems Resource Allocation Algorithms for OFDM Systems Objectives Thesis Overview References Chapter 2 Subcarrier, Bit and Power Allocation for Multiuser OFDMbased Multi-Cell Cognitive Radio Systems Introduction System Model The Optimization Problem The Single CRU Case The Multiple CRU Case iv

5 2.5.1 Reward/Cost Ratio for Each Constraint Efficiency Value for Adding One Bit to Subchannel m of CRBS The Proposed Algorithm Simulation Results Single CRU Case Multiple CRU Case Conclusions References Chapter 3 An Efficient Power Loading Scheme for OFDM-based Cognitive Radio Systems Introduction System Model The Optimization Problem An Approximate Solution for OP Problem Formulation Suboptimal Solution Simulation Results Conclusions References Chapter 4 Performance of Equal Power Allocation in Multiuser OFDMbased Cognitive Radio Systems Introduction System Model Bounds on bit rates for OWF and PEPA Upper Bounds on Achievable Bit Rate for OWF Achievable Bit Rate for PEPA Rayleigh fading channel Opportunistic Subchannel Assignment A Fairer Subchannel Assignment Scheme Numerical Results Conclusions References Chapter 5 Subchannel Power Loading Schemes in Multiuser OFDM Systems Introduction System Model Bit rates for OWF and PEPA An Upper Bound on Achievable Bit Rate for OWF Achievable Bit Rate for Continuous PEPA Achievable Bit Rate for Discrete PEPA Achievable Bit Rate for Improved Discrete PEPA Improved Discrete PEPA Algorithm v

6 5.4 Rayleigh fading channel Opportunistic Subchannel Assignment A Fairer Subchannel Assignment Scheme Simulation Results Conclusions References Chapter 6 Cross-Layer Resource Allocation for OFDM-based Cognitive Radio Systems Introduction System Model Cross-Layer Resource Allocation for RT Services The Optimization Problem Conversion of MAC Layer Requirements to PHY Layer Requirements Proposed Algorithm Estimation of the Number of Available Subchannels in the next m(t 1, t 2 ) slots Cross-Layer Resource Allocation for Mixed Services The Optimization Problem Conversion of MAC Layer Requirements to PHY Layer Requirements A Goal Programming Approach for Improving Feasibility The Cross-Layer Resource Allocation Algorithm Simulation Results RT Services Mixed Services Conclusions References Chapter 7 Resource Allocation for Non-Real-Time Services in OFDMbased Cognitive Radio Systems Introduction System Model Resource Allocation Algorithm Simulation Results Conclusions References Chapter 8 A Distributed Algorithm for Resource Allocation in OFDMbased Cognitive Radio Systems Introduction System Model Distributed Algorithm Determining Achievability of Target Rates Determining HATR in a Resource-limited Situation Determining HATR in a Resource-abundant Situation vi

7 8.3.4 The Proposed Distributed Algorithm Simulation Results Conclusions References Chapter 9 Conclusions Contributions and Discussions Future Wor References Appendix A Optimal Solutions for Optimization Problems in Chapter A.1 Solution for OP A.2 Solution for OP References Appendix B Proofs of Theorems in Chapter B.1 Proof for Theorem B.2 Proof for Theorem References Appendix C Flowcharts for Algorithm in Section Appendix D Derivation of The Results in (7.8) - (7.10) References vii

8 List of Tables 3.1 Actual interference power exceeding I th l with I th 1 = I th, l = 1, 2 by using SUBOPT-APPROX 2 = W, E{H l,m } = 10 14, l = 1, , l = 1, 2 by using SUBOPT-APPROX 2, S = 2.4 W, and E{H l,m } = 10 14, l = 1, Actual interference power exceeding I th l with I th 1 = I th 6.1 Simulation parameters Dropped pacet rates for different values of p n with p n = p a and R RT = 150 bps Dropped pacet rate with respect to video data rate, R RT with p n = p a = Dropped pacet rate with respect to video data rate, R RT. p n = 0.5 and p a = Non-feasible ratio of PHY and HLL Fairness index comparison for three different schedulers Fairness index Subchannel gains ( ) from CRP j s transmitter to CRP i s receiver Number of bits per OFDM symbol and fairness index for each of the three algorithms and four different sets of nominal rate requirements viii

9 List of Figures 1.1 PU active frequency bands, spectrum holes and CRU OFDM subchannels PU active frequency bands, guard bands, spectrum holes and CRU OFDM subchannels in a protective sharing system Spectrum sharing methods Thesis overview PU active frequency bands, spectrum holes and CRU OFDM subchannels Average number of bits per OFDM symbol per subchannel as a function of CRU power S, with interference thresholds set to W Average number of bits per OFDM symbol per subchannel as a function of the interference threshold with S = 0.32 W Simulation topologies: triangles represent CRBSs and circles represent PU transmitters Average number of bits per OFDM symbol per subchannel per CRBS as a function of the CRBS power constraint for Scenario Average number of bits per OFDM symbol per subchannel per CRBS as a function of the CRBS power constraint for Scenario Average number of bits per OFDM symbol per subchannel per CRBS as a function of the CRBS power constraint for Scenario PUP active and non-active bands and CRP OFDM subchannels Average number of bits per OFDM symbol (ANB) for each PUP band as a function of E{H 1,m } with S = 2.4 W, I1 th = I2 th = W and E{H 2,m } = Average number of bits per OFDM symbol (ANB) on the whole PUP bands as a function of S with I1 th = I2 th = W, E{H l,m } = 10 14, l = 1, Average number of bits per OFDM symbol (ANB) on the whole PUP bands as a function of I1 th with S = 2.4 W, I1 th = I2 th, and E{H l,m } = 10 14, l = 1, Marov chain model for the number of available PU bands ABR as a function of number of CRUs for OWF and PEPA. p a = ABR as a function of p a for OWF and PEPA. K = ABR as a function of average SNR γ with K = 12 users for Case A ABR as a function of average SNR γ with K = 12 users for Case B ix

10 5.3 ABR difference between OWF and PEPA as a function of average SNR γ for Case A. K = 12 and M = 64. For simulation curves, T = 1 ms and for theoretical curves, T = ABR difference between OWF and PEPA as a function of average SNR γ for Case B. K = 12 and M = 64. For simulation curves, T = 1 ms and for theoretical curves, T = ABR as a function of number of subchannels M for Case A. K = 12 and γ = 10 db. For simulation curves, T = 1 ms and for theoretical curves, T = ABR as a function of number of subchannels M for Case B. K = 12 and γ = 10 db. For simulation curves, T = 1 ms and for theoretical curves, T = ABR as a function of number of users K for Case A. M = 64 and γ = 10 db. For simulation curves, T = 1 ms and for theoretical curves, T = ABR as a function of number of users K for Case B. M = 64 and γ = 10 db. For simulation curves, T = 1 ms and for theoretical curves, T = Primary users active frequency bands, guard bands, spectrum holes and CRU OFDM subchannels Marov chain model for the number of available PU bands Average CRBS power for U = 1, 2 and 5 slots Transformations and relationships among the optimization prblems Resource allocation time diagram for CRU 4 with p n = p a = 0.99 and R RT = 150 bps Average total power of eight video conference CRUs as a function of p n with p n = p a and R RT = 150 bps Transmit CRU power with p n = p a = 0.9 and R RT = 150 bps Average total power for eight video conference CRUs as a function of video data rate, R RT with p n = p a = Average total power for eight video conference CRUs as function of video data rate, R RT with p n = 0.5 and p a = Resource allocation time diagram for CRUs 3 and Dropped pacet rate of RT CRUs as a function of video bit rate Average throughput of NRT CRUs as a function of video bit rate System throughput as a function of video bit rate System throughput with respect to number of CRUs with R P R = 1, = 1, 2,, K PU active frequency bands, guard bands, spectrum holes and CRU OFDM subchannels Flow chart of the distributed allocation algorithm Average number of bits per OFDM symbol duration per CRP as a function of the number of available subchannels with S = 10 3 W, K = 3, R1 NOM = 25, R2 NOM = 30, R3 NOM = x

11 8.4 Average number of bits per OFDM symbol duration per CRP as a function of the number of available subchannels with S = 10 3 W, K = 3, R1 NOM = 20, R2 NOM = 20, R3 NOM = Average number of bits per OFDM symbol duration per CRP as a function of total power with M CR = 8, K = 3, R1 NOM = 20, R2 NOM = 20, R3 NOM = C.1 Flow chart for the cross layer resource allocation algorithm: Phase 1, the resource-limited phase. Point B refers to the entry point of the resourceabundant phase C.2 Flow chart for the cross layer resource allocation algorithm: Phase 2, the resource-abundant phase C.3 Flow chart for the Assignment algorithm used in the cross layer resource allocation algorithm xi

12 List of Abbreviations ABR ANB AWGN BER BS CDF (cdf) CR CRBS CRP CRU DPR DSL FCC FI HATR HOL ITWF KKT LHS MAC Average Bit Rate. Average Number of Bits per OFDM symbol. Additive White Gaussian Noise. Bit Error Rate. Base Station. Cumulative Distribution Function. Cognitive Radio. Cognitive Radio Base Station. Cognitive Radio transceiver Pair. Cognitive Radio User. Dropped Pacet Rate. Digital Subscriber Line. Federal Communications Commission. Fairness Index. Highest Achievable Target Rate. Head of Line. Iterative Water-Filling. Karush-Kuhn-Tucer. Left Hand Side. Medium Access Control. xii

13 MDKP NABO NRT OFDM OWF PDF (pdf) PEPA PHY PR PSD PU PUP QoS RA RHS RT SINR SNR Multidimensional Knapsac Problem. Non-Active PU Bands Only. Non-Real-Time. Orthogonal Frequency Division Multiplexing. Optimal Water-Filling. Probability Density Function. Plain Equal Power Allocation. Physical. Proportional Rate. Power spectral density. Primary User. Primary User transceiver Pair. Quality of Service. Resource Allocation. Right Hand Side. Real-Time. Signal to Interference plus Noise Ratio. Signal to Noise Ratio. xiii

14 List of Symbols a t,m b,i b m B OW F B P EP A Subchannel assignment indicator function for subchannel m of CRU The length (in bits) of the ith pacet in CRU s buffer The probability of having m available subchannels Average bit rate for OWF Average bit rate for PEPA c,m (i) (c m (i)) Efficiency capacity of subchannel m of CRU (the CRU) for constraint i c t d d m,l (dm l ) f c f,l,m CR CR (fl,m ) f P U,l,m (f P U l,m ) f t g m g m j,i g t,m h m l, (hm l ) Fraction of service lacing for CRU at time slot t Allowed pacet delivery delay after pacet s creation in time slots The power gain for subchannel m from PU l to CRU s receiver (the CRU s receiver) Carrier frequency The interference power introduced by the signal in the mth subchannel of CRU (the CRU) into PU l s frequency band The interference power generated by PU l to the mth subchannel at CRU s receiver (the CRU s receiver) Fairness index at time slot t The power gain for subchannel m from the CRP s transmitter to the CRP s receiver The power gain for subchannel m from CRU i s transmitter to CRU j s receiver The power gain for subchannel m at time slot t from the CRBS to CRU s receiver The power gain for subchannel m from CRU (the CRU) to PU l s receiver xiv

15 h t,m H m (f) I th l I,m The subchannel assignment function for subchannel m of CRU in time slot t The OFDM receiver filter frequency response Interference power threshold for PU l Interference power from other CRPs on subchannel m of CRP s receiver I CR,m (ICR m ) Interference power from the PU transmitters on subchannel m of CRU s receiver (the CRU s receiver) I P U,m (IP U m ) Interference power from the other CRUs on subchannel m of CRU s receiver (the CRU s receiver) K K RT l CR,t L L OW F m CR,t Number of CRUs or CRPs Number of RT service CRUs The number of available PU bands to a CR system at time slot t Number of PUs Water level when using water-filling algorithm The number of subchannels available to a CR system at time slot t m(t 1, t 2 ) Number of available subchannels from time slot t 1 to t 2 M M CR M t ( M t ) p a p n p n m P Q r,m (r m ) r t,m r t,req Number of subchannels (subbands) Expected number of available subchannels The set of available (unavailable) subchannels at time slot t The probability of a PU staying in active state The probability of a PU staying in inactive state Incremental power required to add the nth bit to subchannel m Transition probability matrix for the number of available subchannels Transition probability matrix for the number of available PU bands The number of bits per OFDM symbol that can be supported by subchannel m of CRU (the CRU) The number of bits per OFDM symbol that can be supported by subchannel m of CRU in time slot t Number of bits that need to be transmitted at time slot t for CRU xv

16 r MAX R The maximum number of bits that can be transmitted in time slot t for CRU Total rate over all subchannels of CRU (CRP) R t 1,t 2 Average data rate of CRU from time slot t 1 to t 2 R DAT A R MAX R MAX R NOM R NRT R P R R t,req R T AR s,m (s m ) s t,m S (S) t D,i t S,i T state T symbol, T s T th l u i v w t W l Φ CR (f) The rate at which CRP can reliably transmit data The maximum number of bits that can be allocated on any subchannel Estimated maximum data rate for CRP Nominal rate requirement for CRU (CRP) Rate request of NRT CRU The nominal rate requirement of CRU Minimum number of bits that needs to be transmitted in time slot 1 to t for CRU Target rate of CRP Transmit power for subchannel m of CRU (the CRU) Transmit power for subchannel m of CRU in time slot t Power limit for CRU (the CRU) The delivery time slot of the ith pacet in CRU s buffer The creation time slot of the ith pacet in CRU s buffer The number of time slots between possible state transitions for a PU The duration of an OFDM symbol Interference temperature limit the amount of resource i that has already been consumed Nominal rate degradation for CRP Weight factor of CRU at time slot t Bandwidth of PU l Equivalent baseband power spectral density (PSD) of the CRU OFDM signal for a transmit power of 1 W Φ P U l (f) Power spectral density (PSD) of PU l s signal xvi

17 Γ π l Π σ 2 0 SNR gap parameter which indicates how far the system is operating from capacity The steady-state probability of being in state l The steady-state probability vector for the number of available PU bands Noise power on each subchannel Note: In this thesis, in order to distinguish a random variable from a sample value, the former is denoted by an uppercase letter, whereas the latter is denoted by a lowercase letter. xvii

18 Acnowledgments I would lie to offer my enduring gratitude to the faculty, staff and my fellow students at The University of British Columbia (UBC), all of whom have inspired me to continue my wor in this field. I owe particular thans to my supervisor, Prof. Cyril Leung, who has provided insightful guidelines, constructive comments, and invaluable suggestions throughout this study. Without his support, this wor would not have been possible. It is my very great privilege to have been one of his students. I would lie to express my sincere thans to my parents for their selfless love and caring. Special thans are owed to my husband, who has supported me throughout my years of education. My thans go to the Natural Sciences and Engineering Research Council (NSERC) of Canada for awarding me a Post Graduate Scholarship (PGS) and support under Grant OGP , the Province of British Columbia for awarding me a Pacific Century Graduate Scholarship, UBC for awarding me a University Graduate Fellowship, and the UBC PMC- Sierra Professorship in Networing and Communications for its support. Together, they have provided me with the financial means to engage in and complete this wor. xviii

19 To my parents and family. xix

20 Co-authorship Statement Each of Chapters 2 to 8 is based on manuscripts that have been accepted, submitted, or to be submitted for publication in international peer-reviewed journals. The manuscripts are all co-authored by myself as the first author and my supervisor, Dr. Cyril Leung. In all these wors, I played the primary role in designing and performing the research, doing data analysis, and preparing manuscripts under the supervision of Dr. Cyril Leung. xx

21 Chapter 1 Introduction 1.1 Bacground Cognitive radio (CR) is a new technology that has attracted a lot of attention recently. It was first presented by Mitola [1] as a novel wireless communications approach with the ability to sense the external environment, learn from its history, and mae intelligent decisions in adjusting its transmission parameters based on the current environment. Hayin [2] defines cognitive radio as follows: Cognitive radio is an intelligent wireless communication system that is aware of its surrounding environment (i.e., outside world), and uses the methodology of understanding-by-building to learn from the environment and adapt its internal states to statistical variations in the incoming RF stimuli by maing corresponding changes in certain operating parameters (e.g., transmit-power, carrierfrequency, and modulation strategy) in real-time, with two primary objectives in mind: (1) highly reliable communications whenever and wherever needed; (2) efficient utilization of the radio spectrum. With the ever increasing demand for mobile and wireless applications, the static assignment of radio resources to licensed holders has become a limiting factor in efficient spectrum utilization. In many jurisdictions, there is little spectrum left for exclusive use allocation [3]. However, studies have shown that a large portion of the assigned spectrum is used only sporadically, and that spectrum utilization is generally very low [4]. CR, with its ability 1

22 to sense the unused bandwidth and adjust its transmission parameters accordingly, is an excellent candidate for improving spectrum utilization. Recognizing this, and to alleviate the looming spectrum-shortage crisis, the FCC [5] has suggested the use of CR technology in order to allow unlicensed users to share radio resources with licensed users while not unduly interfering with them. Orthogonal frequency division multiplexing (OFDM) is a frequency division multiplexing (FDM) scheme that uses a large number of closely spaced orthogonal subcarriers to carry data. It has been considered an appropriate modulation candidate for CR systems [6], not only because of its high spectral efficiency, but also its flexibility in dynamically allocating radio resources to multiple users and its low interference between adjacent subcarriers. PU active frequency bands W 1 W 2 Spectrum hole Spectrum hole Spectrum hole Figure 1.1: PU active frequency bands, spectrum holes and CRU OFDM subchannels. Fig. 1.1 shows the spectrum in a typical OFDM-based CR system. The frequency bands that are currently used by the PUs are the shaded areas mared as W 1 and W 2. The remaining areas are not occupied by the primary users (PUs) at this time and this geographic location. These vacant frequency bands, termed spectrum holes, can be used by CR users (CRUs). 2

23 1.2 Scope To implement CR technology, three main tass are involved [2], namely, radio-scene analysis (radio environment estimation and spectrum hole detection), channel identification (channelstate information estimation and channel capacity prediction), and transmit-power control and dynamic spectrum management. In this thesis, we focus on the last tas and aim to design efficient resource allocation (RA) algorithms for OFDM-based CR systems. 1.3 Motivation The introduction of CR technology poses new RA problems that need to be solved. Compared to conventional wireless communication systems, two new issues arise, namely, the interference power to the PU bands should be ept below a certain threshold and good quality of service (QoS) should be provided to CRUs in spite of the time-varying nature of the available spectrum. To mae unlicensed sharing of the licensed spectrum a reality, PU operation must not be compromised. Thus, CRUs should monitor and eep the generated interference to PU bands to an acceptable level. To this end, the FCC Spectrum Policy Tas Force [7] has recommended the use of interference temperature for assessing the level of interference. The specification of an interference temperature limit for a PU corresponds to a maximum allowed level of interference power; CRUs can use PU frequency bands as long as the total generated interference power to the PUs is ept below this limit. In a fading environment, a CRU signal may undergo deep fading and be received with very little power at the PU receiver. As a result, apart from the spectrum holes, CRUs can opportunistically share PU active frequency bands, as long as the total generated interference power at the PU receiver is below the specified interference power threshold. There are two main types of interference generated by CRUs sharing PU bands. One is the co-channel interference generated by CRUs using the PU active frequency bands, and the 3

24 other is the cross-channel interference from the adjacent channels used by CRUs. Because of orthogonality, inter-carrier interference among CRU subcarriers can be ignored. However, since PUs may not be using OFDM, there could be cross-channel interference [8] generated to the PU bands from adjacent CRU bands and to CRU bands from adjacent PU bands. When a CRU shares spectrum holes as well as PU active-frequency bands, the capacity achievable by the CRU is higher than if PU active-frequency bands are left unused [9]. We refer to this type of sharing as aggressive sharing, since any portion of the spectrum may be utilized at any time. To enable aggressive sharing of the spectrum, new RA algorithms that mae efficient use of the radio resource and eep the total generated interference to the PUs below the specified interference power thresholds are necessary. Guard bands W 1 PU active frequency bands W 2 Guard bands Spectrum hole Spectrum hole Spectrum hole Figure 1.2: PU active frequency bands, guard bands, spectrum holes and CRU OFDM subchannels in a protective sharing system. In some practical situations, aggressive sharing may not be possible. This can happen, for example, when the CR system is co-located with a broadcast PU system in which there are so many PU receivers that the probability of eeping the interference power below the specified interference power threshold at all receivers is almost zero. In such situations, PU active frequency bands cannot be shared in order to avoid excessive co-channel interference. To reduce cross-channel interference, appropriate guard bands can be introduced, as indicated in Fig We refer to this type of spectrum sharing, in which interference to PU receivers 4

25 need not be considered in RA, as protective sharing. Fig. 1.3 shows the spectrum which is shared and the types of interference considered in RA for aggressive sharing and protective sharing. Aggressive sharing - shares both active and non-active PU bands - co-channel and cross-channel interference considered in RA Protective sharing - shares non-active PU bands - CRU interference to PU receivers not considered in RA Figure 1.3: Spectrum sharing methods. Note that RA algorithms designed for aggressive sharing systems can be applied to protective sharing systems by setting the PU interference power threshold at each active PU receiver equals to 0. However, the protective sharing model greatly simplifies RA design because CRU interference to PU receivers does not need to be considered. Since interference to the PUs does not need to be considered in protective sharing systems, it might seem that RA schemes designed for conventional OFDM systems apply directly to OFDM-based CR systems. However, in a CR system, besides the fading characteristics of wireless communication channels, the available transmission spectrum also changes over time. RA algorithms designed for conventional OFDM systems assume that the available spectrum is fixed, which is not the case in CR systems. Thus, new RA algorithms that tae into account both the fading characteristics of the transmission channel and the time-varying nature of the available spectrum are needed. 5

26 1.4 Related Previous Wor Resource Allocation Algorithms for OFDM-based CR Systems Algorithms Dealing with Cross-Channel Interference Cross-channel interference is considered in [10] [12]. In [10], the bit and power loading problem is studied for the downlin of an OFDM-based CR system, in which the PU channel is located in the middle of a frequency band available to CRUs: an optimal scheme based on a Lagrange formulation and two suboptimal schemes are proposed assuming that there is only one CRU in the system. A similar model is used in [11] and [12] to study subcarrier, power, and bit allocation for multiple CRUs. Greedy algorithms are proposed based on minimum CRU power and minimum PU interference considerations. Algorithms Dealing with Co-Channel Interference The RA problem with co-channel interference has been studied in [13] [16]. Different optimization problems are formulated and solved based on various interference-temperature-limit considerations. In [13], to simplify the problem, this limit is converted to a power constraint on each PU band by defining a protection area for the PUs. The power constraint is calculated based on a path loss factor and the distance between the edge of the protection area and the CRU transmitter. Interference per subchannel is considered in [14] for the single CRU case and in [15] for multiple CRU case. The optimization problems formulated in [16] for a multiple CRU and multiple PU system use two interference temperature models proposed in [17]. The first model, which assumes a unified interference temperature limit on each subchannel, is translated into an average interference power threshold at the measurement point. The second model, which assumes different interference temperature limits on different PU active frequency bands, is translated into an average interference power threshold at 6

27 each PU receiver. Instead of the interference temperature limit, some other means of protection for PU signals are considered in [18][19] and [20]. Minimum average rate is guaranteed in [18], by assuming that PUs are willing to be cooperative in RA. PU outage probability is ensured in [19]. In [20], the average PU transmission rate is maintained using CRU cooperation. The above-mentioned algorithms, designed for multiple CRUs, all assume that each subchannel can only be used by at most one CRU at any given time. In some situations, e.g., in an ad hoc system or a multicell cellular system, allowing multiple CRUs to share each subchannel can result in a higher spectrum utilization. In [21], a two-phase channel and power allocation scheme is proposed for multi-cell CR networs. In the first phase, resource allocation is done for all base stations (BSs) in a way that ensures that the interference power levels at the PU receivers do not exceed the predefined thresholds. In the second phase, the channels are allocated to the CRUs. In [22][23], CR systems with one channel are considered in which all CRUs access the channel at the same time, while eeping the total generated interference below the predefined interference temperature limit at a single measurement point. Two co-located cellular systems, consisting of one PU system and one CRU system, are studied in [24], in which the average generated interference from the CRUs to the PUs is ensured to be below the interference temperature limit. In [25], the generated interference to the PUs is limited by a per channel power mas, which specifies the highest power that can be used by a CRU on each channel. Algorithms Maing Use of Spectrum Holes Studies assuming the use of spectrum holes appear in [26] [28]. A spectrum-selection scheme is proposed in [26] for ad hoc networs, in which each user pics its channels based on a set of rules. The users try to maximize their own performance with minimal regard to overall system performance. In [27],[28], a game theoretic approach is utilized to solve the channel allocation problem based on the observation that users in CR systems may not be willing 7

28 to cooperate with others but rather may selfishly try to maximize their own performance. A dynamic channel allocation scheme based on a potential game 1 approach is proposed for ad hoc networs in [27]. In [28], a non-cooperative game is formulated to model the multi-channel allocation problem. In [30], although PU active-frequency bands are left unused, the subchannels in spectrum holes are shared among CRUs, with the objective of minimizing the total required power consumption while satisfying the CRUs data rate and bit error rate (BER) requirements. In [31], cross-layer based medium access control (MAC) protocols are proposed to allow CRUs to share the spectrum holes, which are detected by integrated physical (PHY) layer spectrum-sensing policies. The goal in [32] is to minimize CRU throughput variance in a single-user CR system. In [33], the power allocation problem for a single CRU is cast as a rate-maximization problem that considers the ris of losing a certain subchannel due to PU activity Resource Allocation Algorithms for OFDM Systems Centralized Physical (PHY) Layer Approach The bit and power loading problem for single-user OFDM systems can be solved by using the well-nown water-filling [34] algorithm if we assume that the number of bits to be loaded is a real number, or implement a greedy approach that assigns one bit at a time to the subcarrier that requires the least additional power for the integer bit case [35]. To reduce computational complexity for the integer bit case, various low complexity algorithms have been proposed, for both optimal (e.g. [36, 37]) and suboptimal solutions (e.g., [38] [40]). In the case of the downlin transmission of a BS to multiple users, the subchannels need to be assigned to users exclusively [41]. Therefore, RA involves subchannel assignment in addition to power and bit allocation. When the goal is to maximize system throughput, the 1 In game theory, a potential game is one in which the incentive of all players to change their strategies can be expressed in one global function, the potential function [29]. 8

29 problem can be solved in two separate steps [41], namely, assigning each subchannel to the user with the best channel condition, followed by power and bit allocation. When there are QoS or fairness requirements, subchannel, bit, and power allocation becomes more complicated. Since optimal solutions are generally computationally complex, various sub-optimal solutions have been proposed. In [42] [45], suboptimal solutions are proposed to minimize the total transmit power while satisfying rate and BER requirements for real-time (RT) services. For non-real-time (NRT) services, maximizing system throughput while guaranteeing a certain level of fairness among users is a reasonable goal [46] [49]. Most of these suboptimal solutions use a divide-and-conquer approach, in which the subcarrier, power, and bit allocation problem is broen down into two steps, i.e., allocate subcarriers to users and load appropriate power and bits to each subcarrier. During the first step, power is often assumed to be the same across all subcarriers so as to simplify the problem. Centralized Medium Access Control (MAC) Layer Approach RA also occurs in the MAC layer, which is responsible for pacet scheduling. Almost all existing studies extend opportunistic scheduling [50] strategies for the single carrier case to the multiuser OFDM case with multiple subcarriers. For NRT services, some schemes, e.g., [43][51], extend the proportional fair (PF) rule [52], while others (e.g., [53]) extend the modified-largest weighted delay first (M-LWDF) rule [54] for RT traffic. An urgency and efficiency based pacet scheduling (UEPS) algorithm is proposed in [55] for both RT and NRT services using an urgency factor that reflects the urgency of meeting QoS requirements combined with the PF rule to maximize system throughput. The urgency factor approach has previously been used in the single carrier case [56]. Centralized Cross-Layer Approach Some researchers have adopted a cross-layer design approach in allocating system resources. In [57] [59], sub-optimal algorithms for NRT services are proposed; algorithms for both RT 9

30 and NRT services are studied in [60] and [61]. In [60], the QoS for RT applications is improved by giving high priority to users whose head-of-line (HOL) pacet deadlines are approaching. In [61], the MAC layer QoS requirement for each user is converted to a PHY layer fixed rate requirement based on the average user pacet arrival rate and delay constraint. An optimal subchannel and power allocation strategy is proposed that maximizes system throughput subject to a total transmit power limit and user delay requirements. Distributed Approach While centralized RA is suitable for single-cell systems, distributed algorithms may be more appropriate for multi-cell cellular systems or ad hoc systems. Although distributed dynamic channel allocation (DCA) has been studied for multiple cell cellular networs for voice services, it cannot be easily ported to multiuser OFDM systems. This is because traditional DCA schemes assume homogeneous applications with a pre-determined SINR (signal to noise and interference ratio) threshold, and may not efficiently support services with different requirements. To dynamically allocate resources in a multi-cell system or an ad hoc system, subcarriers may be simultaneously shared among served users in order to improve system performance. In this case, co-channel interference has to be considered. In [62], other users signals are treated as noise, and the power allocation problem is viewed as a non-cooperative game. A distributed iterative waterfilling (ITWF) algorithm is proposed for digital subscriber line (DSL) systems. To achieve the optimal power allocation solution, the achievable target rates must be nown. This is not a big problem for DSL systems, but is unrealistic for time-varying wireless channels. To mae ITWF suitable for wireless systems, a scheme is proposed in [63] for multi-cell wireless systems in which a virtual referee is introduced to displace some users out of certain subchannels when necessary, to allow ITWF to converge to good solutions. Power and bit allocation for multiuser OFDM systems with co-channel interference have 10

31 been formulated as a constrained nonlinear programming problem in [64]. To reduce the complexity of finding a solution, a distributed algorithm is proposed that allocates one bit per iteration. 1.5 Objectives The overall goal of the thesis is to design efficient RA algorithms using both aggressive and protective sharing for OFDM-based CR systems. In the category of aggressive sharing, although cross-channel and co-channel interferences to the PUs have been considered by different researchers separately, they have not been considered jointly. To ensure the PUs normal operation, the total generated interference power to the PUs has to be ept below the specified interference power thresholds. Therefore, both cross-channel and co-channel interference have to be taen into account in RA, especially in cases where the PUs do not use OFDM. Our first objective is to Objective 1: Devise efficient RA algorithms to allocate subchannels, powers, and bits in OFDM-based CR systems, which aggressively share both the spectrum holes and PU active frequency bands while guaranteeing that the total generated interference power due to cross-channel and co-channel interference does not exceed the specified interference power threshold of each PU. In the category of protective sharing, most existing studies focus on dynamic channel allocation, and few consider the influence of the time-varying nature of the available spectrum on QoS and fairness of CRUs. In this thesis, our second objective is to Objective 2: Design efficient RA algorithms in OFDM-based CR systems with QoS provisioning and fairness considerations to operate in a fading environment with time-varying spectrum, and protectively share the spectrum holes without generating undue interference to the PUs. 11

32 1.6 Thesis Overview This thesis is written in the manuscript-based format according to the guidelines established by The University of British Columbia. Each chapter has its own reference list. The relationships among the chapters are shown in Fig. 1.4 and described below. RA algorithms that aggressively share PU bands are discussed in Chapters 2 and 3, and RA algorithms that protectively share PU bands are studied in Chapters 4 to 8. Subchannnel, power and bit allocation for multiple CRUs in a multi-cell cellular system is studied in Chapter 2. Each cell is treated as a CRU system consisting of one cognitive radio BS (CRBS) and multiple CRUs. Subchannel allocation is performed within each cell. Power and bit allocation is done across all the cells. Considering co-channel interference among multiple CRUs, as well as cross-channel and co-channel interference resulting from CRU sharing of PU bands, the RA problem is formulated as a multidimensional napsac problem (MDKP). A low-complexity suboptimal solution is proposed for the formulated MDKP problem. In Chapter 3, a simplification of the model proposed in Chapter 2 is formulated which allows for a faster algorithm. The simplification is based on the fact that cross-channel interference from CRUs to PUs is negligible except for a few subchannels adjacent to the PU bands. Assuming that the bandwidth of a PU is much larger than that of a subchannel in an OFDM-based CR system and that there is usually a guard band between two adjacent PU bands, cross-channel interference from any CRU subchannel impacts mostly one PU band, instead of several PU bands as assumed in Chapter 2. In Chapter 4, the performance of the plain equal power allocation (PEPA) algorithm, which allocates the same amount of power to each available subchannel, is studied for the continuous bits case for multiple OFDM-based CR systems. When the goal is to maximize system throughput, the difference between PEPA and the optimal solution is shown to be small. 12

33 Aggressive sharing Main focus: CRU interference power at PU receivers. Chapter 2 - RA for OFDM-based multi-cell CR systems - Both cross-channel and co-channel interference considered in the model - A generalized multidimensional napsac problem formulation - Low-complexity suboptimal solution proposed using a greedy approach Chapter 3 - RA for OFDM-based single-cell CR systems - Low complexity suboptimal solution proposed Protective sharing Main focus: time-varying nature of the available system resource, QoS and fairness among CRUs. Chapter 4 - Performance evaluation of PEPA for multiuser OFDM-based CR systems - Continuous bit case - Two subchannel assignment strategies studied Convex optimization Chapter 7 - RA for a multiuser OFDM-based single-cell CR system - User proportional rates maintained Chapter 5 - Performance evaluation of PEPA for multiuser OFDM systems, suitable for CR systems - Continuous and discrete bit cases - A simple to implement discrete bit PEPA algorithm proposed Cross-layer design Equal subchannel power allocation Chapter 6 - RA for multiuser OFDM-based single-cell CR systems - Dynamic conversion of CRU MAC layer QoS requirements to PHY layer rate requirements - Problem feasibility issue solved using a goal programming approach Chapter 8 - RA for ad hoc or multi-cell OFDM-based CR systems - Distributed algorithm studied - System throughput maximized with user nominal rates achieved if system resource is plentiful - Fair degradation provided if system resource is limited Goal programming RT Services - On-time RT pacet delivery RT and NRT Services - On-time RT pacet delivery - NRT user nominal rates Centralized algorithm Distributed algorithm Figure 1.4: Thesis overview. A simplified model 13

34 The performance difference between PEPA and the optimal solution for both the continuous and discrete bits case in a multiuser OFDM system is examined in Chapter 5. A low-complexity discrete bit PEPA algorithm is proposed that can also be used in an OFDMbased CR system. In Chapter 6, the subchannel, bit and power allocation problems at the PHY layer and QoS requirements at the MAC layer are considered jointly for RT services on the downlin of a multiuser OFDM-based CR system. The proposed algorithm is designed to provide satisfactory QoS to RT applications is spite of the rapidly changing available resources resulting from PU activities. The RT CRU MAC layer QoS requirements are dynamically converted to PHY layer rate requirements; the conversion depends on the delivery status of queued pacets as well as the number of available subchannels. As an extension, the RA problem for a mixture of RT and NRT services is also considered. The time-varying nature of the number of OFDM subchannels available to CRUs gives rise to two resource allocation issues, namely problem feasibility and false urgency. To solve the problem feasibility issue, which arises when resources are insufficient to meet all user QoS requirements, we adopt a goal programming approach. The false urgency issue is effectively avoided by a proposed rate requirement calculation mechanism based on the status of the pacets in queue and system resource availability. A optimization problem is formulated and the optimal solution is provided. In Chapter 7, we study the RA problem in a multiuser OFDM-based CR system for NRT applications in which average user data rates are to be maintained proportionally. In contrast to existing algorithms designed for multiuser OFDM systems, which are unable to guarantee users proportional rates when applied to a CR system, we propose an optimal RA algorithm that ensures CR user rates are maintained in proportion to predefined target rates, while at the same time providing an improved system throughput. The protective sharing RA algorithms in Chapters 4 to 7 are designed for systems in which centralized algorithms are appropriate. In Chapter 8, we consider RA in an ad hoc 14