Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

MAC protocol design and resource management for distributed cognitive radio networks Jha, Satish Chandra 2013

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2013_spring_jha_satish.pdf [ 1.1MB ]
Metadata
JSON: 24-1.0071927.json
JSON-LD: 24-1.0071927-ld.json
RDF/XML (Pretty): 24-1.0071927-rdf.xml
RDF/JSON: 24-1.0071927-rdf.json
Turtle: 24-1.0071927-turtle.txt
N-Triples: 24-1.0071927-rdf-ntriples.txt
Original Record: 24-1.0071927-source.json
Full Text
24-1.0071927-fulltext.txt
Citation
24-1.0071927.ris

Full Text

MAC Protocol Design and Resource Management for Distributed Cognitive Radio Networks by Satish Chandra Jha B. Eng., Tribhuvan University, Nepal, 2004 M. Eng., Asian Institute of Technology, Thailand, 2007 M.Sc., Telecom SudParis, France, 2007  A THESIS SUBMITTED IN PARTIAL FULFILLMENT 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) March 2013 c Satish Chandra Jha, 2013 ⃝  Abstract Cognitive radio (CR) has drawn extensive attention as a promising technique to enable dynamic spectrum allocation in wireless networks in order to increase spectrum utilization. Since coordination and communication over wireless medium is mainly performed at medium access control (MAC) layer, design of a smart MAC is vital for successful deployment of distributed CR network (DCRN). In this thesis, we first investigate research challenges specific to DCRN MAC design and present an overview of current state-of-the-art DCRN MAC protocols. We then propose a MAC protocol called OMC-MAC to address major research issues in DCRN. The distributed architecture and uncertainty in resource availability make QoS provisioning a challenging problem in such networks. Therefore, we incorporate a QoS support module in OMC-MAC in order to provide QoS guarantee to delay sensitive applications. We also investigate adaptive admission control mechanism to limit QoS users in DCRN. Next, we propose resource allocation schemes based on cross-layer interaction between MAC and network layers to minimize network wide resource wastage in multi-hop DCRN. In multi-hop DCRN, loss of a packet after traveling some hops results in wasting all resources allocated to it in previous hops. We propose schemes to allocate transmit power favoring packets which have traveled more hops before reaching resource allocating node. Similar resource allocation schemes are also investigated for multi-hop distributed orthogonal frequency division multiple access network. Finally, we explore research challenges and potential solution approaches in  ii  achieving virtualized future generation cellular networks. Network virtualization (NV) has been envisioned as a promising approach to reshape future generation wireless networks by enabling coexistence of several heterogeneous virtual networks to efficiently share the same physical resources and infrastructure. Successful implementation of NV in cellular systems largely depends on virtualization of their radio access part. Therefore, we investigate research issues in virtualizing radio access of cellular systems. We also discuss the potential of MAC mechanism and resource management developed in this thesis in enabling NV.  iii  Preface The publications which have been resulted from the research conducted in this thesis are given below. A list of other publications which have been resulted from collaboration in various projects have been provided in Appendix C. • Jha, S.C., M.M. Rashid, V.K. Bhargava, and C. Despins, “Medium access control in distributed cognitive radio networks,” IEEE Wireless Commun. Mag., vol. 18, no. 4, pp. 41-51, Aug. 2011 (appears in chapter 2). • Jha, S.C., U. Phuyal, M.M. Rashid and V.K. Bhargava, “Design of OMCMAC: An Opportunistic Multi-channel MAC with QoS Provisioning for Distributed Cognitive Radio Networks,” IEEE Trans. on Wireless Commun., vol. PP, no. 99, pp. 1-12, Aug. 2011 (appears in chapter 3). • Jha, S.C., U. Phuyal and V.K. Bhargava, “Joint Power and Subcarrier Allocation in Multi-hop OFDMA Network: A Cross-Layer Approach,” in Proceedings of AH-ICI2011, Kathmandu, Nepal, pp. 1-6, Nov. 2011 (appears in chapter 4). • Jha, S.C., U. Phuyal and V.K. Bhargava, “Cross-Layer Resource Allocation Approach for Multi-hop Distributed Cognitive Radio Network,” in Proceedings of CWIT’11, Kelowna, BC, Canada, pp. 211-215, May 2011 (appears in chapter 4). • Jha, S.C., M.M. Rashid, V.K. Bhargava, and C. Despins, “OMC-MAC: An Opportunistic Multichannel MAC for Cognitive Radio Networks,” in Proiv  ceedings of IEEE VTC 2009, Anchorage, Alaska, pp. 1-5, Sept. 2009 (appears in chapter 2). In the research contributions made in Chapters 2, 3, 4 and 5, I am the primary author and the researcher. I conducted literature review and identified the research problems. Formulation of the research problems, mathematical analysis and simulations were totally carried out by me. I also wrote the associated manuscripts for publication. Dr. Md. Mamun Rashid is a co-author for contributions in Chapters 2 and 3. He provided some technical feedback during the formulation of the research problems in these chapters. He also provided editorial corrections while writing the associated manuscripts for publication. Umesh Phuyal is a co-author in Chapters 3 and 4. He provided some feedback while formulating the research problems and provided important editorial feedback. Prof. Charles Despins is a co-author for contributions in Chapter 2 and 5. He provided technical suggestions and editorial comments. My supervisor Prof. Vijay K. Bhargava is a co-author for contributions made in Chapters 2, 3, 4 and 5 of this thesis. I consulted him during identification and formulation of the research problems. He also provided editorial feedback for the associated manuscripts.  v  Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  ii  Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  iv  Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  vi  List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  x  List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  xi  Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Motivation and Objectives of the Thesis . . . . . . . . . . . . . . 1.3.1 MAC Design for DCRN . . . . . . . . . . . . . . . . . . 1.3.2 QoS Provisioning for SUs in DCRN . . . . . . . . . . . . 1.3.3 Network-wide Resource Wastage Minimization in Multihop DCRNs and OFDMA Networks . . . . . . . . . . . . 1.3.4 Virtualization of Radio Access in Future Generation Cellular Networks . . . . . . . . . . . . . . . . . . . . . . . 1.4 Selected Assumptions of the Thesis . . . . . . . . . . . . . . . . . vi  1 1 4 8 9 9 10 10 11  1.5  Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . .  2 Design of Medium Access Control Protocol for Distributed CR Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Study of Issues and Research Challenges for MAC Design in DCRNs 2.2 Survey of MAC Proposals for DCRNs . . . . . . . . . . . . . . . 2.3 Design of an Opportunistic Multi-channel MAC (OMC-MAC) for DCRNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Channel Sensing Phase . . . . . . . . . . . . . . . . . . . 2.3.2 Contention Phase . . . . . . . . . . . . . . . . . . . . . . 2.3.3 Contention-free Data Transmission Phase . . . . . . . . . 2.4 Analysis of OMC-MAC . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Contention Phase Length Versus Number of Channels Reserved . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Saturation Throughput . . . . . . . . . . . . . . . . . . . 2.4.3 Effect of Sensing Error on Collision Probability of SU With PU . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Selected Numerical Results . . . . . . . . . . . . . . . . . . . . . 2.5.1 Validation via Simulation . . . . . . . . . . . . . . . . . . 2.6 Distinguishing Features and Advantages of OMC-MAC . . . . . . 3 Enhancement of OMC-MAC and QoS Provisioning in MAC Design for Distributed Cognitive Radio Networks . . . . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Enhancement of OMC-MAC to Further Increase Throughput . . . 3.2.1 Modified Data Transmission Scheme to Improve Throughput 3.2.2 Optimal Length of Contention Phase . . . . . . . . . . . . 3.3 QoS Provisioning in OMC-MAC . . . . . . . . . . . . . . . . . . 3.3.1 Minimum Length of Prioritized Contention Phase Versus QoS Satisfaction . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Admission Control . . . . . . . . . . . . . . . . . . . . . vii  12  15 17 22 29 30 32 34 35 36 38 40 41 44 45  47 47 50 51 52 54 57 58  3.4 3.5  Selected Numerical and Simulation Results . . . . . . . . . . . . Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . .  59 69  4 Resource Optimization in Multi-hop Distributed Networks . . . . . 71 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.2 Resource Optimization in Multi-hop Distributed Cognitive Radio Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.2.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . 76 4.2.2 Hop-count-based Power Allocation Scheme . . . . . . . . 78 4.2.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . 84 4.2.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . 89 4.3 Joint Power and Subcarrier Allocation in Multi-hop OFDMA Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.3.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . 94 4.3.2 Hop-count-based Sub-carrier and Power Allocation Scheme 95 4.3.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . 100 4.3.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . 106 5 Virtualization of Radio Access in Future Generation Cellular Networks: Potentials, Challenges and Solution Approaches . . . . . . . 108 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 5.2 Research Challenges and Implementation Issues . . . . . . . . . . 110 5.2.1 Design of Virtualized BS . . . . . . . . . . . . . . . . . . 111 5.2.2 Coexistence of Multiple VNs and Scalability . . . . . . . 111 5.2.3 Heterogeneity Support and Backward Compatibility . . . . 111 5.2.4 Isolation Between VNs and Overall System Stability . . . 112 5.2.5 Green Communication Support in Modern Cellular Systems112 5.2.6 Fair and Transparent Sharing of Infrastructure and Other Physical Resources among VNs . . . . . . . . . . . . . . 113 5.3 Possible Solution Approaches and Research Scopes . . . . . . . . 114 5.3.1 Implementation of Virtualized BS . . . . . . . . . . . . . 114 viii  5.3.2  5.4  Virtualized MAC Protocol Design and Effective Scheduling & Admission Control Schemes . . . . . . . . . . . . . 5.3.3 Achieving Carbon Efficient Green Communication by Network Virtualization . . . . . . . . . . . . . . . . . . . . . 5.3.4 Co-existence of Non-Cooperative VNs: Cognitive Radio Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.5 Use of Radio Environment Map . . . . . . . . . . . . . . Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . .  6 Conclusions and Future Research Directions 6.1 Conclusions . . . . . . . . . . . . . . . . 6.1.1 Summary of Major Contributions 6.1.2 Concluding Remarks . . . . . . . 6.2 Suggested Future Research Directions . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  115 118 120 121 122 124 124 124 127 128  Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 A Summary of Major Notations Used in Chapter 3 . . . . . . . . . . . 143 B Solution of P4 When XN ,K is Known . . . . . . . . . . . . . . . . . 145 C Other Publications Resulted from Collaboration Work . . . . . . . . 148  ix  List of Tables Table A.1  Summary of major notations used in Chapter 3 . . . . . . . . . 144  x  List of Figures Figure 1.1 Figure 1.2 Figure 1.3  Figure 2.1 Figure 2.2 Figure 2.3 Figure 2.4 Figure 2.5 Figure 2.6 Figure 2.7 Figure 2.8 Figure 2.9 Figure 3.1 Figure 3.2 Figure 3.3 Figure 3.4  A typical scenario of FDM-TDM PU network showing spectrum holes available to SUs for opportunistic use. . . . . . . . Various network architectures of cognitive radio network. . . . A typical scenario of heterogeneous radio resource availability in distributed CR network. . . . . . . . . . . . . . . . . .  2 5 6  Basic functionalities of DCRN MAC. . . . . . . . . . . . . . Classification of existing cognitive MAC protocols. . . . . . . Comparison of MAC protocols for distributed CR networks. . Timing structure of proposed OMC-MAC. . . . . . . . . . . . Channel negotiation and reservation process in OMC-MAC. . Channel busy interval during contention phase. . . . . . . . . Throughput variation with contention phase length and number of channels. . . . . . . . . . . . . . . . . . . . . . . . . . Throughput variation with contention phase length and channel availability. . . . . . . . . . . . . . . . . . . . . . . . . . Acceptable sensing error probability for given PU activity level.  18 23 27 31 33 38  Timing structure of proposed OMC-MAC for QoS provisioning. States of data packets waiting for channel access. . . . . . . . Saturation throughput variation with contention phase length (Tcon ) and number of channels (Nch ). . . . . . . . . . . . . . . Variation of saturation throughput with probability of channel availability (P) for OMC-MAC, OSA-MAC, and HC-MAC. .  55 56  xi  42 43 44  60 61  Figure 3.5  Collision probability of SUs with PUs (Pc ) versus number of secondary flows (N f ) for P = 0.6, and Nch = 4. . . . . . . . . Figure 3.6 Collision probability of SUs with PUs (Pc ) versus sensing error probability (Pe ). . . . . . . . . . . . . . . . . . . . . . . . Figure 3.7 Maximum acceptable sensing error probability (Pˆe ) for given upper bound on collision with PU (Pˆc ) versus number of secondary flows (N f ). . . . . . . . . . . . . . . . . . . . . . . . Figure 3.8 Average channel access delay versus number of channels (Nch ). Figure 3.9 Percentage of QoS unsatisfied users versus number of channels (Nch ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . Figure 3.10 Minimum required length of prioritized contention phase versus maximum acceptable percentage of unsatisfied users (Pˆus ). Figure 3.11 Maximum number of prioritized users in the system (Nˆ f p ) versus maximum acceptable percentage of unsatisfied users (Pˆus ). Figure 4.1 Figure 4.2 Figure 4.3 Figure 4.4 Figure 4.5 Figure 4.6 Figure 4.7 Figure 4.8  A typical scenario of multi-hop distributed cognitive radio network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simplified diagram of a multi-hop DCRN showing connections to and from the node under consideration (N0 ). . . . . . Resource wastage versus number of packets K (β = 1) for different values of power budget Pt . . . . . . . . . . . . . . . Resource wastage versus β (K = 8) for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . . . . . . Outage versus number of packets K (β = 1) for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . . . Outage versus β (K = 8) for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Throughput versus number of packets K (β = 1) for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . Throughput versus β (K = 8) for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii  63 64  65 66 67 68 69  75 77 85 86 87 88 89 90  Figure 4.9 Figure 4.10 Figure 4.11 Figure 4.12 Figure 4.13 Figure 4.14 Figure 4.15  Simplified diagram of a multi-hop OFDMA network showing packet transmissions through the node under consideration (N0 ). 95 Resource wastage versus number of packets K for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . 101 Variation of resource wastage with threshold data rate Rmin for different number of packets K. . . . . . . . . . . . . . . . . . 102 Goodput versus number of packets K for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . . . . . . 103 Variation of goodput with threshold data rate Rmin for different number of packets K. . . . . . . . . . . . . . . . . . . . . . . 104 Outage versus number of packets K for different values of power budget Pt . . . . . . . . . . . . . . . . . . . . . . . . . 105 Outage versus threshold data rate Rmin for different number of packets K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106  xiii  Acknowledgments I would like to profoundly thank my supervisor Prof. Vijay K. Bhargava for his continuous guidance, inspiration and encouragement throughout my PhD studies. In addition, I extend my sincere thanks to Prof. Bhargava for providing me the opportunity to work in the intellectual environment of Information Theory and Systems (ITS) laboratory. I would take this opportunity to thank Prof. Charles ´ Despins from the INRS-EMT, Montr´eal, Qu´ebec, Canada, for his invaluable suggestions, support and encouragement during my thesis. I am thankful to Prof. Robert Schober for serving on my qualifying examination committee; Prof. David G. Michelson and Prof. Alireza Nojeh for serving on my supervisory committee; Prof. Z. Jane Wang and Prof. Sathish Gopalakrishnan for serving on my departmental examination committee; Prof. Victor Leung and Prof. Ryozo Nagamune for serving as University Examiners; Prof. Eric Wohlstadter for serving as Chair of Final Examination committee and Prof. Alexander M. Wyglinski for serving as External Examiner for the Final Examination. Their insightful comments and suggestions have considerably improved the quality of this thesis. I would like to acknowledge the Natural Sciences and Engineering Research Council (NSERC), Canada for the financial support during my PhD studies. I am thankful to members of ITS laboratory for their friendship, encouragement and cooperation during this endeavor. I am also grateful to all other friends who have supported me during my PhD studies. I would like to express special thanks to my parents and family members for  xiv  their love, continuous moral support and encouragement during this thesis work and throughout my life. I would also like to thank my loving wife Swati for her continuous support, encouragement, patience and many sacrifices she has made during this thesis work.  xv  To My Parents . . .  xvi  Chapter 1 Introduction 1.1  Introduction  Radio spectrum is one of the scarcest and valuable resources in wireless communications. Real time measurements have shown that, with traditional static spectrum allocation techniques, significant part of licensed spectrum remains unoccupied most of the time [1]. This massive underutilization of spectrum, in the face of scarcity of spectrum for new applications, is driving the shift of spectrum allocation paradigm from static to dynamic. A promising implementation of this paradigm is called cognitive radio (CR), which can sense the swaths of spectrum and then dynamically tune its transceiver to the identified spare channels [2]–[5]. With CR technology, spectrum utilization can be increased significantly by allowing a group of unlicensed secondary users (SUs) to opportunistically access frequency bands originally licensed to some primary users (PUs). Therefore, CR networks are expected to be one of the prominent future generation wireless network paradigms. Opportunistic use of spectrum requires SUs sensing the spectrum before transmitting data. Spectrum sensing (also called channel sensing) techniques are used to decide whether spectrum is being used by PUs or not at a given time. The parts of spectrum not being used by PUs are available to SUs and are called white/spec1  Available to SUs for Opportunistic Use  Time  Occupied by PU(s)  Frequency Figure 1.1: A typical scenario of FDM-TDM PU network showing spectrum holes available to SUs for opportunistic use. trum holes (or available channels). Figure 1.1 depicts a typical scenario of spectrum holes in a network with frequency division multiplexing (FDM) and time division multiplexing (TDM) implementation. Several spectrum sensing techniques have been explored in the literature [4][6]–[9]. Spectrum sensing techniques are usually Physical layer mechanisms. However, spectrum sensing can include MAC layer and Network layer coordination to increase the reliability of sensing result [10][11]. A database of incumbents can also be used to determine the white space (i.e., spectrum availability) at a particular location [12]–[15]. In this thesis, we assume that there is a reliable sensing mechanism to find out whether a portion of spectrum (or channel) is available to SUs. Since the control and coordination of communication over wireless channels happen mainly at the medium access control (MAC) layer, designing a smart and efficient MAC protocol remains a key requirement for successful deployment of any CR network. It is expected that distributed CR network (DCRN) can be a 2  more practical future generation CR network due to its easier and faster deployment and low cost of implementation. Since DCRN also offers more challenging research issues due to lack of central control unit in such network, in this thesis we focus on the same network. The MAC protocol and resource allocation approach should be able to smartly adapt to the unique features of DCRNs and maintain robust performance in the presence of a highly dynamic environment. Research has shown that MAC solutions designed for traditional distributed wireless systems cannot meet the requirements for DCRNs efficiently [16]. New innovative design of adaptive MAC solutions, therefore, remains of critical importance. Radio spectrum are opportunistically available to SUs which can be relinquished by PUs anytime. Such resource uncertainty makes quality-of-service (QoS) provisioning to SUs an important research issue. Furthermore, efficient resource sharing is crucial to enhance the utilization of resources in CR networks as resources are scarcer and available only opportunistically to SUs. Recently, network virtualization has been envisioned as a promising approach to reshape the future generation wireless networks by enabling coexistence of several heterogeneous virtual networks to efficiently share the same physical resources and infrastructure. It is a novel approach which enhances resource/infrastructure utilization, increases network manageability, and can enable green communication by promoting carbon emission reduction. However, successful implementation of network virtualization in cellular system largely depends on the virtualization of its radio access part. Virtualization of server, node and wired link has been well explored in the literature. Virtualization of radio access is a relatively new research area and significant further research is needed to achieve virtualized radio access in cellular system. In this thesis, we focus on exploring research challenges involved in the MAC protocol design, QoS provisioning for SUs and minimization of network-wide resource wastage in DCRN. In such network, SUs are supposed to co-exist with PUs and hence interference to PUs should be minimized by using only those spectrum portions which are not being used by PUs at any particular time. We provide  3  solutions to some of the major research problems in DCRN MAC design identified in this thesis. In later part of this thesis, we explore and identify research potentials, challenges and solution approaches in virtualizing radio access part of cellular networks to achieve fully virtualized future generation cellular networks. Furthermore, we analyze the potential of cognitive radio technique in co-existence of non-cooperative virtual networks in context of network virtualization of future cellular network.  1.2  Background  In general, CR networks can be formed with two different architectures: centralized and distributed as depicted in Figure 1.2. Usually, centralized architecture is infrastructure-based, while distributed one is an ad-hoc implementation. An infrastructure-based CR network such as IEEE 802.22 wireless regional area network (WRAN) [17] consists of a central unit called an SU base station (BS), which gathers information about the radio environment and manages communication among the SUs within its coverage area. Due to presence of BS, spectrum sensing, spectrum management and coordination among SUs become easier. However, implementation of such a system requires extensive planning and huge investment. Given that spectrum is available only opportunistically, wireless communication operators may not be eager to invest in installing such a network. A distributed CR network i.e., DCRN, on the other hand, usually forms a peer-topeer network among SUs with no central unit to gather information about the radio environment. With respect to network architecture, DCRN is very similar to conventional distributed networks. However, nodes of a DCRN have additional features such as capability of sensing channels, and negotiating a common available channel with intended receiver (Rx) whenever needed. Some of the main characteristics of DCRN can be summarized as follows: • The channels available to SUs are function of time as well as of geographical location of SU nodes. At a given location, a channel available at a 4  PU, BS PU, BS  SU SU  SU  SU  SU SU  SU, BS SU  SU  SU  SU Infrastructure-based CR Network  PU, BS  Distributed CR Network  Figure 1.2: Various network architectures of cognitive radio network (SU = secondary user, PU = primary user, BS = base station). particular time may not be available after a while, because of PU activity. Similarly, at a given time, a channel available at a certain location may not be available at another location, depending on the behavior of the PUs at those locations, as shown in Figure 1.3. • The available channels (i.e., spectrum not being used by PUs) are generally discontinuous and may lie anywhere in the entire spectrum. Therefore, SU nodes are usually capable of switching over a wide range of frequency. SUs’ access to the PU spectrum is opportunistic only. Therefore, the SUs must have the capability to cope with the scenarios that involve relinquishing transmission over the channels reoccupied by any PU. • There is no central unit to coordinate channel sensing, channel access and synchronization among SUs in a DCRN. Furthermore, a DCRN is a multichannel network, with two basic differences from the existing distributed multi-channel wireless networks. The number of available channels varies 5  Ch1 Occupied by PU System PU, BS  Ch2, Ch3 Ch1, Ch2, Ch3 Ch1, Ch3  Ch2 Occupied by PU System PU, BS Figure 1.3: A typical scenario of heterogeneous radio resource availability in distributed CR network at a particular time instant. (Chx: represents a channel X which can be a combination of frequency band (or code) and/or time slot based on PU system implementation) with time and the sets of available channels are different at different nodes in a DCRN. Since DCRN is easy to deploy, has a self-organizing capability, has a lower cost of implementation, and has low complexity, it seems to be more practical future CR network. Therefore, in this thesis, we are mainly focussed on DCRN which does not have any central control entity such as BS. Lack of central coordinating unit combined with dynamic radio environment bring several research challenges in DCRN. More concisely, resource sharing and medium access con6  trol become prime factors for successful operation of such network. Our focus in this thesis is at MAC layer protocol design and resource management. Dynamic spectrum environment may also bring research challenges at network layer routing in the multi-hop DCRNs (MHDCRNs). Various routing approaches have been explored in the literature [18]–[20] for MHDCRNs. In this thesis, while dealing with MHDCRNs, we assume that a routing protocol is in operation to decide about the next hop node for a packet being processed at any node. Routing protocol is a network layer mechanism to find the route for a packet from a source node to a destination node through various intermediate nodes in a multi-hop network. Each node usually maintains a table of next hops for various destinations. The routing table can be established in advance (e.g. in proactive routing protocols) or alternatively a route to any destination can be found on demand by route discovery methods (e.g. in reactive routing protocols) [21]–[24]. Therefore, next hop is known to a node for any packet. The next hop can be considered as intended receiver for the packet processing node when considering MAC layer mechanism. The research conducted so far indicates that it may be difficult to propose a general routing solution for multi-hop cognitive environment as the cognitive environment differs from network to network depending on the PUs activities [25]. For example, if the PU channels, once available, remain available for very long duration (e.g., for several communication durations), the MHDCRNs are then very similar to multichannel multi-hop mesh networks regarding routing problems. Routing solutions proposed for multi-hop mesh networks are applicable to such static MHDCRNs. However, if the spectrum availability is more dynamic, CR-specific routing approaches should be used to keep the performance acceptable. In some cases, the PU channel availability are so dynamic that they are on average available for a shorter duration than communication duration. Since the path can not be considered to be stable for a flow duration, such networks may require per packet routing solutions. Opportunistic packet forwarding based on instantaneous channel information may be preferable to replace the end-to-end routing approaches in a highly dynamic environment. The detail discussion on  7  routing in MHDCRNs is out of scope of this thesis. The literature review showing the state-of-the-art research development in the field of MAC design, QoS provisioning, resource allocation to minimize the network-wide resource wastage for CR networks and network virtualization of future generation cellular systems are presented in relevant chapters.  1.3  Motivation and Objectives of the Thesis  A DCRN is a multi-channel network, with two basic differences from the existing distributed multi-channel wireless networks: (i) The number of available channels varies with time in a DCRN. However, it is assumed to be fixed in existing wireless networks. (ii) The set of available channels is different for different nodes in a DCRN. In contrast, in traditional distributed wireless networks, all nodes belonging to a single network are assumed to have the same set of channels available to them. Due to these two differences, protocol stacks developed for traditional multi-channel wireless networks may not work efficiently for DCRN. Specifically, MAC mechanism which defines medium access approaches must be developed customized to DCRN challenges. The broad objective of this thesis is to design an efficient MAC protocol and develop resource sharing mechanism which address the distributed CR network environment in order to provide a smart coordination and cooperation among CR nodes1 over the air interface. The other objective of this thesis is to explore the challenges in virtualizing the radio access part of cellular systems and investigate potential solution approaches to achieve network virtualization in future generation cellular systems. The specific objectives of this thesis can be described as follows: 1 Throughout  the thesis, we use the terms CR nodes, SU and SU nodes invariably.  8  1.3.1  MAC Design for DCRN  Although several recent MAC proposals have been reported in the literature, there still remain a number of major issues that have not been addressed very effectively. For example, minimal interference to PUs, efficient spectrum utilization, good coordination among SUs and resolution of multichannel hidden terminal problem [26] while keeping the node complexity lower are major challenges which a MAC design should address efficiently in DCRN. These issues demand further research in the same field. In this thesis, therefore, we explore the state-of-the-art MAC design for DCRN in literature with the aim of proposing an efficient MAC protocol for the same network. Our focuss is to first identify the critical design issues and open research challenges in designing DCRN MACs. Then, we propose a DCRN MAC protocol called OMC-MAC to address the major open research challenges identified in this thesis.  1.3.2  QoS Provisioning for SUs in DCRN  Most of the DCRN MAC proposals in literature have not considered QoS provisioning issue. The radio resource available for SUs is opportunistic and entirely depends on PU activity. Since exact behavior of PUs is difficult to predict, radio resource environment is highly dynamic in DCRN. As a result, QoS is a challenging problem. In this thesis, we incorporate a QoS mechanism in the proposed OMC-MAC to provide a QoS guarantee to prioritized SUs. The prioritized SUs can be defined as SUs with stringent QoS requirements such as SUs with delay sensitive applications. The proposed mechanism ensures higher channel access opportunities for prioritized SUs in order to decrease end-to-end packet transmission delay.  9  1.3.3  Network-wide Resource Wastage Minimization in Multi-hop DCRNs and OFDMA Networks  Several resource allocation schemes for CR networks have been proposed in literature to achieve various objectives such as sum rate maximization, interference to PUs minimization and mitigating the effect of sensing error [27]–[33]. Most of these schemes optimize the resource allocation considering single hop network performance parameters. These schemes may not be well suited to optimize the performance of multi-hop distributed CR networks (MHDCRNs). The fact that resources are allocated at each hop to a packet which travels multiple hops before reaching the final destination in MHDCRNs is crucial in evaluating the overall utilization of resources in these networks. Loss of a packet with higher hop-count results in wastage of all resources it has used in previous hops. Therefore, during resource allocation at a particular node, all packets should not be treated equally. Number of hops a packet has already completed and the total resources already invested on it in previous hops must be taken into account to optimize the resource allocation in the network. In this thesis, we propose novel cross-layer power allocation schemes for MHDCRNs which favor packets that have already traveled higher number of hops. We also investigate the similar resource allocation schemes for multi-hop orthogonal frequency division multiple access (OFDMA) networks.  1.3.4  Virtualization of Radio Access in Future Generation Cellular Networks  Network virtualization (NV) is a novel approach enabling existence of multiple virtual networks (VNs), which may have different network interfaces or architectures, on a common physical infrastructure and resources [34]–[36]. NV allows multiple operators (or several networks with different interfaces belonging to the same operator) to share same physical resources in a more flexible and dynamic manner, utilizing the resources efficiently. However, it needs the physical resources/infrastructure to be virtualized into a number of virtual resources which 10  are then offered to different VNs. Although significant amount of work can be found on virtualization of servers, nodes and wire line links in the literature [37] –[39], only a few works have been reported on the virtualization of radio access part of wireless networks. Virtualization of radio access part is crucial for successful implementation of NV concept. We investigate selected research challenges, implementation issues and potential solution approaches for the practical deployment of NV concept in cellular systems. The solution approaches discussed in this thesis may serve as a basis to accelerate research work in the same area.  1.4  Selected Assumptions of the Thesis  In Chapters 2 and 3, we have considered MAC design for distributed CR networks, where SU nodes sense PU channels/spectrum during sensing phase. Spectrum sensing techniques are usually Physical layer mechanisms. A database of incumbents can also be used to increase the reliability of spectrum sensing or to determine the white space (i.e., spectrum availability) at a particular location. In this thesis, we assume that there is a reliable Physical layer sensing mechanism to find out whether a portion of spectrum (or a channel) is available to SUs. Sensing error during spectrum sensing usually consists of two kinds of errors: miss detection and false alarm. Miss detection is making a decision that spectrum is available for SUs, although PUs are present on the spectrum. Miss detection, therefore, can cause collision between a SU and a PU. False alarm, on the other hand, is the situation when we make a decision that PUs are present on a spectrum band, although the spectrum is not occupied by PUs. In case of false alarm, SU system looses transmission opportunity. Since collision of SUs with PUs is affected only by miss detection, in Chapters 2 and 3 of this thesis, we have considered the case of only miss detection as sensing error while analyzing the collision probability of SUs with PUs. In Chapter 4, we have considered a multi-hop ad hoc CR network with distributed architecture while proposing resource allocation schemes. The routing protocol at network layer selects the end-to-end route between source and des11  tination for any data packet. We have assumed that there is a routing protocol in operation, which decides end-to-end route for packets between a source and a destination. There is a defined multi-hop route from source to destination for each data packet in the network based on the routing protocol implemented in the network. The link layer resource allocator at a node, therefore, knows the next hop for each packet. We have not assumed any specific routing protocol. The performance of the proposed scheme is independent of implemented routing protocols. To compare the performance of proposed resource allocation scheme with that of classical scheme, the channel coefficients between the nodes of a multi-hop DCRN are modeled as zero-mean Rayleigh distributed complex random variables in Chapter 4. Usually, there is no line of sight signal path between nodes in most of the modern wireless communication networks. Therefore, the assumption of Rayleigh faded channel coefficients between the nodes should not be unreasonable for numerical analysis. Since the results presented in this chapter are comparative results obtained for the classical and the proposed schemes with similar channel conditions, the assumption on channel coefficient should not affect the nature of final conclusions drawn from the comparative results.  1.5  Outline of the Thesis  Rest of the thesis is organized as follows: • In Chapter 2, we present a MAC protocol design for DCRN. To clarify relevant research challenges and issues in design of MAC protocol, we first provide a detailed study of critical design issues and an overview of current state-of-the-art MAC protocols proposed for DCRN. A classification of existing proposals is provided, and their salient features, advantages and limitations are discussed. We then propose a novel MAC protocol called OMC-MAC that addresses major research issues better than existing solutions. The results achieved from the analytical model and validated by simulation, show that our simple yet efficient MAC design achieves excellent spectrum utilization, shows robustness in presence of sensing error and 12  handles multichannel hidden terminal problem very effectively. • In Chapter 3, we propose an enhancement to OMC-MAC in order to further improve the network throughput. We compare the performance of OMCMAC with those of two existing DCRN MAC protocols. The comparison results show that OMC-MAC outperforms these existing MAC protocols by providing higher throughput and lower collision probability with PUs in case of sensing error. We also introduce a QoS module in our MAC design to address QoS requirements of delay sensitive applications by defining higher priority to such applications during channel reservation. The proposed QoS provisioning module efficiently reduces average delay of delay sensitive applications. It also analyzes admission control criteria to guarantee a predefined QoS requirement for such applications in terms of maximum acceptable delay. • In Chapter 4, we present a cross-layer resource allocation scheme for multihop DCRN in order to minimize the network-wide resource wastage and hence to improve resource utilization in the network. The proposed scheme identifies and utilizes the fact that loss of a packet after traveling some hops in a multi-hop network results in waste of all the resources allocated to it in previous hops. Traditional resource allocation schemes ignore this fact and hence may not be efficient in optimizing the overall resource utilization. Simulation results show that the proposed scheme is capable of minimizing wastage of network resources used by packets in their previous hops without any degradation in throughput and outage performance. • In Chapter 5, we discuss the selected research challenges and implementation issues for the practical deployment of network virtualization concept in wireless communication systems particularly in cellular networks. Virtualization of radio access part of cellular network is crucial to implement a virtualized cellular network in the future. Since there has not been significant work on virtualization of radio access, we focus on it in this chapter. 13  We also present solution approaches to the identified issues and challenges. Moreover, we discuss the potential of cognitive radio technique in respect to enabling network virtualization. • In Chapter 6, we conclude the thesis by summarizing our contributions and presenting concluding remarks. Future research directions are also presented.  14  Chapter 2 Design of Medium Access Control Protocol for Distributed CR Network1 Radio spectrum is a limited resource that is not growing with the increased demand from continuously expanding reach of wireless communications. In the face of this rapidly increasing demand, it is becoming increasingly important to make the best possible usage of this valuable resource. Unfortunately, traditional static spectrum allocation techniques use the spectrum very inefficiently as significant part of licensed spectrum remains unoccupied most of the time [1]. Underutilization of spectrum and scarcity of spectrum for new applications, drive the shift of spectrum allocation paradigm from static to dynamic, resulting in the development of cognitive radio (CR) technology [2][4]. This promising technology allows a group of unlicensed secondary users (SUs) to opportunistically access 1 The  papers based on the research work presented in this chapter have been published as: Jha, S.C., M.M. Rashid, V.K. Bhargava, and C. Despins, “OMC-MAC: An Opportunistic Multichannel MAC for Cognitive Radio Networks,” in Proceedings of the IEEE VTC 2009, Anchorage, Alaska, pp. 1-5, Sept. 2009 and Jha, S.C., M.M. Rashid, V.K. Bhargava, and C. Despins, “Medium access control in distributed cognitive radio networks,” IEEE Wireless Commun. Mag., vol. 18, no. 4, pp. 41-51, Aug. 2011.  15  frequency bands originally licensed to some primary users (PUs). Therefore, a CR network encompasses a number of SUs that use sophisticated spectrum sensing and access techniques to utilize spare bandwidth from PU networks within transmission range. Since the control and coordination of communication over wireless channels happen at the medium access control (MAC) layer, designing a smart and efficient MAC protocol remains a key requirement for successful deployment of any CR network. Compared to its infrastructure-based counterpart, a distributed CR network can be a more practical choice because of its easier and faster deployment, lower system complexity and lower cost of implementation. Therefore, in this thesis, we focus on distributed CR networks (DCRNs), which also offers more challenging research issues due to lack of a central control unit. A MAC protocol for a distributed CR network should be able to smartly adapt to the unique features of such networks and maintain robust performance in the presence of a highly dynamic system environment. MAC design for CR networks is an exciting new research area, and a number of MAC protocols have been proposed in the literature. We briefly review the state-of-the-art in MAC proposals for DCRNs shortly. Although several recent MAC proposals have been reported in the literature, there still remain a number of major issues that have not been addressed very effectively such as minimal interference to PUs, efficient spectrum utilization, and resolution of multichannel hidden terminal problem [26] while keeping the node complexity lower. Therefore, it is interesting to explore the MAC design for distributed CR networks in order to devise an efficient MAC which addresses major research challenges of the same networks. The organization of this chapter is as follows. Section 2.1 presents major research challenges for MAC design in DCRNs. State-of-the-art MAC protocol proposals for DCRNs in literature are investigated in Section 2.2. We classify selected MAC proposals from the literature based on the major design approaches. We also overview the merits and demerits of these approaches in addressing key design issues. Pros and Cons of these existing MAC proposals have been evaluated based  16  on their capabilities to address the research issues identified in Section 2.1. Next, in Section 2.3, we introduce a novel MAC design, referred to as Opportunistic Multi-Channel MAC (OMC-MAC), to better address a number of design issues identified in this thesis. We also analyze the performance of OMC-MAC by deriving expressions for important performance measures in Section 2.4. Section 2.5 presents selected numerical and simulation results which show that OMC-MAC is very efficient in providing higher throughput and lower collision probability with PUs due to sensing error at SU nodes. Finally, we conclude the chapter by discussing distinguished features and advantages of the proposed MAC in Section 2.6.  2.1  Study of Issues and Research Challenges for MAC Design in DCRNs  A MAC protocol design for a DCRN must address some unique challenges that accompany the highly dynamic nature of resource availability in DCRNs. In the following, we summarize the main critical design issues and research challenges for the design of a DCRN MAC [40]. • Resource Availability. The MAC must be able to maintain its robustness and efficiency in the presence of uncertainty about the amount of radio resource that might be available to the CR network over time. The amount of radio resource available in a CR network depends on the activity of PUs. Because SUs cannot predict communications among PUs, MAC has to allocate resources based on the current available information about PU activities. Although channel sensing can help to mitigate the resource uncertainty, it cannot entirely solve the problem because sensing only gives current snapshot of PU activity– not exact prediction of future activity. Moreover, ideal spectrum sensing is not possible in practice and therefore, some amount of sensing error is bound to occur. Any amount of sensing error will add uncertainty to the already dynamic nature of resource availability. 17  Sensing Techniques/ Mechanisms at PHY layer  Scheduling and Management of Channel Sensing  Negotiation and Reservation of Common Available Channels Between Transmitter and Intended Receiver  Data Transmission With No Interference to PUs. Abandon Transmission Upon PUs Come Back  Figure 2.1: Basic functionalities of DCRN MAC. • Interference to PU and Sensing Error. Although SUs opportunistically access only spare bandwidth from PU network, they are likely to cause some level of interference to the data transmission among PUs. The interference occurs mainly due to sensing error that any practical CR network has to endure. If MAC is not designed to effectively counter effect of sensing error, transmissions among SUs can collide with communications among PUs and cause significant performance degradation in the PU network. Due to hardware limitations in a practical CR node, sensing error cannot be completely avoided and therefore, the goal of the MAC design should be to minimize 18  the level of interference as much as possible. Only a few MAC proposals (such as [41] and [42]) have considered effect of the sensing error in their proposals. Since sensing error cannot be completely eliminated, how to minimize interference to PUs in presence of sensing error remains an open research problem. • Channel Sensing Period. Sensing PU channels is necessary but it comes with the price of increased overhead. Due to hardware constraints of the SU nodes, it takes certain time to scan each PU channel, and therefore, scanning all of these channels might not be possible in practice. Besides, scanning different sets of PU channels could yield different amount of resource gains. The distributed architecture makes the issue even more complicated because there is no central authority to coordinate the channel sensing by different SUs in the network. The length and frequency of sensing phase of the MAC, therefore, need to be designed carefully so that the resource utilization could be improved with a tradeoff between any particular sensing strategy and its incurred overhead. • Channel Negotiation. Since a transmitter (Tx) and its intended receiver (Rx) may see different sets of channels available to them and there is no central authority in a distributed CR network, it becomes necessary for the SUs to undergo a channel negotiation mechanism among themselves to select common available channels before transmission [43]. Therefore, a MAC design must incorporate an efficient negotiation mechanism for the proper coordination among the SUs. If not designed properly, the negotiation mechanism could consume significant amount of resources itself due to messaging and time overhead. How to design a channel negotiation mechanism in the MAC protocol so that resources available for data transmission are minimally affected, remains a critical issue for DCRNs. • Time Synchronization. Due to the lack of central authority, some type of time synchronization among SUs (particularly between Tx and Rx before 19  data transmission begins) is critical for a DCRN MAC to function properly. Time synchronization is required for channel negotiation, network establishment, and coordination among SUs. Besides, time synchronization is critical to ensure a global quiet period during sensing phase of a neighboring SU. Achieving a network-wide time synchronization in practice without a central unit is a difficult research problem itself and the presence of heterogeneous spectrum availability in a CR network makes it even more challenging. • QoS Provisioning. QoS provisioning in a wireless network is not possible without necessary support from the MAC protocol, which is responsible for coordinating access over the available radio resources. Such support usually comes in the form of necessary signaling, resource scheduling and admission control mechanisms in the MAC layer. In addition to the distributed architecture, time varying uncertainty of spectrum availability makes it challenging to provide scheduling or admission control support for QoS provisioning in a DCRN. The signaling schemes for QoS support also have to be designed carefully. These signaling schemes must be robust against uncertainty in message exchange due to time varying availability of resource to communicate the QoS-relevant messages. Although, meeting QoS requirement of individual connections/applications in DCRNs is discussed in a few works (e.g., [44] and [45]), it remains an open research issue for DCRNs. • Multi-Channel Hidden Terminal Problem (MHTP). A data channel selected for transmission by a Tx-Rx pair may not be known to the neighbors, therefore, there is always a possibility that one or more neighbors can select the same data channel for transmission, resulting in collision. This phenomenon in multi-channel distributed networks is termed as MHTP. For example, suppose nodes A and B choose a channel Ch1 and start transmitting data on it. Let us also assume that nodes C and D, which are neighbor of A and B, choose a channel Ch2 while A and B are communicating on Ch1.  20  Therefore, A and B do not know about the reservation made by nodes C and D. After completing the current transmission, A and B can choose channel Ch2 for next transmission, although Ch2 is already in use by their neighbor. This results in a collision, and is called the MHTP. If not handled properly, MHTP can drastically degrade the system performance in a DCRN. Since a node is hidden to all transmissions going on the channels to which it is not currently tuned, MHTP arises frequently in multi-channel networks. In DCRNs, an SU may not be aware of the channel negotiation or reservation by neighbors if it is not tuned to the same channel. As a result, the SU may reserve the same channel for transmission and cause collision. This problem needs to be tackled at the MAC layer by proper synchronization and signaling mechanism. It can be addressed better in a multi-transceiver MAC as shown in a few DCRN MACs (e.g., [44] and [46]). However, the presence of multiple transceivers at each node makes the SU node more complex and expensive. The SU nodes in DCRNs are end users, so complexity of SU nodes and hence their cost must be kept as low as possible. The single-transceiver cognitive MAC seems to be more practical solution to keep the node complexity lower. However, the current single-transceiver MAC solutions either do not address the issue at all or handle the issue very inefficiently due to high control overhead. • Network Coordination and Reconfiguration. Heterogeneous spectrum availability in a DCRN is a major challenge for maintaining a good coordination among SUs as well as during network initialization and reconfiguration. There is no known common medium among neighbors for the initial message exchange to start any communication. In the traditional multi-channel wireless networks, if a channel is available to a node, the node can assume that the same channel must be available to its neighbors and hence can start initial message exchange for negotiation of intended upcoming transmission. Once initial contact is established between transmitter and intended receiver, they can agree upon time, channel and other information to be 21  used for the intended transmission. However, due to heterogeneous channel availability in DCRN, initial contact may not be guaranteed. Existence of a common control channel (CCC) owned by DCRN and a network-wide time synchronization may help DCRN MAC to solve the problem by providing a message exchange mechanism among SUs. Assuring a good network coordination and reconfiguration mechanism may be costly in terms of overheads such as control signaling, time taken for the process, and disruption to the entire network. Therefore, this problem must be efficiently addressed while designing a DCRN MAC. Figure 2.1 shows basic DCRN specific functionalities that MAC protocol should provide to ensure opportunistic use of PU channels and to cope with heterogeneous channel availability at SU nodes.  2.2  Survey of MAC Proposals for DCRNs  Since MAC design for CR networks is an exciting new research area, a number of MAC protocols have been proposed in the literature. In this section, we review the state-of-the art in MAC design for DCRNs. This section is based on our published work [40]. To facilitate the discussion, we start with a general classification of existing MAC proposals, as shown in Figure 2.2. CR MAC can be broadly divided into two classes: centralized and distributed– depending on the architecture of the CR network in consideration. In a DCRN, design of a MAC protocol largely depends on number of transceivers needed at each SU node to support the MAC operations. Therefore, DCRN MACs can be categorized into two major groups: single-transceiver and multi-transceiver. Both of these design approaches have their benefits as well as drawbacks. An SU equipped with a single transceiver can listen to only a single channel at any time and therefore, can miss control messages when its transceiver is busy transmitting or receiving data. As a result, single-transceiver MACs are more vulnerable to MHTP. On the other hand, if equipped with multiple transceivers, an 22  Cognitive MAC  Distributed  Centralized  Multi-Transceiver  Single-Transceiver  Without CCC  Without CCC  With CCC  With CCC  SYN-MAC  DC-MAC  With Global Time Synchronization  Without Global Time Synchronization  Dedicated Global CCC  OSA-MAC, C-MAC  HC-MAC  O-MAC  Dynamically Configurable CCC  Global CCC  Local CCC  DOSS  CogMesh  Figure 2.2: Classification of existing cognitive MAC protocols (CCC = common control channel, DC-MAC [41], OSA-MAC [42], O-MAC [44], SYN-MAC [46], C-MAC [47], HC-MAC [48], DOSS [49], CogMesh [50]). SU node can listen to both data and control messages simultaneously. This can be achieved by keeping one of the transceivers continuously tuned to the control messages, while other transceivers are being used for data communications. Hence an SU node is aware of exchange of control messages among neighboring SUs and MHTP becomes easier to handle with a multi-transceiver MAC. However, compared to multi-transceiver MACs, a single-transceiver MAC can work with SU nodes that are much cheaper and less complex to implement. One of the important design assumptions for both single- and multi-transceiver 23  MACs is the presence or absence of a CCC. As discussed in section 2.1, exchange of control messages for channel negotiation and coordination is a crucial MAC issue. In literature, control message exchange is performed either on PU data channels or on a CCC. A CCC is a channel allocated solely for control message exchange. A DCRN can have a single CCC shared by all SUs or multiple CCCs each of which is shared by many SUs. Presence of a CCC simplifies the design of the MAC protocol and offers several advantages including better coordination among SUs and reduced collision of their data transmissions. Most of the existing DCRN MACs are designed assuming the presence of a CCC. Reliance on a CCC, however, may lead to a phenomenon called CCC saturation. It occurs when many or all SUs try to transmit control messages on the CCC at the same time. This might cause the MAC to suffer from performance degradation, especially in a large network or when the traffic load is high. Furthermore, an adversary node can disrupt the operation of the entire network by attacking the CCC (e.g., by flooding control messages on CCC persistently). To avoid these problems, some DCRN MACs are designed without a CCC. For example, DC-MAC and SYN-MAC are proposed to work without a CCC for single- and multi-transceiver systems respectively. The exchange of control messages in these MACs are usually performed on PU data channels themselves. Without CCC, the MAC design becomes more challenging because there is no predefined channel to start the exchange of control messages with intended Rx or neighbors. To address this challenge, Tx continuously visits the channels where intended Rx is expected to be tuned with higher probability in DC-MAC. Once Tx and Rx are successful to communicate, they exchange the initial control messages which contain required information for further coordination regarding upcoming transmission among them. Similarly, an SU node has to visit many channels randomly to collect the updated information about all of its neighbors. On the other hand, time is divided into a number of slots, each of which is assigned for the control message exchange on a particular data channel in SYN-MAC. When a node wants to start communication with a neighbor (i.e., intended Rx), it chooses one  24  of the common data channels between them, waits for beginning of the time slot representing that channel, and then starts exchanging control messages. Each SU node in SYN-MAC has two transceivers or radios: listening radio dedicated to listen to control messages and data radio for data transmission. Since listening radios of SU nodes are tuned to the respective data channel in each time slot, exchange of any control message is known to all neighbors. Consequently, SYN-MAC can provide a good coordination among SUs and avoid the multi-channel hidden terminal problem to a large extent. The cost paid is requirement of global/network-wide time synchronization and waiting for a particular time slot to exchange control message on the respective data channel. Time synchronization becomes an important issue for single transceiver MACs that are designed with a CCC (e.g., HC-MAC, OSA-MAC and C-MAC). In these MACs, SUs cannot remain tuned to the CCC all the time because the same transceiver is used for data transmission as well. Time synchronization among SUs is thus needed to make sure that the SUs know when they have to tune to the CCC for control message exchanges and when they can resume data transmission. This can be achieved either through local or global time-synchronization. Local timesynchronization is usually preferred in DCRNs where coordination is limited to neighboring nodes. For example, HC-MAC, assumes synchronization among only neighboring SUs for its operation. The time frame is divided into three phases in HC-MAC: contention, sensing and transmission. All the SUs remain tuned to the CCC whenever they are idle (i.e., not transmitting/receiving data). Only one Tx-Rx pair can go to sensing phase in a neighborhood if they have succeeded in contention based CCC access during the contention phase. Once a Tx-Rx pair succeeds in contention, the sensing phase begins, and the next contention phase can only begin after this Tx-Rx pair goes through data transmission phase. Some single-transceiver MACs, on the other hand, rely on a network-wide or global time-synchronization. Periods of data transmission, channel negotiation and channel sensing are predefined and known to all SUs in the network. Examples of such MACs include OSA-MAC and C-MAC, where multiple Tx-Rx pairs  25  negotiate common data channels during the channel negotiation phase. The TxRx pairs then transmit data on these negotiated channels during data transmission phase. With global time-synchronization sensing phase of all SUs is simultaneous, which is free of any transmission from SUs. Such global quiet period during channel sensing is not possible with local time synchronization. In multi-transceiver MACs, a CCC can be either dedicated or configured dynamically. In O-MAC, the CCC is assumed to be a dedicated spectrum that either belongs to the DCRN system itself or to an license-exempt (ISM) band. The dedicated CCC is always available to SUs, which simplifies the MAC operation. But, it also requires an extra spectrum band solely for control message exchange. This may lead to wastage of valuable spectrum. To avoid such wastage, a PU data channel is dynamically selected as CCC in DOSS and CogMesh [50]. The dynamically selected CCC can be available either locally to a neighborhood or globally to all SUs. Cluster based DCRNs such as CogMesh usually select a local CCC in each cluster. For example, in CogMesh, a node initiates cluster formation with a locally available PU channel as the CCC. The node then becomes the leader of its cluster. The neighbors to whom the selected CCC is available too, can join the cluster. On the other hand, DOSS selects a PU channel available to all SUs as a global CCC. It is easier to reconfigure a Local CCC than a global one in case PUs reclaim the current CCC. Even in a highly dynamic spectrum availability scenario, SUs within a geographical neighborhood can see a few channels available to all of them. They can select one of these channels as a local CCC. A PU channel being available to all SUs at the same time is less probable. Moreover, the effect of CCC saturation or attack on CCC by an adversary node is less severe in case of a local CCC. Only a part of network is affected by these situations. However, it is difficult to have a global time synchronization and good network coordination in a DCRN with local CCCs. As discussed earlier, a sensing period with a global quiet period (i.e., period with no transmission from SU nodes) cannot be defined without global time synchronization. A common problem with the MACs that use dynamically configured CCC is the interruptions of their MAC operations by  26  PUs. These interruptions can occur quite frequently since the PU network can reclaim the CCC at any time. Since DCRNs do not have central support to gather the topology information, they rely on local coordination to gather the topological information. The existence of a dedicated CCC makes it much easier to gather topology information compared to the case of DCRNs with dynamically configurable CCC. Interference To PUs  MultiChannel Hidden Terminal Problem  Time Synchronization  Sensing Error  DC-MAC Considered  Addressed  Global  Considered  C-MAC  Not Considered  Addressed  Global  Not Considered  HC-MAC  Not Considered  Addressed  Local  Not Considered  OSAMAC  Considered  Addressed  Global  Considered  SYNMAC  Not Considered  Addressed  Global  Not Considered  O-MAC  Not Considered  Not Addressed  Global  Not Dedicated, Considered Global /High  DOSS  Not Considered  Addressed but Inefficient  Global  Not Considered  CogMesh  Not Considered  Not Addressed  Global Not (Partially Considered Addressed)  Network CCC/ ReconfigNetwork uration Coordination Overhead  QoS Provision  Not Not Required High Addressed /Low Dynamic, Not Global High Addressed /Moderate Dedicated, Not Global / Low Addressed High Dedicated, Not Low Global /High Addressed Not Not Required Moderate Addressed /High Low  Partially Addressed  Dynamic, Global /Moderate  Moderate  Not Addressed  Dynamic, Local /Low  Moderate  Not Addressed  Figure 2.3: Comparison of MAC protocols for distributed CR networks (PU = primary user, CCC = common control channel, DC-MAC [41], OSA-MAC [42], O-MAC [44], SYN-MAC [46], C-MAC [47], HCMAC [48], DOSS [49], CogMesh [50]). In addition to above MAC proposals, game theoretic approaches have been pursued recently by a number of researchers, for example in [51] and [52]. Authors in [52] have introduced a simple MAC framework which incorporates game theoretic spectrum allocation strategies. Most of the works on this topic, how27  ever, focus on game theory based dynamic spectrum allocation issues rather than on proposing an entire MAC framework. More research is needed to develop a practical MAC framework that can successfully embed game theoretic spectrum allocation strategies. Figure 2.3 summarizes the salient features of selected MAC proposals for DCRNs on the basis of various operational and performance parameters. Although the level of interference to PUs is an important issue to consider in designing a DCRN MAC, it has not been addressed in most of the existing proposals. In DC-MAC and OSA-MAC, the authors have considered this aspect to some extent in their analysis. Multichannel hidden terminal problem has been addressed in most of the MAC proposals under consideration. This has been achieved usually through exchanges of certain control messages. DOSS utilizes a busy tone approach, which could be inefficient as the busy tone spectrum can be occupied by PU at any time. Most of the DCRN MACs require global time synchronization for their operation. Some MACs such as HC-MAC work with local time synchronization, which could result in exposed terminal problem as mentioned in [48]. Among these existing MAC schemes, only DC-MAC has considered the possibility of sensing error in its design. As discussed earlier, such errors are unavoidable in practice and should be considered in design or analysis of CR MACs. Although considered to some extent in the analysis of OSA-MAC, this issue has remained mostly unaddressed in existing DCRN MACs. HC-MAC, OSA-MAC and O-MAC can achieve better coordination among SUs with the presence of a global dedicated CCC. C-MAC and DOSS cannot achieve the same level of coordination using a dynamically configurable global CCC. But they still achieve a moderate level of network coordination, which is better than that achieved by DC-MAC without a CCC and CogMesh with a dynamic local CCC. SYN-MAC does not have a CCC, however, it can achieve a moderate network coordination due to the presence of a listening radio solely dedicated for listening to control messages. Similarly, the network reconfiguration overhead is lower in presence of dedicated CCC com-  28  pared to that with a dynamic CCC. DC-MAC, on the other hand, has higher reconfiguration overhead because it does not incorporate any CCC. In C-MAC, a complicated mechanism to reconfigure the dynamic global CCC increases its reconfiguration overhead. Most of the MAC proposals discussed have not considered the QoS provisioning in their work. Only few MAC proposals (such as O-MAC, GB-MAC [45]) have included QoS in their MAC operation. However, they propose simplistic strategies to provide QoS guarantee. Similarly, none of the considered MAC proposals has discussed the effect of spectrum mobility/handover problem in DCRNs. Only few MAC proposals (such as OSA-MAC, HC-MAC, DC-MAC) have proposed simplistic sensing strategies. However, how to optimize the length of sensing phase to maximize the overall system performance has not been considered by most of the MAC proposals. HC-MAC has proposed a optimal length of sensing phase constrained by hardware and transmission constraints of SUs.  2.3  Design of an Opportunistic Multi-channel MAC (OMC-MAC) for DCRNs  In this section, we introduce a distributed MAC protocol that efficiently addresses some of the major issues of DCRNs identified in Section 2.1. The proposed protocol is called opportunistic multi-channel MAC (referred as OMC-MAC) and has been published as [53]. The key features of OMC-MAC are as follows: it efficiently handles mullti-channel hidden terminal problem using a simple control message exchange mechanism, the collision probability of SUs with PU is independent of SU node density, there is no collision among SUs during data transmission, and it maintains a good coordination among SUs. The novel contributions of the OMC-MAC proposal lie in the following facts. It integrates channel sensing, negotiation and data transmission phases in an effective way to propose a complete working MAC framework. It addresses multichannel hidden terminal problems using simple SU nodes each having a single half-duplex transceiver. It is achieved by developing a smart message exchange 29  mechanism during channel negotiation period. It ensures a data transmission phase free of any collision among SUs which improves utilization of opportunistically available radio spectrum. It keeps the collision probability of SUs with PUs minimal by reducing the collision with PUs in case of sensing error (miss detection). OMC-MAC reduces collision due to miss detection by allowing any channel to be reserved by only one Tx–Rx SU pair. Furthermore, each of Tx and Rx senses the channel independently and hence reduces the collision with PUs because collision can occur only when both of them miss detect the channel. A brief overview of OMC-MAC is presented below. We assume a single half duplex transceiver per node to make the SU node simple. The protocol assumes a dedicated global CCC owned by DCRN to exchange the control messages. The time frame is divided into periodic beacon intervals. We assume that SU nodes are synchronized by the periodic beacon signal. Whenever a node joins the network, it first listens to beacon signal for at least one beacon duration on the CCC and synchronizes itself with rest of the network. Since CCC is always available and known to SUs in OMC-MAC, the network initialization, reconfiguration and coordination is easier and reliable. Therefore, if the joining node does not hear any beacon message, it starts sending periodic beacon message assuming it is the first node in the network. The time synchronization and network association in distributed networks are well addressed areas in the literature and are out of scope of this thesis. The beacon interval is further divided into three phases: sensing, contention and data transmission as depicted in Figure 2.4. Below, we provide a brief overview of the working operation of the protocol during these phases.  2.3.1  Channel Sensing Phase  OMC-MAC incorporates channel sensing before channel selection as a part of MAC framework. Therefore, it can benefit from up-to-date sensing information before going into data transmission. It also adds the flexibility of choosing appropriate ratio between the durations of the sensing phase and the data transmission phase to achieve a desired tradeoff between sensing overhead and system utiliza30  Beacon  Sensing Phase  Contention Phase  (Scanning channels)  (Negotiation & Coordination among SUs, Channel Reservation)  Data Transmission Phase  Beacon Period  Figure 2.4: Timing structure of proposed OMC-MAC. tion. During sensing phase, SUs are not allowed to transmit; therefore, sensing result is free from interference due to neighboring SUs. Each SU senses the channels independently and maintains a list of available channels by the end of sensing period. It may not be possible to scan all the available data channels in a given sensing period due to hardware limitations of SUs. Therefore, only a part of the total data channels may be scanned by each SU. For example, SUs can keep track of previous channel scanning results and try to scan those channels which were available most of the time based on transmission history. Similarly, SUs can scan the channels which are likely to be available to themselves as well as to their neighbors. Partial channel sensing issues have been addressed in detail in literature (e.g., [7][54][55] and references therein). In the subsequent sections of this chapter, we assume that each SU senses all PU channels during sensing phase. However, performance of our protocol can be further improved from partial channel sensing strategy and is a topic for future work.  31  2.3.2  Contention Phase  In this phase, each SU having data to send contends for the CCC in order to participate in a channel negotiation and reservation mechanism. The contention resolution is based on backoff mechanism similar to that in the conventional IEEE 802.11 distributed coordination function (DCF).2 Figure 2.5 shows how this channel negotiation and reservation take place, for example, when nodes Q and R want to send data to nodes U and S, respectively. Let us assume that Q becomes successful in the contention to get access to CCC and sends a request to send (RTS) message to the intended Rx, i.e., U. The RTS message primarily contains the list of available channels at node Q. Upon successful reception of the RTS, U selects a common available channel X1 between itself and Q and sends a clear to send (CTS) message to Q with the identity of X1 after a short inter-frame space (SIFS) period. If no common channel is available between them, U will just ignore RTS and CCC will be considered idle after distributed inter-frame space (DIFS) period. If Q gets CTS, it then confirms the reservation of channel X1 by sending a confirm RTS (CRTS) message back to U after SIFS period. Channel X1 is hence reserved for transmission from Q to U during the following data transmission phase. Each Tx–Rx pair reserves at most one data channel. All neighbors of U (namely nodes R, S, and T in Figure 2.5) set network allocation vector (NAV) [56] for the reserved channel X1 after receiving the CTS message. On the other hand, the neighbors of Q (i.e., nodes R and T ) set their NAV after receiving the CRTS message. Note that if a handshake message is lost, the channel will be considered free after DIFS period. In this example, we assume that R ends its backoff in the current contention period and also succeeds in reserving channel X2. Note that channel X1 is no longer considered for negotiation between R and S during the current BI because they have already set NAV for X1 . All nodes clear their NAVs in the beginning of next BI. If the last channel reservation goes beyond the contention phase, the involved Tx–Rx pair will simply ignore the reservation process and wait for next contention phase. 2 Description  of IEEE 802.11 DCF can be found in [56] and [57].  32  R  S  U  Q  T  Available Channels: Q (Tx)  X1, X3, X4, X9  U (Rx)  X1, X5, X7  R (Tx)  X1, X2, X5, X9  S (Rx)  T  X2, X5, X7 X1, X2, X4, X7  Sensing Phase  RTS CRTS, Ch X1  NAV: Ch X2 Transmit on Ch X1 NAV: Ch X2  CTS, Ch X1 RTS  Transmit on Ch X2 CRTS, Ch X2 NAV: Ch X1  CTS, Ch X2 NAV: Ch X1  Receive on Ch X1  NAV: Ch X1 Receive on Ch X2  NAV: Ch X1, X2  Contention Phase  Data Transmission Phase  Time  Figure 2.5: Channel negotiation and reservation process in OMC-MAC. In case of multiple common data channels available between a Tx–Rx pair, we assume that the Rx chooses a channel randomly during negotiation. However, OMC-MAC can incorporate other channel selection strategies as well. For example, a Tx–Rx pair could choose the best available common channel. Channel assignment strategies in CR networks in order to optimize performance parameters like throughput and fairness exist in literature (e.g., [58]–[60] and references therein) which are out of scope of this thesis. Our protocol can also benefit  33  from better channel selection strategies compared to the random channel selection scheme assumed in this chapter. Note that the proposed design of contention phase can effectively address MHTP because it provides a smart coordination among neighbors during channel reservation process. Whenever a channel is reserved by a Tx–Rx pair, all nodes which are neighbors of Tx or Rx set NAV for the channel. Now, if any of these neighbors needs to reserve a channel, that particular channel will not be reserved again as the NAV is already set for it. Since all SUs remain tuned to the CCC for the entire negotiation phase, a node is able to set NAV for all channels reserved by its neighbors. Therefore, a channel cannot be reserved by more than one Tx–Rx pair and, consequently, MHTP does not occur. The mechanism used in OMC-MAC to combat MHTP is simpler and more efficient compared to those in other protocols such as HC-MAC, SYN-MAC, C-MAC, and OSA-MAC. If a node is not visible to both Tx and Rx, it cannot set NAV and can reserve the same channel. However, the transmission from such a node which is not visible to Tx as well as Rx will not cause collision and hence there will be no MHTP.  2.3.3  Contention-free Data Transmission Phase  During this phase, each Tx node which has successfully reserved a channel during contention phase starts transmitting data to its intended Rx node using the reserved channel. We assume that MAC packet data unit (MPDU) is of fixed size and one or more MPDUs can be transmitted during data transmission phase (DTP). Each MPDU transmission is followed by a short acknowledgement (ACK) message. If the Tx does not receive ACK, it will consider the data transmission unsuccessful. DTP is assumed to be multiple of MPDU plus ACK transmission time. Note that data transmission of different Tx–Rx pairs occur in parallel and are free of contention from other transmitting SUs. If both nodes in a Tx–Rx pair happen to suffer from sensing error during sensing phase, a collision with PU can happen during DTP. However, this probability is usually very low in practice. In case such a collision occurs, the Tx needs to go through sensing and contention phases in 34  the next BI.  2.4  Analysis of OMC-MAC  We assume an overlay architecture where SU nodes belonging to a DCRN opportunistically exploit the spectrum holes in the band licensed to a PU network. For network topology, we assume that the SU nodes are distributed uniformly across the DCRN. Furthermore, for simplicity of the analysis, a homogeneous system is considered where occupancy on each of the PU channels is independently and identically distributed across the DCRN. Over time (several transmission slots), distribution of PU channel occupancy may follow any distribution (e.g., Geometric, Exponential, and Gamma). However, for simplicity we assume that only the average occupancy of channel j by PUs over time is known, and is equal to 1 − Pj , where Pj is the average probability of channel j being free from PU transmission. In each transmission slot, the availability of PU channel j is then assumed to follow Bernoulli distribution with parameter Pj . Each SU node is assumed to have similar transmission range. For our analysis, those neighbors of an SU which are not contending for data transmission can be ignored without loss of generality. The average number of contending neighbors of an SU node is denoted by N f , which depends on the assumed transmission range and the node density in the DCRN. We also assume that each node is capable of handling maximum of one traffic flow or application at a time. Therefore, we use the terms SU node, user, and traffic flow interchangeably throughout rest of this chapter. With these assumptions, we will derive a number of important performance measures for OMC-MAC below. Note that the proposed protocol does not require homogeneous channel availability and data rate assumptions for its operation. In this section, we have made such assumptions only for the sake of simplicity and to make the analysis mathematically tractable. However, similar analysis can be performed for other cases such as when PU channels are available with different probabilities, average channel availability varies with location and time, and channels are on different spec35  trum bands supporting different transmission ranges and data rates. The proposed protocol senses the channels at the beginning of each frame, the Tx–Rx pairs negotiate common channels between them and transmit data on the negotiated channels. These operations work well with all these cases of homogeneous and heterogeneous channels availabilities, transmission ranges and data rates. In heterogeneous channel availability case, if partial sensing is implemented, a statistical sensing strategy which favors sensing of channels with higher availability may improve the protocol performance. Similarly, in case of channels with heterogeneous data rates, channel negotiation may be modified to achieve other advantages. For example, selecting channels with higher data rate can enhance the system throughput whereas allowing SUs with stringent QoS requirements to reserve better channels can be beneficial in QoS provisioning. Notation: In this Chapter, unless stated otherwise, the maximum, average and minimum values of a variable x are denoted by x, ˆ x, ¯ and x, ˇ respectively. ⌊x⌋ represents the integer part of x.  2.4.1  Contention Phase Length Versus Number of Channels Reserved  SUs exchange control messages on the CCC using access mechanism similar to IEEE 802.11 DCF during the contention phase. The contending SU takes part in contention trial which may consist of one or more channel busy periods, one or more empty slots, one or more unsuccessful message exchanges, and a successful message exchange [57][61]. Busy period represents the time when channel is occupied by neighbor(s). SUs freeze their backoff counters during the busy period. Note that time is slotted when CCC is idle. We denote length of each idle slot by σ . A slot will be empty if no node attempts to exchange message at the beginning of the slot; otherwise, it will be the beginning of either a successful message exchange period (Ts ) (if only one node attempts sending a message), or an unsuccessful message exchange period (Tc ) (if more than one node attempt sending message). Let Pt be the probability that a node attempts to send a control message 36  in a randomly chosen slot, Pcm be the probability that control messages collide during contention phase and Pb be the channel busy probability. Then, based on two dimensional Markov model as described in [61], Pt can be written as Ψ1 , (1 − Pb )(1 − Pcm )Ψ1 + (1 − 2Pcm )(Wˇ + 1)Ψ2 + PcmWˇ (1 − (2Pcm )m )Ψ2 (2.1) where Ψ1 = 2(1 − Pb )(1 − 2Pcm ), Ψ2 = Pb + Pcm (1 − Pb ), Wˇ is minimum backoff contention window size and m is the maximum backoff stage such that maximum window size is 2mWˇ . The probability that there is at least one transmission in the considered slot (i.e., channel busy probability) is given by Pt =  Pb = 1 − (1 − Pt )N f .  (2.2)  Also, the conditional probability of success Pcs , which is the probability that one and only one SU transmits in a slot, is given by Pcs =  N f Pt (1 − Pt )N f −1 . Pb  (2.3)  Therefore, the average length of one cycle of successful message exchange for channel reservation, Tsi , can be approximated as Tsi =  Pempty σ + Ps Ts + Pc Tc , Ps  (2.4)  where Ps = Pb Pcs and Pc = Pb (1 − Pcs ) are the probabilities that a transmission in a slot becomes successful and unsuccessful, respectively, and Pempty = (1 − Pb ). It can be shown that Tc = tDIFS + tRT S and Ts = Tc + 2tSIFS + tCT S + tCRT S , where tDIFS , tRT S , tSIFS , tCT S , and tCRT S are durations of DIFS, RTS, SIFS, CTS, and CRTS messages, respectively as shown in Figure 2.6. If Tcon is the length of contention phase, expected number of successful message exchanges can be approximated as Tcon /Tsi . Hence, from (2.4), the maximum number of successful message exchanges per BI is given by  37  Ts DIFS  RTS  SIFS  CTS  SIFS  CRTS  Successful message exchange Tc DIFS  RTS  Unsuccessful message exchange – RTS is collided or No common data channel between transmitter and receiver  Figure 2.6: Channel busy interval during contention phase.  ⌊  ⌋ ⌊ ⌋ Tcon Tcon Ps Nsucc = = . Tsi Pempty σ + Ps Ts + Pc Tc  (2.5)  We assume that channel occupancy by PUs at different data channels is homogeneous and denote the average probability of any channel to be free from PU transmission by P. Denoting total number of data channels by Nch , the average number of available data channels is given by Nch P. The average length of contention phase required for all available channels to be successfully reserved can thus be approximated as Tsi Nch P.  2.4.2  Saturation Throughput  As saturation throughput is a major performance measure to evaluate MAC protocols [62], we analyze it in this section under the assumption that SUs always have data packets in their transmit queue. When the number of available channels is sufficiently large, the number of successful message exchange given by (2.5) will be equal to the number of channels reserved in the system. However, in practice, when the available number of channels is small, the contention phase may be empty after few initial channel reservations due to lack of channels available to be reserved. Similarly, if the number of channels is very high, all channels may not be reserved in a contention phase of given duration. This factor must be 38  considered while calculating system throughput. Therefore, the average number of channels that can be reserved during Tcon is given by ( ) Nch Nch i ¯ N = ∑ min(i, Nsucc ) P (1 − P)Nch −i . i i=1  (2.6)  The saturation throughput of the system averaged over several BIs normalized by per channel data rate can be calculated as ¯ DT /TBI , S¯ = NT  (2.7)  where TDT and TBI are durations of DTP and BI, respectively. It is easily seen from Figure 2.4 that TDT = TBI − Tsensing − Tcon ,  (2.8)  where Tsensing represents the duration of sensing phase. It should be noted that the duration of each acknowledgement message is generally very short compared to that of data packets. Thus, their overhead during DTP can be ignored while ¯ calculating S. Note that in the calculation of saturation throughput we have not considered the case of false alarm. False alarm is a case when SU senses a PU channel occupied by PUs although the PU channel is free from PU transmission. In the above analysis, false alarm cases are considered as PU channels being occupied by PUs. Therefore, the actual achievable saturation throughput by the proposed protocol will be higher when there is no false alarm. False alarm does not contribute towards increasing collision probability with PUs. It only reduces the transmission opportunity.  39  2.4.3  Effect of Sensing Error on Collision Probability of SU With PU  Many of the existing protocols (e.g., DOSS, C-MAC, and SYN-MAC) seem to overlook the effect of sensing error in MAC design. Although OSA-MAC [42] addresses the issue to some extent, the effect of sensing error grows with number of traffic flows. Decentralized cognitive MAC (DC-MAC) [41] has analyzed the effect of sensing error in a probabilistic way using partially observable Markov decision process. However, actual implementation of DC-MAC may be difficult in practice. On the other hand, proposed OMC-MAC effectively addresses this issue in a more practical way. It decreases the collision of SUs with PU due to sensing error arising from miss detection by allowing at most one Tx–Rx pair to reserve a channel for transmission and ensuring no collision when only one node among the Tx–Rx pair suffers from such error. Let us represent the average probability that an SU fails to detect PU’s presence in a data channel by Pe . We assume that Pe is same for all SUs and all channels. Each SU node is assumed to make sensing decision independently. Therefore, a Tx–Rx pair selects channel j for transmission only when it is sensed as available by both the Tx and Rx. Since each channel can be reserved by at most one Tx–Rx pair, the collision probability with PU on channel j can be written as Pc j = Pe2 (1 − Pj ), where Pj is the probability that channel j is available. Assuming homogenous availability of channels (i.e., Pj = P ∀ j), the collision probability is given by Pc = Pe2 (1 − P).  (2.9)  From (2.11), it is apparent that the collision of SUs with PU is independent of number of SUs in a neighborhood. Consequently, OMC-MAC is expected to show highly scalable performance in minimizing interference to the PUs. Moreover, for a specified upper bound on Pc (i.e., Pˆc ), proposed MAC can tolerate maximum Pe  40  given by  √ Pˆe =  Pˆc . 1−P  (2.10)  Although we have assumed homogeneous availability of data channels, the heterogeneous channel availability case can also be covered by deriving Pˆc as ˇ Pˆc = maximize Pc j = Pe2 (1 − P), j  (2.11)  where Pˇ is the minimum probability of any data channel being unoccupied by PU.  2.5  Selected Numerical Results  In this section, we evaluate the performance of OMC-MAC using the analytical methods presented in the previous section and compare the selected numerical results with our simulation results. Unless otherwise specified, we use TBI = 100ms, sensing phase length Tsensing = 10ms, Pt = 0.02, N f = 50, W = 32, m = 5 and control channel bit rate of 1 mbps for the numerical and simulation results. We have assumed Tsensing = 10ms, however its actual value depends on many factors like sensing strategy, hardware configuration of SU nodes, number of PU data channels and so on. The value of Tsensing does not change the nature of results, but it simply shifts the curves in the graphs plotted. We also assume DIFS = 128 µ s, SIFS = 28 µ s, σ = 50 µ s for the message exchange on CCC during contention phase. Figure 2.7 shows that, for a given Nch and an increasing Tcon , throughput increases first, reaches a maximum value and then starts decreasing. The maximum throughput is reached when Tcon becomes just enough to reserve all available channels. Further increase of Tcon negatively affects the throughput because it reduces TDT , while the number of reserved channels remains unchanged (Nch P). Varying P while keeping Nch fixed produces similar results as shown in Figure 2.8. From Figure 2.7 and 2.8, it is apparent that OMC-MAC is very efficient in utilizing the spare radio resources from PU network. For example, when Nch = 10, 41  14 Simulation Results Numerical Results  N  ch  12  Throughput( × Data Channel Rate)  N  ch  10  =55  =40  8 Nch =25 6  4 Nch =10 2  0  0  10  20 30 40 Contention Period length (% of Beacon Interval)  50  60  Figure 2.7: Throughput variation with contention phase length and number of channels Nch (P = 0.5, N f = 45). P = 0.5, subtracting the unavoidable sensing overhead (we keep it at 10% of TBI ), the expected data rate available for use by SUs is Rexp = Nch P × 0.9 × Rc = 4.5Rc . Here Rc is the data rate of a single PU channel. According to Figure 2.7, for same parameter values, the OMC-MAC has an average throughput of about 4Rc , which is close to Rexp . Figure 2.9 shows the maximum acceptable sensing error probability with given channel availability and tolerable interference to PUs. Analyzing the plot, we can see that OMC-MAC has a very high tolerance to sensing error. For example, collision probability with PUs remains below 1% even if the sensing error probability rises to 10%, which is a significant level for sensing error. OMC-MAC can offer this excellent performance since it does not cause collision to PUs when only the 42  12 Simulation Results Numerical Results 10 Throughput ( × Data Channel Rate)  P =0.55  8 P =0.40  6 P =0.25  4  2  0 0  P =0.10  10  20 30 40 Contention Period Length (% of Beacon Interval)  50  60  Figure 2.8: Throughput variation with contention phase length and channel availability(Nch = 30, N f = 45). TX or Rx suffers from sensing error. In OMC-MAC, both the Tx and Rx have to make sensing error at the same time to cause collision to PUs. Furthermore, when PUs are less active (i.e., when P is higher), required sensing accuracy can be relaxed even more. The performance of proposed OMC-MAC has been compared with those of two existing MAC proposals called OSA-MAC and HC-MAC in Chapter 3. The proposed MAC provides higher saturation throughput compared to those provided by OSA-MAC and HC-MAC. Also, proposed MAC is able to lower the effect of sensing error in terms of collision with PUs compared to that in OSA-MAC.  43  0.25  Maximum Acceptable Sensing Error (Pˆe )  Pˆc = 0.002 0.2  Pˆc = 0.004 Pˆc = 0.006 Pˆc = 0.008 Pˆc = 0.01  0.15  0.1  0.05  0 0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  Probability of Data Channel Availability (P)  Figure 2.9: Acceptable sensing error probability for given PU activity level and maximum limit on collision with PU.  2.5.1  Validation via Simulation  To validate our analytical method, we have simulated our proposed OMC-MAC protocol using a discrete event simulator. For the simulation runs, we use same values for the system parameters as were used to gather the numerical results. We plot the simulation results alongside our analytical results in Figs. 2.7 and 2.8. The vertical error bars on the simulation curves are drawn for 95% confidence interval. As we can see from those plots, the curves depicting the simulation results match very closely with the curves drawn from numerical results. Therefore, we are confident about the validity of our analytical method in evaluating the performance of OMC-MAC. 44  2.6  Distinguishing Features and Advantages of OMC-MAC  Here, we discuss some distinguishing features of OMC-MAC to highlight its benefits compared to existing protocols. Many of the protocols (e.g., DOSS, C-MAC, SYN-MAC) seem to overlook the effect of sensing error in the MAC design. In contrast, we recognize the importance of the issue and effectively address it in OMC-MAC. Although OSA-MAC addresses the issue to some extent, the affect of sensing error grows with the node density of the CR network. In OMC-MAC, probability of collision with PUs due to sensing error is very low and the effect is independent of the SU node density or the average number of nodes within the neighborhood of an SU node. Unlike existing protocols, OMCMAC effectively addresses the spectrum utilization issue using simple nodes. It ensures no collision among SUs during data transmission phase to improve spectrum utilization. For example, HC-MAC [48] suffers from significant spectrum underutilization because it allows only one Tx-Rx pair to transmit data at any time instant even though some channels remain unused. On the other hand, SYNMAC tries to reduce collision among SUs during data transmission phase at the cost of node complexity as it requires two transceivers per node. In contrast, OMC-MAC avoids such node complexity and can still deliver excellent spectrum utilization–as shown in our presented numerical results. The mechanism used in OMC-MAC to combat MHTP is simpler and more efficient compared to those in other protocols such as HC-MAC, SYN-MAC, CMAC, and OSA-MAC. OSA-MAC uses IEEE 802.11 CSMA/CA based four way handshake messages: RTS/CTS/DATA/ACK to eliminate hidden terminal problem during data transmission. Use of IEEE 802.11 mechanism on data channels is not bandwidth efficient. SYN-MAC relies on more complex nodes with two transceivers while C-MAC proposes an algorithm that has high implementation complexity. Although HC-MAC eliminates the problem, but it does so at the cost of significant wastage of spectrum opportunity. Moreover, CCC is always known and available to SUs in OMC-MAC which provides a reliable medium for initial 45  message exchange among SUs. Therefore, the network initialization, reconfiguration and coordination is easier and reliable in OMC-MAC.  46  Chapter 3 Enhancement of OMC-MAC and QoS Provisioning in MAC Design for Distributed Cognitive Radio Networks1 3.1  Introduction  In Chapter 2, we proposed an opportunistic MAC protocol called OMC-MAC to address some of the major challenges in distributed CR networks (DCRNs). In this chapter, we first propose an enhancement for negotiation and data transmission phases of OMC-MAC to increase the achievable thoughput with OMC-MAC in operation. We then incorporate a QoS mechanism in the OMC-MAC in order to address QoS requirement of delay sensitive applications. We published the work described in this chapter as [63]. 1 The paper based on the research work presented in this chapter has been published as: Jha, S.C., U. Phuyal, M.M. Rashid and V.K. Bhargava, “Design of OMC-MAC: An Opportunistic Multi-channel MAC with QoS Provisioning for Distributed Cognitive Radio Networks,” IEEE Trans. on Wireless Commun., vol. PP, no. 99, pp. 1-12, Aug. 2011.  47  The highly dynamic system environment in terms of resource availability and lack of central control unit makes the quality of service (QoS) provisioning a challenging problem in DCRNs. Utilizing opportunistically available spectrum efficiently is another crucial challenge for MAC design in such networks. Avoiding collision among SUs and with PUs, combating multi-channel hidden terminal problem (MHTP) with simple SU nodes and keeping a good coordination among SUs in heterogeneous spectrum environment are other major challenges in DCRN MAC design which have been addressed by the OMC-MAC as described in Chapter 2. Designing QoS provisioning and further improvement of system throughput by increasing the spectrum utilization are main focuses of this chapter. As MAC design for DCRN is an exciting new research area, a number of MAC protocols have been proposed in the literature [41]–[48]. Most of the DCRN MAC proposals in literature have not considered QoS provisioning issue. Only a few MAC proposals, such as opportunistic MAC (O-MAC) [44] and group based MAC (GB-MAC) [45], have included a discussion on this issue. Authors in [44] have analyzed throughput and average transmission delay for two different sensing policies with the assumption that there are two transceivers at each node. Although the authors claim that their analysis can serve as a basis for QoS provisioning, the paper lacks any detail on it. Furthermore, the assumption that a single user will transmit on all available channels at any particular time may not be true in practice due to hardware constraints of user node. This might render inefficient utilization of spectrum. On the other hand, GB-MAC [45] proposes cluster based architecture which divides users into different groups and allocates the available channels among the groups based on their bandwidth requirements. Each group is controlled by a group leader and the communication among group leaders is coordinated by one of the leaders selected as manager. Since only one pair of users can communicate at any time in a group, the channel set allocated to a group may not be fully used in GB-MAC. Authors in [45] have discussed QoS for SUs, however, GB-MAC lacks any module specifically designed to provide QoS guarantee to SUs.  48  In this chapter, we enhance opportunistic multi-channel cognitive MAC (OMCMAC) for DCRN which proposes QoS provisioning and admission control modules in addition to addressing a few other major issues. The data transmission phase and negotiation phases have been modified to boost the spectrum utilization and increase the system throughput. The effectiveness of enhanced OMC-MAC lies in the following facts: i) It connects channel sensing, negotiation and data transmission phases in an integrated way to propose a complete working MAC framework. ii) It addresses multi-channel hidden terminal problems using simple SU nodes each having a single half-duplex transceiver. We devise a smart message exchange mechanism during channel negotiation to achieve it. iii) It proposes a data transmission phase free of any collision among SUs which improves utilization of opportunistically available radio spectrum. Since radio resource is available opportunistically in CR networks, therefore, whenever it is free from PU, we should try not to waste it due to collision among SU nodes themselves. It helps to enhance the radio resource (spectrum) utilization. iv) It assures a minimal collision probability to PUs. SUs may collide with PUs during the data transmission phase as users from both networks may use the same channels during this phase. Collision occurs in two cases: (a) PUs come back on a channel being used by an SU and (b) SUs miss-detect (one kind of sensing error) the channel. OMC-MAC addresses the collision due to missdetection. To reduce collision due to miss detection, OMC-MAC allows any channel to be reserved by only one Tx–Rx SU pair. Furthermore, each of Tx and Rx senses the channel independently and hence reduces the collision with PUs because collision can occur only when both of them miss detect the channel. v) It proposes a QoS provisioning module for delay sensitive applications which 49  reduces their average delay. It also analyzes admission control criteria to guarantee a predefined QoS requirement for such applications in terms of maximum acceptable delay. The remainder of this chapter is organized as follows. We present the proposed enhancement to OMC-MAC protocol in Section 3.2. We also develop analytical methods to derive some important performance parameters of enhanced OMCMAC in Section 3.2. A QoS module is introduced in OMC-MAC to address the QoS requirements of delay sensitive applications in Section 3.3. Section 3.4 presents and discusses selected numerical and simulation results from our analysis and compares performance of OMC-MAC with those of two existing protocols. The comparison results show that OMC-MAC outperforms these existing schemes in terms of providing higher throughput and lower collision probability with PUs in case of sensing error. Furthermore, the results illustrate effectiveness of QoS module in decreasing channel access delay for delay sensitive applications. The chapter is concluded in Section 3.5 with some concluding remarks.  3.2  Enhancement of OMC-MAC to Further Increase Throughput  As discussed in Chapter 2, OMC-MAC assumes a single half-duplex transceiver per node and a dedicated global CCC owned by DCRN to exchange the control messages. The time frame is divided into periodic beacon intervals (BIs). SUs are synchronized by a periodic beacon signal. BI is further divided into three phases: sensing, contention, and data transmission phase (DTP). The data transmission starts only after the contention phase and is of a fixed duration which is same for all transmitter-receiver (Tx-Rx) pairs. In the following, we discuss an enhancement to OMC-MAC in order to increase the data transmission duration for throughput enhancement. The enhancement makes the data transmission duration of variable size and is different for various Tx-Rx pairs. Moreover, the enhancement ensures that the data transmission duration for any Tx-Rx pair is equal to at 50  least DTP of OMC-MAC before enhancement. Notation: In this chapter, unless stated otherwise, the maximum, average and minimum values of a variable x are denoted by x, ˆ x, ¯ and x, ˇ respectively. ⌊x⌋ represents the integer part of x. A list of major symbols used in this chapter are summarized in Table A.1.  3.2.1  Modified Data Transmission Scheme to Improve Throughput  In Chapter 2, we have assumed that data transmission starts only after contention phase is ended. However, once a Tx–Rx pair reserves a channel, the Tx can immediately start transmitting data on the reserved channel without affecting the operation of our protocol. Here, the assumption is that MPDU is of variable length so that Tx node can transmit one or more MPDUs until the end of DTP. The first Tx-Rx pair which successfully reserves the data channel during contention phase, therefore, gets the longest transmission opportunity, while last Tx-Rx pair succeeding to reserve the data channel gets smallest transmission opportunity. Note that the smallest transmission opportunity is greater than or equal to DTP. In such scenario, the gain in average transmission time per reserved channel in a BI can be expressed as ¯ 1 ¯ si )) = Tcon − (N + 1)Tsi . ∆TDT = ¯ ((Tcon − Tsi ) + (Tcon − 2Tsi ) + · · · + (Tcon − NT 2 N (3.1) Here, N¯ is given by (2.6) and Tsi is given by (2.4). From (2.7) and (3.1), the improved average normalized saturation throughput of the system is given by [ ( )] ∆TDT N¯ (N¯ + 1)Tsi ¯ ¯ ¯ Sim = S + N = TDT + Tcon − . TBI TBI 2  (3.2)  ¯ si ≤ Tcon < (N¯ +1)Tsi . If we choose Tcon as an integer multiple In (3.2), we have NT ¯ si . With this approximaof Tsi , then Tcon can be made approximately equal to NT  51  tion, (3.2) can be simplified as 1 S¯im = TBI  ( ) T con ¯ DT + NT (N¯ − 1) . 2  (3.3)  (3.2) and (3.3) provide expressions for new improved saturation throughput. From these expressions, one can note that S¯im increases with increase in length of contention phase. However, increase in contention phase decreases the length of data transmission phase which, on the other hand, decreases S¯im . Therefore, the length of the contention phase should be selected carefully to maximize the overall saturation throughput. In the following, we explore the optimal length of contention phase to achieve maximum S¯im .  3.2.2  Optimal Length of Contention Phase  Average length of contention phase depends on various parameters like number of SUs with data packets in the system and average number of channels available. For given Pt and N f , average time for one channel reservation, Tsi , can be calculated from (2.4). Let us analyze the effect of contention phase length on system throughput for given Nch and P. From (2.6), for small value of Tcon , increase in Tcon increases throughput by enabling more and more channels to be reserved. After reaching a maximum value, throughput starts decreasing with the further increase in Tcon because, from (2.8), for a given Tsensing , increase in Tcon decreases TDT . Assuming other parameters such as N f , Pt , Nch and P to be known, optimal Tcon can be calculated to maximize the throughput by solving the following optimization problem: 1 maximize S¯im = Tcon TBI  ( ) T con ¯ DT + NT (N¯ − 1) , 2  (3.4)  subject to Tcon ≥ 0, where N¯ is given by (2.6). The individual channel reservation takes place at an interval of Tsi , so the objective function in (3.4) decreases slightly during the time 52  between successive channel reservations. However, the envelope of the objective function is concave [64] and the throughput is maximized by such value of Tcon which is an integer multiple of Tsi , i.e., Tcon = nTsi . Since TBI is a constant, the problem (3.4) transforms to the following integer programming problem: ( ) nT nTsi si maximize f (n) = N¯ TBI − Tsensing − − , n 2 2  (3.5)  subject to n ∈ N , +  where N+ represents set of nonnegative integers. If n∗ is optimal value of n, we can say that n∗ ≤ Nch . Therefore, even extensive search-based algorithm is practical to solve this problem.⌊ For example, Algorithm 1 outlines steps to solve this ⌋ Nch +1 problem in maximum of iterations. We can start the search with initial 2 value of n = ⌊Nch /2⌋. As f (n) is concave for integer n, we can find the search direction (∆n) and continue on that direction ⌊ ⌋ until the maximum of f (n) is reached. Nch +1 Note that there may be maximum iterations where each iteration involves 2 constant computational complexity independent of Nch . Therefore, Algorithm 1 has O(Nch ) time complexity. Algorithm 1 Calculate optimum Tcon 1: 2: 3: 4: 5: 6: 7: 8: 9: 10:  n ← ⌊Nch /2⌋ if f (n + 1) > f (n) then ∆n ← 1 else ∆n ← −1 end if while f (n + ∆n) > f (n) and 0 < n < Nch do n ← n + ∆n end while ∗ ← nT return Optimal Tcon : Tcon si  53  3.3  QoS Provisioning in OMC-MAC  QoS requirements of applications differ from each other. An application may need to fulfil one or more QoS requirements such as maximum acceptable delay, minimum data rate, and maximum acceptable packet loss rate. In this section, we mainly focus on the QoS provisioning for delay sensitive applications. We refer to delay sensitive applications as prioritized applications and users with such applications as prioritized users. We call other applications (users) as non-prioritized or general applications (users). The contention phase is divided into two parts: 1) contention phase for prioritized users, and 2) contention phase for general users, as shown in Figure 3.1. We have assumed that QoS requirement of prioritized users is specified in terms of maximum acceptable end-to-end delay, while general users do not have any such restriction. We have also assumed that each prioritized user has at most one prioritized data packet at any time. Under this assumption, the end-to-end delay of prioritized user can be simply approximated as channel access delay plus TDT . Therefore, in rest of this section, we analyze the QoS performance in terms of channel access delay. In the proposed scheme, prioritized users get opportunity to reserve channels first during the first part of contention phase. If there is no more prioritized user, general users get access to CCC. This can be implemented, for example, by defining longer DIFS for general users. The CCC which is always available during data transmission phase in our proposed MAC can also be used to transmit data by prioritized users with stringent QoS requirements. Moreover, scheduler and admission control module can be developed at each SU node due to distributed architecture of the proposed system model. Each SU can have data packets from applications with various QoS requirements. An SU has full flexibility to schedule the packets from various applications waiting at its queue based on the instantaneous resource information. Similarly, admission controller at each SU can decide to accept or reject a connection request based on the QoS requirements of the connection/application and the instantaneous resource availability. We analyze the delay suffered by a prioritized data packet with the help of 54  Contention Phase for Prioritized Users  Sensing Phase  Contention Phase for Non-prioritized Users  Beacon  Data Transmission Phase  Contention Phase (Channel Negotiation and Reservation among SUs)  Beacon Interval  Figure 3.1: Timing structure of proposed OMC-MAC for QoS provisioning. state diagram as shown in Figure 3.2, where each state represents number of BIs a packet has been waiting in queue. The probability of state k is represented by Pk . Let Dˆ be the maximum acceptable channel access delay for a prioritized data packet to guarantee QoS. When a packet arrives to the transmission queue of a user, it takes the initial state (i.e., state 0). We assume that the user gets transmission opportunity in each BI with a probability Pap , and waits for next BI with the probability Pnap = 1 − Pap ,  (3.6)  n . as depicted in Figure 3.2. Therefore, we can write Pn = P0 Pnap The probability that a packet does not get transmission opportunity before reaching state Dˆ + 1 is the probability of being QoS unsatisfied, Pus , which is given by ˆ (D+1)  Pus = PDˆ Pnap = P0 Pnap  ,  (3.7)  and the probability of a packet being QoS satisfied, Psat = P0 − Pus . The average 55  Pnap  Pnap  0  1  2  Pap  Pap  Pnap  D̂  .  QoS Unsatisfied  Pap  Pap QoS Satisfied  Figure 3.2: States of data packets waiting for channel access. The state represents number of beacon intervals a packet has been waiting to get access to channel. delay of QoS satisfied packets (Ds ) in terms of number of BIs can be expressed as Dˆ  Ds =  ∑  k kP0 Pap Pnap  k=0  ) P0 Pnap ( Dˆ ˆ = 1 − Pnap (1 + Pap D) . Pap  (3.8)  The average delay experienced by general users can be expressed as ∞  Dg =  k = ∑ kP0PagPnag  k=0  P0 Pnag , Pag  (3.9)  where Pag is channel access probability for general users and Pnag = 1 − Pag . Therefore, average delay of all satisfied users (including QoS satisfied prioritized users and general users) in terms of number of BIs can be expressed as DQoS =  Psat N f p Ds + N f g Dg . Psat N f p + N f g  (3.10)  Now, let us consider the case where there is no requirement to guarantee QoS. In such a case, let Pa be the probability of channel access for any user per BI and Pna = 1 − Pa . The average channel access delay can be shown as Dno−QoS = P0 Pna /Pa . 56  (3.11)  Out of N f users who have data packets to transmit, assume N f p users have packets with QoS requirements and the rest N f g = N f − N f p are general users. In case of no QoS provisioning, N f users will use the entire contention phase of length Tcon , while in case of QoS provisioning, prioritized users use Tconp period as dedicated contention phase, and Tcong = Tcon − Tconp is used by non-prioritized users. Note that prioritized contention phase will be utilized by non-prioritized users in case there is no more prioritized users contending for channel reservation. The channel access probabilities for prioritized users, non-prioritized users, and for the case of no QoS provisioning can be approximated as Pap = N¯ p /N f p ,  (3.12)  Pag = N¯ g /N f g ,  (3.13)  ¯ f, Pa = N/N  (3.14)  and  respectively, where N¯ p and N¯ g are average number of channels that can be reserved in the contention phase of lengths Tconp and Tcong , respectively, by N f p and N f g users. N¯ can be calculated using (2.6), and N¯ p and N¯ g can be calculated by using a similar approach.  3.3.1  Minimum Length of Prioritized Contention Phase Versus QoS Satisfaction  Assuming average number of channels available Nch P is greater than N¯ p , N¯ p can be approximated as number of successful message exchanges possible in Tconp , i.e., N¯ p = Tconp /Tsip ,  57  (3.15)  where Tsip is average time taken by a prioritized user to reserve a channel and can be calculated using an approach similar to (2.4). The length of Tconp which keeps the fraction of unsatisfied users less than a pre-specified limit Pˆus can be found using the inequality: Pus ≤ Pˆus . Then, using (3.6), (3.7), (3.12), and (3.15), we get ( ) ) ˆ1 ( D+1 ˆ Tconp ≥ N f p Tsip 1 − Pus /P0 .  3.3.2  (3.16)  Admission Control  For a given Pt , using (2.2)–(2.4), Tsip can be expressed in terms of N f p as follows: (1 − Pt )(Tc − σ ) Tc + (Ts − Tc ) + N f p Pt N f p Pt (1 − Pt )(N f p −1) C A , +B+ = − Nf p N f p E (N f p −1)  Tsip = −  (3.17)  where A = (1 − Pt )(Tc − σ )Pt , B = (Ts − Tc ), C = Tc /Pt , and E = (1 − Pt ). Since the prioritized users reserve channels only during Tconp , from (3.16) and (3.17), we can write ( Tconp ≥ −A + BN f p +  )(  C E (N f p −1)  (  1 − Pˆus /P0  ) ˆ1  D+1  ) .  (3.18)  From (3.18), an upper bound for the maximum number of prioritized users that can be accepted in the system can be calculated as ⌊  1 Nˆ f p = L log E  ( ) ⌋ C log E 0, − F−1 + F , BE  (3.19)  where L (·) is the Lambert W function [65] and ( 1 F= B  A+  )  Tconp 1  ˆ 1 − (Pˆus /P0 ) D+1  58  .  (3.20)  This result can be used to design an admission control module for the system. For example, if number of prioritized users is known, a maximum arrival rate of prioritized data packet per user can be determined. A distributed admission control module at each user can accept or reject new prioritized data packet based on calculated maximum average arrival rate per user. Admission control in distributed network needs cooperation among the SUs. SUs needs to be truthful to each other. Estimating total number of prioritized users in the neighborhood may be difficult in such architecture. Note that CCC is not used for control message exchange during data transmission phase in the proposed protocol. Hence, the proposed protocol can easily be modified to utilize the CCC for data transmission during this phase. It may help SUs with stringent QoS constraints to fulfill their transmission requirement.  3.4  Selected Numerical and Simulation Results  In this section, we evaluate the performance of enhanced OMC-MAC and compare it with selected existing MAC proposals using computer simulations. For numerical and simulation analysis, we use TBI = 100 ms, data rate (R) of 1 mbps per data channel, control channel bit rate of 1 mbps, tDIFS = 128 µ s, tSIFS = 28 µ s, and σ = 50 µ s. However, the actual values of these parameters depend on the physical layer techniques used for transmission. Figure 3.3 shows that, other parameters (such as Nch , P, N f ) being fixed, with an increase in Tcon , throughput increases first, reaches a maximum and then starts decreasing. The maximum throughput is reached when Tcon becomes just enough to reserve all available channels. Further increase in Tcon negatively affects the throughput by reducing TDT while the average number of reserved channels (Nch P) remains unchanged. It is clear from Figure 3.3 that there is an optimal value of Tcon to maximize the throughput for each value of Nch , as discussed in Section 3.2.2. The maximum achievable throughput increases with the increase in Nch and P as expected. Similarly, OMC-MAC with variable TDT performs better than that 59  14 Variable T , Simulation DT  N =25 ch  Fixed T , Simulation DT  12  Variable T  DT  Fixed T  DT  Throughput (mbps)  10  8  Nch=15  6  4  N =5 ch  2  0 5  10  15  20  25  30  Contention phase length (T  35  40  45  50  ) (% of beacon interval)  con  Figure 3.3: Saturation throughput variation with contention phase length (Tcon ) and number of channels (Nch ) for P = 0.6, N f = 60, Pt = 0.0155, and Tsensing = 10 ms. with fixed TDT because the former utilizes a portion of contention phase for data transmission. From Figure 3.3, it is apparent that OMC-MAC is very efficient in utilizing the spare radio resource from PU network. For example, when Nch = 15 and P = 0.6, subtracting the unavoidable sensing overhead (assumed to be 10% of TBI ), the expected maximum data rate for SUs is Rˆ exp = 0.9RNch P = 8.1 mbps. In Figure 3.3, for the same parameter values, we see that OMC-MAC has an average throughput of about 7.3 mbps for variable TDT and 6.74 mbps for fixed TDT , which is close to Rˆ exp . 60  12 N =5, OMC Variable T ch  DT  Nch=5, OMC Fixed TDT Nch=5, HC  10  N =5, OSA ch  N =15, OMC Variable T ch  DT  Nch=15, OMC Fixed TDT  Throughput (mbps)  8  Nch=15, HC, Nch=15, OSA Nch=25, OMC Variable TDT Nch=25, OMC Fixed TDT  6  Nch=25, HC N =25, OSA ch  4  2  0 0.2  0.3  0.4  0.5  0.6  0.7  0.8  Probability of channel availability (P)  Figure 3.4: Variation of saturation throughput with probability of channel availability (P) for OMC-MAC, OSA-MAC, and HC-MAC for N f = 35, Pt = 0.0155, and Tspc = 1 ms. To validate our analytical method, we simulated proposed OMC-MAC protocol using a discrete event simulator, using same values for the system parameters as in the numerical analysis. The simulation results are also plotted in Figure 3.3, where the vertical error bars represent 95% confidence interval. As the simulation results match very closely with the numerical results, this confirms the validity of our analytical method in evaluating the performance of OMC-MAC. Note that in this Section, numerical results are obtained using MATLAB, while to validate the numerical results we have developed a simulator using C-based discrete-event  61  simulation language ‘PARSEC’ (PARallel Simulation Environment for Complex systems [66]). In Figs. 3.4, 3.5, 3.6, and 3.7, the performance of proposed OMC-MAC is compared with two other MACs existing in the literature, namely OSA-MAC [42] and HC-MAC [48]. All of these MACs assume a dedicated CCC and a single half duplex transceiver at each SU node. The time required to reserve a channel during contention based channel access is assumed to be same for each MAC with a fixed backoff window size of 128 and sensing time per channel (Tspc ) is assumed to be 1 ms. To make a fair comparison, it is assumed that OMC-MAC senses all channels during the sensing phase. For HC-MAC, we use the same assumptions as suggested in [48], such as maximum bandwidth spread = 6, maximum fragments = 2, ratio of data transmission to signaling overhead = 10, and 2-stage look-ahead for stopping criterion. For OSA-MAC, the backoff window size for DTP is 64 [42]. For simplicity, results in Figs. 3.3 and 3.4 are obtained with Pt = 0.0155 assuming a fixed backoff window size W . The actual value of Pt depends on Pcm , P, W and N f . However, for larger N f , the value of Pt mainly depends on W and the chosen value of Pt should be reasonably close to its actual value when W = 128. In Figure 3.4, comparative throughput performance of proposed MAC is presented. The optimal Tcon given by (3.4) is used for both cases of fixed and variable TDT in OMC-MAC. From the figure, it is obvious that proposed OMC-MAC outperforms the existing MACs by providing higher throughput and hence utilizing the available radio spectrum efficiently. Since HC-MAC allows only one user to transmit data even when there are available data channels not being used by that user, its throughput performance degrades with increase in number of available channels compared to the proposed MAC. In OSA-MAC, multiple users may select a channel before channel sensing; as a result, these users either go for contention during DTP if the selected channel is unoccupied by PUs, or otherwise wait for next BI. Contention free DTP in OMC-MAC enables it to utilize the opportunistically available spectrum more efficiently. Furthermore, it should be noted that, since the proposed OMC-MAC is assumed to sense all the channels,  62  −1  10  Probability of collision with PUs due to sensing error  OSA−MAC −2  10  −3  10  N =5 ch  Nch=15  −4  10  N =25 ch  OMC−MAC  −5  10  −6  10  10  15  20  25 30 35 Number of secondary flows (Nf)  40  45  50  Figure 3.5: Collision probability of SUs with PUs (Pc ) versus number of secondary flows (N f ) for P = 0.6, and Nch = 4. its total sensing period may become too long (e.g., Nch Tspc = 25 ms for Nch = 25), thereby reducing the overall throughput. Throughput performance of OMC-MAC can be further enhanced by implementing smarter sensing strategies to scan only a portion of total channels, details of which is out of the scope of this thesis. In terms of decreasing the collision probability with PUs due to sensing error, OMC-MAC is highly robust compared to OSA-MAC as depicted in Figs. 3.5, 3.6 and 3.7. In OMC-MAC, both the Tx and Rx have to make sensing error at the same time to cause collision to PUs. Therefore, the proposed protocol keeps the interference to PUs very low and hence can help to achieve QoS in PU network 63  −1  Probalility of collision with PUs due to sensing error  10  −2  10  Nf=40  −3  10  OSA−MAC  Nf=25 Nf=10  −4  10  −5  10  OMC−MAC −6  10  −7  10  1  2  3  4 5 6 7 Sensing error probability (Pe)  8  9  10 x 10  −3  Figure 3.6: Collision probability of SUs with PUs (Pc ) versus sensing error probability (Pe ) for P = 0.6, and Nch = 4. as well. Furthermore, analyzing the plots in Figs. 3.5 and 3.6, it is obvious that collision probability with PUs due to sensing error does not depend on number of SUs (or traffic flows) in the neighborhood, i.e., the performance of OMC-MAC is highly scalable in this respect. As shown in Figure 3.7, for a specified upper bound on acceptable maximum collision probability with PU (Pˆc ), OMC-MAC can tolerate significantly higher sensing error at each SU compared to OSA-MAC. This excellent performance is offered by OMC-MAC because it does not cause collision to PUs when only the Tx or Rx suffers from sensing error. We now present the performance results of proposed MAC with QoS provi64  0  10  Maximum acceptable sensing error probability  OMC−MAC  −1  10  Pˆc = 5 × 10−3 Pˆc = 3 × 10−3 −2  10  OSA−MAC  Pˆc = 10−3  −3  10  −4  10  10  15  20  25  30  35  40  45  50  Number of secondary flows (Nf)  Figure 3.7: Maximum acceptable sensing error probability (Pˆe ) for given upper bound on collision with PU (Pˆc ) versus number of secondary flows (N f ) for P = 0.6 and Nch = 4.  65  3  10  General Users, QoS Case  P=0.5  Average delay in terms of beacon intervals  P=0.6 P=0.7 2  10  1  10  No QoS Case 0  10  Prioritized Users, QoS Case 5  10  15 Number of channels (N )  20  25  ch  Figure 3.8: Average channel access delay versus number of channels (Nch ) for Tcon = 20 ms, N f = 30, Dˆ = 4, N f p = 0.3N f , Tconp = 0.3Tcon , and P0 = 1. sioning. Figure 3.8 depicts the improvement of channel access delay of prioritized users after defining a prioritized contention phase. In most of the cases, average channel access delay of QoS satisfied prioritized users remains less than one BI. The ‘No QoS case’ indicates the average delay of users when all users get the same priority. It serves as a reference to evaluate the delay improvement of prioritized users after implementing QoS module. However, the delay of general users increases after implementing QoS provisioning in the system particularly when total number of available channels (Nch P) is lower. Our proposed scheme efficiently decreases the transmission delay of delay sensitive applications with-  66  20 18  Percentage of QoS unsatisfied users  16  P=0.5 P=0.6 P=0.7  14 12 10 8 6 4 2 0 5  10  15 Number of channels (Nch)  20  25  Figure 3.9: Percentage of QoS unsatisfied users versus number of channels (Nch ) for Tcon = 20 ms, N f = 30, Dˆ = 4, N f p = 0.3N f , Tconp = 0.3Tcon , and P0 = 1. out much increase in delay of general users when there is higher number of data channels or higher probability of data channels being available. Figure 3.9 shows the percentage of prioritized users whose QoS requirements could not be satisfied. It is obvious from the figure that proposed QoS scheme satisfies the QoS requirements of more than 98% users when the number of available channels is reasonably high. Variation of minimum length of prioritized contention phase required for specified upper bounds on acceptable percentage of unsatisfied users and acceptable delay of prioritized users is shown in Figure 3.10. Relaxation on either of these 67  Minimum Length of prioritized contention phase (ms)  10 ˆ =2 D  9  ˆ =3 D ˆ =4 D  8  7  6  5  4 1  2 3 4 5 6 7 8 9 10 ˆ Maximum acceptable percentage of unsatisfied users (Pus )  Figure 3.10: Minimum required length of prioritized contention phase versus maximum acceptable percentage of unsatisfied users (Pˆus ) for N f p = 12, P = 0.6, P0 = 1, and Nch P ≥ N¯ p . bounds decreases the length of required Tconp . When one of the two bounds Dˆ and Pˆus is stringent for a fixed Tconp , the other should be flexible. For example, when Tconp = 7 ms, Pˆus ≤ 2% can be achieved for Dˆ = 4, but not for Dˆ = 3 or 2. Figure 3.11 shows the variation of maximum number of acceptable prioritized traffic flows in the system with constraints Dˆ and Pˆus when Tconp = 10 ms. It is obvious from the graph that more simultaneous prioritized flows are acceptable when the bounds are loose. These results can be used to devise effective admission control criteria in DCRN.  68  ˆfp ) Maximum number of prioritized users (N  24 22  ˆ =2 D  20  ˆ =3 D ˆ =4 D  18 16 14 12 10 8 1 2 3 4 5 6 7 8 9 10 ˆ Maximum acceptable percentage of unsatisfied users (Pus )  Figure 3.11: Maximum number of prioritized users in the system (Nˆ f p ) versus maximum acceptable percentage of unsatisfied users (Pˆus ) for Tconp = 10 ms, P0 = 1, and Nch P ≥ N¯ p .  3.5  Concluding Remarks  In this chapter, we first proposed an enhancement to the proposed OMC-MAC protocol to further increase the system throughput. We then incorporated a QoS provisioning in OMC-MAC. The proposed QoS module in OMC-MAC ensures priority of delay sensitive applications over general users. We analyzed the proposed QoS scheme and showed that it efficiently fulfils the QoS requirements of delay sensitive applications by significantly reducing their channel access delay. We proposed a method to calculate the maximum number of prioritized applications in the system based on their QoS requirements which can be used to develop admission control modules. The numerical results further revealed that the en69  hanced OMC-MAC performs better than existing protocols in providing higher throughput and keeping the collision with PUs lower in case of erroneous channel sensing.  70  Chapter 4 Resource Optimization in Multi-hop Distributed Networks1 4.1  Introduction  In multi-hop distributed networks, link layer resource allocation must consider the information about number of hops packets have already traveled in the network in order to optimize the overall network resources utilization. The loss of a packet after traveling some hops results in the wastage of all resources allocated to it in previous hops. The classical resource allocation schemes at link layer which emphasize single hop performance optimization may not provide optimal resource utilization in multi-hop distributed networks. Therefore, in this chapter, we explore schemes to allocate radio resources to different packets favoring those packets which have traveled more hops before reaching a particular node. We propose cross-layer approaches in which link layer gets the hop-count informa1 The  papers based on the research work presented in this chapter have been published as:Jha, S.C., U. Phuyal and V.K. Bhargava, “Cross-Layer Resource Allocation Approach for Multi-hop Distributed Cognitive Radio Network,” in Proceedings of the CWIT’11, Kelowna, BC, Canada, pp. 211-215, May 2011 and Jha, S.C., U. Phuyal and V.K. Bhargava, “Joint Power and Subcarrier Allocation in Multi-hop OFDMA Network: A Cross-Layer Approach,” in Proceedings of the AHICI2011, Kathmandu, Nepal, pp. 1-6, Nov. 2011.  71  tion from the network layer module. Distributed implementation is possible with the proposed schemes because link layer of each node can access this information from higher layer. In the following, we first explore effectiveness of hop-count based resource allocation schemes in multi-hop distributed cognitive radio networks. We then analyze the similar resource allocation schemes for multi-hop orthogonal frequency division multiple access networks.  4.2  Resource Optimization in Multi-hop Distributed Cognitive Radio Networks  Most of the resource allocation proposals for CR networks in literature optimize the power allocation focusing on single hop performance [67][68]. However, these schemes may not be well suited to optimize the performance of a multi-hop DCRN (MHDCRN). The fact that resources are allocated at each hop to a packet which travels multiple hops before reaching the final destination in MHDCRNs is crucial in evaluating the overall utilization of resources in these networks. We refer to the number of hops a packet traveled before reaching a node under consideration as hop-count. Loss of a packet with higher hop-count results in wastage of all resources it has used in previous hops. Therefore, this information must be taken into account during resource allocation at a particular node to maximize the resource utilization in the network. Only a little work has been done in the area of resource allocation in MHDCRNs (e.g., [69][70]). Authors in [69] propose and analyze joint route selection and resource allocation scheme for OFDMA-based multi-hop CR network. They have proposed an iterative scheme to optimize transmit power and allocated bandwidth jointly with route selection. A distributed spectrum allocation and power control strategy has been developed considering inter-flow interferences and the cumulative interference temperature in [70]. Authors in [70] have adopted the TDM-based power control strategy to maximize end-to-end throughput by choosing multi-hop route and spectrum combination appropriately for each flow. However, both of these proposals mainly focus on route selection. To the best of our 72  knowledge, consideration of hop-count in power allocation strategy has not been efficiently addressed in literature so far. Note that our focus in this chapter is on link layer (MAC layer) resource allocation, however, dynamic spectrum environment also brings research challenges at network layer routing in the multi-hop DCRNs. The research conducted so far indicates that it is very difficult to propose a general routing solution for multihop cognitive environment as the cognitive environment differs from network to network depending on the PUs activities [25]. For example, if the PU channels, once available, remain available for very long duration (e.g., for several communication durations), the MHDCRNs are then very similar to multichannel multi-hop mesh networks regarding routing problems. Routing solutions proposed for multi-hop mesh networks are applicable to such static MHDCRNs. However, if the spectrum availability is more dynamic, CR-specific routing approaches should be used to keep the performance acceptable. In some cases, the PU channel availability are so dynamic that they are on average available for a shorter duration than communication duration. Since the path cannot be considered to be stable for a flow duration, such networks may require per packet routing solutions. Opportunistic packet forwarding based on instantaneous channel information may be preferable to replace the end-to-end routing approaches in highly dynamic environment. Various routing approaches have been explored in the literature [18]–[20][71]–[77]. Although the focus of this thesis is not on routing protocol, in the following we briefly discuss various routing protocol in literature for MHDCRN. For detail understanding about routing in MHDCRN, interested readers are directed to the provided references. Dynamic routing protocols have been proposed in the literature to combat the routing issues inherent in MHDCRNs. Authors in [76] have proposed a routing scheme based on a probabilistic metric which takes into account the behavior of PUs. The link cost is determined based on interference from PUs, occupancy rates of PUs on the channels observed on that link and received signal strength. The most probable path to destination is then selected by running a Dijkstra-like algo-  73  rithm. A spectrum aware mesh routing (SAMER) has been proposed in [18] which forms a forwarding mesh around the long-term shortest path and adjusts it periodically according to spectrum dynamics in distributed multi-hop CR Networks. It computes all the paths of at most H hops (H is set higher than shortest path length) and from these paths, it selects the one with the highest spectrum availability to consider short-term performance gain. It tries to balance between long-term route stability and short-term route performance. Building runtime mesh and its periodic updating can be expensive in DCRNs as they lack common broadcasting medium because of heterogeneous spectrum availability. Some other spectrum aware routing protocols have been proposed in literature [20][71][72]. The operations of all these protocols depend on a CCC and two transceivers at each node. On the other hand, the spectrum aware routing proposed in [73] uses a single transceiver with no CCC. It sends route request over all channels to the destination. The destination then selects spectrum for the shortest path based on analytical estimates of various parameters like spectrum switching time and channel contention. It does not use a CCC at the cost of broadcasting route request on all channels, which may result in very high overhead. All these spectrum aware protocols [20][71]–[73] find route on-demand by flooding network with route request. Therefore, route discovery and recovery may have high latency and overhead. To avoid the drawbacks of on-demand routing solutions, Xia et al. have proposed a reinforcement learning based routing in [74]. It is inspired by existing routing algorithms proposed for wired packet-switched networks: QRouting [78] and Dual Reinforcement Learning [79]. Each node stores a table of Q values which show the number of available channels on the routes to various destinations. The Q values are continuously updated while routing the packets to different destinations. To find out the common channels with a neighbor, a node transmits RTS on all of its available channels in succession. Reinforcement learning based routing has also been proposed in [19]. In this thesis, while dealing with MHDCRNs, we assume that a routing protocol is in operation to decide about the next hop node for a packet being pro-  74  S1  N3  N2 N1  D3 S2  N0 N4  D2  N6 S3  D1  N5  Figure 4.1: A typical scenario of multi-hop distributed cognitive radio network. cessed at any node. The processing node, therefore, knows the next hop of the packet which leads forwarding packet toward its final destination. We propose a cross-layer power allocation scheme for MHDCRNs which favors packets that have already traveled higher number of hops. The work described in this section has been published as [80]. The main contributions of this work lie in the facts that the proposed scheme i) minimizes the wastage of resources a packet has already consumed in previous hops by providing higher priority to packets based on hop-counts, and ii) guarantees QoS in terms of minimum signal to interferenceplus-noise ratio (SINR) to each packet. The rest of this section is organized as follows. Subsection 4.2.1 presents system model and discusses various assumptions for the proposed model. Problem formulation and the proposed schemes are presented in Subsection 4.2.2. Then, in Subsection 4.2.3, we provide selected simulation results to evaluate the pro-  75  posed scheme. Finally, the section is concluded with some concluding remarks in Subsection 4.2.4. Notation: In this section, we represent column vectors by boldface alphabets and the corresponding regular alphabet represents its element. The column vector with unit elements is represented by 1. The transpose of a vector x is represented by x T and ≽ represents element-wise inequality.  4.2.1  System Model  We consider a multi-hop ad hoc CR network with distributed architecture, where each node is capable of transmitting data on multiple channels simultaneously. This assumption is valid under different scenarios, for example when each user node has multiple full-duplex transceivers as in [69], or when each user has single half-duplex or full-duplex transceiver capable of implementing discontinuous OFDM [81] as discussed in [48] and the references therein or non-continuous OFDM as discussed in [82]. The routing protocol at network layer selects the end-to-end route between source–destination for any data packet. Therefore, there is a defined multi-hop route from source to destination for each data packet in the network based on the routing protocol implemented in the network. We have not assumed any specific routing protocol; performance of the proposed scheme is independent of implemented routing protocols. Figure 4.1 shows a typical MHDCRN at a particular time instant. It depicts the end-to-end routes for three packets between source–destination pairs S1 –D1 , S2 –D2 , and S3 –D3 , all of which pass through node N0 . Without loss of generality, the packets travelling in the network through the node under consideration N0 is depicted in Figure 4.2. Here we study a resource allocation scenario at N0 which receives data packets originated from K sources S1 , . . . , SK routed towards K destinations D1 , . . . , DK . For simplicity, let us represent the set of connection indices by K = {1, . . . , K}. PHk and NHk respectively represent the previous and next hop nodes with respect to the node N0 for data packet from source k. We assume that hop-counts for all packets reaching N0 are known to the resource 76  S1  D1 NH1  PH1 S2  D2 PH2  SK  N0  PHK  NH2  NHK  DK  Figure 4.2: Simplified diagram of a multi-hop DCRN showing connections to and from the node under consideration (N0 ). allocator at N0 , and are equal to wk ∀k ∈ K . For example, w1 = 3, w2 = 1, and w3 = 2 for the scenario shown in Figure 4.1. Note that hop-count information is usually available in the packet header at network layer and can be shared to resource allocator at link layer by a cross-layer cooperation between network and MAC layer modules. The node N0 reserves a channel for each packet. We assume that a MAC protocol is in operation which coordinates and controls the channel sensing, and channel selection and reservation in the network. A survey of various MAC protocols for distributed multi-hop CR networks can be found in [40]. Typically, the MAC protocol allocates a periodic sensing phase in which each node senses many channels and prepares a list of available channels, i.e., channels currently not occupied by PUs. After that each transmitter–receiver pair reserves channels for upcoming transmission. We assume that N0 has the knowledge of channel gains for all common channels between N0 and NHk for all k. Also note that N0 is the only transmitter and NHk , ∀k are receivers. Therefore, N0 decides which channel to use for each packet. Let AN0 and ANHk denote the channel sets available at nodes N0 and NHk as  77  determined by the channel sensing process. For simplicity, we assume that N0 assigns the best available channel to the packet that has already traveled maximum number of hops starting from the one with the highest hop-count. The best available channel between N0 and NHk is given by ck =  arg max i∈{AN0 ∩ANHk ∩A¯R }  |hik |2 ,  where AR represents the set of channels reserved for other packets in previous steps and hik is the channel coefficient for channel i on the link N0 –NHk . Now, ck is added to the set AR before repeating the above steps for next packet. Our focus in this section is to distribute power among packets after channel reservation and before upcoming transmission at node N0 . Let hk denote channel coefficient of the selected N0 –NHk channel. The received SINR at node NHk can be written as  γk = αk Pk ,  (4.1)  where αk = |hσk2| , Pk is the transmit power for kth packet, and σk2 is the variance of k noise at the kth receiver which includes thermal noise as well as interference from PU transmitters. 2  4.2.2  Hop-count-based Power Allocation Scheme  Other parameters remaining unchanged, the achievable data rate is monotonically increasing function of transmit power. Let us define a user perceived utility function R(·) as a concave monotonically increasing function of transmit powers PK = [P1 , . . . , PK ]T . Our goal is to obtain optimal power allocation so as to maximize such a function. In the later part of this section, we describe two utility functions, one of which is the proposed utility function in order to prioritize the packet that has already traveled more hops, while the other is existing classical function independent of the hop-counts.  78  The problem of optimizing the available transmission power to maximize the utility while satisfying minimum QoS guarantee for transmitted packets can be written as P1: R(PK , xK )  max  PK ,xxK  γk ≥ xk γk,min ,  subject to  ∀k ∈ K  (4.2)  1T PK ≤ Pt , PK ≽ 0, xk ∈ {0, 1},  ∀k ∈ K ,  where x K is an indicator vector whose element xk is 1 when kth packet is selected for transmission and 0 otherwise, γk is the SINR of received signal at NHk , and γk,min is the minimum SINR required by kth packet for QoS satisfaction. For simplicity, we assume that γk,min = γmin for all packets. Note that the utility function R is usually chosen to be a non-linear function of allocated powers. For example, it contains logarithmic terms when our goal is to maximize the achievable sum rate. Therefore, P1 is a mixed integer non-linear programming problem, for which an analytical solution is not straightforward. In the following, we present a two-step iterative approach to find optimal solution of P1. Before going to the solution, let us define a simplified problem for a set of users S chosen from K , (i.e., S ⊆ K ) relaxing constraint (4.2) and setting indicator variables to 1, as P2: RS (PS ) = R(PS , 1)  max PS  1T PS ≤ Pt ,  subject to  (4.3)  PS ≽ 0, where vector PS is formed by elements Pk ∀k ∈ S . The steps to solve problem P1 iteratively are outlined in Algorithm 2. The  79  algorithm initially starts with S = K . Problem P2 is solved for S . In the next step, we check constraint (4.2) for each packet. If one or more packets do not satisfy constraint (4.2), the packet with worst next-hop SINR will be excluded from S . Note that such case arises because it may not be possible to achieve SINRs above threshold for all packets within given power budget. The above steps are repeated until either all packets in S satisfy constraint (4.2) or there is no packet for next iteration of power allocation. The packets which are not allocated any power during this process due to limited power budget contribute to the system outage. Algorithm 2 Optimal Solution of P1 1: Given K , Pt , γmin 2: Set S ← K 3: while |S | ≥ 1 do 4: Solve P2 to obtain P˜k ∀k ∈ S . 5: Calculate γk ∀k ∈ S using (4.1). 6: if min γk < γmin then k∈S  7:  l ← arg min γk k∈S  8: S ← S \{l} 9: xl∗ ← 0, Pl∗ ← 0 10: else 11: xk∗ ← 1, Pk∗ ← P˜k 12: Break while loop 13: end if 14: end while 15: return P∗K , x ∗K  ∀k ∈ S  In the following, we define different utility functions and present corresponding analytical solutions of P2, each of which is required in Step 4 of Algorithm 2 for each case.  80  Proposed hop-count-based utility function We formulate a utility function in order to maximize the overall system capacity with higher priority on next-hop SINR for data packets with higher hop-counts. Hop-count-based weight is applied to SINR of each packet by defining the utility function as R1 (PK , x K ) = ∑ xk log (1 +Wk αk Pk ) , (4.4) k∈K  where Wk represents the respective weight given to each packet. This weight can be defined in various ways. We define Wk as Wk = β min(wk , wmax ), where scaling factor β is a design parameter which can take only positive values. To combat false hop-information from adversary nodes and reduce the effect of unintended routing loop, the value of wk is capped by wmax . It is worthwhile to mention that for the packets originated at N0 , the node transmits it with the highest priority. This may be implemented by: i) allocating required power for such packet to ensure QoS before running allocation algorithm for rest of the packets, or ii) by setting wk = wmax for such packet. We present the solution of P2 in the following theorem. Theorem 1 The optimal power allocation solution to problem P2 when R = R1 is given by ( ) 1 1 ˜ Pk = max 0, − , ∀k ∈ S , (4.5) ν˜ Wk αk where ν˜ is chosen such that 1T P˜ S = Pt . Proof: For a set of users S , the objective function for P2 in this case can be written as R1S (PS ) = ∑ log (1 +Wk αk Pk ) , k∈S  81  Since R1S (PS ) is concave on the optimization domain, P2 is a convex optimization problem. We obtain its analytical solution by using KKT-conditions. The Lagrangian of the problem can be written as [64] ( ) L (PS , ν , λ ) = −R1S (PS ) + ν 1T PS − Pt − λ T PS , where ν and λ = [λk ∀k ∈ S ]T are known as Lagrange multipliers. Representing ˜ the KKT conditions are the value of a variable at optimal solution point by (·), given by 1T P˜ S ≤ Pt ,  −  ∂ R1S ∂ Pk  Pk =P˜k  P˜ S ≽ 0,  (4.6)  ν˜ ≥ 0, λ˜ ≽ 0, ( T ) ν˜ 1 P˜ S − Pt = 0,  (4.7) (4.8)  λ˜ k P˜k = 0,  ∀k ∈ S ,  (4.9)  + ν˜ − λ˜ k = 0,  ∀k ∈ S .  (4.10)  After some simplification, (4.10) can be re-written as  ν˜ = λ˜ k +  Wk αk , 1 +Wk αk P˜k  ∀k ∈ S ,  (4.11)  and using (4.7), this can further be reduced to  ν˜ ≥  Wk αk , 1 +Wk αk P˜k  ∀k ∈ S .  (4.12)  Therefore, ν˜ > 0, and using (4.8), 1T P˜ S = Pt ,  82  (4.13)  i.e., constraint (4.3) becomes tight at the optimal point. Let us first consider the case when ν˜ ≤ αkWk . In this case, inequality (4.12) holds only for P˜k ̸= 0. Therefore, from (4.9), λ˜ k = 0. Using this on (4.11), we can write 1 1 P˜k = − , ∀k ∈ S . (4.14) ν˜ Wk αk Note that value of P˜k obtained from (4.14) is always nonnegative satisfying (4.6). Now, consider the case when ν˜ > αkWk . Under this condition, from equation (4.11), λ˜ k > 0. From (4.9), this means P˜k = 0 for this condition. Combining both the results, we obtain { P˜k =  1 ν˜  if ν˜ ≤ αkWk  − Wk1αk  0  otherwise,  which can be re-written as (4.5). Finally, as the calculated values of P˜k depend on ν˜ , we choose ν˜ such that (4.13) is satisfied. The power allocation scheme with utilization function R1 is referred to as proposed hop-count-based power allocation (HCB-PA) scheme in rest of the Section 4.2. Utility function independent of hop-count We also analyze the system performance when the utility function is a classical one which is independent of hop-counts and compare the performance in this case with that of the proposed HCB-PA scheme. For this case, the utility function R2 may be obtained by setting Wk = 1 ∀k in R1 given by (4.4), i.e., R2 (PK , x K ) =  ∑  xk log (1 + αk Pk ) .  k∈K  For this case, the solution of P2 may be obtained from (4.5) by setting Wk = 1 ∀k. In rest of the Section 4.2, the power allocation scheme with utility function R2 is referred to as classical hop-count independent power allocation (HCI-PA) scheme. 83  4.2.3  Simulation Results  Selected results obtained via computer simulations are presented in this subsection to analyze the performance of the proposed resource allocation scheme. We assume that the nodes are distributed uniformly throughout the area under consideration. Therefore, without loss of generality we normalize the distance between the nodes such that average distance is unity. The channel coefficients between the nodes are modeled as zero-mean Rayleigh distributed complex random variables. Moreover, since all the channels in such a case undergo similar path-loss and fading, we normalize the variance of all channels to unity. Usually, there is no line of sight signal path between nodes in most of the modern wireless communication networks. Therefore, the assumption of Rayleigh faded channel coefficients between the nodes should be reasonable for numerical analysis. Furthermore, the results presented in this section are comparative results obtained for the classical and the proposed schemes, therefore the assumption on channel coefficient should not affect the nature of final conclusions. We also assume noise variance on each channel to be unity, i.e., σk2 = 1 ∀k. Since channel coefficients change dynamically, we have taken each data point in the graphs as an average over 10, 000 independent channel realizations. The node under consideration N0 may have multiple available channels common with next hop of each of the packets. Channel assignment is done following the strategy described in Subsection 4.2.1. For all the simulations, we assume wmax = 6, γmin = 1, and network diameter (i.e., maximum length of end-to-end route) = 10. Performance metrics We compare the performance of the proposed HCB-PA with a classical HCI-PA scheme. We have selected three performance metrics: one-hop throughput, resource wastage in the network, and outage at the node under consideration for performance evaluation. One-hop throughput represents the throughput achievable within the power budget Pt considering next hop (with respect to N0 ) per84  45 40  Proposed HCB−PA Classical HCI−PA  Resource wastage (units)  35 30 P = 5 units t  25 Pt = 3 units 20 15 10 5 Pt = 7 units 0 2  4  6  8 10 Number of packets  12  14  16  Figure 4.3: Resource wastage versus number of packets K (β = 1) for different values of power budget Pt . formance only. In a multi-hop network, resources used by a packet in previous hops may be assumed to be proportional to the hop-count. Therefore, if a scheme fails to guarantee the required next-hop SINR to a packet with hop-count wk , we consider a resource wastage of wk units for the entire network. This assumption is reasonable when no retransmission is done by link layer. Outage represents the fraction of total number of packets which are not allocated sufficient power by a scheme to achieve a QoS in terms of predefined SINR threshold.  85  16 Pt = 3 units  Resource wastage (units)  14  12  10  Proposed HCB−PA Classical HCI−PA  P = 5 units t  8  6 P = 7 units t  4  2 0.2  0.4  0.6  0.8  1  β  1.2  1.4  1.6  1.8  2  Figure 4.4: Resource wastage versus β (K = 8) for different values of power budget Pt . Selected results In Figs. 4.3 and 4.4, we compare the resource wastage in network with the proposed HCB-PA and the classical HCI-PA schemes in operation. It can be seen from Figure 4.3 that the proposed scheme outperforms HCI-PA for all values of available power budget. This is achieved because the proposed scheme favors packets with higher hop-counts. Performance of the proposed scheme improves with increase in the number of packets to be transmitted. With an increase in value of design parameter β , the packets with higher hop-counts are favored more. Owing to this, resource wastage is further decreased by the proposed scheme for  86  0.7 Pt = 3 units Proposed HCB−PA Classical HCI−PA  0.6  One−hop outage  0.5 Pt = 5 units  0.4  0.3  0.2 Pt = 7 units 0.1  0 2  4  6  8 10 Number of packets  12  14  16  Figure 4.5: Outage versus number of packets K (β = 1) for different values of power budget Pt . higher value of β , which is evident from Figure 4.4. Figs. 4.5 and 4.6 depict the one-hop outage performance of the proposed HCB-PA scheme compared to the classical HCI-PA scheme. It is clear that the proposed scheme decreases the outage probability efficiently. The outage performance of the proposed scheme improves with higher number of packets as well as higher value of β . This is because, using the classical scheme, packets with lower channel gains get lower transmit power which may not be sufficient to guarantee the predefined QoS. In contrary, some of the packets may get sufficient power by the proposed power allocation if the packets have higher Wk .  87  0.6 Pt = 3 units  0.55 0.5  One−hop outage  0.45 Pt = 5 units  0.4  Proposed HCB−PA Classical HCI−PA  0.35 0.3 0.25  Pt = 7 units  0.2 0.15 0.1 0.2  0.4  0.6  0.8  1  β  1.2  1.4  1.6  1.8  2  Figure 4.6: Outage versus β (K = 8) for different values of power budget Pt . Next, in Figs. 4.7 and 4.8, we present the throughput performance of both schemes. We observe that the throughput increases with increase in power budget and number of packets for both schemes. It is obvious that throughput is independent of β for HCI-PA. We also observe that for very low β , throughput is slightly better with HCI-PA, but an increase in β provides opportunity for HCB-PA in outperforming HCI-PA. Even though there is not much difference between the throughput obtained by either schemes, HCB-PA throughput is slightly higher in most of the cases. This is because of the lower outage offered by the proposed HCB-PA as discussed above. When value of β is 1 or more, the proposed HCB-PA scheme outperforms the 88  16 Proposed HCB−PA Classical HCI−PA  One−hop throughput (bit/s/Hz)  14  12  P = 7 units t  10 Pt = 5 units 8  6 P = 3 units t  4  2 2  4  6  8 10 Number of packets  12  14  16  Figure 4.7: Throughput versus number of packets K (β = 1) for different values of power budget Pt . classical HCI-PA in all respects, i.e., throughput, resource wastage and outage. It is apparent that the proposed scheme is able to decrease the outage and resource wastage in the network without decreasing the one-hop throughput. Therefore, the proposed scheme can improve the network-wide resource utilization while improving one-hop performance at the same time.  4.2.4  Concluding Remarks  In Section 4.2, we formulated a power allocation problem based on cross-layer approach for a multi-hop distributed cognitive radio network in order to decrease 89  12 P = 7 units t  One−hop throughput (bit/s/Hz)  11 Proposed HCB−PA Classical HCI−PA  10 Pt = 5 units 9  8  7 P = 3 units t  6 0.2  0.4  0.6  0.8  1  β  1.2  1.4  1.6  1.8  2  Figure 4.8: Throughput versus β (K = 8) for different values of power budget Pt . wastage of network resources. The proposed scheme favors the data packets which have already used higher amount of resources in previous hops. We considered the number of hops traveled by a packet as a basis to estimate the amount of network resources already invested on the packet. A design parameter was also introduced which decides the priority level for packets with higher hop-counts. Simulation results showed that the proposed scheme reduces resource wastage and QoS outage in the network without any decrease in system throughput.  90  4.3  Joint Power and Subcarrier Allocation in Multi-hop OFDMA Network  Orthogonal frequency division multiplexing (OFDM) has been identified as the transmission technology for next generation wireless systems due to its key advantages over other widely used wireless access techniques such as time division multiple access (TDMA), frequency division multiple access (FDMA) and code division multiple access (CDMA) [83][84]. The main advantage of OFDM is that the wireless channel is divided into many narrow-band, low-rate, frequencynonselective subcarriers so that multiple symbols can be transmitted in parallel while maintaining a high spectral efficiency. In orthogonal frequency division multiple access (OFDMA), each subcarrier may carry data intended for a different user. This results in a simple multiple access scheme called OFDMA which also enables the transmission of multiple media with varying quality-of-service (QoS) requirements within the same radio link by efficient allocation of radio resources (power and subcarriers). Different modulation and/or coding scheme may be employed for different subcarriers and therefore for different users depending upon their instantaneous channel gain of the particular subcarrier. Besides the implementation flexibility, the low complexity required in transmission and reception, robustness against frequency-selective fading channels (using OFDM with channel coding) and robustness against the inter-symbol interference (ISI) (using a cyclic prefix) present OFDM as a highly attractive candidate for high datarate communications over time-varying frequency-selective wireless environment [83]. Recently, OFDMA has emerged as one of the prime multiple access techniques to enhance data rates in future wireless systems operating in frequencyselective fading environments [84]–[86]. The main advantage of OFDMA lies in its capability to exploit the multiuser diversity through subcarrier allocation [86][87]. Compared to centralized architecture based multi-hop wireless networks, a distributed one can be more practical choice due to its ability of easier and faster deployment, lower system complexity and lower cost of implementation 91  [40][63]. Therefore, in this section, we focus on distributed multi-hop OFDMA (MHOFDMA) wireless networks. The use of OFDMA introduces the issue of subcarrier allocation to various packets/users and also the power allocation to multiple orthogonal frequency subcarriers used in the transmission. These resource allocation (RA) issues have been well addressed in the literature [86][88] and the references therein. However, most of the work in literature focuses on RA schemes optimizing single hop performance. The link layer resource allocators at a node in these schemes assume all packets to be identical and ignore the fact that various packets may have traveled different number of hops before arriving that particular node. Therefore, most of the existing schemes may not be well suited to optimize performance of a MHOFDMA network. As discussed in Section 4.2, losing a packet with higher hop-count causes more resource wastage in such networks as resources need to be allocated at each hop to a packet. Therefore, this information must be taken into account during resource allocation at a particular node to maximize the resource utilization. In the following section, we explore hop-count based resource allocation schemes for Multi-hop OFDMA Networks. Only a little work has been done in the area of RA in MHOFDMA networks (e.g., [69][89]–[92]). However, none of them have considered the hop-count in their RA schemes. For example, authors in [69] propose and analyze joint route selection and resource allocation scheme for an OFDMA-based multi-hop cognitive radio network. They have proposed an iterative scheme to optimize transmit power and allocated bandwidth jointly with route selection. However, authors mainly focus on route selection. An RA scheme is proposed in [89] to maximize end-to-end transmission data rate in an OFDM based linear multi-hop network. The proposed scheme allocates power on each subcarrier over each hop and transmission time for each hop at the beginning of each frame to ensure equal instantaneous transmission rate over all hops. However, the scheme has impractical assumptions such as one dimensional  92  linear network structure and only one packet transmission at any time. Schedular knows the global channel state information in advance, and channel state on each subcarrier remains invariant during the entire time frame. Also the authors in [89] have optimized the resource utilization for only one packet transmission in the network at any time. Similarly, authors in [92] maximize the minimum throughput among all nodes under the routing and physical layer constrains for a cooperative relaying-enabled MHOFDMA network. The proposed scheme requires a centralized network controller and therefore, may not be suitable for distributed MHOFDMA networks. To the best of our knowledge, consideration of hop-count for resource allocation in MHOFDMA has not been efficiently addressed in literature so far. In this Section, we propose a cross-layer power and subcarrier allocation scheme for MHOFDMA which favors packets having higher hop-counts. We published the work described in this section as [93]. The proposed scheme i) minimizes the wastage of resources a packet has already consumed in previous hops in MHOFDMA by providing higher priority to packet based on hop-counts, and ii) addresses joint allocation of subcarriers and power to optimize the system performance. The rest of the Section 4.3 is organized as follows. Subsection 4.3.1 presents system model and discusses various assumptions for the proposed model. Problem formulation and the proposed scheme are presented in Subsection 4.3.2. Then, in Subsection 4.3.3 we provide selected simulation results to evaluate the proposed scheme. Finally, the section is concluded in Subsection 4.3.4. Notation: In this Section, we represent matrices and vectors by boldface alphabets and the corresponding regular alphabet represents their element. The column vector with unit elements is represented by 1. The transpose of a vector x (or matrix X) is represented by x T (or XT ) whereas ≽ and ≼ represent element-wise inequalities.  93  4.3.1  System Model  We consider a multi-hop ad hoc OFDMA network with a distributed architecture. We assume that the available bandwidth is divided into N frequency-flat subcarriers so that the broadband communication between nodes takes place using OFDM technique. Let N = {1, . . . , N} denotes the set of subcarriers. Therefore, at link layer a node can transmit multiple data packets to different nodes simultaneously by assigning one or more subcarriers to each of them. The routing protocol at network layer selects the end-to-end route between source–destination for any data packet. Therefore, there is a defined multi-hop route from source to destination for each data packet in the network based on the implemented routing protocol. We have not assumed any specific routing protocol; performance of the proposed scheme is independent of implemented routing protocols. Consider a typical distributed MHOFDMA network at a particular time instant when there are three packet transmission sessions between source–destination pairs S1 –D1 , S2 –D2 , and S3 –D3 . Also assume that the end-to-end routes between these pairs pass through a node N0 . Without loss of generality, the packets routed in the network through the node under consideration (N0 ) are depicted in Figure 4.9. Here we study a resource allocation scenario at N0 which receives data packets originated from K sources S1 , . . . , SK routed towards K destinations D1 , . . . , DK . For simplicity, let us represent the set of connection indices by K = {1, . . . , K}. PHk and NHk respectively represent the previous and next hop nodes with respect to the node N0 for data packet from source k. Let φk ∀k ∈ K represent hop-counts for all packets reaching N0 which are known to the resource allocator at N0 . Note that hop-count information is usually available in the packet header at network layer and can be shared to resource allocator at link layer by a cross-layer cooperation between network and MAC layer modules. We assume that N0 has the knowledge of channel gains for all subcarriers between N0 and NHk for all k. Let hn,k denote channel coefficient of N0 –NHk link  94  S1  Previous Hops  Next Hops D1  ϕ1  S2  D2  ϕ2 ϕK  N0 DK  SK  Figure 4.9: Simplified diagram of a multi-hop OFDMA network showing packet transmissions through the node under consideration (N0 ). on subcarrier n. The received SNR at node NHk can be written as  γk = αn,k Pn,k , where αn,k =  |hn,k |2 , σk2  (4.15)  Pn,k is the transmit power for kth packet on subcarrier n, and  σk2 is the variance of noise at the kth receiver. Let PN ,K be an N × K matrix formed by the elements Pn,k , ∀n ∈ N , k ∈ K .  4.3.2  Hop-count-based Sub-carrier and Power Allocation Scheme  Other parameters remaining unchanged, the achievable data rate is usually monotonically increasing function of allocated subcarriers and transmit power. Let us define a user perceived utility function U(·) as a concave monotonically increasing function of allocated subcarriers and transmit powers. Our goal is to obtain optimal joint subcarrier and power allocation so as to maximize such function. It is customary to define U(·) as the summation of normalized channel capacity for all users. In the later part of this section, we describe two utility functions, 95  one of which is the proposed utility function in order to prioritize the packet that has already traveled more hops while the other is existing classical function independent of the hop-counts. The problem of jointly optimizing the available subcarriers and transmission power allocation to maximize the utility within the given power budget can be written as P3: max  PN ,K ,XN ,K  subject to  U(PN ,K , XN ,K ) 1T PN ,K 1 ≤ Pt , XN ,K 1 ≼ 1, PN ,K ≽ 0, xn,k ∈ {0, 1},  ∀xn,k ∈ XN ,K ,  (4.16)  where XN ,K is an N × K indicator matrix whose element xn,k is 1 when nth subcarrier is assigned to kth packet for transmission and 0 otherwise, and Pt is the total transmission power budget for node N0 . Note that the utility function U(·) is usually chosen to be a non-linear function of allocated subcarriers and powers. For example, it contains logarithmic terms when our goal is to maximize the achievable sum rate. Therefore, P3 is a mixed integer non-linear programming problem, for which an analytical solution is not straightforward. In the following, we present a heuristic based approach to find suboptimal solution of P3. We first relax integer constraint (4.16) allowing xn,k to be continuous variable and define a non-linear problem as P4: max  PN ,K ,XN ,K  subject to  U(PN ,K , XN ,K ) 1T PN ,K 1 ≤ Pt , XN ,K 1 ≼ 1, PN ,K ≽ 0, XN ,K ≽ 0.  96  (4.17)  If utility function U(·) is concave, Problem P4 is a convex optimization problem and there exist approaches in literature to solve it numerically [64]. We will provide analytical solutions for Problem P4 in special cases shortly. Algorithm 3 outlines the steps to find suboptimal solution to Problem P3. The algorithm divides the problem into two steps. Initially, it finds the subcarrier allocation to different packets. In this step, each subcarrier is allocated to at most one packet. Problem P4 is solved to get XN ,K assuming equal power allocation. The nth subcarrier is then assigned to the packet k for which xn,k is highest. In case of ties (i.e., when xn,k are equal for two or more packets), the subcarrier is allocated to the packet which has least number of subcarriers allocated at that point of time. Further ties are resolved by random approach. Once the subcarrier allocation (i.e., XN ,K ) is known, PN ,K is calculated by solving P4 in second step. Algorithm 3 Suboptimal Solution of P3 Subcarrier Allocation: Pt 1: Calculate PN ,K using Pn,k = NK ∀Pn,k ∈ PN ,K 2: Given PN ,K , Solve P4 to obtain XN ,K . 3: for n ∈ N do 4: if max XN ,K > 0 then k∈K  5:  i ← arg max XN ,K k∈K  6: xn,i ← 1 7: for k ∈ K \{i} do 8: xn,k ← 0 9: end for 10: end if 11: end for  Power Allocation: 12: Given XN ,K , Solve P4 to Obtain PN ,K 13: return PN ,K , XN ,K  97  Proposed Hop-Count-Based Utility Function In this section, we propose a utility function which favors packets with higher hopcounts. The proposed utility function maximizes the overall system capacity by ensuring more priority on next-hop SNR for data packets with higher hop-counts. Hop-count-based weight is applied to SNR of each packet by defining the utility function as U1 (PN ,K , XN ,K ) =  ∑ ∑  ) ( log 1 + βk αn,k Pn,k xn,k ,  (4.18)  n∈N k∈K  where βk represents the respective weight given to packet k ∀ k ∈ K . This weight can be defined in various ways. We define βk as  βk = η min(φk , βmax ), where scaling factor η is a design parameter which can take only positive values. It defines the level of priority given to the packets with higher hop-counts. Higher value of η provides more priority to such packets. To combat false hopinformation from adversary nodes and reduce the effect of unintended routing loop, the value of βk is capped by βmax . It is worthwhile to mention that for the packets originated at N0 , the node transmits it with the highest priority. This may be implemented by: i) allocating required subcarriers and power for such packet before running allocation algorithm for rest of the packets, or ii) by setting βk = βmax for such a packet. Theorem 2 When subcarrier allocation (i.e., XN ,K ) is known, the optimal power allocation (PN ,K ) can be obtained by solving problem P4 and is given by ( ) 1 1 ˜ Pn,k = max 0, − , ν˜ βk αn,k xn,k  ∀n ∈ N , k ∈ K ,  (4.19)  where ν˜ is chosen such that 1T P˜ N ,K 1 = Pt . Eq. (4.19) is required in step 12 of Algorithm 3 to calculate PN ,K . 98  Proof: Proof is provided in Appendix B. Theorem 3 When power allocation (i.e., PN ,K ) is known, problem P4 can be solved to obtain XN ,K as follows: ) ( 1 1 , ∀n ∈ N , k ∈ K , x˜n,k = max 0, − λ˜ n βk αn,k Pn,k where λ˜ n ∀n is chosen such that  ∑  (4.20)  x˜n,k = 1. Note that step 2 of Algorithm 3  k∈K  needs (4.20) to calculate XN ,K . Proof: Proof can be derived using a similar approach as in the Proof of Theorem 2. In the rest of the Section 4.3, we refer to resource allocation scheme with utilization function U1 as proposed hop-count-based resource allocation (HCBRA) scheme. Classical Hop-Count Independent Utility Function We also analyze the system performance when the utility function is a classical one which is independent of hop-counts and compare its performance with that of the proposed HCB-RA scheme. For this case, the utility function U2 is independent of βk ∀k ∈ K and is given by U2 (PN ,K , XN ,K ) =  ∑ ∑  ( ) log 1 + αn,k Pn,k xn,k .  (4.21)  n∈N k∈K  In this case, the solution of P4 may be obtained from (4.19) and (4.20) by setting βk = 1 ∀k ∈ K . In the rest of the Section 4.3, the resource allocation scheme with utility function U2 is referred to as classical hop-count independent resource allocation (HCI-RA) scheme.  99  4.3.3  Simulation Results  Selected results obtained via computer simulations are presented in this section to analyze the performance of the proposed resource allocation scheme. We assume that the nodes are distributed uniformly throughout the area under consideration. Therefore, without loss of generality we normalize the distance between the nodes such that average distance is unity. The channel coefficients between the nodes are modeled as zero-mean Rayleigh distributed complex random variables. Moreover, since all the channels in such a case undergo similar path-loss and fading, we normalize the variance of all channels to unity. We also assume noise variance on 2 = 1, ∀n ∈ N , k ∈ K . Usually, there is no line each channel to be unity, i.e., σn,k of sight signal path between nodes in most of the modern wireless communication networks. Therefore, the assumption of Rayleigh faded channel coefficients between the nodes should be reasonable for numerical analysis. Furthermore, the results presented in this section are comparative results obtained for the classical and the proposed schemes, therefore the assumption on channel coefficient should not affect the nature of final conclusions. Since channel coefficients change dynamically, we have taken each data point in the graphs as an average over 10, 000 independent channel realizations. We assume βmax = 6, and network diameter (i.e., maximum length of end-to-end route) = 10 for all simulations. In addition, βk is normalized by βmax so that βk ≤ 1. Performance metrics We have selected three parameters to compare performance of the proposed HCBRA with that of the classical HCI-RA scheme. The selected parameters are system outage, resource wastage in the network, and one-hop goodput. We assume that for a packet being successfully transmitted to the next hop, it must be guaranteed with a predefined minimum data rate of Rmin . The system outage in this section represents the fraction of total number of packets which are not allocated sufficient subcarriers and power by a scheme to achieve a minimum data rate of Rmin . We have analyzed resource wastage as an index to estimate the effect of re100  50 45  Resource wastage (units)  40 35  Proposed HCB−RA, Pt=3 units Classical HCI−RA, Pt=3 units Proposed HCB−RA, Pt=4 units Classical HCI−RA, Pt=4 units Proposed HCB−RA, P =5 units t  30  Classical HCI−RA, Pt=5 units  25 20 15 Rmin=1, η=1, N=16  10 5 4  5  6 7 8 Number of packets (K)  9  10  Figure 4.10: Resource wastage versus number of packets K for different values of power budget Pt . source allocation at a node to the entire network. In a multi-hop network, resources used by a packet in previous hops may be assumed to be proportional to the hop-count for simplicity. Note that in practical networks, resource allocations to a packet at different previous hops are not the same, and hence total resources used by a packet in previous hops may not be exactly proportional to the hop-count. If a scheme fails to guarantee the required minimum data rate of Rmin to a packet with hop-count φk , we consider network resource wastage of φk units. This assumption is reasonable when no retransmission is done by the link layer. Goodput represents the total throughput of only those packets which have data rate more than Rmin . One-hop goodput represents the useful throughput achievable within the power budget Pt considering next hop’s (with respect to N0 )  101  50 45  Proposed HCB−RA Classical HCI−RA  Resource wastage (units)  40 35 K=9  30 25  K=7 20 15 pt=5, η=1, N=16  10 5 1  K=5 1.2  1.4 1.6 Threshold data rate (Rmin)  1.8  2  Figure 4.11: Variation of resource wastage with threshold data rate Rmin for different number of packets K. performance only. Selected results In Figs. 4.10 and 4.11, we compare the network resource wastage with the proposed HCB-RA and the classical HCI-RA schemes in operation. It can be seen from these figures that the proposed scheme outperforms HCI-RA for all values of available power budget for a given Rmin and vice-versa. This is achieved because the proposed scheme favors packets with higher hop-counts. Performance of the proposed scheme improves with increase in the number of packets to be transmitted. This is because with higher number of packets, resources (subcarrier and power) become insufficient to satisfy transmission requirements of all pack102  5.5 5  P =5  One−hop goodput (bit/s/Hz)  t  Proposed HCB−RA Classical HCI−RA  4.5 4 Pt=4 3.5  R  =1, η=1, N=16  min  3 Pt=3 2.5 2 4  5  6 7 8 Number of packets (K)  9  10  Figure 4.12: Goodput versus number of packets K for different values of power budget Pt . ets and the proposed scheme smartly favors packets with higher hop-counts under such situation minimizing the network-wide resource wastage. Next, we present the one-hop goodput performance of both schemes in Figs. 4.12 and 4.13. From Figure 4.12, it is clear that the goodput of both schemes increases with increase in power budget for a specified Rmin as more packets can achieve the data rate higher than Rmin . Due to the same reason, the goodput of both schemes increases with relaxing the Rmin requirement for a given power budget as shown in Figure 4.13. When the resources are sufficient (i.e., power budget is higher for a given number of packets) and Rmin requirement is not stringent (i.e., lower Rmin ), the classical scheme works well. However, once the resources are not sufficient or Rmin requirement is stringent for given number of packets,  103  5.5 p =5, η=1, N=16  One−hop goodput (bit/s/Hz)  5  t  4.5 4 3.5 3 2.5 2 1.5 1 1  Proposed HCB−RA, K=5 Classical HCI−RA, K=5 Proposed HCB−RA, K=7 Classical HCI−RA, K=7 Proposed HCB−RA, K=9 Classical HCI−RA, K=9  1.2  1.4 1.6 Threshold data rate (Rmin)  1.8  2  Figure 4.13: Variation of goodput with threshold data rate Rmin for different number of packets K. the proposed scheme outperforms HCI-RA. This is because of the lower outage offered by the proposed HCB-RA under such a condition as explained below. Figs. 4.14 and 4.15 depict the one-hop outage performance of the proposed HCB-RA scheme compared to the classical HCI-RA scheme. When the resources are sufficient and Rmin requirement is lower, the classical HCI-RA performs slightly better. The reason is that under such a scenario, favoring packets with higher βk in the proposed scheme may result in starvation of some packets with lower hop-counts which may get more resources in the classical HCI-RA. However, the proposed scheme decreases the outage probability more efficiently when the resources are not sufficient and starts outperforming HCI-RA. The outage performance of the proposed scheme improves with higher number of packets as well  104  1 Proposed HCB−RA Classical HCI−RA  0.9 R One−hop outage  0.8  P =3  =1, η=1, N=16  t  min  0.7 0.6  Pt=4  0.5 Pt=5 0.4 0.3 4  5  6 7 8 Number of packets (K)  9  10  Figure 4.14: Outage versus number of packets K for different values of power budget Pt . as higher value of Rmin . This is because, using the classical scheme, packets with lower channel gains get lower transmit power and less number of subcarrier which may not be sufficient to guarantee the predefined Rmin . In contrary, some of such packets may get sufficient power and subcarriers by the proposed scheme if the packets have higher βk . It is apparent that the proposed scheme is able to decrease the resource wastage in the network while keeping the one-hop goodput and outage comparable with those in the classical scheme. Therefore, the proposed scheme can improve the network-wide resource utilization without degrading one-hop performance at the same time. When resources are not sufficient or Rmin requirement is higher, the proposed HCB-RA scheme outperforms the classical HCI-RA in all three consid-  105  1  0.9  One−hop outage  K=9 0.8  0.7 K=7 0.6  Proposed HCB−RA Classical HCI−RA  0.5  p =5, η=1, N=16 t  K=5 0.4 1  1.2  1.4 1.6 Threshold data rate (Rmin)  1.8  2  Figure 4.15: Outage versus threshold data rate Rmin for different number of packets K. ered aspects, i.e., resource wastage, goodput and outage. Note that resources are limited compared to traffic demand most of the time in modern wireless networks and hence the proposed scheme should be a better option compared to the classical HCI-RA.  4.3.4  Concluding Remarks  In this Section, we proposed a hop-count based resource allocation scheme to minimize the network-wide resource wastage in a multi-hop OFDMA network. The proposed scheme jointly allocates subcarrier and power providing higher priority to packets which have already utilized more resources in the previous hops. A design parameter has also been introduced to customize the level of priority given 106  to packets with higher hop-counts. We first modeled the resource allocation criteria as a mixed integer non-linear optimization problem, and then proposed a low complexity algorithm to allocate the subcarriers and power suboptimally. In the suboptimal formulation, we relaxed the integer constraint in order to obtain a convex optimization problem and solved it analytically using Lagrangian dual. The simulation results showed that the proposed scheme is very effective in minimizing the resource wastage without affecting the one-hop useful throughput and without notable degradation in system outage.  107  Chapter 5 Virtualization of Radio Access in Future Generation Cellular Networks: Potentials, Challenges and Solution Approaches 5.1  Introduction  In Chapters 2, 3 and 4, we proposed MAC framework, QoS provisioning module, admission control mechanism and resource allocation schemes for cognitive radio networks. In this chapter, we explore the virtualization of radio access part of the future generation cellular networks. Virtualization of radio access part of cellular system is crucial to implement the concept of network virtualization in future generation cellular system. The MAC framework, QoS provisioning module, admission control mechanism and resource allocation schemes proposed in previous chapters can be customized to develop various modules enabling virtualization of radio access part of the future generation cellular networks. Therefore, virtualization of radio access may be considered as a specific use case of research work presented in previous chapters. 108  Network virtualization (NV) is a novel approach enabling existence of multiple virtual networks (VNs), which may have different network interfaces or architectures, on a common physical infrastructure and resources [34]–[36]. It allows multiple operators (or several networks with different interfaces belonging to the same operator) to share the same physical resources in a more flexible and dynamic manner utilizing the resources more efficiently. However, it needs the physical resources/infrastructure to be virtualized into a number of virtual resources which are then offered to different VNs. Since NV has tremendous potential to enhance resource utilization and enable coexistence of heterogeneous wireless networks, it has attracted the attention of a huge research community as a candidate for next-generation networking paradigm [34]. NV is now being considered as a potential technique to develop future generation wireless networks. As a result, several projects have been initiated on it globally such as 4WARD [35], VINI [94], Cabernet [95], VN architecture in [96] and the references therein. In addition to enhancing the utilization of physical infrastructure/resources, NV can reduce carbon emission and energy consumption, provide opportunity to build up networks from multiple operators (or with multiple interfaces) on the same physical infrastructure, and decrease overall interference in the networks. Virtualization is a technique which efficiently controls and coordinates the interaction between users, applications, and computing resources independent of physical specification of resources [97]. Virtualization itself is not a very new concept. For example, significant amount of work can be found on virtualization of servers, nodes and wire line links in the literature [37] –[39]. However, only a few works have been reported on the virtualization of radio access part of the wireless networks. Since virtualization of radio access part is crucial for the successful implementation of NV concept, significant research is essential to develop virtual radio access in these networks. In this chapter, we will discuss the selected research challenges, implementation issues and solution approaches for the practical deployment of NV concept in wireless communication systems, particularly in cellular networks. We also discuss the potential of the cognitive radio technique  109  in respect to enabling network virtualization. The rest of the chapter is organized as follows. Section 5.2 investigates and identifies various research challenges and implementation issues for enabling NV. Then in Section 5.3, we present potential solution approaches to address the research issues identified in the previous section. This section also explores selected major approaches to achieve carbon efficient green communication by NV. Furthermore, we have discussed the potential of the cognitive radio technique in co-existence of non-cooperative VNs. The Chapter is concluded by concluding remarks in Section 5.4.  5.2  Research Challenges and Implementation Issues  NV enables sharing of over-provisioned wireless infrastructure whose usage in traditional approach is much lower than the resources it consists of [98]. NV can change the role of traditional wireless operators and decouple its role into two entities: (i) infrastructure providers and service providers. Infrastructure providers manage physical resources and infrastructure, while service providers can create VNs by leasing resources from one or more infrastructure providers [34]. Hence, NV enables a networking environment which consists of several heterogeneous network architectures from different service providers. NV provides flexibility, promotes diversity, increases manageability, enhances infrastructure utilization and can also enable green communication by promoting carbon emission reduction. However, full potential of NV of cellular systems hinges on the development of advanced techniques and protocols required for successful virtualization of radio access part of cellular systems. In the following, we discuss some of the major challenges related to the implementation of NV in cellular systems.  110  5.2.1  Design of Virtualized BS  In cellular systems, key factor to enable NV is design/implementation of a virtualized BS (VBS). Since BS handles the access to radio channel and scheduling of air interface resources, the BS must be virtualized in order to virtualize the air interface in the cellular systems [36]. VBS allows multiple VNs to share the same set of physical resources and implement separate customized control protocols on them. Since VBS handles multiple heterogeneous VNs, it must have multiple virtual BS instances so that each VN will be assigned to one of them. Implementation of efficient VBS is the main challenge in virtualization of existing cellular systems and is an open research problem.  5.2.2  Coexistence of Multiple VNs and Scalability  NV must allow coexistence of multiple VNs from same or different service providers on shared physical resources. Each VN may span over part or full underlying physical network infrastructure from one or more infrastructure providers. Scalability support is a crucial feature needed in the NV. The reason is that multiple VNs may not be implemented at the same time. The NV must support addition/removal of new/existing VNs in the future without affecting the performance of other VNs [34].  5.2.3  Heterogeneity Support and Backward Compatibility  Coexistence of heterogenous VNs are inevitable in the future generation wireless networks [36]. Therefore, heterogeneity support is unavoidable in NV. Moreover, NV must ensure the compatibility with the existing cellular network considering it as one of the VNs. However, how to achieve backward compatibility is still an open research challenge.  111  5.2.4  Isolation Between VNs and Overall System Stability  NV should ensure the isolation between VNs. The fault, error and reconfiguration in one VN should not affect or destabilize the operation of other existing VNs. However, fault in physical infrastructure may affect the operation of all VNs. Therefore, there should be some mechanisms in the VNs which can bring them back to the stable states in case of instability generated by fault in the infrastructure [34].  5.2.5  Green Communication Support in Modern Cellular Systems  Energy consumption in wireless communication is increasing every year and this trend seems to continue in the near future. The increasing energy consumption also increases the carbon emission in the environment. Therefore, recently, green wireless communication has attracted attention of a huge research community due to increasing operational cost of wireless networks and its adverse effect on the environment [99]–[101]. In this section, we will discuss the issues and challenges related to the carbon efficient ( and/or energy efficient) cellular system design. Carbon Emission Measurement Quantitative measure of carbon emission and classification of devices/components based on its carbon reduction efficiency is necessary to propose a green network architecture. Only quantitative measure of energy/power consumption by a device does not specify the related carbon emission quantity. Power supply source and the power generation techniques play a vital role in quantifying the carbon emission. For example, use of a chargeable battery which can be charged by solar power should be greener than electricity supply generated by coal. Carbon emission by electricity again depends on the technique used to generate the electricity. Electricity generated by hydro-power, wind energy and solar energy are far greener than that generated by burning coal. Therefore, our objective should  112  be utilizing the network components run by greener energy sources as much as possible in order to minimize the overall carbon emission. Most of the works in literature ignore the contribution of power generation techniques in the carbon reduction (green) schemes. Based on these factors, each device should have a carbon emission level (CEL) assigned to it. Performance Metrics for Carbon Efficient Networks Performance metrics are indicators of efficiency and effectiveness of a scheme or design approach to achieve an objective. Understanding and selection of these metrics are vital for the quantitative evaluation of the scheme or design approach. In the context of green performance of cellular networks, the green metrics can be defined at component, device and system/network levels [99]. The metrics at component and device levels indicate the green performance of individual component and device, while green metric at network level evaluate the performance at system/network level. Energy efficiency (EE) [99][102] and carbon reduction efficiency (CRE) are related to each other. However, EE alone may not be sufficient to quantity the CRE. CEL of a device/component has significant effect on CRE. Energy efficiency can be measured using EE metrics such as power consumption per unit throughput (J/bit), power needed per unit coverage area to operate a cellular system (W /m2 ), power consumption per subscriber served, and ratio of output power to the input power for a component [99][102]. The CRE metrics can be defined in the similar way such as in total carbon emission per unit system throughput, carbon emission to serve a unit coverage area (suitable for low traffic rural area) and carbon emission per served subscriber (suitable for heavy traffic urban area).  5.2.6  Fair and Transparent Sharing of Infrastructure and Other Physical Resources among VNs  Since each VN operator pays money to utilize the common physical resources and infrastructure in the network vitualization concept, there must be a fair and 113  efficient scheme at VBS to share the air resources such as spectram/subcarriers/codewords, time slots and power [36]. Sharing resources based on VN operator’s contract and payment is challenging. However, it is crucial in the practical network.  5.3  Possible Solution Approaches and Research Scopes  In this section, we will discuss approaches to address selected research challenges identified in the previous section.  5.3.1  Implementation of Virtualized BS  As discussed in Section 5.2.1, how to implement a VBS concept is the key hinderance in virtualization of air interface of cellular systems. In this section, we discuss one of the ways to devise VBS. Generally, virtualization of the BS is a kind of node virtualization. There are many ways to virtualize a node. For example, the physical resources (such as processor, memory and I/O devices) of the node (BS) can be shared between multiple instances of virtual operating systems [36]. Let us call the module, responsible for scheduling these physical resources of BS, as virtualization Enabler (VE). VE will be responsible for virtualizing the BS into a number of virtual BS instances, each of which is assigned to one VN. Access to physical resources (such as processor, memory and I/O devices) of physical BS by these virtual BS instances will be controlled and scheduled by VE. Furthermore, the allocation of radio resources (such as frequency/subcarriers/codewords, time slots and transmit power) among these virtual BS instances will also be controlled and scheduled by VE. Usually, VE collects the information such as channel conditions, traffic load, QoS requirements, contract related information, and carbon efficiency information from all VNs. Then, based on these information, VE schedules the air interface resources to the individual VN. Note that VNs do not need to be aware of the virtualization  114  of BS. In practice, VBS can simply be considered as the module VE. Moreover, a single VE can be connected to multiple physical BSs. These physical BSs may have either the same or different air interfaces. Therefore, VBS can be considered as a new module in virtualized cellular systems which controls the usage of a set of physical BSs. In the rest of this chapter, we use VBS and VE interchangeably. VBS may need to have a powerful processor, huge memory and several I/O interfaces to which all users/mobile stations from various VNs will be connected to get the service. Technically, VBS is a processor which controls several physical BSs with same/different air-interfaces. VBS will be the main unit/module in the virtualized cellular systems which is responsible for efficient implementation of resource allocation, scheduling, admission control, handover, and network wide cooperation and coordination. Therefore, further research is required to design a VBS carefully to optimize the system performance by integrating these operations.  5.3.2  Virtualized MAC Protocol Design and Effective Scheduling & Admission Control Schemes  The current MAC protocols and schemes designed for cellular systems will not be enough to coordinate access in virtualized wireless networks. New MAC frame structure, control messages and signaling protocol have to be defined. We should design the key building blocks for the MAC layer exclusively customized for virtual wireless networking architecture. After designing the MAC protocol, the next problems that we should handle is the MAC scheduling scheme and QoS provisioning framework that will be needed to meet the QoS requirement of users from various VNs. This involves designing QoS signaling schemes in the MAC protocol, making the scheduling scheme QoS aware and integrating QoS-aware admission control scheme with the MAC protocol. Therefore, MAC layer should be designed to integrate resource allocation schemes, scheduling and admission control schemes together for QoS provisioning. 115  Design of Smart Virtualized MAC In developing the virtualized MAC protocol, we should design its essential building blocks exclusively for NV such as frame structure and timing, MAC management messages and details of their usage, network entry and initial ranging, construction of MPDUs from packets received from higher layers at each VN, signaling between VBS and VNs for bandwidth/resource request and allocations, and signaling for adding new connection/VN or removing any existing connection/VN. NV may need multiple virtual MAC protocols at the VBS to handle the diversity of service requests from heterogeneous VNs. Since network coordination and cooperation are primarily performed at MAC layer, design of efficient virtualized MAC protocol is essential for successful implementation of NV concept. In the following, we describe other key modules which are integral parts of MAC operation. Resource Allocation and Scheduling Efficient resource allocation and scheduling of infrastructure and other physical resources among multiple VN requests is of extreme importance in order to maximize resource/infrastructure utilization, the number of coexisting VNs, carbon reduction efficiency, and revenue [34]. However, some of these objectives of resource allocation and scheduling may be conflicting to each other and a trade off between them are inevitable in practice. For example, making the network green by minimizing carbon emission and revenue maximization may not be achieved simultaneously in traditional resource allocation and scheduling approaches. New approaches should be investigated to find a balance between these conflicting objectives. The scheduling and resource allocation may also be based on the contracts signed with VNs or the different amount of payments made by VNs. The VN which pays more money, may want higher priority during resource allocation and scheduling in the commercial wireless networks. We need to consider all these factors while proposing resource allocation and scheduling schemes for virtualized cellular systems. 116  Admission Control Since physical resources are shared by several VNs with different service demands (QoS requirements), resources may not be sufficient to fulfil the QoS of all VNs. Therefore, there should be an admission control mechanism which accepts new VN only if physical resources are sufficient to fulfil the QoS requirement of existing and the new VNs. Admission control should be developed at the VN level rather than for a node or link. These admission control module can be incorporated at the VBS to decide the acceptance/rejection of service request from VNs. Alternately, VBS can send the estimate of physical resources portion available to each VN and then VN can implement an admission control to accept/reject different kinds of service requests within the VN. Since same physical resources are shared by several VNs, admission control mechanism is essential to maintain fairness among VNs and to assure a QoS guarantee to end-users of each VN. Such an admission control also requires message/information exchange among several modules of the network which may incur very high overhead. How to efficiently incorporate an admission control mechanism in NV is a difficult research problem. Fairness Among VNs Resource allocation, scheduling and admission control are key factors to ensure efficient sharing of infrastructure and physical resources in the virtualized cellular systems. The resource sharing can be devised based on several factors such as amount of money operators pay, carbon credits accumulated by networks, and type of green applications networks have. Generally, efficient resource utilization comes at the cost of degraded fairness among VNs or users. Therefore, resource allocation, scheduling and admission control mechanisms must maintain a reasonable level of fairness among the VNs. For example, these mechanisms may first guarantee a minimum level of service (in terms of minimum data rate, minimum number of service connections and so on) to each VN, and then remaining resources can be allocated to maximize the utilization. Since Virtualized cellular system may consist of several VNs (probably from different service providers), 117  they can claim a certain portion of physical resources based on the payment they paid. Fairness is, therefore, critical issue to enable coexistence of multiple VNs and is not easy to achieve in practice.  5.3.3  Achieving Carbon Efficient Green Communication by Network Virtualization  One of the most promising potentials of NV is its ability to enable green communication in the modern wireless networks. NV can modify the resource sharing approach and reduce the amount of BS equipment in order to decrease the required energy to run wireless networks and carbon emission [36]. In the following, we discuss some of the approaches tackling the carbon efficient green communication by NV. Energy Efficient Versus Carbon Efficient Resource Allocation As mentioned in Section 5.2.5, energy efficiency may not exactly quantify the carbon reduction efficiency. Therefore, carbon emission reduction should be minimized by optimizing carbon reduction efficiency metrics rather than energy efficiency metrics to achieve a carbon efficient green cellular systems. Optimizing carbon reduction efficiency metrics such as total carbon emission per unit system throughput, carbon emission to serve a unit coverage area and carbon emission per served subscriber can significantly improve the green performance of the system. However, improvement of green performance may degrade the overall system throughput and other non-green performance parameters. A trade off between green performance and system capacity is often required. Incentive for Carbon Efficient Transmission– Carbon Credit Versus Money Making a network carbon efficient may come at the cost of degraded system throughput. Therefore, VN operators must need some kind of incentive for reducing the carbon emission. Providing carbon credits to VNs can motivate them to achieve higher CRE. The carbon credit should be then sold for money to other 118  VNs or for higher priority during resource and infrastructure sharing. Carbon/Energy-Aware Handover Strategies When there are heterogeneous VNs, all of them do not have the same CRE. Some networks, which may have lower transmission range or coverage, are capable of providing higher data rate with less transmit power and less carbon emission. Presence of heterogeneous networks hence provide opportunity to optimize the overall green performance of system by connecting the virtual users to most green physical networks/units most of the time. It can be possible by developing a handover strategy based of CRE of networks. For example, when a virtual user is under the coverage of two or more networks with different CREs, it can be switched to the network with highest CRE if the current network is not green enough. However, handover always incurs overhead on the network, therefore, the handover scheme should be designed carefully. Otherwise, the overhead incurred can degrade the system performance in terms of throughput, delay, load balancing among networks as well as CRE. Energy Saving Scheme for Low Traffic Scenario In high traffic scenarios (e.g., busy hours or high traffic geographical zone), most of the physical resources such as BSs are used to serve VNs. However, in case of a low traffic scenario, the objective of resource allocation and scheduling can be devised to use lowest number of physical BSs in order to save the energy. That is, the objective should be to turn off as much BSs as possible to make the overall network green. In the traditional cellular systems, this is not possible as each operator has its own physical BS running all the time. Note that around 60-80% of energy consumption takes place at BS in wireless cellular systems [100][101]. Furthermore, handover strategy can also be modified to consider the idea of reducing number of running BSs when instantaneous total traffic from all VNs can be served by lower number of BSs (in case of low traffic). It can be achieved by switching the users to new BSs so that most of the BSs can be turned 119  off.  5.3.4  Co-existence of Non-Cooperative VNs: Cognitive Radio Approach  In the NV, usually VBS coordinates among VNs and provides a basis for sharing the physical resources and infrastructure fairly and efficiently. However, such operation is based on the assumptions that all VNs cooperate with each other through VBS. In NV, a VN is not supposed to be known to other VNs. Only VBS should know about the existence of all VNs. However, when VNs do not coordinate to VBS, the resource sharing must be then performed in such a way that VNs do not disturb/interfare each other. Opportunistic resource sharing, i.e., a resource can be accessed by a VN only when it is not being used by any other VN, can be an attractive approach. Cognitive Radio (CR) is one of the techniques that can be used to share the physical resources and infrastructure opportunistically by multiple networks without interfering each other. CR is a novel approach which enables a group of secondary users (unlicensed users) to opportunistically share the spectrum which is allocated to another group of licensed users (primary users) [2][3]. However, significant amount of work is needed to incorporate CR concept in NV due to basic differences in virtual cellular networks and CR networks. In CR networks, spectrum sensing is required, while in cellular network, VNs or VN nodes may get the information about occupancy of physical resources by contacting the VBS. Opportunistic VNs If the physical resources are more than that required by existing/registered VNs, these resources may remain unoccupied/unused by these registered networks for a significant amount of time. Therefore, secondary use of physical resources should be allowed to one or more unregistered opportunistic networks or users. The opportunistic use of physical resources by other opportunistic networks (such as VN or non-VN CR networks) can further enhance the resource utilization. Here, we 120  call VNs, which are known or subscribed to VBS, as registered/primary networks, while networks using the resources only opportunistically as opportunistic/secondary networks (SNs). Obviously, primary networks have higher priority over secondary networks to utilize the resources. As discussed above, CR technique can help achieving opportunistic use of physical resources. CR VN should be able to access physical resources opportunistically without hampering the operation of other VNs or keeping the interference to primary networks below a predefined level. VBS should be able to coordinate with SNs, and monitor and control the interference level of SNs. In this case, the existing primary VNs do not need to know the presence of virtual CR network. Since physical resources are owned by operator in commercial cellular networks, secondary use of resources can be motivated by charging lower prices to SNs. The price for opportunistic use of resources can be negotiated between VBS and SN instantly based on the resource availability. Game theoretic based approaches can be incorporated for the negotiation.  5.3.5  Use of Radio Environment Map  REM is an integrated database which stores diverse information about the surroundings of the wireless network. REM consists of multi-domain radio environment information such as geographical patterns, available networks/services, location, radio activities, traffic pattern at a location during various time of the day, past experience, and spectrum policies and usage pattern [103]–[105]. Moreover, these information can be updated frequently, if needed, by the users, VBS and physical BSs. REM can help VNs and VBS in fast adaptation to the current radio scenario by leveraging prior knowledge of the radio environment. REM can play a vital role in resource management and performance optimization in NV. VBS has knowledge of service request from various VNs. This knowledge can be used with REM information to optimize the radio resource utilization [103]. REM also helps to discover the location of the user [104]. Knowing location of users and the traffic 121  load on different VNs at that location at different times of a day, user may be switched to a particular VN with less traffic load or higher carbon efficiency or higher data rate to optimize various objectives. It can be considered as REM based handover. VBS consists of several physical BSs with different radio access techniques. REM information can be used to connect the user to the best physical BS at a particular time and a particular location. However, to achieve all these benefits there should be a REM information exchange mechanism in the network. Use of REM certainly needs exchange of information between VBS, user and VNs which incurs overhead. The message exchange, therefore, should be designed carefully to minimize such overhead. Otherwise, performance gain by the use of REM may not be justified. One approach to reduce the overhead is to perform analysis of most of the REM data at VBS (or at data acquiring nodes) and then exchange only useful statistics in the network. The exchange of raw REM data over the network may consume a huge part of radio resources, particularly when raw data set is very large [104]. Exchange of statistics only is far more efficient than exchanging the raw data. REM server is usually implemented as centralized architecture with less complex REM agent at the user nodes. Implementation of REM agent, which is needed to exchange the REM data to/from VBS, may increase the node complexity of user node.  5.4  Concluding Remarks  Network Virtualization is a promising design paradigm which can reshape the future generation wireless networks in order to achieve significant improvement in the utilization of physical resources/infrastructure. Network Virtualization can enable coexistence of multiple heterogeneous virtual networks to share the same physical resource and infrastructure in the modern cellular networks. VN has tremendous potential to increase the resource utilization, provide flexibility, explore diversity, enhance network manageability, and enable green communication by promoting carbon emission reduction. Most researchers now realize that a redesign of future wireless (including cellular) networks is a bare necessity, not a 122  luxury and network virtualization can be a leading technique to achieve it. However, successful deployment of network virtualization needs to address several issues as mentioned in this chapter. Addressing these issues may not be easy and needs more research work in this field.  123  Chapter 6 Conclusions and Future Research Directions In this chapter, we summarize the main contributions and the results obtained in this thesis. Moreover, the directions for future research are also presented.  6.1  Conclusions  In this thesis, we have considered the design of medium access control (MAC) protocol and resource allocation schemes for distributed cognitive radio networks (DCRNs) with the intention of improving radio resource management and providing better control and coordination over wireless medium for secondary users (SUs). The proposed MAC protocol and resource allocation schemes will play a significant role in DCRNs to boost the practical implementation of distributed ad-hoc cognitive radio (CR) networks, where the radio environment is highly dynamic.  6.1.1  Summary of Major Contributions  In particular, we have made the following major contributions in this dissertation. First, we proposed a novel MAC design for distributed cognitive radio net124  works. The proposed OMC-MAC connects channel sensing, negotiation and data transmission phases in an integrated way to propose a complete working MAC framework. It incorporates efficient channel negotiation and contention free data transmission to efficiently utilize spare bandwidth from PU networks. The simulation results further revealed that the proposed MAC performs better than existing protocols in providing higher throughput and keeping the collision with PU lower in case of erroneous channel sensing. It ensures no collision among SUs during data transmission phase, which helps to utilize the opportunistically available spectrum efficiently. It also avoids multi-channel hidden terminal problem without increasing SU node complexity by implementing a smart control message exchange mechanism. Moreover, common control channel is always known and available to SUs in proposed OMC-MAC which provides a reliable medium for initial message exchange among SUs. Therefore, the network initialization, reconfiguration and coordination is easier and reliable even in presence of heterogeneous spectrum availability. The work has been published in [53]. We also explored MAC design issues for DCRNs and surveyed state-of-theart research efforts to address these issues. Although DCRNs have some similarities with the existing multi-channel distributed wireless networks, research has shown that the MAC protocols proposed for these networks cannot be used for DCRNs without modifications. The most critical issues for such MAC designs are minimizing the effect of sensing error, avoiding MHTP, keeping interference to the PUs as minimum as possible, assuring good network coordination with heterogeneous spectrum availability and maximizing the utilization of the spare bandwidth available from the PU network. We surveyed selected state-of-the-art DCRN MAC protocols, analyzed their comparative benefits and limitations, and discussed how effectively these MACs can address the critical design issues that we outlined. This survey can provide a framework for future research in this area. The work has been published in [40]. Second, we devised a QoS mechanism in proposed OMC-MAC protocol for distributed CR networks in order to provide a QoS assurance to the prioritized  125  SUs such as SU with delay sensitive applications in a highly dynamic CR environment. The incorporated QoS mechanism in OMC-MAC ensures priority of delay sensitive applications over general users. We analyzed the proposed scheme and showed that it efficiently fulfils the QoS requirements of delay sensitive applications by significantly reducing their channel access delay. We proposed a method to calculate the maximum number of prioritized applications in the system based on their QoS requirements which can serve as a basis to develop admission control modules in DCRN. The work has been published in [63]. Third, we proposed a hop-count based resource allocation scheme to minimize the network-wide resource wastage in a multi-hop distributed cognitive radio network (MHDCRN). A cross-layer approach was considered in which link layer resource allocator obtained the hop-count information from network layer module. Distributed implementation of proposed scheme is possible because each node can access hop-count information from the higher layer. The formulated power allocation problem favors the data packets which have already used higher amount of resources in previous hops in order to decrease overall wastage of network resources. We considered the number of hops traveled by a packet as a basis to estimate the amount of network resource already invested on the packet. A design parameter was also introduced which decides the priority level for packets with higher hop-counts. Simulation results showed that proposed scheme reduces resource wastage and QoS outage in the MHDCRN without any decrease in system throughput. The work has been published in [80]. We also extended hop-count based resource allocation scheme to minimize the network-wide resource wastage in a multi-hop OFDMA (MHOFDMA) networks. In multi-hop OFDMA networks, the proposed scheme jointly allocates subcarrier and power providing higher priority to packets with higher hop-counts. We first modeled the resource allocation criteria as a mixed integer non-linear optimization problem, and then proposed a low complexity algorithm to allocate the subcarriers and power suboptimally. In the suboptimal formulation, we relaxed the integer constraint in order to obtain a convex optimization problem and solved it analyt-  126  ically using Lagrangian dual. The simulation results showed that the proposed scheme is very effective in minimizing the resource wastage without affecting the one-hop useful throughput and without notable degradation in system outage in the MHOFDMA network. The work has been published in [93]. Fourth, we explored network virtualization (NV) concept as a potential technique to redesign the future generation wireless networks such as cellular networks in order to enhance utilization of physical resources/infrastructure. NV can enable co-existence of several heterogeneous virtual networks to share same physical resources and infrastructures in modern cellular networks. We first presented a survey of state-of-the-art research in the field of Network virtualization of cellular systems. We then pointed out promising benefits and potential capabilities of NV concept in the context of cellular systems. NV has tremendous potential to increase the resource utilization, provide resource scheduling flexibility, explore diversity, enhance network manageability, and enable green communication by promoting carbon emission reduction. We also explored research challenges and implementation issues for successful realization of virtualizing radio access of cellular systems. Note that although virtualization of server, node and wired link has been explored in the literature, virtualization of radio access is a relatively new research area. We also investigated the solution approaches to address major research and implementation issues identified in this thesis. Furthermore, we discussed the potential of CR technique to enhance NV.  6.1.2  Concluding Remarks  In this thesis, we proposed an opportunistic multi-channel MAC protocol (OMCMAC) which efficiently integrates channel sensing, channel negotiation and data transmission phases to provide a complete MAC framework for distributed CR networks. The proposed OMC-MAC outperforms existing MAC proposals considered in this thesis by utilizing opportunistically available spectrum more efficiently and reducing effect of sensing error on PUs during SU’s data transmission phase. It handles multi-channel hidden terminal problem effectively and also en127  sures no collision among SUs during SU’s data transmission phase. We then introduced a QoS module in the proposed OMC-MAC. The QoS module significantly reduces the channel access delay of prioritized (i.e., delay sensitive) applications in a highly dynamic CR environment. The maximum number of prioritized applications that can be accommodated in the system was explored for given QoS requirements of these applications, which can serve as a basis to develop admission control module in DCRNs. We also proposed a hop-count based resource allocation scheme to minimize the network-wide resource wastage in multi-hop distributed cognitive radio networks by formulating power allocation problem which favors the data packets which have already used higher amount of resources in previous hops. It has been found that incorporating hop-count information in the resource allocation formulation reduces resource wastage and QoS outage in the multi-hop DCRNs without any decrease in system throughput. We explored similar resource allocation scheme in multi-hop OFDMA networks by proposing a scheme which jointly allocates subcarrier and power providing higher priority to packets with higher hop-counts. The simulation results showed that the proposed scheme is very effective in minimizing the resource wastage in the multi-hop OFDMA networks as well. Therefore, a link layer resource allocator at a node should consider previous resource usage by packets to maximize network-wide resource utilization in multi-hop DCRNs and OFDMA networks. The proposed MAC framework and resource allocation schemes can play significant roles to accelerate the practical implementation of distributed ad-hoc cognitive radio networks.  6.2  Suggested Future Research Directions  The research work conducted in this thesis has given rise to several interesting challenges and research problems. Our current works are addressing various of these problems. In the following, we discuss some of the research issues which can be investigated as potential future work:  128  • Adaptation of Partial Channel Sensing to Increase Data Transmission Period. In Chapter 2, we assumed that all channels are scanned during the sensing phase. Sensing is a required part of cognitive MAC protocol design to identify data channels unoccupied by the PUs. Sensing ensures no or minimal collision with PUs. It may not be possible to scan all the available data channels in a given sensing period due to hardware limitations of SUs. Sensing a channel takes certain amount of time, and hence very long sensing period may require to sense all channels. Longer sensing time decreases the data transmission duration and hence the system capacity. On the other hand, sensing only a portion of total PU channels may decrease the probability of finding a common available channel between transmitter and intended receiver. Partial channel sensing therefore needs to be designed carefully. For example, SUs can keep track of previous channel sensing results and try to scan those channels which were available most of the time based on transmission history. Similarly, SUs can scan the channels which are likely to be available to themselves as well as to their neighbors with higher probability. Performance of OMC-MAC can be further improved by partial channel sensing strategy and is a topic for future work. • Integrating Channel Assignment Strategies in OMC-MAC. As discussed in Chapters 2 and 3, multiple data channels can be common between transmitter and intended receiver during channel negotiation. In case of multiple common data channels available between a Tx–Rx pair, we assumed that the Rx chooses one of these common channels randomly during negotiation. However, OMC-MAC can incorporate better channel selection strategies as well. For example, a Tx–Rx pair could choose the best available common channel. Channel assignment can be performed in CR networks in order to optimize performance parameters like throughput and fairness. Also, allocation of better channels to delay sensitive applications can help assuring faster transmission of these applications. As a result, it may help to provide better QoS assurance to such applications. Our protocol can also benefit 129  from better channel selection strategies compared to the random channel selection scheme assumed in this thesis. Various channel assignment strategies can be integrated in the proposed OMC-MAC to investigate the fullfledged benefits of the proposed protocol, which is an interesting area to explore. • Estimation of Resource Consumed in Previous Hops. In our accomplished work in Chapter 4, we assumed that resources consumed by a packet is proportional to the number of hops it has travelled (i.e., hop-count). However, actual values of resources allocated to the packets at previous hops may be quite different based on various factors such as instantaneous link conditions. More realistic estimation of resources consumed by a packet in previous hops can further enhanced the performance of hop-count based power allocation schemes proposed in this thesis and is an interesting topic for future research.  130  Bibliography [1] “Federal communications commission, spectrum policy task force report,” ET Docket no. 02-135, Nov. 2002. → pages 1, 15 [2] I. Mitola, J. and J. Maguire, G.Q., “Cognitive radio: making software radios more personal,” IEEE Personal Commun., vol. 6, no. 4, pp. 13–18, Aug 1999. → pages 1, 15, 120 [3] S. Haykin, “Cognitive radio: brain-empowered wireless communications,” IEEE J. Select. Areas Commun., vol. 23, no. 2, pp. 201–220, Feb. 2005. → pages 120 [4] A. M. Wyglinski, M. Nekovee, and T. Hou, Eds., Cognitive Radio Communications and Networks: Principles and Practice. Academic Press, 2009. → pages 2, 15 [5] E. Hossain and V. K. Bhargava, Eds., Cognitive Wireless Communication Networks. Springer, 2007. → pages 1 [6] E. Axell, G. Leus, E. Larsson, and H. Poor, “Spectrum sensing for cognitive radio : State-of-the-art and recent advances,” IEEE Signal Process. Mag., vol. 29, no. 3, pp. 101 –116, May 2012. → pages 2 [7] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithms for cognitive radio applications,” IEEE Commun. Surveys Tuts., vol. 11, no. 1, pp. 116 –130, Mar. 2009. → pages 31 [8] F. Zeng, C. Li, and Z. Tian, “Distributed compressive spectrum sensing in cooperative multihop cognitive networks,” IEEE J. Sel. Topics Signal Process., vol. 5, no. 1, pp. 37 –48, Feb. 2011. → pages  131  [9] O. Fatemieh, R. Chandra, and C. Gunter, “Secure collaborative sensing for crowd sourcing spectrum data in white space networks,” in Symp. IEEE DySPAN’10, Apr. 2010, pp. 1 –12. → pages 2 [10] H. Vu-Van and I. Koo, “Cooperative spectrum sensing with collaborative users using individual sensing credibility for cognitive radio network,” IEEE Trans. Consum. Electron., vol. 57, no. 2, pp. 320 –326, May 2011. → pages 2 [11] R. Yu, Y. Zhang, L. Yi, S. Xie, L. Song, and M. Guizani, “Secondary users cooperation in cognitive radio networks: balancing sensing accuracy and efficiency,” Wireless Commun., vol. 19, no. 2, pp. 30 –37, Apr. 2012. → pages 2 [12] R. Murty, R. Chandra, T. Moscibroda, and P. Bahl, “SenseLess: A database-driven white spaces network,” IEEE Trans. Mobile Comput., vol. 11, no. 2, pp. 189 –203, Feb. 2012. → pages 2 [13] H. R. Imam, K. Inage, M. Ohta, and T. Fujii, “Measurement based radio environment database using spectrum sensing in cognitive radio,” in Mobile and Wireless Networking (iCOST), 2011 International Conference on Selected Topics in, 2011, pp. 110–115. → pages [14] M. Mueck, M. Di Renzo, M. Debbah, and T. Renk, “Combination of centralized and decentralized database and terminal-based spectrum sensing for secondary spectrum access,” in Proc. IEEE ICWITS’10, 2010, pp. 1–4. → pages [15] D. Denkovski, V. Rakovic, M. Pavloski, K. Chomu, V. Atanasovski, and L. Gavrilovska, “Integration of heterogeneous spectrum sensing devices towards accurate REM construction,” in Proc. IEEE WCNC’12, 2012, pp. 798–802. → pages 2 [16] C. Cormio and K. R. Chowdhury, “A survey on MAC protocols for cognitive radio networks,” Ad Hoc Networks, vol. 7, no. 7, pp. 1315 – 1329, 2009. → pages 3 [17] (2008, Oct) The IEEE 802.22 WRAN working group website. [Online]. Available: http://www.ieee802.org/22/ → pages 4  132  [18] I. Pefkianakis, S. Wong, and S. Lu, “SAMER: Spectrum aware mesh routing in cognitive radio networks,” in Proc. IEEE DySPAN’08, Oct. 2008, pp. 1 –5. → pages 7, 73, 74 [19] B. Li, D. Li, Q. hui Wu, and H. Li, “ASAR: Ant-based spectrum aware routing for cognitive radio networks,” in Proc. WCSP’09, Nov. 2009, pp. 1 –5. → pages 74 [20] G. Cheng, W. Liu, Y. Li, and W. Cheng, “Spectrum aware on-demand routing in cognitive radio networks,” in Proc. IEEE DySPAN’07, Apr. 2007, pp. 571 –574. → pages 7, 73, 74 [21] P. Parvathi, “Comparative analysis of CBRP, AODV, DSDV routing protocols in mobile ad-hoc networks,” in Proc. ICCCA’12, 2012, pp. 1–4. → pages 7 [22] M. Rahman, F. Anwar, J. Naeem, and M. Abedin, “A simulation based performance comparison of routing protocol on mobile ad-hoc network (proactive, reactive and hybrid),” in Computer and Communication Engineering (ICCCE), 2010 International Conference on, 2010, pp. 1–5. → pages [23] S.-J. Lee, M. Gerla, and C.-K. Toh, “A simulation study of table-driven and on-demand routing protocols for mobile ad hoc networks,” Network, IEEE, vol. 13, no. 4, pp. 48–54, 1999. → pages [24] L. Abusalah, A. Khokhar, and M. Guizani, “A survey of secure mobile ad hoc routing protocols,” Communications Surveys Tutorials, IEEE, vol. 10, no. 4, pp. 78–93, 2008. → pages 7 [25] H. Khalife, N. Malouch, and S. Fdida, “Multihop cognitive radio networks: to route or not to route,” Network, IEEE, vol. 23, no. 4, pp. 20 –25, Jul.Aug. 2009. → pages 7, 73 [26] J. So and N. H. Vaidya, “Multi-channel MAC for ad hoc networks: handling multi-channel hidden terminals using a single transceiver,” in Proc. ACM MobiHoc’04, 2004, pp. 222–233. → pages 9, 16 [27] S.-J. Kim and G. Giannakis, “Optimal resource allocation for MIMO ad hoc cognitive radio networks,” in Annu. Conf. Commun., Control and Comput. 2008, Sept. 2008, pp. 39 –45. → pages 10 133  [28] H.-J. Lim, D.-Y. Seol, and G.-H. Im, “Resource allocation for mitigating the effect of sensing errors in cognitive radio networks,” IEEE Commun. Lett., vol. 14, no. 12, pp. 1119 –1121, Dec. 2010. → pages [29] S. Almalfouh and G. Stuber, “Interference-aware radio resource allocation in OFDMA-based cognitive radio networks,” IEEE Trans. Veh. Technol., vol. 60, no. 4, pp. 1699 –1713, May 2011. → pages [30] L. Akter and B. Natarajan, “Distributed approach for power and rate allocation to secondary users in cognitive radio networks,” IEEE Trans. Veh. Technol., vol. 60, no. 4, pp. 1526 –1538, May 2011. → pages [31] X. Zhou, G. Li, D. Li, D. Wang, and A. Soong, “Probabilistic resource allocation for opportunistic spectrum access,” Wireless Communications, IEEE Transactions on, vol. 9, no. 9, pp. 2870 –2879, Sept. 2010. → pages [32] L. Ding, T. Melodia, S. Batalama, J. Matyjas, and M. Medley, “Crosslayer routing and dynamic spectrum allocation in cognitive radio ad hoc networks,” IEEE Trans. Veh. Technol., vol. 59, no. 4, pp. 1969 –1979, May 2010. → pages [33] J. Mietzner, L. Lampe, and R. Schober, “Distributed transmit power allocation for multihop cognitive-radio systems,” IEEE Trans. Wireless Commun., vol. 8, no. 10, pp. 5187–5201, 2009. → pages 10 [34] N. Chowdhury and R. Boutaba, “Network virtualization: state of the art and research challenges,” IEEE Commun. Mag., vol. 47, no. 7, pp. 20 –26, Jul. 2009. → pages 10, 109, 110, 111, 112, 116 [35] N. Niebert, S. Baucke, I. El-Khayat, M. Johnsson, B. Ohlman, H. Abramowicz, K. Wuenstel, H. Woesner, J. Quittek, and L. Correia, “The way 4WARD to the creation of a future internet,” in Symp. IEEE PIMRC’08, Sept. 2008, pp. 1 –5. → pages 109 [36] Y. Zaki, L. Zhao, C. Goerg, and A. Timm-Giel, “LTE wireless virtualization and spectrum management,” in Conf. IFIP WMNC’10, Oct. 2010, pp. 1 –6. → pages 10, 109, 111, 114, 118 [37] R. Morris, E. Kohler, J. Jannotti, and M. F. Kaashoek, “The Click modular router,” in Proc. ACM SOSP’99, 1999, pp. 217–231. → pages 11, 109 134  [38] S. Bhatia, M. Motiwala, W. Mhlbauer, V. Valancius, A. Bavier, N. Feamster, L. Peterson, and J. Rexford, “Hosting virtual networks on commodity hardware,” in Georgia Tech. University., Tech. Rep. GT-CS-07-10, 2008. → pages [39] “Cisco VN-Link: Virtualization-aware networking,” in White paper, Cisco Systems, Inc., 2009, pp. 1 –8. → pages 11, 109 [40] S. C. Jha, M. M. Rashid, V. K. Bhargava, and C. Despins, “Medium access control in distributed cognitive radio networks,” IEEE Wireless Commun. Mag., vol. 18, no. 4, pp. 41–51, Aug. 2011. → pages 17, 22, 77, 92, 125 [41] Q. Zhao, L. Tong, A. Swami, and Y. Chen, “Decentralized cognitive MAC for opportunistic spectrum access in ad hoc networks: A POMDP framework,” IEEE J. Select. Areas Commun., vol. 25, no. 3, pp. 589–600, Apr. 2007. → pages 19, 23, 27, 40, 48 [42] L. Le and E. Hossain, “OSA-MAC: A MAC protocol for opportunistic spectrum access in cognitive radio networks,” Proc. IEEE WCNC ’08, pp. 1426–1430, Apr. 2008. → pages 19, 23, 27, 40, 62 [43] Y. Yuan, P. Bahl, R. Chandra, P. Chou, J. Ferrell, T. Moscibroda, S. Narlanka, and Y. Wu, “KNOWS: Cognitive radio networks over white spaces,” in Proc. IEEE DySPAN’07, Apr. 2007, pp. 416 –427. → pages 19 [44] H. Su and X. Zhang, “Cross-layer based opportunistic MAC protocols for QoS provisionings over cognitive radio wireless networks,” IEEE J. Select. Areas Commun., vol. 26, no. 1, pp. 118–129, Jan. 2008. → pages 20, 21, 23, 27, 48 [45] H. Song and X. Lin, “A group based MAC protocols for QoS provisioning in cognitive radio networks,” in Proc. IEEE ICCS’08, Nov. 2008, pp. 1489– 1493. → pages 20, 29, 48 [46] Y. Kondareddy and P. Agrawal, “Synchronized MAC protocol for multihop cognitive radio networks,” Proc. IEEE ICC ’08, pp. 3198–3202, May 2008. → pages 21, 23, 27 [47] C. Cordeiro and K. Challapali, “C-MAC: A cognitive MAC protocol for multi-channel wireless networks,” Apr. 2007, pp. 147–157. → pages 23, 27 135  [48] J. Jia, Q. Zhang, and X. Shen, “HC-MAC: A hardware-constrained cognitive MAC for efficient spectrum management,” IEEE J. Select. Areas Commun., vol. 26, no. 1, pp. 106–117, Jan. 2008. → pages 23, 27, 28, 45, 48, 62, 76 [49] L. Ma, X. Han, and C.-C. Shen, “Dynamic open spectrum sharing MAC protocol for wireless ad hoc networks,” Proc. IEEE DySPAN’05, pp. 203– 213, Nov. 2005. → pages 23, 27 [50] T. Chen, H. Zhang, G. Maggio, and I. Chlamtac, “Cogmesh: A clusterbased cognitive radio network,” in Proc. IEEE DySPAN’07, Apr. 2007, pp. 168–178. → pages 23, 26, 27 [51] M. Maskery, V. Krishnamurthy, and Q. Zhao, “Decentralized dynamic spectrum access for cognitive radios: cooperative design of a noncooperative game,” IEEE Trans. Commun., vol. 57, no. 2, pp. 459–469, Feb. 2009. → pages 27 [52] C. Zou and C. Chigan, “A game theoretic DSA-driven MAC framework for cognitive radio networks,” in Proc. IEEE ICC ’08, May 2008, pp. 4165– 4169 4165–4169. → pages 27 [53] S. Jha, M. Rashid, V. Bhargava, and C. Despins, “OMC-MAC: an opportunistic multichannel MAC for cognitive radio networks,” in Proc. IEEE VTC’09-Fall, Sept. 2009, pp. 1–5. → pages 29, 125 [54] G. Yuan, R. Grammenos, Y. Yang, and W. Wang, “Performance analysis of selective opportunistic spectrum access with traffic prediction,” IEEE Trans. Veh. Technol., vol. 59, no. 4, pp. 1949 –1959, May 2010. → pages 31 [55] P. Wang, Z. Zhang, H. Huang, K. Wu, G. Yu, and A. Huang, “Multi-channel cooperative spectrum sensing based on belief propagation algorithm,” in Proc. IEEE VTC’09-Fall, Sep. 2009, pp. 1 –5. → pages 31 [56] (2007) IEEE standard for wireless LAN medium access control (MAC) and physical layer (PHY) specifications (revised). → pages 32 [57] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination function,” IEEE J. Select. Areas Commun., vol. 18, no. 3, pp. 535– 547, Mar. 2000. → pages 32, 36 136  [58] H. Ahmad and B. Salameh, “Throughput-oriented channel assignment for opportunistic spectrum access networks,” Mathematical and Computer Modelling, vol. 53, no. 1112, pp. 2108 – 2118, 2011, recent Advances in Simulation and Mathematical Modeling of Wireless Networks. → pages 33 [59] C. Peng, H. Zheng, and B. Y. Zhao, “Utilization and fairness in spectrum assignment for opportunistic spectrum access,” Mob. Netw. Appl., vol. 11, no. 4, pp. 555–576, 2006. → pages [60] L. Cao and H. Zheng, “Distributed spectrum allocation via local bargaining,” in Proc. IEEE SECON’05, Sept., 2005, pp. 475 – 486. → pages 33 [61] E. Ziouva and T. Antonakopoulos, “CSMA/CA performance under high traffic conditions: throughput and delay analysis,” Computer Communications, vol. 25, no. 3, pp. 313 – 321, 2002. → pages 36, 37 [62] J. Mo, H.-S. So, and J. Walrand, “Comparison of multichannel MAC protocols,” IEEE Trans. Mobile Comput., vol. 7, no. 1, pp. 50–65, Jan. 2008. → pages 38 [63] S. Jha, U. Phuyal, M. Rashid, and V. Bhargava, “Design of OMC-MAC: An opportunistic multi-channel MAC with QoS provisioning for distributed cognitive radio networks,” IEEE Trans. Wireless Commun., vol. PP, no. 99, pp. 1 –12, Aug. 2011. → pages 47, 92, 126 [64] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge University Press, Mar. 2004. → pages 53, 82, 97, 145 [65] R. M. Corless, G. H. Gonnet, D. E. G. Hare, D. J. Jeffrey, and D. E. Knuth, “On the Lambert W function,” in ADVANCES IN COMPUTATIONAL MATHEMATICS, vol. 5, 1996, pp. 329–359. → pages 58 [66] “PARSEC: PARallel Simulation Environment for Complex systems,” UCLA Parallel Computing Laboratory. [Online]. Available: http://pcl.cs.ucla.edu/projects/parsec/ → pages 62 [67] A. Guha and V. Ganapathy, “Power allocation schemes for cognitive radios,” in Proc. COMSWARE’08, Jan. 2008, pp. 51–56. → pages 72  137  [68] R. Zhang, Y. chang Liang, and S. Cui, “Dynamic resource allocation in cognitive radio networks,” IEEE Signal Process. Mag, vol. 27, no. 3, pp. 102–114, May 2010. → pages 72 [69] Q. Lu, T. Peng, W. Wang, W. Wang, and C. Hu, “Optimal route selection and resource allocation in multi-hop cognitive radio networks,” in Proc. IEEE GLOBECOM’09, 2009, pp. 1–6. → pages 72, 76, 92 [70] W. Wang, K. Shin, and W. Wang, “Joint spectrum allocation and power control for multi-hop cognitive radio networks,” IEEE Trans. Mobile Comput., vol. PP, no. 99, p. 1, 2010. → pages 72 [71] H. Song and X. Lin, “Spectrum aware highly reliable routing in multihop cognitive radio networks,” in Proc. WCSP’09, Nov. 2009, pp. 1 –5. → pages 73, 74 [72] G. Cheng, W. Liu, Y. Li, and W. Cheng, “Joint on-demand routing and spectrum assignment in cognitive radio networks,” in Proc. IEEE ICC’07, Jun. 2007, pp. 6499 –6503. → pages 74 [73] H. Ma, L. Zheng, X. Ma, and Y. luo, “Spectrum aware routing for multi-hop cognitive radio networks with a single transceiver,” in Proc. CrownCom’08, May 2008, pp. 1 –6. → pages 74 [74] B. Xia, M. Wahab, Y. Yang, Z. Fan, and M. Sooriyabandara, “Reinforcement learning based spectrum-aware routing in multi-hop cognitive radio networks,” in Proc. CROWNCOM’09, Jun. 2009, pp. 1 –5. → pages 74 [75] C. Xin, L. Ma, and C.-C. Shen, “Path-centric channel assignment in cognitive radio wireless networks,” in Proc. CrownCom’07, Aug. 2007, pp. 313 –320. → pages [76] H. Khalife, S. Ahuja, N. Malouch, and M. Krunz, “Probabilistic path selection in opportunistic cognitive radio networks,” in Proc. IEEE GLOBECOM’08, Dec. 2008, pp. 1 –5. → pages 73 [77] J. Jia, J. Zhang, and Q. Zhang, “Relay-assisted routing in cognitive radio networks,” in Proc. IEEE ICC’09, Jun. 2009, pp. 1 –5. → pages 73  138  [78] J. A. Boyan and M. L. Littman, “Packet routing in dynamically changing networks: A reinforcement learning approach,” in Advances in Neural Information Processing Systems 6. Morgan Kaufmann, 1994, pp. 671–678. → pages 74 [79] S. Kumar and R. Miikkulainen, “Dual reinforcement Q-Routing: An online adaptive routing algorithm,” in Proc. ANNIE’97. ASME Press, 1997, pp. 231–238. → pages 74 [80] S. C. Jha, U. Phuyal, and V. K. Bhargava, “Cross-layer resource allocation approach for multi-hop distributed cognitive radio network,” in Proc. CWIT’11, May 2011, pp. 211 –215. → pages 75, 126 [81] J. Poston and W. Horne, “Discontiguous OFDM considerations for dynamic spectrum access in idle TV channels,” in Proc. IEEE DySPAN’05, Nov. 2005, pp. 607–610. → pages 76 [82] R. Rajbanshi, A. M. Wyglinski, and G. J. Minden, “An efficient implementation of NC-OFDM transceivers for cognitive radios,” in Proc. CROWNCOM’06, Jun. 2006, pp. 1–5. → pages 76 [83] M. Jiang and L. Hanzo, “Multiuser MIMO-OFDM for next-generation wireless systems,” Proc. IEEE, vol. 95, no. 7, pp. 1430–1469, Jul. 2007. → pages 91 [84] M. Salem, A. Adinoyi, M. Rahman, H. Yanikomeroglu, D. Falconer, Y.-D. Kim, E. Kim, and Y.-C. Cheong, “An overview of radio resource management in relay-enhanced OFDMA-based networks,” IEEE Commun. Surveys Tuts., vol. 12, no. 3, pp. 422 –438, 2010. → pages 91 [85] Y. Pan, A. Nix, and M. Beach, “Distributed resource allocation for OFDMA-based relay networks,” IEEE Trans. Veh. Technol., vol. 60, no. 3, pp. 919 –931, Mar. 2011. → pages [86] M. Salem, A. Adinoyi, H. Yanikomeroglu, and D. Falconer, “Opportunities and challenges in OFDMA-based cellular relay networks: A radio resource management perspective,” IEEE Trans. Veh. Technol., vol. 59, no. 5, pp. 2496 –2510, 2010. → pages 91, 92  139  [87] B. Bai, W. Chen, Z. Cao, and K. Ben Letaief, “Max-matching diversity in OFDMA systems,” IEEE Trans. Commun., vol. 58, no. 4, pp. 1161 –1171, Apr. 2010. → pages 91 [88] J. Leinonen, J. Hamalainen, and M. Juntti, “Performance analysis of downlink OFDMA resource allocation with limited feedback,” IEEE Trans. Wireless Commun., vol. 8, no. 6, pp. 2927 –2937, Jun. 2009. → pages 92 [89] X. Zhang, W. Jiao, and M. Tao, “End-to-end resource allocation in OFDM based linear multi-hop networks,” in Proc. IEEE INFOCOM’08, Apr. 2008, pp. 879 –887. → pages 92, 93 [90] H. Li, “OFDMA resource allocation in hybrid wireless network,” in Proc. IEEE VTC’09-Fall, Sept. 2009, pp. 1 –5. → pages [91] P. Zhou, G. Miao, and B. Bing, “Cross-layer congestion control and scheduling in multi-hop OFDMA wireless networks,” in Proc. IEEE GLOBECOM’09, Dec. 2009, pp. 1 –6. → pages [92] S.-J. Kim, X. Wang, and M. Madihian, “Optimal resource allocation in multi-hop OFDMA wireless networks with cooperative relay,” IEEE Trans. Wireless Commun., vol. 7, no. 5, pp. 1833 –1838, May 2008. → pages 92, 93 [93] S. C. Jha, U. Phuyal, and V. K. Bhargava, “Joint power and subcarrier allocation in multi-hop OFDMA network: A cross-layer approach,” in Proc. AH-ICI’11, Nov. 2011, pp. 1 –6. → pages 93, 127 [94] A. Bavier, N. Feamster, M. Huang, L. Peterson, and J. Rexford, “In VINI veritas: realistic and controlled network experimentation,” ACM SIGCOMM Comput. Commun. Rev., vol. 36, pp. 3–14, Aug. 2006. → pages 109 [95] Y. Zhu, R. Zhang-shen, S. Rangarajan, and J. Rexford, “Cabernet: Connectivity architecture for better network services abstract,” in Proc. ACM ReArch08, Dec. 2008, pp. 1–6. → pages 109 [96] G. Schaffrath, C. Werle, P. Papadimitriou, A. Feldmann, R. Bless, A. Greenhalgh, A. Wundsam, M. Kind, O. Maennel, and L. Mathy, “Network virtualization architecture: proposal and initial prototype,” in Proc. ACM VISA’09, 2009, pp. 63–72. → pages 109 140  [97] S.-W. Ahn and C. Yoo, “Network interface virtualization in wireless communication for multi-streaming service,” in Symp. IEEE ISCE’11, Jun. 2011, pp. 67 –70. → pages 109 [98] S. Perez, J. Cabero, and E. Miguel, “Virtualization of the wireless medium: A simulation-based study,” in Proc. IEEE VTC’09-Spring, Apr. 2009, pp. 1 –5. → pages 110 [99] L. Correia, D. Zeller, O. Blume, D. Ferling, Y. Jading, I. G´odor, G. Auer, and L. Van Der Perre, “Challenges and enabling technologies for energy aware mobile radio networks,” IEEE Commun. Mag., vol. 48, no. 11, pp. 66–72, Nov. 2010. → pages 112, 113 [100] E. Oh, B. Krishnamachari, X. Liu, and Z. Niu, “Toward dynamic energyefficient operation of cellular network infrastructure,” IEEE Commun. Mag., vol. 49, no. 6, pp. 56 –61, Jun. 2011. → pages 119 [101] C. Han, T. Harrold, S. Armour, I. Krikidis, S. Videv, P. Grant, H. Haas, J. Thompson, I. Ku, C.-X. Wang, T. A. Le, M. Nakhai, J. Zhang, and L. Hanzo, “Green radio: radio techniques to enable energy-efficient wireless networks,” IEEE Commun. Mag., vol. 49, no. 6, pp. 46 –54, Jun. 2011. → pages 112, 119 [102] T. Chen, H. Kim, and Y. Yang, “Energy efficiency metrics for green wireless communications,” in Proc. WCSP’10, Oct. 2010, pp. 1–6. → pages 113 [103] Y. Zhao, L. Morales, J. Gaeddert, K. Bae, J.-S. Um, and J. Reed, “Applying radio environment maps to cognitive wireless regional area networks,” in Proc. IEEE DySPAN’07, Apr. 2007, pp. 115 –118. → pages 121 [104] J. Riihijarvi, P. Mahonen, M. Petrova, and V. Kolar, “Enhancing cognitive radios with spatial statistics: From radio environment maps to topology engine,” in Proc. CROWNCOM’09., Jun. 2009, pp. 1 –6. → pages 121, 122 [105] D.-Y. Seol, H.-J. Lim, and G.-H. Im, “Optimal threshold adaptation with radio environment map for cognitive radio networks,” in Symp. IEEE ISIT’09, Jul. 2009, pp. 2527 –2531. → pages 121  141  [106] U. Phuyal, S. C. Jha, and V. K. Bhargava, “Joint zero-forcing based precoder design for QoS-aware power allocation in MIMO cooperative cellular network,” IEEE J. Sel. Areas Commun., vol. 30, no. 2, pp. 350 –358, Feb. 2012. → pages 148 [107] R. Devarajan, S. C. Jha, U. Phuyal, and V. K. Bhargava, “Energy-aware resource allocation for cooperative cellular network using multi-objective optimization approach,” IEEE Trans. Wireless Commun., vol. 11, no. 5, pp. 1797 –1807, May 2012. → pages [108] U. Phuyal, S. C. Jha, and V. Bhargava, “Resource allocation for green communication in relay-based cellular networks,” in Green Radio Communication Networks, E. Hossain, V. K. Bhargava, and G. Fettweis, Eds. Cambridge: UK, 2012, pp. 331–356. → pages [109] C.-Y. Chiang, S. C. Jha, and V. K. Bhargava, “A superposition modulation based cooperative transmission scheme in regenerate-and-forward relaying networks,” in Proc. AH-ICI’11, Nov. 2011, pp. 1 –5. → pages [110] U. Phuyal, S. C. Jha, and V. K. Bhargava, “QoS guaranteed resource allocation in cooperative cellular network with MIMO-based relays,” in Proc. IEEE ICC 2011, june 2011, pp. 1 –6. → pages [111] R. Devarajan, S. C. Jha, U. Phuyal, and V. K. Bhargava, “Energy-aware user selection and power allocation for cooperative communication system with guaranteed quality-of-service,” in Proc. CWIT’11, May 2011, pp. 216 –220. → pages [112] U. Phuyal, S. Jha, and V. Bhargava, “Green resource allocation with qos provisioning for cooperative cellular network,” in Proc. CWIT’11, May 2011, pp. 206 –210. → pages [113] S. N. Pradhan, R. Devarajan, S. C. Jha, and V. Bhargava, “Uplink power allocation schemes in heterogeneous cellular networks,” in Proc. NCC’13, 2013, pp. 1–5. → pages [114] U. Phuyal, S. Jha, and V. Bhargava, “Optimal and suboptimal relay selection and power allocation in multi-relay cooperative network,” in Proc. AH-ICI’11, Nov. 2011, pp. 1 –6. → pages 148  142  Appendix A Summary of Major Notations Used in Chapter 3  143  Table A.1: Summary of major notations used in Chapter 3 Symbol  Description  Dˆ N¯  Maximum acceptable channel access delay for a prioritized data packet  Nch  Total number of PU data channels  N f /N f p /N f g N¯ p /N¯ g  Number of total/prioritized/non-prioritized (general) users  Nsucc  Maximum number of successful message exchanges per BI  P  Average channel availability probability for channels  P0  Probability of being at stage 0 of Markov chain  Pa /Pna  Probability of getting/not getting Channel access per BI in no QoS case  Pap /Pag  Probability of getting Channel access per BI for prioritized/general users  Pc  Collision probability of SUs with PU  Pe  Average sensing error (miss detection) probability  Pnap /Pnag  Probability of not getting Channel access per BI for prioritized/general users  Psat /Pus  probability of a prioritized user being QoS satisfied/unsatisfied  Pt S¯  Control message transmission probability by an SU in a randomly chosen slot  S¯im  Improved average normalized saturation throughput of the system  TBI  Length of beacon period  Tcon  Length of contention phase  Tconp /Tcong  Length of prioritized/non-prioritized contention phase  TDT  Length of data transmission phase  Tsensing  Length of sensing phase  Tsi  Average time taken to reserve a single channel during contention phase  Tsip  Average channel reservation time during Tconp by prioritized users  ⌊x⌋  Integer part of x  x/ ˆ x/ ¯ xˇ  Maximum (upper bound)/average/minimum (lower bound) value of a variable x  Average number of channels reserved  Average number of channel reservations during Tconp and Tcong  Average normalized saturation throughput of the system  144  Appendix B Solution of P4 When XN ,K is Known When utility function U(·) of problem P4 is defined as U1 (PN ,K ) in (4.18) with given XN ,K , the lagrangian of problem P4 can be given as [64] ( ) ( ) L (PN ,K , ν , µ N ,K ) = −U1 PN ,K + ν 1T PN ,K 1 − Pt ) ( − tr PN ,K µ TN ,K ,  (B.1) (B.2)  where ν and µn,k ∈ µ N ,K , ∀n ∈ N , k ∈ K are known as Lagrange multipliers. ˜ the KKT Representing the value of a variable at optimal solution point by (·), conditions are given by 1T P˜ N ,K 1 ≤ Pt , P˜ N ,K ≽ 0,  (B.3)  ν˜ ≥ 0, µ˜ N ,K ≽ 0, ) ν˜ 1T P˜ N ,K 1 − Pt = 0, (  µ˜ n,k P˜n,k = 0,  145  ∀n ∈ N , k ∈ K ,  (B.4) (B.5) (B.6)  −  ∂ U1 ∂ Pn,k  Pn,k =P˜n,k  + ν˜ − µ˜ n,k = 0,  ∀n ∈ N , k ∈ K .  (B.7)  After some simplification, (B.7) can be re-written as  ν˜ = µ˜ n,k +  βk αn,k xn,k , 1 + βk αn,k xn,k P˜n,k  ∀n ∈ N , k ∈ K ,  (B.8)  and using (B.4), this can further be reduced to  ν˜ ≥  βk αn,k xn,k , 1 + βk αn,k xn,k P˜n,k  ∀n ∈ N , k ∈ K .  (B.9)  In practical communication systems, βk > 0 and αn,k and xn,k cannot be zero, ∀n ∈ N , k ∈ K . Therefore, ν˜ > 0, and using (B.5), 1T P˜ N ,K 1 = Pt ,  (B.10)  i.e., constraint (4.17) becomes tight at the optimal point. Let us first consider the case when ν˜ < βk αn,k xn,k . In this case, inequality (B.9) holds only when P˜n,k > 0. Therefore, from (B.6), µ˜ n,k = 0. Using this on (B.8), we can write 1 1 P˜n,k = − , ˜ν βk αn,k xn,k  ∀n ∈ N , k ∈ K .  (B.11)  Note that value of P˜n,k obtained from (B.11) is always nonnegative satisfying (B.3). Now, consider the case when ν˜ > βk αn,k xn,k . Under this condition, from (B.8), µ˜ n,k > 0. From (B.6), this means P˜n,k = 0 for this condition. Similarly, when ν˜ = βk αn,k xn,k , we get µ˜ n,k = 0 and P˜n,k = 0 from (B.8). Combining these results, we obtain  P˜n,k =     1 ν˜   0  −β  1  k αn,k xn,k  if ν˜ < βk αn,k xn,k otherwise,  146  which can be re-written as (4.19). Finally, as the calculated values of P˜n,k depend on ν˜ , which is chosen such that (B.10) is satisfied.  147  Appendix C Other Publications Resulted from Collaboration Work The publications [106]–[114] which have been resulted as a result of collaboration in various projects are as follows: • Phuyal, U., S. C. Jha and V. K. Bhargava, “Green communication by cooperative transmission in cellular networks” in Green Radio Communication Networks, E. Hossain, V.K. Bhargava, and G. Fettweis (Eds.), Cambridge University Press, UK, July 2012. • Devarajan, R., S. C. Jha, U. Phuyal and V. K. Bhargava, “Energy-Aware Resource Allocation for Cooperative Cellular Network: A Multi-Objective Optimization Based Approach,” IEEE Trans. on Wireless Commun., pp. 1797-1807 May 2012. • Phuyal, U., S. C. Jha and V. K. Bhargava, “Joint Zero-Forcing Based Precoder Design for QoS-Aware Power Allocation in MIMO Cooperative Cellular Network,” IEEE J. Sel. Areas Commun., pp. 350-358, Jan. 2012. • Phuyal, U., S. C. Jha and V. K. Bhargava, “Optimal and Suboptimal Relay  148  Selection and Power Allocation in Multi-relay Cooperative Network,” presented in IEEE/IFIP AH-ICI2011, Kathmandu, Nepal, pp. 1-5, Nov. 2011. • Chiang, C-Y, S. C. Jha and V. K. Bhargava, “A Superposition Modulation Based Cooperative Transmission Scheme in Regenerate-and-Forward Relaying Networks,” presented in IEEE/IFIP AH-ICI2011, Kathmandu, Nepal, pp. 1-5, Nov. 2011. • Phuyal, U., S. C. Jha and V. K. Bhargava, “Green Resource Allocation With QoS Provisioning for Cooperative Cellular Network,” in Proceedings of the IEEE CWIT’11, Kelowna, BC, Canada, pp. 206-210, May 2011. • Devarajan, R., S. C. Jha, U. Phuyal and V. K. Bhargava, “Energy-Aware User Selection and Power Allocation for Cooperative Communications System with Guaranteed Quality-of-Service,” in Proceedings of the IEEE CWIT’11, Kelowna, BC, Canada, pp. 216-220, May 2011. • Phuyal, U., S. C. Jha and V. K. Bhargava, “QoS Guaranteed Resource Allocation in Cooperative Cellular Network with MIMO-based Relays,” in proceedings of the IEEE ICC 2011, Kyoto, Japan, pp. 1-6, Jun. 2011. • Pradhan, S. N., R. Devarajan, S. C. Jha, and V. K. Bhargava, Uplink Power Allocation Schemes in Heterogeneous Cellular Networks, Accepted, NCC’13, Delhi, India, Feb. 2013.  149  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0071927/manifest

Comment

Related Items