ADVANCED WIRELESS NETWORKS Cognitive, Cooperative and Opportunistic 4G Technology Second Edition Savo Glisic Beatriz Lorenzo University of Oulu, Finland A John Wiley and Sons, Ltd., Publication

ADVANCED WIRELESS NETWORKS

ADVANCED WIRELESS NETWORKS Cognitive, Cooperative and Opportunistic 4G Technology Second Edition Savo Glisic Beatriz Lorenzo University of Oulu, Finland A John Wiley and Sons, Ltd., Publication

This edition rst published 2009 C 2009 John Wiley & Sons Ltd., Registered ofce John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United Kingdom For details of our global editorial ofces, for customer services and for information about how to apply for permission to reuse the copyright material in this book please see our website at www.wiley.com. The right of the author to be identied as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher. Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books. Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book. This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold on the understanding that the publisher is not engaged in rendering professional services. If professional advice or other expert assistance is required, the services of a competent professional should be sought. Library of Congress Cataloging-in-Publication Data Glisic, Savo G. Advanced wireless networks : 4G technologies / Savo Glisic, Beatriz Lorenzo Veiga. 2nd ed. p. cm. Includes bibliographical references and index. ISBN 978-0-470-74250-1 (cloth) 1. Wireless communication systems. I. Veiga, Beatriz Lorenzo. II. Title. TK5103.2.G553 2009 621.384dc22 2009001817 A catalogue record for this book is available from the British Library. ISBN 978-0-470-74250-1 (H/B) Typeset in 9/11 Times by Laserwords Private Limited, Chennai, India Printed in Singapore by Markono Print Media Pte Ltd

To our families

Contents Preface to the Second Edition xix 1 Fundamentals 1.1 4G Networks and Composite Radio Environment 1.2 Protocol Boosters 1.2.1 One-element error detection booster for UDP 1.2.2 One-element ACK compression booster for TCP 1.2.3 One-element congestion control booster for TCP 1.2.4 One-element ARQ booster for TCP 1.2.5 A forward erasure correction booster for IP or TCP 1.2.6 Two-element jitter control booster for IP 1.2.7 Two-element selective ARQ booster for IP or TCP 1.3 Green Wireless Networks References 1 1 7 9 9 9 9 10 10 11 11 11 2 Opportunistic Communications 2.1 Multiuser Diversity 2.2 Proportional Fair Scheduling 2.3 Opportunistic Beamforming 2.4 Opportunistic Nulling in Cellular Systems 2.5 Network Cooperation and Opportunistic Communications 2.5.1 Performance example 2.6 Multiuser Diversity in Wireless Ad Hoc Networks 2.6.1 Multiple-output and multiple-input link diversity 2.6.2 Localized opportunistic transmission 2.6.3 Multiuser diversity-driven clustering 2.6.4 Opportunistic MAC with timeshare fairness 2.6.5 CDF-based K-ary opportunistic splitting algorithm 2.6.6 Throughput 2.6.7 Optimal opportunistic MAC 15 15 16 19 20 22 25 27 29 30 31 34 34 37 37

viii CONTENTS 2.7 2.8 2.6.8 Contention resolution between clusters 2.6.9 Performance examples Mobility-Assisted Opportunistic Scheduling (MAOS) 2.7.1 Mobility models 2.7.2 Optimal MAOS algorithm 2.7.3 Suboptimum MAOS algorithm 2.7.4 Mobility estimation and prediction 2.7.5 Estimation of Lagrange multipliers 2.7.6 Performance examples Opportunistic and Cooperative Cognitive Wireless Networks 2.8.1 The system model 2.8.2 The outage probability 2.8.3 Cellular trafc shaping 2.8.4 User mobility modeling 2.8.5 Absorbing Markov chain system model 2.8.6 Throughput analysis 2.8.7 Collision resolution 2.8.8 Opportunistic transmission with intercell interference awareness 2.8.9 Performance examples References 38 40 46 48 49 51 51 52 52 53 53 57 58 59 61 62 65 65 68 70 3 Relaying and Mesh Networks 3.1 Relaying Strategies in Cooperative Cellular Networks 3.1.1 The system model 3.1.2 System optimization 3.1.3 Relay strategy selection optimization 3.1.4 Performance example 3.2 Mesh/Relay Networks 3.2.1 The system model 3.2.2 Exhaustive sleep 3.2.3 Practical applications 3.2.4 Performance example 3.3 Opportunistic Ad Hoc Relaying For Multicast 3.3.1 The system model 3.3.2 Proxy discovery and route interference 3.3.3 Near-optimal multicast and approximations 3.3.4 Performance examples References 73 73 73 75 79 84 85 86 88 94 95 97 98 99 101 103 107 4 Topology Control 4.1 Local Minimum Spanning Tree (LMST) Topology Control 4.1.1 Basics of MST topology control 4.1.2 Performance examples 4.2 Joint Topology Control, Resource Allocation and Routing 4.2.1 JTCR algorithm 4.3 Fault-Tolerant Topology 4.3.1 The system model 4.3.2 Fault-tolerant topology design 4.3.3 -Approximation algorithms 4.3.4 Performance examples 113 115 115 118 118 121 123 124 124 127 132

CONTENTS 4.4 ix Topology Control in Directed Graphs 4.4.1 The system model 4.4.2 Minimum-weight-based algorithms 4.4.3 Augmentation-based algorithms 4.4.4 Performance examples Adjustable Topology Control 4.5.1 The system model 4.5.2 The r -neighborhood graph Self-Conguring Topologies 4.6.1 SCT performance References 132 133 133 135 138 138 140 142 143 145 148 5 Adaptive Medium Access Control 5.1 WLAN Enhanced Distributed Coordination Function 5.2 Adaptive MAC for WLAN with Adaptive Antennas 5.2.1 Description of the protocols 5.3 MAC for Wireless Sensor Networks 5.3.1 S-MAC protocol design 5.3.2 Periodic listen and sleep 5.3.3 Collision avoidance 5.3.4 Coordinated sleeping 5.3.5 Choosing and maintaining schedules 5.3.6 Maintaining synchronization 5.3.7 Adaptive listening 5.3.8 Overhearing avoidance and message passing 5.3.9 Overhearing avoidance 5.3.10 Message passing 5.4 MAC for Ad Hoc Networks 5.4.1 Carrier sense wireless networks 5.4.2 Interaction with upper layers References 157 157 160 160 166 167 168 168 169 169 170 170 172 172 172 174 176 179 180 6 Teletrafc Modeling and Analysis 6.1 Channel Holding Time in PCS Networks References 183 183 191 7 Adaptive Network Layer 7.1 Graphs and Routing Protocols 7.1.1 Elementary concepts 7.1.2 Directed graph 7.1.3 Undirected graph 7.1.4 Degree of a vertex 7.1.5 Weighted graph 7.1.6 Walks and paths 7.1.7 Connected graphs 7.1.8 Trees 7.1.9 Spanning tree 7.1.10 MST computation 7.1.11 Shortest path spanning tree 7.2 Graph Theory 193 193 193 193 194 194 195 195 195 196 197 199 201 212 4.5 4.6

x CONTENTS 7.3 7.4 Routing with Topology Aggregation Network and Aggregation Models 7.4.1 Line segment representation 7.4.2 QoS-aware topology aggregation 7.4.3 Mesh formation 7.4.4 Star formation 7.4.5 Line-segment routing algorithm 7.4.6 Performance measure 7.4.7 Performance example References 214 215 217 220 220 221 222 224 225 228 8 Effective Capacity 8.1 Effective Trafc Source Parameters 8.1.1 Effective trafc source 8.1.2 Shaping probability 8.1.3 Shaping delay 8.1.4 Performance example 8.2 Effective Link Layer Capacity 8.2.1 Link-layer channel model 8.2.2 Effective capacity model of wireless channels 8.2.3 Physical layer vs link-layer channel model 8.2.4 Performance examples References 235 235 237 238 238 241 243 244 246 249 251 254 9 Adaptive TCP Layer 9.1 Introduction 9.1.1 A large bandwidth-delay product 9.1.2 Buffer size 9.1.3 Round-trip time 9.1.4 Unfairness problem at the TCP layer 9.1.5 Noncongestion losses 9.1.6 End-to-end solutions 9.1.7 Bandwidth asymmetry 9.2 TCP Operation and Performance 9.2.1 The TCP transmitter 9.2.2 Retransmission timeout 9.2.3 Window adaptation 9.2.4 Packet loss recovery 9.2.5 TCP-OldTahoe (timeout recovery) 9.2.6 TCP-Tahoe (fast retransmit) 9.2.7 TCP-Reno fast retransmit, fast (but conservative) recovery 9.2.8 TCP-NewReno (fast retransmit, fast recovery) 9.2.9 Spurious retransmissions 9.2.10 Modeling of TCP operation 9.3 TCP for Mobile Cellular Networks 9.3.1 Improving TCP in mobile environments 9.3.2 Mobile TCP design 9.3.3 The SH-TCP client 9.3.4 The M-TCP protocol 9.3.5 Performance examples 257 257 258 259 260 261 262 262 263 264 264 265 265 265 265 265 265 266 267 267 268 269 270 272 273 275

CONTENTS 9.4 9.5 Random Early Detection Gateways for Congestion Avoidance 9.4.1 The RED algorithm 9.4.2 Performance example TCP for Mobile Ad Hoc Networks 9.5.1 Effect of route recomputations 9.5.2 Effect of network partitions 9.5.3 Effect of multipath routing 9.5.4 ATCP sublayer 9.5.5 ATCP protocol design 9.5.6 Performance examples References 10 Network Optimization Theory 10.1 Introduction 10.2 Layering as Optimization Decomposition 10.2.1 TCP congestion control 10.2.2 TCP Reno/RED 10.2.3 TCP Vegas/Drop Tail 10.2.4 Optimization of the MAC protocol 10.2.5 Utility optimal MAC protocol/social optimum 10.3 Crosslayer Optimization 10.3.1 Congestion control and routing 10.3.2 Congestion control and physical resource allocation 10.3.3 Congestion and contention control 10.3.4 Congestion control, routing and scheduling 10.4 Optimization Problem Decomposition Methods 10.4.1 Decoupling coupled constraints 10.4.2 Dual decomposition of the basic NUM 10.4.3 Coupling constraints 10.4.4 Decoupling coupled objectives 10.4.5 Alternative decompositions 10.4.6 Application example of decomposition techniques to distributed crosslayer optimization 10.5 Optimization of Distributed Rate Allocation for Inelastic Utility Flows 10.5.1 Nonconcave utility ows 10.5.2 Capacity provisioning for convergence of the basic algorithm 10.6 Nonconvex Optimization Problem in Network with QoS Provisioning 10.6.1 The system model 10.6.2 Solving the nonconvex optimization problem for joint congestioncontention control 10.7 Optimization of Layered Multicast by Using Integer and Dynamic Programming 10.7.1 The system model 10.7.2 Lagrangian relaxation for integer programs 10.7.3 Group prot maximization by dynamic programming 10.8 QoS Optimization in Time-Varying Channels 10.8.1 The system model 10.8.2 Dynamic control algorithm 10.9 Network Optimization by Geometric Programming 10.9.1 Power control by geometric programming: high SNR 10.9.2 Power control by geometric programming: low SNR 10.10 QoS Scheduling by Geometric Programming xi 276 276 277 280 280 280 280 281 282 287 287 289 289 290 290 291 292 292 295 298 298 301 303 306 307 307 308 310 310 313 315 319 319 322 323 323 325 326 327 329 329 331 331 332 337 338 340 340

xii CONTENTS 10.10.1 Optimization of OFDM system by GP 10.10.2 Maximum weight matching scheduling by GP 10.10.3 Opportunistic scheduling by GP 10.10.4 Rescue scheduling by GP References 344 344 345 345 346 11 Mobility Management 11.1 Introduction 11.1.1 Mobility management in cellular networks 11.1.2 Location registration and call delivery in 4G 11.2 Cellular Systems with Prioritized Handoff 11.2.1 Channel assignment priority schemes 11.2.2 Channel reservation CR handoffs 11.2.3 Channel reservation with queueing CRQ handoffs 11.2.4 Performance examples 11.3 Cell Residing Time Distribution 11.4 Mobility Prediction in Pico- and MicroCellular Networks 11.4.1 PST-QoS guarantees framework 11.4.2 Most likely cluster model Appendix: Distance Calculation in an Intermediate Cell References 351 351 353 355 374 377 377 378 382 383 388 390 391 398 403 12 Cognitive Radio Resource Management 12.1 Channel Assignment Schemes 12.1.1 Different channel allocation schemes 12.1.2 Fixed channel allocation 12.1.3 Channel borrowing schemes 12.1.4 Simple channel borrowing schemes 12.1.5 Hybrid channel borrowing schemes 12.1.6 Dynamic channel allocation 12.1.7 Centralized DCA schemes 12.1.8 Cell-based distributed DCA schemes 12.1.9 Signal strength measurement-based distributed DCA schemes 12.1.10 One-dimensional cellular systems 12.1.11 Reuse partitioning (RUP) 12.2 Dynamic Channel Allocation with SDMA 12.2.1 Single-cell environment 12.2.2 Resource allocation 12.2.3 Performance examples 12.3 Packet-Switched SDMA/TDMA Networks 12.3.1 The system model 12.3.2 Multibeam SDMA/TDMA capacity and slot allocation 12.3.3 SDMA/TDMA slot allocation algorithms 12.3.4 SDMA/TDMA performance examples 12.4 SDMA/OFDM Networks with Adaptive Data Rate 12.4.1 The system model 12.4.2 Resource allocation algorithm 12.4.3 Impact of OFDM/SDMA system specications on resource allocations 12.4.4 Performance examples 12.5 Intercell Interference Cancellation SP Separability 407 407 409 410 410 411 412 414 415 417 419 420 422 426 426 430 435 435 437 439 441 445 446 446 448 450 453 454

CONTENTS 12.6 12.7 12.8 12.9 12.5.1 Channel and cellular system model 12.5.2 Turbo spacetime multiuser detection for intracell communications 12.5.3 Multiuser detection in the presence of intercell interference 12.5.4 Performance examples Intercell Interference Avoidance in SDMA Systems 12.6.1 The BOW scheme 12.6.2 Generating beam-off sequences 12.6.3 Constrained QRA-IA Multilayer RRM 12.7.1 The SRA protocol 12.7.2 The ESRA protocol Resource Allocation with Power Preassignment (RAPpA) 12.8.1 Resource assignment protocol 12.8.2 Analytical modeling of RAPpA Cognitive and Cooperative Dynamic Radio Resource Allocation 12.9.1 Signal-to-interference ratio 12.9.2 System performance 12.9.3 Multicell operation 12.9.4 Performance examples Appendix 12A: Power Control, CD Protocol, in the Presence of Fading Appendix 12B: Average Intercell Throughput References 13 Ad Hoc Networks 13.1 Routing Protocols 13.1.1 Routing protocols 13.1.2 Reactive protocols 13.2 Hybrid routing protocol 13.2.1 Loop-back termination 13.2.2 Early termination 13.2.3 Selective broadcasting (SBC) 13.3 Scalable Routing Strategies 13.3.1 Hierarchical routing protocols 13.3.2 Performance examples 13.3.3 FSR (sheye routing) protocol 13.4 Multipath Routing 13.5 Clustering Protocols 13.5.1 Introduction 13.5.2 Clustering algorithm 13.5.3 Clustering with prediction 13.6 Cashing Schemes for Routing 13.6.1 Cache management 13.7 Distributed QoS Routing 13.7.1 Wireless links reliability 13.7.2 Routing 13.7.3 Routing information 13.7.4 Token-based routing 13.7.5 Delay-constrained routing 13.7.6 Tokens 13.7.7 Forwarding the received tokens 13.7.8 Bandwidth-constrained routing xiii 455 457 459 460 461 467 468 468 470 471 473 475 476 479 484 486 488 491 492 494 498 499 505 505 507 512 524 526 527 528 531 531 533 534 537 539 539 541 542 549 549 558 558 558 559 559 560 561 562 562

xiv CONTENTS 13.7.9 Forwarding the received tickets 13.7.10 Performance example References 562 564 567 14 Sensor Networks 14.1 Introduction 14.2 Sensor Networks Parameters 14.2.1 Pre-deployment and deployment phase 14.2.2 Post-deployment phase 14.2.3 Re-deployment of additional nodes phase 14.3 Sensor networks architecture 14.3.1 Physical layer 14.3.2 Data link layer 14.3.3 Network layer 14.3.4 Transport layer 14.3.5 Application layer 14.4 Mobile Sensor Networks Deployment 14.5 Directed Diffusion 14.5.1 Data propagation 14.5.2 Reinforcement 14.6 Aggregation in Wireless Sensor Networks 14.7 Boundary Estimation 14.7.1 Number of RDPs in P 14.7.2 Kraft inequality 14.7.3 Upper bounds on achievable accuracy 14.7.4 System optimization 14.8 Optimal Transmission Radius in Sensor Networks 14.8.1 Back-off phenomenon 14.9 Data Funneling 14.10 Equivalent Transport Control Protocol in Sensor Networks References 573 573 575 576 576 577 577 578 578 581 585 586 587 590 591 593 593 596 598 598 599 600 602 606 607 610 613 15 Security 15.1 Authentication 15.1.1 Attacks on simple cryptographic authentication 15.1.2 Canonical authentication protocol 15.2 Security Architecture 15.3 Key Management 15.3.1 Encipherment 15.3.2 Modication detection codes 15.3.3 Replay detection codes 15.3.4 Proof of knowledge of a key 15.3.5 Point-to-point key distribution 15.4 Security management in GSM networks 15.5 Security management in UMTS 15.6 Security architecture for UMTS/WLAN Interworking 15.7 Security in Ad Hoc Networks 15.7.1 Self-organized key management 15.8 Security in Sensor Networks References 623 623 625 629 631 635 637 637 637 637 638 639 643 645 647 651 652 654

CONTENTS xv 16 Active Networks 16.1 Introduction 16.2 Programable Networks Reference Models 16.2.1 IETF ForCES 16.2.2 Active networks reference architecture 16.3 Evolution to 4G Wireless Networks 16.4 Programmable 4G Mobile Network Architecture 16.5 Cognitive Packet Networks 16.5.1 Adaptation by cognitive packets 16.5.2 The random neural networks-based algorithms 16.6 Game Theory Models in Cognitive Radio Networks 16.6.1 Cognitive radio networks as a game 16.7 Biologically Inspired Networks 16.7.1 Bio-analogies 16.7.2 Bionet architecture References 659 659 661 662 662 665 667 670 672 673 675 678 682 682 684 686 17 Network Deployment 17.1 Cellular Systems with Overlapping Coverage 17.2 Imbedded Microcell in CDMA Macrocell Network 17.2.1 Macrocell and microcell link budget 17.2.2 Performance example 17.3 Multitier Wireless Cellular Networks 17.3.1 The network model 17.3.2 Performance example 17.4 Local Multipoint Distribution Service 17.4.1 Interference estimations 17.4.2 Alternating polarization 17.5 Self-Organization in 4G Networks 17.5.1 Motivation 17.5.2 Networks self-organizing technologies References 693 693 698 699 702 703 704 708 709 711 711 713 713 715 717 18 Network Management 18.1 The Simple Network Management Protocol 18.2 Distributed Network Management 18.3 Mobile Agent-Based Network Management 18.3.1 Mobile agent platform 18.3.2 Mobile agents in multioperator networks 18.3.3 Integration of routing algorithm and mobile agents 18.4 Ad Hoc Network Management 18.4.1 Heterogeneous environments 18.4.2 Time varying topology 18.4.3 Energy constraints 18.4.4 Network partitioning 18.4.5 Variation of signal quality 18.4.6 Eavesdropping 18.4.7 Ad hoc network management protocol functions 18.4.8 ANMP architecture References 721 721 725 726 728 728 730 735 735 735 736 736 736 736 736 738 743

xvi CONTENTS 19 Network Information Theory 19.1 Effective Capacity of Advanced Cellular Networks 19.1.1 4G cellular network system model 19.1.2 The received signal 19.1.3 Multipath channel: nearfar effect and power control 19.1.4 Multipath channel: pointer tracking error, rake receiver and interference canceling 19.1.5 Interference canceler modeling: nonlinear multiuser detectors 19.1.6 Approximations 19.1.7 Outage probability 19.2 Capacity of Ad Hoc Networks 19.2.1 Arbitrary networks 19.2.2 Random networks 19.2.3 Arbitrary networks: an upper bound on transport capacity 19.2.4 Arbitrary networks: lower bound on transport capacity 19.2.5 Random networks: lower bound on throughput capacity 19.3 Information Theory and Network Architectures 19.3.1 Network architecture 19.3.2 Denition of feasible rate vectors 19.3.3 The transport capacity 19.3.4 Upper bounds under high attenuation 19.3.5 Multihop and feasible lower bounds under high attenuation 19.3.6 The low-attenuation regime 19.3.7 The Gaussian multiple-relay channel 19.4 Cooperative Transmission in Wireless Multihop Ad Hoc Networks 19.4.1 Transmission strategy and error propagation 19.4.2 OLA ooding algorithm 19.4.3 Simulation environment 19.5 Network Coding 19.5.1 Max-ow min-cut theorem (mfmcT) 19.5.2 Achieving the max-ow bound through a generic LCM 19.5.3 The transmission scheme associated with an LCM 19.5.4 Memoryless communication network 19.5.5 Network with memory 19.5.6 Construction of a generic LCM on an acyclic network 19.5.7 Time-invariant LCM and heuristic construction 19.6 Capacity of Wireless Networks Using MIMO Technology 19.6.1 Capacity metrics 19.7 Capacity of Sensor Networks with Many-to-One Transmissions 19.7.1 Network architecture 19.7.2 Capacity results References 747 747 749 750 752 753 755 757 757 761 762 764 765 768 769 773 773 775 776 776 777 778 779 780 783 784 784 787 788 789 792 793 794 794 795 798 800 805 805 807 809 20 Energy-efcient Wireless Networks 20.1 Energy Cost Function 20.2 Minimum Energy Routing 20.3 Maximizing Network Lifetime 20.4 Energy-efcient MAC in Sensor Networks 20.4.1 Staggered wakeup schedule References 813 813 815 816 821 821 823

CONTENTS xvii 21 Quality-of-Service Management 21.1 Blind QoS Assessment System 21.1.1 System modeling 21.2 QoS Provisioning in WLAN 21.2.1 Contention-based multipolling 21.2.2 Polling efciency 21.3 Dynamic Scheduling on RLC/MAC Layer 21.3.1 DSMC functional blocks 21.3.2 Calculating the high service rate 21.3.3 Heading-block delay 21.3.4 Interference model 21.3.5 Normal delay of a newly arrived block 21.3.6 High service rate of a session 21.4 QoS in OFDMA-Based Broadband Wireless Access Systems 21.4.1 Iterative solution 21.4.2 Resource allocation to maximize capacity 21.5 Predictive Flow Control and QoS 21.5.1 Predictive ow control model References 827 827 829 831 831 832 835 837 838 840 841 841 842 842 846 848 849 850 854 Index 859

Preface to the Second Edition Although the rst edition of the book was not published long ago, a constant progress in research in the eld of wireless networks has resulted in a signicant accumulation of new results that urge the extension and modication of its content. The major additions in the book are the following new chapters: Chapter 1: Fundamentals, Chapter 2: Opportunistic Communications, Chapter 3: Relaying and Mesh Networks, Chapter 4: Topology Control, Chapter 10: Network Optimization and Chapter 12: Cognitive Radio Resource Management. OPPORTUNISTIC COMMUNICATIONS Multiuser diversity is a form of diversity inherent in a wireless network, provided by independent time-varying channels across the different users. The diversity benet is exploited by tracking the channel uctuations of the users and scheduling transmissions to users when their instantaneous channel quality is near the peak. The diversity gain increases with the dynamic range of the uctuations and is thus limited in environments with little scattering and/or slow fading. In such environments, the multiple transmit antennas can be used to induce large and fast channel uctuations so that multiuser diversity can still be exploited. The scheme can be interpreted as opportunistic beamforming and true beamforming gains can be achieved when there are sufcient users, even though very limited channel feedback is needed. Furthermore, in a cellular system, the scheme plays an additional role of opportunistic nulling of the interference created on users of adjacent cells. This chapter discusses the design implications of implementing this scheme in a wireless system. RELAYING AND MESH NETWORKS In a wireless network with many sourcedestination pairs, cooperative transmission by relay nodes has the potential to improve the overall network performance. In a distributed multihop mesh/relay network (e.g. wireless ad hoc/sensor network, cellular multihop network), each node acts as a relay node to forward data packets from other nodes. These nodes are often energy-limited and also have limited buffer space. Therefore, efcient power-saving mechanisms (e.g. sleeping mechanisms) are

xx PREFACE TO THE SECOND EDITION required so that the lifetime of these nodes can be extended while at the same time the quality of service (QoS) requirements (e.g. packet delay and packet loss rate) for the relayed packets can be satised. In Chapter 3, a queuing analytical framework is presented to study the tradeoffs between the energy saving and the QoS at a relay node as well as relaying strategies in cooperative cellular networks. In addition integrated cellular and ad hoc multicast, which increases multicast throughput through opportunistic use of ad hoc relays, is also discussed. NETWORK TOPOLOGY CONTROL Energy efciency and network capacity are perhaps two of the most important issues in wireless ad hoc networks and sensor networks. Topology control algorithms have been proposed to maintain network connectivity while reducing energy consumption and improving network capacity. The key idea to topology control is that, instead of transmitting with maximal power, nodes in a wireless multihop network collaboratively determine their transmission power and dene the network topology by forming the proper neighbour relation under certain criteria. The topology control affects network spatial reuse and contention for the medium. A number of topology control algorithms have been proposed to create a power-efcient network topology in wireless multihop networks with limited mobility. In Chapter 4, we summarize existing work in this eld. Some of the algorithms require explicit propagation channel models, while others incur signicant message exchanges. Their ability to maintain the topology in the case of mobility is also rather limited. The chapter will discuss the tradeoffs between these opposing requirements. NETWORK OPTIMIZATION Network protocols in layered architectures have traditionally been obtained on an ad hoc basis, and many of the recent crosslayer designs are also conducted through piecemeal approaches. Network protocol stacks may instead be systematically analyzed and designed as distributed solutions to some global optimization problems. Chapter 10 presents a survey of the recent efforts toward a systematic understanding of layering as optimization decomposition, where the overall communication network is modelled by a generalized network utility maximization problem, where each layer corresponds to a decomposed subproblem and the interfaces among layers are quantied as functions of the optimization variables coordinating the subproblems. There can be many alternative decompositions, leading to a choice of different layering architectures. This chapter will survey the current status of horizontal decomposition into distributed computation and vertical decomposition into functional modules such as congestion control, routing, scheduling, random access, power control and channel coding. Key results are summarized and open issues discussed. Through case studies, it is illustrated how layering as optimization decomposition provides a common language to modularization, a unifying, top-down approach to design protocol stacks and a mathematical theory of network architectures. COGNITIVE RADIO RESOURCE MANAGEMENT Network optimization, including radio resource management, discussed in Chapter 10, provides algorithms that optimize system performance dened by a given utility function. In Chapter 12, we present suboptimum solutions for resource management that include high level of cognition and cooperation to mitigate intercell interference. An important segment of this topic dealing with the

PREFACE TO THE SECOND EDITION xxi exible spectra sharing is covered in another of our books on Advanced Wireless Communications focusing more on the physical layer, published by John Wiley & Sons, Ltd in 2007. In addition to the new chapters, which represent about 40 % of the book, other chapters have been also updated with latest results. Savo Glisic Beatriz Lorenzo

1 Fundamentals 1.1 4G NETWORKS AND COMPOSITE RADIO ENVIRONMENT In the wireless communications community we are witnessing more and more the existence of the composite radio environment (CRE ) and as a consequence the need for recongurability concepts based on cognitive, cooperative and opportunistic algorithms. The CRE assumes that different radio networks can be cooperating components in a heterogeneous wireless access infrastructure, through which network providers can more efciently achieve the required capacity and quality of service (QoS) levels. Recongurability enables terminals and network elements dynamically to select and adapt to the most appropriate radio access technologies for handling conditions encountered in specic service area regions and time zones of the day. Both concepts pose new requirements on the management of wireless systems. Nowadays, a multiplicity of radio access technology (RAT) standards are used in wireless communications. As shown in Figure 1.1, these technologies can be roughly categorized into four sets: Cellular networks that include second-generation (2G) mobile systems, such as Global System for Mobile Communications (GSM) [1], and their evolutions, often called 2.5G systems, such as enhanced digital GSM evolution (EDGE), General Packet Radio Service (GPRS) [2] and IS 136 in the US. These systems are based on TDMA technology. Third-generation (3G) mobile networks, known as Universal Mobile Telecommunications Systems (UMTS) (WCDMA and cdma2000) [3] are based on CDMA technology that provides up to 2 Mbit/s. Long-term evolution (LTE) [412] of these systems is expected to evolve into a 4G system providing up to 100 Mbit/s on the uplink and up to 1 Gbit/s on the downlink. The solutions will be based on a combination of multicarrier and spacetime signal formats. The network architectures include macro, micro and pico cellular networks and home (HAN) and personal area networks (PAN). Broadband radio access networks (BRANs) [13] or wireless local area networks (WLANs) [14] which are expected to provide up to 1 Gbit/s in 4G. These technologies are based on OFDMA and spacetime coding. Digital video broadcasting (DVB) [15] and satellite communications. Ad hoc and sensor networks with emerging applications. Advanced Wireless Technologies: Cognitive, Cooperative & Opportunistic 4G Technology Second Edition Savo G. Glisic C 2009 John Wiley & Sons, Ltd.

2 FUNDAMENTALS Sensor networks (self configur ation) PLMN PSTN Ad hoc networks IP Network (mobile user agents) Private Network Cellular multihop network macro/micro/ Pico/PAN/BAN Space-time frequency coding (100Mb) Cellular multihop network Access Network Reconfiguration & Dynamic Spectra Allocation BRAN/ WLAN/mesh Access Space-timefrequency coding, (1Gbit) DVB satellite Reconfigurable Mobile Terminals (Cognitive, Cooperative and Opportunistic) Figure 1.1 Composite radio environment in cognitive, cooperative and opportunistic 4G networks. In order to increase the spectral efciency further, besides the spacetime frequency coding in the physical layer, the new paradigms like cognitive [1620], cooperative [2132] and opportunistic [3338] solutions will be used. Although 4G is open for new multiple access schemes, the CRE concept remains attractive for increasing the service provision efciency and the exploitation possibilities of the available RATs. The main assumption is that the different radio networks, GPRS, UMTS, BRAN/WLAN, DVB and so on, can be components of a heterogeneous wireless access infrastructure. A network provider (NP) can own several components of the CR infrastructure (in other words, can own licenses for deploying and operating different RATs), and can also cooperate with afliated NPs. In any case, an NP can rely on several alternate radio networks and technologies, for achieving the required capacity and QoS levels, in a cost-efcient manner. Users are directed to the most appropriate radio networks and technologies, at different service area regions and time zones of the day, based on prole requirements and network performance criteria. The various RATs are thus used in a

4G NETWORKS AND COMPOSITE RADIO ENVIRONMENT 3 complementary manner rather than competing each other. Even nowadays a mobile handset can make a handoff between different RATs. The deployment of CRE systems can be facilitated by the recongurability concept, which is an evolution of a software-dened radio [39, 40]. The CRE requires terminals that are able to work with different RATs, and the existence of multiple radio networks offering alternate wireless access capabilities to service area regions. Recongurability supports the CRE concept by providing essential technologies that enable terminals and network elements dynamically (transparently and securely) to select and adapt to the set of RATs that are most appropriate for the conditions encountered in specic service area regions and time zones of the day. According to the recongurability concept, RAT selection is not restricted to those that are pre-installed in the network element. In fact, the required software components can be dynamically downloaded, installed and validated. This makes it different from the static paradigm regarding the capabilities of terminals and network elements. The networks provide wireless access to IP (Internet protocols)-based applications and service continuity in the light of intrasystem mobility. Integration of the network segments in the CR infrastructure is achieved through the management system for the CRE (MS-CRE) component attached to each network. The management system in each network manages a specic radio technology; however, the platforms can cooperate. The xed (core and backbone) network will consist of public and private segments based on IPv4- and IPv6-based infrastructures. A mobile IP (MIP) will enable the maintenance of IP-level connectivity regardless of the likely changes in the underlying radio technologies used that will be imposed by the CRE concept. Figures 1.2 and 1.3 depict the architecture of a terminal that is capable of operating in a CRE context. The terminals include software and hardware components (layer 1 and 2 functionalities) for operating with different systems. The higher protocol layers, in accordance with their peer entities in the network, support continuous access to IP-based applications. Different protocol busters can further enhance the efciency of the protocol stack. There is a need to provide the best possible IP performance over wireless links, including legacy systems. Within the performance implications of link characteristics (PILC) of the IETF group, the concept of a performance-enhancing proxy Terminal management system Network discovery support Network selection Mobility management intersystem (vertical) handover QoS monitoring Profile management user preferences, terminal characteristics Application Enhanced for TMS interactions and Information flow synchronization Transport layer TCP/UDP Network layer protocol boosters & conversion IP Mobile IP bandwidth reasignment GPRS support protocol Layers 2/1 UMTS support protocol Layers 2/1 WLAN/BRAN Support protocol Layers 2/1 DVB-T Support protocol Layers 2/1 Figure 1.2 Architecture of a terminal that operates in a composite radio environment.

4 FUNDAMENTALS Application Enhanced for TMS interactions and Information flow synchronization Terminal management System Network discovery support Network selection Mobility management intersystem (vertical) handovers QoS monitoring Profile management Functionality for software download, instalation, validation Security, fault/error recovery Transport layer TCP/UDP protocol boosters & conversion Network layer IP, Mobile IP Active configurations Repository Reconfigurable modem Interface Reconfiguration commands Monitoring information Software components for communication through The selected RATs bandwidth reasignment Rat-specific and generic software components an parameters Figure 1.3 Architecture of a terminal that operates in the recongurability context. (PEP) [4144] has been chosen to refer to a set of methods used to improve the performance of Internet protocols on network paths where native TCP/IP performance is degraded due to characteristics of a link. Different types of PEPs, depending on their basic functioning, are also distinguished. Some of them try to compensate for the poor performance by modifying the protocols themselves. In contrast, a symmetric/asymmetric boosting approach, transparent to the upper layers, is often both more efcient and exible. A common framework to house a number of different protocol boosters provides high exibility, as it may adapt to both the characteristics of the trafc being delivered and the particular conditions of the links. In this sense, a control plane for easing the required information sharing (cross-layer communication and congurability) is needed. Furthermore, another requirement comes from the appearance of multihop communications, as PEPs have been traditionally used over the last hop, so they should be adapted to the multihop scenario. Most communications networks are subject to time and regional variations in trafc demands, which lead to variations in the degree to which the spectrum is utilized. Therefore, a services radio spectrum can be underused at certain times or geographical areas, while another service may experience a shortage at the same time/place. Given the high economic value placed on the radio spectrum and the importance of spectrum efciency, it is clear that wastage of radio spectrum must be avoided. These issues provide the motivation for a scheme called dynamic spectrum allocation (DSA), which aims to manage the spectrum utilized by a converged radio system and share it between participating radio networks over space and time to increase overall spectrum efciency, as shown in Figures 1.4 and 1.5. Composite radio systems and recongurability, discussed above, are potential enablers of DSA systems. Composite radio systems allow seamless delivery of services through the most appropriate

4G NETWORKS AND COMPOSITE RADIO ENVIRONMENT Time or region RAN3 RAN1 RAN3 RAN2 RAN3 RAN3 RAN2 RAN2 RAN1 RAN2 RAN1 RAN3 RAN3 Time or region RAN1 RAN1 RAN2 RAN1 RAN3 RAN2 RAN1 RAN1 RAN1 RAN1 RAN1 Frequency RAN2 RAN2 RAN2 RAN2 RAN1 RAN2 RAN1 RAN2 RAN2 RAN1 RAN1 RAN2 RAN1 RAN2 RAN2 RAN1 Fragmented Frequency RAN2 RAN1 Frequency RAN2 Contiguous RAN1 Fixed 5 Time or region Figure 1.4 Fixed spectrum allocation compared to contiguous and fragmented DSA. 2483 2400 2200 1900 1880 1710 960 880 854 470 230 217 WLAN UMTS GSM GSM Analog TV and DVB-T DAB Contiguous DSA (a) Contiguous DSA WLAN Analog TV and DVB-T GSM GSM UMTS WLAN Analog TV and DVB-T DAB Fragmented DSA (b) Contiguous DSA WLAN GSM UMTS WLAN GSM GSM UMTS Analog TV and DVB-T DAB Fragmented DSA (c) Figure 1.5 DSA operation congurations: (a) static (current spectrum allocations); (b) continuous DSA operations; (c) discrete DSA operations.

6 FUNDAMENTALS access network, and close network cooperation can facilitate the sharing not only of services but also of spectrum. Recongurability is also a very important issue, since with a DSA system a radio access network could potentially be allocated any frequency at any time in any location. It should be noted that the application layer is enhanced with the means to synchronize various information streams of the same application, which could be transported simultaneously over different RATs. The terminal management system (TMS) is essential for providing functionality that exploits the CR environment. On the user/terminal side, the main focus is on the determination of the networks that provide, in a cost-efcient manner, the best QoS levels for the set of active applications. A rst requirement is that the MS-CRE should exploit the capabilities of the CR infrastructure. This can be done in a reactive or proactive manner. Reactively, the MS-CRE reacts to new service area conditions, such as the unexpected emergence of hot spots. Proactively, the management system can anticipate changes in the demand pattern. Such situations can be alleviated by using alternate components of the CR infrastructure to achieve the required capacity and QoS levels. The second requirement is that the MS-CRE should provide resource brokerage functionality to enable the cooperation of the networks of the CR infrastructure. Finally, parts of the MS-CRE should be capable of directing users to the most appropriate networks of the CR infrastructure, where they will obtain services efciently in terms of cost and QoS. To achieve the above requirements the MS architecture shown in Figure 1.6 is required. The architecture consists of three main logical entities: Monitoring, service-level information and resource brokerage (MSRB). Resource management strategies (RMS). Session managers (SMs). Short-term operation Mid-term operation MS-CRE Profile and service-level information Resource brokerage Session manager Management Plane interface Service configuration traffic distribution Netwotk configuration Status monitoring MSRB Management Plane interface Mobile terminal Managed network (component of CR infrastructure) Legacy element and network management systems User and control plane interface Figure 1.6 Architecture of the MS-CRE. RMS Management Plane interface

PROTOCOL BOOSTERS Session manager MSRB RMS 7 MS-CRE 1. Identification of new condition in service area 2. Extraction of status of Network and of SLAs 3a. Offer request 3b. Offer request 3c. Offer request 4a. Optimization request 4b. Determination of new service provision pattern (QoS levels, traffic distribution to networks) Computation of Tentative reconfigurations 4c. Reply 5. Solution acceptance phase. Reconfiguration of managed Network and managed components Figure 1.7 MS-CRE operation scenario. The MSRB entity identies the triggers (events) that should be handled by the MS-CRE and provides corresponding auxiliary (supporting) functionality. The RMS entity provides the necessary optimization functionality. The SM entity is in charge of interacting with the active subscribed users/terminals. The operation steps and cooperation of the RMS components are shown in Figures 1.7 and 1.8, respectively. In order to gain an insight into the scope and range of possible recongurations, we review the network and protocol stack architectures of the basic CRE components as indicated in Figure 1.1. 1.2 PROTOCOL BOOSTERS As pointed out in Figure 1.2, an element of the reconguration in 4G networks are protocol boosters. A protocol booster is a software or hardware module that transparently improves protocol performance. The booster can reside anywhere in the network or end systems, and may operate independently (one-element booster) or in cooperation with other protocol boosters (multielement booster). Protocol boosters provide an architectural alternative to existing protocol adaptation techniques, such as protocol conversion. A protocol booster is a supporting agent that by itself is not a protocol. It may add, delete or delay protocol messages, but never originates, terminates or converts that protocol. A multielement protocol booster may dene new protocol messages to exchange among themselves, but these protocols are originated and terminated by protocol booster elements, and are not visible or

8 FUNDAMENTALS MSRB Service configuration traffic distribution Network configuration 1. Optimization request 2. Service configuration and traffic distribution: Allocation to QoS and networks 3a. Request for checking the feasibility of solution 3b. Computation of Tentative network reconfiguration 3c. Reply on feasibility of solution 4. Selection of best Feasible solution 5. Reply 6. Solution acceptance phase 7. Network configuration Figure 1.8 Cooperation of the RMS components. Protocol messages Host X Booster A Booster B Host Y Booster messages Figure 1.9 Two-element booster. meaningful external to the booster. Figure 1.9 shows the information ow in a generic two-element booster. A protocol booster is transparent to the protocol being boosted. Thus, the elimination of a protocol booster will not prevent end-to-end communication, as would, for example, the removal of one end of a conversion (e.g. a TCP/IP header compression unit). In what follows we will present examples of protocol busters.

PROTOCOL BOOSTERS 9 1.2.1 One-element error detection booster for UDP UDP has an optional 16-bit checksum eld in the header. If it contains the value zero, it means that the checksum was not computed by the source. Computing this checksum may be wasteful on a reliable LAN. On the other hand, if errors are possible, the checksum greatly improves data integrity. A transmitter sending data does not compute a checksum for either local or remote destinations. For reliable local communication, this saves the checksum computation (at the source and destination). For wide-area communication, the single-element error detection booster computes the checksum and puts it into the UDP header. The booster could be located either in the source host (below the level of UDP) or in a gateway machine. 1.2.2 One-element ACK compression booster for TCP On a system with asymmetric channel speeds, such as broadcast satellite, the forward (data) channel may be considerably faster than the return (ACK) channel. On such a system, many TCP ACKs may build up in a queue, increasing round-trip time and thus reducing the transmission rate for a given TCP window size. The nature of TCPs cumulative ACKs means that any ACK acknowledges at least as many bytes of data as any earlier ACK. Consequently, if several ACKs are in a queue, it is necessary to keep only the ACK that has arrived most recently. A simple ACK compression booster could ensure that only a single ACK exists in the queue for each TCP connection. (A more sophisticated ACK compression booster allows some duplicate ACKs to pass, allowing the TCP transmitter to get a better picture of network congestion.) The booster increases the protocol performance because it reduces the ACK latency and allows faster transmission for a given window size. 1.2.3 One-element congestion control booster for TCP Congestion control reduces buffer overow loss by reducing the transmission rate at the source when the network is congested. A TCP transmitter deduces information about network congestion by examining acknowledgments (ACKs) sent by the TCP receiver. If the transmitter sees several ACKs with the same sequence number, then it assumes that network congestion caused a loss of data messages. If congestion is noted in a subnet, then a congestion control booster could articially produce duplicate ACKs. The TCP receiver would think that data messages have been lost because of congestion, and would reduce its window size, thus reducing the amount of data it injects into the network. 1.2.4 One-element ARQ booster for TCP TCP uses ARQ to retransmit data unacknowledged by the receiver when a packet loss is suspected, such as after a retransmission timeout expires. If we assume the network of Figure 1.9 (except that Booster B does not exist), then an ARQ booster for TCP will: (a) cache packets from Host Y; (b) if it sees a duplicate acknowledgment arrive from Host X and it has the next packet in the cache; then it deletes the acknowledgment and retransmits the next packet (because a packet must have been lost between the booster and Host X); and (c) delete packets retransmitted from Host Y that have been acknowledged by Host X. The ARQ booster improves performance by shortening the retransmission path. A typical application would be if Host X were on a wireless network and the booster were on the interface between the wireless and wireline networks.

10 FUNDAMENTALS 1.2.5 A forward erasure correction booster for IP or TCP For many real-time and multicast applications, forward error correction coding is desirable. The two-element FZC booster uses a packet forward error correction code and erasure decoding. The FZC booster at the transmitter side of the network adds parity packets. The FZC booster at the receiver side removes the parity packets and regenerates missing data packets. The FZC booster can be applied between any two points in a network (including the end systems). If applied to an IP, then a sequence number booster adds sequence number information to the data packets before the rst FZC booster. If applied to TCP (or any protocol with sequence number information), then the FZC booster can be more efcient because: (1) it does not need to add sequence numbers and (2) it could add new parity information on TCP retransmissions (rather than repeating the same parities). At the receiver side, the FZC booster could combine information from multiple TCP retransmissions for FZC decoding. 1.2.6 Two-element jitter control booster for IP For real-time communication, we may be interested in bounding the amount of jitter that occurs in the network. A jitter control booster can be used to reduce jitter at the expense of increased latency. At the rst booster element, timestamps are generated for each data message that passes. These (a) (b) Figure 1.10 Three-dimensional amplitude patterns of a two-element uniform amplitude array for d D 2, directioned towards (a) 0 D 0 , (b) 0 D 60 .

GREEN WIRELESS NETWORKS (a) (c) (b) 11 (d) Figure 1.11 Three-dimensional amplitude patterns of a ten-element uniform amplitude array for d D =4, directioned towards (a) 0 D 0 , (b) 0 D 30 , (c) 0 D 60 , (d) 0 D 90 . timestamps are transmitted to the second booster element, which delays messages and attempts to reproduce the intermessage interval that was measured by the rst booster element. 1.2.7 Two-element selective ARQ booster for IP or TCP For links with signicant error rates using a selective ARQ protocol (with selective acknowledgment and selective retransmission) can signicantly improve the efciency compared to using TCPs ARQ (with cumulative acknowledgment and possibly go-back-N retransmission). The two-element ARQ booster uses a selective ARQ booster to supplement TCP by: (1) caching packets in the upstream booster, (2) sending negative acknowledgments when gaps are detected in the downstream booster and (3) selectively retransmitting the packets requested in the negative acknowledgments (if they are in the cache). 1.3 GREEN WIRELESS NETWORKS 4G wireless networks might be using a spatial notching (angle ) to suppress completely antenna radiation towards the user, as illustrated in Figures 1.10 and 1.11. These solutions will be referred to as green wireless networks for obvious reasons. In order to ensure the connectivity in the case when the antenna lobe is not directed towards the access point, a multihop communication, with the possibility of relaying, is required. In addition, to reduce the overall transmit power a cooperative transmit diversity, discussed in Section 19.4, and adaptive MAC protocol, discussed in Chapter 5 can be used. REFERENCES [1] M. Mouly and M.-B. Pautet, The GSM System for Mobile Communications, Palaiseau, France, 1992.

12 FUNDAMENTALS [2] R. Kalden, I. Meirick and M. Meyer, Wireless Internet access based on GPRS, IEEE Pers. Commun., vol. 7, no. 2, April 2000, pp. 818. [3] 3rd Generation Partnership Project (3GPP), http://www.3gpp.org. [4] P. Mogensen, Wei Na, I. Z. Kovacs, F. Frederiksen, A. Pokhariyal, K. I. Pedersen, T. Kolding, K. Hugl and M. Kuusela, LTE capacity compared to the Shannon bound, in Vehicular Technology Conf., 2225 April 2007, pp. 12341238. [5] H. Holma, A. Toskala, K. Ranta-aho, and J. Pirskanen, High-speed packet access evolution in 3GPP Release 7 [Topics in Radio Communications], IEEE Commun. Mag., vol. 45, no. 12, December 2007, pp. 2935. [6] A. Hoikkanen, A techno-cconomic analysis of 3G long-term evolution for broadband access, in Conf. on Telecommun. Techno-Economics, 1415 June 2007, pp. 17. [7] A. Racz, A. Temesvary and N. Reider, Handover performance in 3GPP long term evolution (LTE) systems, in Mobile and Wireless Commun. Summit, 16th IST, 15 July 2007, pp. 15. [8] M. Valkama, L. Anttila and M. Renfors, Some radio implementation challenges in 3G-LTE context, in Signal Processing Advances in Wireless Commun., IEEE 8th Workshop on SPAWC 2007, 1720 June 2007, pp. 15. [9] J. J. Sanchez, D. Morales-Jimenez, G. Gomez and J. T. Enbrambasaguas, Physical layer performance of long term evolution in cellular technology, in 16th IST Mobile and Wireless Commun. Summit, 15 July 2007, pp. 15. [10] J. Berkmann, C. Carbonelli, F. Dietrich, C. Drewes and Wen Xu, On 3G LTE terminal implementation standards, algorithms, complexities and challenges, in International Wireless Commun. and Mobile Computing Conf., IWCMC 08, 68 August 2008, pp. 970975. [11] C. Spiegel, J. Berkmann, Zijian Bai, T. Scholand and C. Drewes, MIMO schemes in UTRA LTE, a comparison, in IEEE Vehicular Technology Conf., 1114 May 2008, pp. 22282232. [12] S.-E. Elayoubi, O. Ben Haddada and B. Fourestie, Performance evaluation of frequency planning schemes in OFDMA-based networks, IEEE Trans. Wireless Commun., vol. 7, no. 5, part 1, May 2008, pp. 16231633. [13] J. Khun-Jush et al., HiperLAN2: broadband wireless communications at 5 GHz, IEEE Commun. Mag., vol. 40, no. 6, June 2002, pp. 130137. [14] U. Varshney, The status and future of 802.11-based WLANs, IEEE Comp., vol. 36, no. 6, June 2003, pp. 102105. [15] Digital Video Broadcasting (DVB), http://www.dvb.org, January 2002. [16] V. Stavroulaki, P. Demestichas, A. Katidiotis and D. Petromanolakis, Evolution in equipment management concepts: from recongurable to cognitive wireless terminals, in 16th IST Mobile and Wireless Commun. Summit, 15 July 2007, pp. 15. [17] R. Muraleedharan and L. A. Osadciw, Increasing QoS and security in 4G networks using cognitive intelligence, in IEEE Globecom Workshops, 2630 November 2007, pp. 16. [18] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran and S. Mohanty, Next generation/dynamic spectrum access/cognitive radio wireless networks: a survey, Computer Networks J., Elsevier, vol. 50, September 2006, pp. 21272159. [19] M. Muck, D. Bourse, K. Moessner, N. Alonistioti, P. Demestichas, E. Nicollet, E. Buracchini, D. Bateman, Z. Boudis, E. Patouni, V. Stavroulaki, A. Trogolo and P. Goria, End-to-end recongurability in heterogeneous wireless systems software and cognitive radio solutions enriched by policy- and context-based decision making, in 16th IST Mobile and Wireless Commun. Summit, 15 July 2007, pp. 15. [20] B. Aazhang, J. Lilleberg and G. Middleton, Spectrum sharing in a cellular system, in 2004 IEEE Eighth Int. Symp. on Spread Spectrum Techniques and Applications, 30 August2 September 2004, pp. 355359. [21] K. Doppler, A. Osseiran, M. Wodczak and P. Rost, On the integration of cooperative relaying into the WINNER system concept, in 16th IST Mobile and Wireless Commun. Summit, 15 July 2007, pp. 15.

REFERENCES 13 [22] Qian Zhang, Qing Chen, Fan Yang, Xuemin Shen and Zhisheng Niu, Cooperative and opportunistic transmission for wireless ad hoc networks, IEEE Network , vol. 21, no. 1, January February 2007, pp. 1420. [23] M. Katz and F. H. P. Fitzek, Cooperative techniques and principles enabling future 4G wireless networks, in EUROCON 2005, The Int. Conf. on Computer as a Tool, vol. 1, 2005, pp. 2124. [24] H. Paloheimo, J. Manner, J. Nieminen and A. Yla-Jaaski, Challenges in packet scheduling in 4G wireless networks, in 2006 IEEE 17th Int. Symp. on Personal, Indoor and Mobile Radio Commun., September 2006, pp. 16. [25] Carlos Leonel Flores Mayorga,, Francescantonio della Rosa, Satya Ardhy Wardana, Gianluca Simone,, Marie Claire Naima Raynal, Joao Figueiras and Simone Frattasi, Cooperative positioning techniques for mobile localization in 4G cellular networks, in IEEE Int. Conf. on Pervasive Services, 1520 July 2007, pp. 3944. [26] D. Niyato and E. Hossain, A cooperative game framework for bandwidth allocation in 4G heterogeneous wireless networks, in 2006 IEEE Int. Conf. on Commun., vol. 9, June 2006, pp. 43574362. [27] M. Dohler, D.-E. Meddour, S.-M. Senouci and A. Saadani, Cooperation in 4G hype or ripe?, in IEEE Technology and Society Mag., vol. 27, no. 1, Spring 2008, pp. 1317. [28] S. Frattasi, M. Monti and R. Prasad, A cooperative localization scheme for 4G wireless communications, in 2006 IEEE Radio and Wireless Symp., 1719 January 2006, pp. 287290. [29] C. Politis, T. Oda, S. Dixit, A. Schieder, H.-Y. Lach, M. I. Smirnov, S. Uskela and R. Tafazolli, Cooperative networks for the future wireless world, IEEE Commun. Mag., vol. 42, no. 9, September 2004, pp. 7079. [30] V. Marques, R. L. Aguiar, C. Garcia, J. I. Moreno, C. Beaujean, E. Melin and M. Liebsch, An IP-based QoS architecture for 4G operator scenarios, IEEE Wireless Commun. [see also IEEE Pers. Commun.], vol. 10, no. 3, June 2003, pp. 5462. [31] Luan Huang, Kar Ann Chew and R. Tafazolli, Network selection for one-to-many services in 3G-broadcasting cooperative networks, in 2005 IEEE 61st Vehicular Technology Conf., vol. 5, 30 May1 June 2005, pp. 29993003. [32] S. M. S. Masajedian and H. Khoshbin, Cooperative location management method in next generation cellular networks, in Ninth Int. Symp. on Computers and Communications, ISCC 2004, vol. 1, 28 June1 July 2004, pp. 525530. [33] S. Sorrentino and U. Spagnolini, A predictive opportunistic scheduler for 4G wireless systems, in 16th IST Mobile and Wireless Commun. Summit, 15 July 2007, pp. 15. [34] A. K. F. Khattab and K. M. F. Elsayed, Opportunistic scheduling of delay sensitive trafc in OFDMA-based wireless networks, in Int. Symp. on a World of Wireless, Mobile and Multimedia Networks, WoWMoM 2006, 2629 June 2006, 10 pp. [35] Yu Wu, Ji-bo Wei, Yong Xi, Byung-Seo Kim and Sung Won Kim, Opportunistic scheduling with statistical fairness guarantee in wireless networks, in IEEE 18th Int. Symp. on Personal, Indoor and Mobile Radio Commun., PIMRC 2007, 37 September 2007, pp. 15. [36] S. Mangold, Zhun Zhong, K. Challapali and Chun-Ting Chou, Spectrum agile radio: radio resource measurements for opportunistic spectrum usage, in IEEE Global Telecommun. Conf. 2004, GLOBECOM 04, vol. 6, 29 November3 December 2004, pp. 34673471. [37] S. Sanayei, A. Nosratinia and N. Aldhahir, Opportunistic dynamic subchannel allocation in multiuser OFDM networks with limited feedback, in IEEE Information Theory Workshop, 2004, 2429 October 2004, pp. 182186. [38] W. Ajib and D. Haccoun, An overview of scheduling algorithms in MIMO-based fourthgeneration wireless systems, IEEE Network , vol. 19, no. 5, SeptemberOctober 2005, pp. 4348. [39] S. Glisic, Advanced Wireless Communications: 4G Cognitive and Cooperative Broadband Technology, 2nd edition. John Wiley & Sons, Ltd: Chichester, London, 2007.

14 FUNDAMENTALS [40] J. Mitola III and G. Maguire Jr, Cognitive radio: making software radios more personal, IEEE Pers. Commun., vol. 6, no. 4, August 1999, pp. 1318. [41] J. Border et al., Performance enhancing proxies intended to mitigate link-related degradations, RFC 3135, June 2001. [42] D. C. Feldmeier et al., Protocol boosters, IEEE J. Selected Areas Commun., vol. 16, no. 3, April 1998, pp. 437444. [43] M. Garca et al., An experimental study of snoop TCP performance over the IEEE 802.11b WLAN, in 5th Int. Symp. on Wireless Personal Multimedia Commun., vol. III, Honolulu, HA, October 2002, pp. 10681072. [44] L. Mu oz et al., Optimizing Internet ows over IEEE 802.11b wireless local area networks: n a performance enhancing proxy based on forward error correction, IEEE Commun. Mag., vol. 39, no. 12, December 2001, pp. 6067.

2 Opportunistic Communications As pointed out in Chapter 1, opportunistic signaling will be used in 4G networks to increase further the spectral efciency of these systems. In this chapter we discuss a number of different solutions that are based on that principle. 2.1 MULTIUSER DIVERSITY Multiuser diversity is provided in wireless networks by independent time-varying channels across the different users. The diversity benet is exploited by tracking the channel uctuations of the users and scheduling transmissions to users when their instantaneous channel quality is highest. The diversity gain increases with the dynamic range of the uctuations and is thus limited in environments with slow fading. In such environments, multiple-transmit antennas can be used to induce large and fast channel uctuations so that multiuser diversity can be improved [1]. The scheme can be interpreted as opportunistic beamforming and beamforming gains can be achieved when there are sufcient users. In a cellular system, the scheme plays an additional role of opportunistic nulling of the interference created on users of adjacent cells. Let us assume a simple model of the downlink of a cellular wireless communication system with a base station (transmitter) having a single antenna communicating with K users (receivers). The time-slotted block-fading channel model in baseband is given by yk (t) D h k (t)x(t) C zk (t); T k D 1; 2; : : : ; K (2.1) T In Equation (2.1), x(t) 2 C is the vector of T transmitted symbols, yk (t) 2 C is the vector of T received symbols of user k, h k (t) 2 C is the fading channel gain from the transmitter to receiver k and fzk (t)gt is an independent and identically distributed (i.i.d.) sequence of zero mean circularsymmetric Gaussian random vectors CN (0; 2 IT ). Here we assume that the channel is constant over time slots of length T samples and that the transmit power level is P D const at all times, i.e. E[kx(t)k2 ] D P T . Advanced Wireless Technologies: Cognitive, Cooperative & Opportunistic 4G Technology Second Edition Savo G. Glisic C 2009 John Wiley & Sons, Ltd.

16 OPPORTUNISTIC COMMUNICATIONS Total Throughput in bps/Hz 2,2 AWGN Channel Rayleigh Fading 2 1,8 1,6 1,4 1,2 1 0,8 0 2 4 6 8 10 Number of Users 12 14 Figure 2.1 Sum capacity of two channels, Rayleigh fading and AWGN, with average SNR D 0 dB. We can view this downlink channel as a set of parallel Gaussian channels, one for each fading state. The sum capacity of this channel, dened by the maximum achievable sum of long-term average data rates transmitted to all the users, can be achieved by a simple time division multiple access (TDMA) strategy: at each fading state, transmit to the user with the strongest channel [2]. The sum capacity of the downlink channel is presented in Figure 2.1, as a function of the number of users, for the case when users undergo independent Rayleigh fading with average received signal-to-noise ratio (SNR) D 0 dB. One can see that the sum capacity increases with the number of users in the system. On the other hand, the sum capacity of a nonfaded downlink channel, where each user has a xed additive white Gaussian noise (AWGN) channel with SNR = 0 dB, is constant irrespective of the number of users. In a system with many users with independently varying channels, it is likely that at any time there is a user with a channel much stronger than the average SNR. By transmitting to users with strong channels at all times, the overall spectral efciency of the system can be made high, signicantly higher than that of a nonfaded channel with the same average SNR. In order to exploit such multiuser diversity it is necessary that: Each receiver should track its own channel SNR through, say, common downlink pilot, and feed back the instantaneous channel quality to the base station. The base station has the ability to schedule transmissions among the users as well as to adapt the data rate as a function of the instantaneous channel quality. 2.2 PROPORTIONAL FAIR SCHEDULING The concept of multiuser diversity brings about two issues: fairness and delay. When users fading statistics are the same, the strategy above maximizes not only the total capacity of the system but also the throughput of individual users. When the statistics are not symmetrical, the multiuser diversity concept provides maximum long-term average throughputs. In practice, there are latency requirements, in which case the average throughputs and limited delay is the performance metric of interest. In the sequel the objective will be to address these issues while at the same time exploiting the multiuser diversity gain inherent in a system, with users having independent, uctuating channel conditions. In a further discussion the feedback of the channel quality of user k in time slot t

PROPORTIONAL FAIR SCHEDULING 17 Requested rates in bps/Hz 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0 50 100 150 200 250 Time Slots Requested rates in bps/Hz Figure 2.2 For symmetric channel statistics of users, the scheduling algorithm reduces to serving each user with the largest requested rate. 1,1 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0 50 100 150 200 250 Time Slots Figure 2.3 In general, with asymmetric user channel statistics, the scheduling algorithm serves each user when it is near its peak within the latency time scale tc . to the base station will be expressed in terms of a requested data rate Rk (t). This is the data rate that the kth users channel can support at the time. The scheduling algorithm keeps track of the average throughput Tk (t) of each user in a past window of length tc , and in time slot t the scheduling algorithm transmits to the user k with the largest Rk (t)=Tk (t) among all active users in the system. The average throughputs Tk (t) can be updated using an exponentially weighted low-pass lter Tk (t C 1) D (1 (1 1=tc )Tk (t) C Rk (t)=tc ; 1=tc )Tk (t); k D k : k 6D k (2.2) The operation of the algorithm is illustrated in Figures 2.2 and 2.3. The sample paths of the requested data rates of two users are plotted as a function of time slots (each time slot is 1.67 ms). In Figure 2.2, the two users have identical fading statistics. If the scheduling time scale tc is much larger than the correlation time scale of the fading dynamics, then by symmetry the throughput of each user Tk (t) converges to the same quantity. The scheduling algorithm reduces to always

OPPORTUNISTIC COMMUNICATIONS Average Throughput in bps/Hz 18 2,2 2 1,8 1,6 1,4 1,2 Mobile Fixed Equal time scheduling 1 0,8 0 5 10 15 Number of Users 20 25 Figure 2.4 Multiuser diversity gain in xed and mobile environments. picking the user with the highest requested rate. Thus, each user is scheduled when its channel is good and at the same time the scheduling algorithm is perfectly fair in the long term. In Figure 2.3, one users channel is much stronger than the other users on the average, although both channels uctuate due to multipath fading. Always picking the user with the highest requested rate means giving all the system resources to the statistically stronger user and would be unfair. Under the scheduling algorithm dened by Equation (2.2), users compete for resources not only based on their requested rates but after normalization by their respective average throughputs. The user with the statistically stronger channel will have a higher average throughput. Thus, the algorithm schedules a user when its instantaneous channel quality is high relative to its own average channel condition over the period tc . In other words, data are transmitted to a user when the channel is near its own peaks. Multiuser diversity benet can still be exploited because channels of different users uctuate independently so that if there is a sufcient number of users in the system, there is likely to be a user near its peak at any one time. The parameter tc is tied to the latency time scale of the application. Peaks are dened with respect to this time scale. If the latency time scale is large, then the throughput is averaged over a longer time scale and the scheduler can afford to wait longer before scheduling a user when its channel hits a really high peak. The theoretical properties of this scheduling algorithm are further explored later in this chapter. It will be shown that this algorithm guarantees a fairness property called proportional fairness. Figure 2.4 gives some insights into the issues involved in realizing multiuser diversity benets in practice. The plot shows the total throughput of the downlink under the proportional fair scheduling algorithm in the following two simulated environments: Fixed. Users are static but there are movements of objects around them (2 Hz Rician, def D E direct =E specular D 5). Mobile. Users move at walking speeds (3 km/h, Rayleigh). The total throughput increases with the number of users in both the xed and mobile environments, but the increase is more dramatic in the mobile case. While the channel fades in both cases, the dynamic range and the rate of the variations is larger in the mobile environment than in the xed one. This means that over the latency time scale (1.67 s in these examples), the peaks of the channel uctuations are likely to be higher in the mobile environment, and the peaks are what determines the performance of the scheduling algorithm. Thus, the inherent multiuser diversity is more limited in the xed environment.

OPPORTUNISTIC BEAMFORMING 2.3 19 OPPORTUNISTIC BEAMFORMING The effectiveness of multiuser diversity depends on the rate and dynamic range of channel uctuations. In environments where the channel uctuations are small, the multiuser diversity gain can be increased by inducing faster and larger uctuations. In Reference [1] multiple-transmit antennas at the base station are used for these purposes, as illustrated in Figure 2.5. For such a system with N transmit antennas at the base station, let h nk (t) be the complex channel gain from antenna n to the kth user in time slot t. In time slot t, the same block of symbols x(t) is p transmitted from all of the antennas except that it is multiplied by a complex number n (t)e jn (t) N at antenna n, for n D 1; : : : ; N , such that nD1 n (t) D 1, preserving the total transmit power. The received signal at user k is given by N n (t)e jn (t) h nk (t) x(t) C zk (t) yk (t) D (2.3) nD1 Thus, the overall channel gain seen by receiver k is now N n (t)e jn (t) h nk (t) h k (t) :D (2.4) nD1 In Equation (2.4), the n (t) denote the fractions of power allocated to each of the transmit antennas and the n (t) the phase shifts applied at each antenna to the signal. By varying over time n (t) from 0 to 1 and n (t) from 0 to 2, uctuations in the overall channel can be induced even if the physical channel gains h nk (t) do not uctuate much. Each receiver k reports back the value of x (t) a(t) 1a(t)e jq(t) h2k (t) h1k (t) User k Figure 2.5 The same signal is transmitted over the two antennas with time-varying phase and powers.

20 OPPORTUNISTIC COMMUNICATIONS Average Throughput in bps/Hz 2,2 2 1,8 1,6 1,4 Mobile Fixed, Opp, BF Fixed Equal time scheduling 1,2 1 0,8 0 5 10 15 20 25 30 Number of Users Figure 2.6 Amplication in multiuser diversity gain with opportunistic beamforming in a xed environment. SNRjh k (t)j2 = 2 of its own channel to the base station and the base station schedules transmissions to users accordingly. The rate of variation of fn (t)g and fn (t)g in time is a design parameter of the system. On one side it should be as fast as possible to provide full channel uctuations within the latency time scale of interest. On the other hand, the variation should be slow enough to allow the channel to be reliably estimated by the users and the SNR information feedback. Further, the variation should be slow enough to ensure that the channel seen by the users does not change abruptly and thus maintains stability of the channel tracking loop. To illustrate the performance of this scheme, we now consider the xed environment of Figure 2.5 with two antennas of equal and constant (over time) power split and phase rotation over [0; 2 ] (with one complete rotation in 30 ms as in Reference [1]). Figure 2.6 plots the improved performance as a function of number of users. This improvement is due to the fact that the channel is changing faster and the dynamic range of variation is larger over the time scale of scheduling (1.67 s in this example). 2.4 OPPORTUNISTIC NULLING IN CELLULAR SYSTEMS For wide-band cellular systems with full frequency reuse, it is important to consider the effect of intercell interference on the performance of the system, particularly in interference-limited scenarios. In a cellular system, the channel quality of a user is measured by the signal-to-interference plus noise ratio (SINR). In a fading environment, the energies in both the received signal and the received interference uctuate over time. Since the multiuser diversity scheduling algorithm allocates resources based on the channel SINR (which depends on both the channel amplitude and the amplitude of the interference), it automatically exploits both the uctuations in the energy of the received signal as well as that of the interference: the algorithm tries to schedule resource to a user whose instantaneous channel is good and the interference is weak. Thus, multiuser diversity naturally takes advantage of the time-varying interference to increase the spatial reuse of the network. From this point of view, power and phase randomization at the base station transmit antennas plays an additional role: it increases not only the amount of uctuations of the received signal

OPPORTUNISTIC NULLING IN CELLULAR SYSTEMS 21 to the intended users within the cells, but also the amount of the uctuations of the interference the base station causes in adjacent cells. Hence, opportunistic beamforming has a dual benet in an interference-limited cellular system. In fact, opportunistic beamforming performs opportunistic nulling simultaneously, while randomization of power and phase in the transmitted signals from the antennas allows near-coherent beamforming to some user within the cell; it will create near nulls at some other user in adjacent cells. This in effect allows interference avoidance for that user if it is currently being scheduled. In a slow at fading scenario under power and phase randomization at all base stations, the received signal of a typical user being interfered by J adjacent base stations is given by J g j (t)u j (t) C z(t) y(t) D h(t)x(t) C (2.5) jD1 where x(t) is the signal of interest, u j (t) is the interference from the jth base station and z(t) is additive Gaussian noise. All base stations have the same transmit power P and N transmit antennas and are performing power and phase randomization independently; h(t) and g j (t) are the overall channel gains from the base stations N n (t)e jn (t) h n h(t) :D (2.6) nD1 N n j (t) D e j g j (t) :D n j (t) gn j (2.7) nD1 where h n and gn j are the slow fading channel gains to the user from the nth transmit antenna of the base station of interest and the interfering base station j, respectively. Averaging over the signal x(t) and the interference u j (t), the (time-varying) SINR of the user can be computed to be Pjh(t)j2 SINR(t) D J P (2.8) jg j (t)j2 C 2 jD1 The SINR varies because of both the variations of the overall gain from the base station of interest as well as those from the interfering base station. In a system with many other users, the N proportional fair scheduler will serve this user while its SINR is at its peak P nD1 jh n (t)j2 = 2 , i.e. when the received signal is the strongest and the interference is completely nulled out. Thus, the opportunistic nulling and beamforming technique has the potential to shift a user from a low-SNR interference-limited regime to a high-SNR noise-limited regime. How close the performance of opportunistic beamforming and nulling in a nite-size system is to this asymptotic limit depends on the probability that the received signal is near beamformed and all the interference is near null. In the interference-limited regime when P= 2 1, the performance depends mainly on the probability of the latter event. This probability is larger when there are only one or two base stations contributing most of the interference, as is typically the case. In contrast, when there is interference from many base stations, interference averaging occurs and the probability that the total interference is near null is much smaller. Interference averaging, which is good for CDMA networks, is actually unfavorable for the opportunistic scheme described here, since it reduces the likelihood of the nulling of the interference and hence the likelihood of the peaks of the SINR. In a typical cell, there will be a distribution of users, some closer to the base station and some closer to the cell boundaries. Users close to the base station are at high SNR and are noise-limited; the contribution of the intercell interference is relatively small. These users benet

22 OPPORTUNISTIC COMMUNICATIONS mainly from opportunistic beamforming (diversity gain plus a 3 dB power gain if there are two transmit antennas). Users close to the cell boundaries, on the other hand, are at low SNR and are interference-limited; the average interference power can be much larger than the background noise. These users benet both from opportunistic beamforming and from opportunity nulling of intercell interference. Thus, the cell-edge users benet more in this system than users in the interior. This is rather desirable from a system fairness point of view, as the cell-edge users tend to have poorer service. This feature is particularly important for a system without soft handoff (which is difcult to implement in a packet data scheduling system). To maximize the opportunistic nulling benets, the transmit power at the base station should be set as large as possible, subject to regulatory and hardware constraints. 2.5 NETWORK COOPERATION AND OPPORTUNISTIC COMMUNICATIONS In this section we consider a network architecture with B cooperating and B noncooperating base stations (BSs). The cluster of cooperating BSs cooperate to improve the capacity with fairness to the users in the network. The noncooperating BSs do not cooperate with BSs in this cluster and cause interference to the users, which cannot be mitigated. As an example, in Figure 2.7 [3], there are three BSs (marked with circles), which cooperate and provide service to the users in the shaded region. The other BSs cause interference to these users. In the sequel we focus only on the activities of the cooperating BSs. During every time slot each coordinating BS selects one beam out of the set of L beams available to it and services one user. In this way, the cluster of B cooperating BSs supports B users simultaneously. The system operates in two steps. In the rst step, each BS transmits a pilot signal using a certain beam and collects the SINR reports from the users. The BS then transmits these reports to a centralized unit which, in the second step, schedules users based on their current and past SINR reports for data packet transmission. This process is repeated for the entire time period over which data are transmitted. The number of users K being serviced during this time period is assumed constant. Figure 2.7 Three-base intercell coordination scenario.

NETWORK COOPERATION AND OPPORTUNISTIC COMMUNICATIONS 23 In a given timeslot n during the rst step, each BS transmits a pilot signal using one of the L beams. The received signal power at user i from the BS b on the lth beam is Si [n] D P(i;b) jh (i;b;l) [n]j2 , where h (i;b;l) [n] is the channel gain from BS b to the user i on beam l and P(i;b) is the received signal power at the user i with the combined effect of path loss and shadowing. The interference plus noise power is INi [n] D Ii [n] C Noi [n], where I () is the interference power from the noncooperating BSs and No() is the thermal noise power. The user measures the received SINR, (i;b;l) [n] D Si [n]=INi [n] and transmits this SINR value to the BS. The centralized scheduler collects the SINR reports from all the BSs and obtains the matrix K B , where the (i; b)th element represents the SINR feedback by user i on the beam from BS b. In the second phase the centralized scheduler using these data opportunistically schedules B users on each of the B BSs, and each BS then transmits to the scheduled user. The data rate R(i;b;l) , a BS b, provides a given user i on beam l and is calculated by R(i;b;l) D log2 1 C Si [n] D log2 1 C B I Ni [n] C (i;b;l) [n] 1 C 2 P(i;b) h (i;b;l) [n] b6Db B (i;b;l) [n] (2.9) b6Db In a scenario where the BSs cooperate and opportunistically beamform and schedule users the main steps are [3]: (1) Choose U users to serve: (kU k D K u ; B < K u < K ) using one of the following criteria: a) Select users that experience large inter-packet delay, i [n]. b) Select users who have low short-term throughput Ti [n]. c) Select users who have the largest i [n]=Ti [n] ratio. (2) The centralized scheduler using the user-base-beam selection algorithm described below selects, a subset of B users from the above chosen group of K u users, and the corresponding BSs and beams on which they may be serviced. This algorithm generates a set of B triplets where the triplet (i j ; b j ; l j ) represents the case where user i j is best served by BS b j on the beam l j ; j 2 f1 Bg: (3) Each BS generates the above assigned beam and collects the SINR reports from all the users. All the BSs transmit these reports to the centralized scheduler. (4) The centralized scheduler using the K B SINR matrix and the short-term throughput vector T schedules B users for service. Steps 1,2 and 3 constitute the rst phase of the system and step 4 constitutes the second phase. In the user-base-beam selection (UBBS) algorithm, mentioned above, B users from the initially chosen set of K u users and the indices of the BSs and the beams on which these users may be serviced are chosen. The beams must be chosen such that they least interfere with the B selected users. One possible approach to attain this objective is through the instantaneous system throughput, which is the sum of the achievable user rates. For B D 3 we state this formally as arg max i 1 ;i 2 ;i 3 2f1K u g;b1 ;b2 ;b3 2f1Bg;l1 ;l2 ;l3 2f1Lg f (i; b; l) (2.10)

24 OPPORTUNISTIC COMMUNICATIONS with i D fi 1 ; i 2 ; i 3 g, b D fb1 ; b2 ; b3 g, l D fl1 ; l2 ; l3 g and f () is the sum of the rates of the three users. By using Equation (2.9) we have f (i; b; l) D log2 1C (i 1 ;b1 ;l1 ) 1C 1C (i 1 ;b2 ;l2 ) C 1C (i 1 ;b3 ;l3 ) (i 2 ;b2 ;l2 ) 1C (i 2 ;b1 ;l1 ) C (i 2 ;b3 ;l3 ) (i 3 ;b3 ;l3 ) 1C (i 3 ;b1 ;l1 ) C (i 3 ;b2 ;l2 ) In the absence of current SINRs, the short-term averaged SINRs (i;b;l) can be used instead of in Equation (2.10). The short-term SINR is computed as (i;b;l) [n] D n 1 tD0 n t 1 (i;b;l) [t]((i;b) [t]; l)a n 1 n t 1 tD0 ((i;b) [t]; l)a (i;b;l) (2.11) where (i;b;l) [t] represents the index of the beam on which user i was serviced by base station b at time instant t. Thus, this alternative criterion for selecting the set of triplets is stated as arg max i 1 ;i 2 ;i 3 2f1K u g;b1 ;b2 ;b3 2f1Bg;l1 ;l2 ;l3 2f1Lg (i; b; l) f (2.12) with (i; b; l) given by Equation (2.10) using (i;b;l) instead of (i;b;l) . f The optimization of Equation (2.12) is practically infeasible when B and L are large. The following, computationally less intensive, modied UBBS algorithm was proposed in Reference [3] for scenarios where B and L are large. A group of users U is initially chosen and arranged in order of preference for service using one of the criteria stated earlier. The main steps of this modied procedure are: (1) Choose the rst user i 1 and obtain the BS and beam index fb1 ; l1 g on which the user i 1 experiences maximum throughput. Formally these indices are obtained as (b1 ; l1 ) D arg max (i;b;l) (2.13) b2f1Bg;l2f1Lg 1 D (i1 ;b1 ;l1 ) (2.14) (2) Choose the second user i 2 from the remainder of the set (U the beam and BS using the following criterion: arg max i 1 ), and the indices for 2 (i; b; l); f (2.15) b2 2f1Bg;l2 2f1Lg with i D fi 1 ; i 2 g and b D fb1 ; b2 g; l D fl1 ; l2 g and 2 (i; b; l) D f 1C (i1 ;b1 ;l1 ) 1 C (i1 ;b2 ;l2 ) 1C (i2 ;b2 ;l2 ) 1 C (i2 ;b1 ;l1 ) f 2 D 2 (i; b; l) In this way the newly chosen user i 2 causes minimal interference to user i 1 and vice versa. The terms 1 , 2 represent the virtual SINR of the system as the system is loaded with users and BSs. We also use the constraint that 2 > 1 to ensure that the addition of a new user does not decrease the overall virtual SINR of the system. In the same way, the remaining set of users and the beams on which they are serviced by the BSs are chosen. When this constraint of increasing the virtual SINR of the system fails, the beams are chosen at random. The scheduler obtains the terms (i;b;l) , which are the SINR reports of the users from the BSs on the beams given by the UBBS algorithm. Then the rate R(i;b;l) that user i obtains from BS b using

NETWORK COOPERATION AND OPPORTUNISTIC COMMUNICATIONS 25 beam l is computed. Using this information, B users are chosen whose weighted sum of rates is maximum, with the weights given by the reciprocal of the users short-term throughputs T []. This criterion is also computationally intensive, so the following alternative is considered. The users are ordered according to their weighted rates as (R=T )i1 ;b1 (R=T )i2 ;b2 (R=T )i K ;bk and the rst B users are chosen while maintaining the constraint that only one user is scheduled from a BS. 2.5.1 Performance example We will illustrate the performance of the above methods for two different hexagonal cellular networks, as shown in Figures 2.7 and 2.8. The number of coordinating BSs B in these two networks are 3 and 9 respectively. The hexagonal cells are divided into three 120 sectors and each of these sectors are covered by a BS with a smart antenna system. The BSs are located at the centre of each of the hexagonal cells and the users are distributed randomly in the shaded regions. The received signal at the user is modelled as in Section 2.5 to take into account path loss, shadow fading and correlated Rayleigh fading effects. The path loss is based on the Hata urban propagation model with a path loss coefcient of 3.5. A lognormal shadow fading process with zero mean and a standard deviation of 8 dB was used to characterize the variations in the signal due to environmental clutter. Two different correlated Rayleigh fading channels corresponding to Doppler rates with user velocities of 1 m/s and 8 m/s are analyzed. The antennas at the BSs consist of a uniform linear array (ULA) of antenna elements. Each of these elements is fed equal currents and only their phases are varied. Eleven such equispaced radiation patterns were generated using a four-element array. The normalized array factor considered in the simulations is j A( )j D jsin(N =2)= sin( =2)j=N , where N is the number of antenna array elements and is the azimuth in radians. The radiation pattern of each element is shown by the dashed line in Figure 2.9. The