Smart packet access and call admission control for ...jultika.oulu.fi/files/isbn9514276779.pdf · Phan, Vinh V., Smart packet access and call admission control for efficient resource

SMART PACKET ACCESS AND CALL ADMISSION CONTROL FOR EFFICIENT RESOURCE MANAGEMENT IN ADVANCED WIRELESS NETWORKS

VINH V.PHAN

Faculty of Technology,Department of Electrical and

Information Engineering,University of Oulu

OULU 2005

VINH V. PHAN

SMART PACKET ACCESS AND CALL ADMISSION CONTROL FOR EFFICIENT RESOURCE MANAGEMENT IN ADVANCED WIRELESS NETWORKS

Academic Dissertation to be presented with the assent ofthe Faculty of Technology, University of Oulu, for publicdiscussion in Raahensali (Auditorium L10), Linnanmaa, on April 22nd, 2005, at 12 noon

OULUN YLIOPISTO, OULU 2005

Copyright © 2005University of Oulu, 2005

Supervised byProfessor Savo G. Glisic

Reviewed byProfessor Abbas JamalipourProfessor Timo O. Korhonen

ISBN 951-42-7676-0 (nid.)ISBN 951-42-7677-9 (PDF) http://herkules.oulu.fi/isbn9514276779/

ISSN 0355-3213 http://herkules.oulu.fi/issn03553213/

OULU UNIVERSITY PRESSOULU 2005

Phan, Vinh V., Smart packet access and call admission control for efficient resourcemanagement in advanced wireless networks Faculty of Technology, University of Oulu, P.O.Box 4000, FIN-90014 University of Oulu, Finland,Department of Electrical and Information Engineering, University of Oulu, P.O.Box 4500, FIN-90014 University of Oulu, Finland 2005Oulu, Finland

AbstractEfficient management of rather limited resources, including radio spectrum and mobile-terminalbattery power, has been the fundamental design challenge of wireless networks and one of the mostwidespread research problems over the years. MAC (Medium Access Control) for packet access andCAC (Call Admission Control) for connection-oriented service domains are commonly used aseffective tools to manage radio resources, capacity and performance of wireless networks whileproviding adequate QoS (Quality of Service) to mobile users. Hence, analysis and synthesis ofefficient MAC and CAC schemes for advanced wireless networks have significant academic andpractical values. This dissertation addresses that topic and presents seven separate contributions ofthe author: four on adaptive MAC schemes for centralized PRN (Packet Radio Networks), referredto as SPA (Smart Packet Access) and three on CAC schemes for cellular networks, referred to as SCA(Smart Call Admission). These contributions are published in eighteen original papers by the author,which are listed and referred to as Papers I–XVIII in this thesis.

In SPA, the first contribution, reported in Papers II and IV, studies implementation losses ofadaptive feedback-control MAC schemes for the uplink of DS-CDMA (Direct-Sequence CodeDivision Multiple Access) PRN in the presence of various system imperfections. The secondcontribution, reported in Papers XI, XII, XV and XVI, proposes a bit-rate adaptive MAC scheme forDS-CDMA PRN, referred to as SPR (Smart Packet Rate). The third contribution, reported in PapersIII, XIII and XIV, develops two alternative MAC schemes with adaptive packet-length overcorrelated fading channels in DS-CDMA PRN, referred to as SPL (Smart Packet Length). The fourthcontribution, reported in Papers XVII and XVIII, develops alternative adaptive MAC schemes foroptimal trade-offs between throughput and energy consumption of TCP (Transmission ControlProtocol) applications in advanced cellular networks. These include a so-called SPD (Smart PacketDispatching) for HSPA (High Speed Packet Access) and, again, SPL for LSPA (Low Speed PacketAccess).

Moving on to SCA, the first contribution, reported in Papers V and VII, provides a simple andaccurate analytical method for performance evaluation of a class of fixed-assignment CAC schemeswith generic guard-channel policy and queuing priority handoffs in cellular networks. The secondcontribution, reported in Papers VI, IX and X, proposes a simple and effective SCAC (Soft-decisionCAC) scheme for CDMA cellular networks. This is evaluated against fixed-assignment andmeasurement-based CAC schemes with a simple and reliable method provided as a part of thecontribution. The third contribution, reported in Papers I and VIII, incorporates alternative QoSdifferentiation paradigms and resource partitioning into CAC, defines GoS (Grade of Service) formultimedia cellular networks, and provides an in-hand tool for efficient capacity and GoSmanagement.

Keywords: call admission control (CAC), energy efficiency, medium access control(MAC), resource management, spectral efficiency, wireless networks

To my mom, như món quà nhỏ cho đời mẹ cao cả, hy sinh

Preface

I came to Oulu, Finland in November 1997 to work for Nokia. I had by then considered this trip rather a one-year break, as I just graduated and got a full scholarship to continue with PhD study from the Technical University of Budapest, Hungary. Then, the warm welcome, honesty of the people I have met here; the freshness, clean, peacefulness of the nature I have breathed; the flexible working environment, interesting job, admirable culture I have been offered by Nokia; and the unique opportunity I have got to pursue the doctoral degree at the University of Oulu have made that one year of mine a bit too long, but worthwhile.

I officially enrolled for this study program at the Telecommunication Laboratory, Department of Electrical and Information Engineering, Faculty of Technology, University of Oulu, in autumn 1998 under the supervision of Prof. Savo Glisic. For the first couple of years, I managed to sit most of exams, study literature, and define focus areas of interest. In the following year, I started learning to get familiar with scientific research and publication processes, putting through few first papers of mine. I keep on learning and trying to conduct, publish, and pattern my research results ever since, also serve as a referee of technical papers for various international conferences, magazines, and journals in the field of wireless communications. I suppose I have learned to appreciate being a scientist who is able to ask questions and have his own idea in seeing technical problems. This thesis marks the final stage of my study towards the doctoral degree and, hopefully, a new chapter of my career with many more contributions to come.

I wish to express my deepest gratitude to my supervisor, Prof. Savo Glisic, for his guidance, encouragement, and help over the years. His pursuit for the highest standards in research is a great inspiration for my work. I am grateful to him for what I have learned and achieved throughout the course of this study.

I thank the Department of Electrical and Information Engineering, Faculty of Technology, University of Oulu and, in particular, the Telecommunication Laboratory headed by Prof. Pentti Leppänen, for accepting and providing me with an excellent and flexible environment to carry out this study and research.

I am grateful to the reviewers of this thesis, Prof. Abbas Jamalipour of the University of Sydney, Australia, and Prof. Timo O. Korhonen of Helsinki University of Technology, Finland. Their invaluable effort and kindness as well as their constructive comments and

suggestions to improve the manuscript are highly appreciated. I am thankful to Prof. Stephen McLaughlin of the University of Edinburgh, Scotland, and, again, Prof. Timo O. Korhonen for taking the crucial roles, the opponents, in the public defense of my thesis.

I want to take this opportunity to thank my employer, Nokia, for hosting my stay, work, and supporting my study in Finland. I am grateful to the former managers of mine, especially Tarmo Kumpula and Timo Kononen, for inviting me to join Nokia at the start of my journey, and Heikki Ojaniemi, for his understanding and constant supports. I am thankful to the current managers of mine for providing me with flexible, yet challenging opportunities to work and study, and generous financial supports. In particular, my special thanks are due to Markku Vainikka, also for doing the honor of writing a short abstract in Finnish for this thesis.

I should like to express my warmest regards to numerous teachers, friends, and colleagues of mine for the enjoyable years of living, working, and studying abroad. I am grateful to my former supervisor, Prof. Katalin Tarnay, and advisor, Dr. Janos Miskolczi, at the Technical University of Budapest, Hungary, for their constant encouragement. My special thanks are due to Dr. Ramin Baghaie, a former Nokia colleague of mine, Dr. Ling Yu of Nokia, Docent Dr. Tien Do of the Technical University of Budapest, Hungary, for reading and commenting on the introductory chapter of my thesis, and Dung Luong for his friendship and cooperation.

I highly appreciate the helpful advice and warm assistance of the following personnel of the Telecommunication Laboratory: Pekka Pirinen in taking the exam courses during this study, Antero Kangas in obtaining the course materials, and Laila Kuhalampi in arranging the public defense of my thesis. I also highly appreciate the effort of the editorial staff of Acta Universitatis Ouluensis in publishing this thesis book.

I dedicate this thesis to my mom whose sacrifices and wishes are the greatest source of my inspiration and motivation. I wish to express my warmest gratitude to my family for their loves and supports.

Oulu, March 2005 Vinh Phan

List of original papers

This thesis is based on eighteen original, peer-reviewed papers, which are listed below with Roman numerals and referred to as Papers I – XVIII in the text. The listing is based on separating book chapters, journals, and conferences, and in order of year/months of publication. All the listed papers have been conceived and written by the author during his doctorate study and under the supervision of Professor Savo G. Glisic. Professor Savo G. Glisic provided opportunities, motivations, reviews, criticism, and suggestions related to technical issues, editorial corrections, term inventions, and guidance in the study and publication processes. In the case of Papers III, X, XVII and XVII, the co-author Dung D. Luong carried out the computer simulations under the instructions of the author. In the case of Paper XVII, the co-author Dr. Ramin Baghaie proofread the manuscript and assisted in getting financial support. Other than that, the author has conducted all the modeling, analyses and results presented in the publications. Seven of the most representative papers, Papers II, III, VII, VIII, IX, XII and XVIII covering the main contributions of the author, are selected for reprint in the appendices of the thesis. I Phan VV & Glisic S (2002) Radio access control in wireless IP networks. In: Dixit

S & Prasad R (eds) Wireless IP and Building the Mobile Internet, Artech House Publishers, Boston, 227-254.

II Phan VV & Glisic S (2001) Estimation of implementation losses in MAC protocols for wireless CDMA networks. International Journal of Wireless Information Networks, 8: 115-132.

III Phan VV, Glisic S & Luong DD (2004) Packet-length adaptive CLSP/DS-CDMA: performance in burst-error correlated fading channels. IEEE Transactions on Wireless Communications, 3: 147-158.

IV Glisic S & Phan VV (2000) Sensitivity function of soft decision carrier sense MAC protocols for wireless CDMA networks with specified QoS. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’00), Invited Paper, London, UK, 1: 205-211.

V Phan VV & Glisic S (2001) An analytical modeling for a class of queueing priority handoff and call admission control schemes in micro-cellular PCN’s. Proc. IEEE Semiannual Vehicular Technology Conference (VTC’01-Spring), Rhodes, Greece, 2: 901-905.

VI Phan VV (2001) Enhanced model for teletraffic performance evaluation of WCDMA cellular PCN’s with effective capacity and call admission control design. Proc. International Conference on Third Generation Wireless and Beyond (3Gwireless’01), San Francisco, USA, 308-314.

VII Phan VV & Glisic S (2001) Performance analysis of queueing schemes for priority handoff and admission control in mobile cellular radio networks. Proc. IEEE International Conference on Communications (ICC’01), Helsinki, Finland, 7: 2291-2295.

VIII Phan VV & Glisic S (2001) Radio resource management in CDMA cellular segments of multimedia wireless IP networks. Proc. International Symposium on Wireless Personal Multimedia Communications (WPMC’01), Invited Paper, Aalborg, Denmark, 1: 57-73.

IX Phan VV (2001) Soft decision CAC versus fixed channel assignment and interference measurement based policies for WCDMA cellular PCN’s. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’01), San Diego, USA, 2: E27-31.

X Phan VV & Luong DD (2001) Capacity enhancement with simple and robust soft decision CAC for WCDMA mobile cellular PCN’s. Proc. IEEE Semiannual Vehicular Technology Conference (VTC’01-Fall), Atlantic City, USA, 3: 1349-1353.

XI Phan VV (2002) Throughput-delay improvement of unslotted CDMA packet radio networks using rate adaptive CLSP. Proc. IEEE Wireless Communication and Networking Conference (WCNC’02), Orlando, USA, 2: 537-541.

XII Phan VV & Glisic S (2002) Unslotted DS/CDMA packet radio network using rate/space adaptive CLSP. Proc. IEEE International Conference on Communications (ICC’02), New York, USA, 5: 3382-3387.

XIII Phan VV (2002) Unslotted DS/CDMA packet radio network using CLSP and packet-length adaptation in Rayleigh fading channel. Proc. IEEE Semiannual Vehicular Technology Conference (VTC’02-Spring), Birmingham, USA, 2: 782-786.

XIV Phan VV & Glisic S (2002) MAC layer packet-length adaptive CLSP/DS-CDMA radio networks: performance in flat Rayleigh fading channel. Proc. IEEE Symposium on Computers and Communications (ISCC’02), Taormina, Italy, 161-166.

XV Phan VV & Glisic S (2002) Green wireless network: new concept for adaptive mobile communications. Proc. IEEE Workshop on Applications and Services in Wireless Networks (ASWN’02), Paris, France.

XVI Phan VV & Glisic S (2002) Mobile-location aware rate adaptive unslotted CLSP/DS-CDMA packet radio networks: performance in a heavily-correlated fading channel. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’02), Lisbon, Portugal, 2: 605-609.

XVII Phan VV, Luong DD, Baghaie R & Glisic S (2003) Energy-efficient high-speed packet access in WCDMA systems with smart packet dispatching: performance with TCP/IP based applications. Proc. IEEE Semiannual Vehicular Technology Conference (VTC’03-Fall), Orlando, USA, 3: 1632-1636.

XVIII Phan VV, Glisic S & Luong DD (2005) Energy-efficient smart packet access in WCDMA systems: performance with TCP/IP based applications. Proc. IEEE Hawaii International Conference on System Sciences (HICSS-38), Hawaii, USA, 288-297.

List of symbols and abbreviations*

a Coefficient A Offered traffic intensity in Erlang Area measure Coefficient A0 Matrix of admission probabilities for QoS differentiation As Matrix of admission probabilities AFD Average-fade-duration parameter aj Activity factor of data source of user j ajk Admission probability of traffic class-(j, k) av Activity factor of voice user As Matrix of admission probabilities for SCAC b Integer index of base stations in B Coefficient B Blocking probability Coefficient B Set {1, 2, ..., Nb} for indexing different base stations Bc Bandwidth per RF channel Bguard Guard bandwidth in FDMA Bjk Blocking probability of class-(j, k) traffic Bt Total RF bandwidth c Channel capacity System load state in terms of load factor C Channel capacity C Set {1, 2, ..., Nc} for indexing different channels c1 Coefficient C1 Cost wasted c2 Coefficient C2 Cost wasted

* Some symbols may have multiple meanings. Each symbol, however, is used unambiguously in its context and section.

cavrg Average cell capacity in number of users Cavrg Average cell capacity in term of load factor cj Channel or waveform of user j in C Cj Capacity index of class-j users cl Lower bound of cell capacity in number of users Cl Lower bound of cell capacity in term of load factor Cp Constant coefficient of free-space path loss cu Upper bound of cell capacity in number of users Cu Upper bound of cell capacity in term of load factor d Distance from the user terminal to the base station D Average packet delay Detection rate of correct channel states D0 Reference average packet delay DP Average packet delay Ds Delay sensitivity Eb/I0 Bit energy per interference density f Other-to-own cell interference ratio F Fade margin Handoff failure probability False-alarm rate fd Maximal Doppler frequency Fjk Handoff failure probability of class-(j, k) traffic g Path gain gb,j Path gain of user j toward base station b G Goodput Ga Average offered load gf Short-term fast fading Gl Lower bound of UL load factor Gother UL load factor caused by other cells gp Free-space path loss gs Long-term shadowing fading Gsf Goodput of packet transmission Gu Upper bound of UL load factor H Length of packet header i Integer index Iother Interference from other cells or inter-cell interference level Iown Interference of own cell or intra-cell Itotal Total received interference at the base station j Integer index of user in U J Number of subscription classes J Set {1, 2, ..., J} for indexing different subscription classes k Number of cells in a cluster of frequency-reuse pattern Integer index K Channel threshold K Set {1, 2, ..., M} for indexing different call services Kd Capacity index of data-user class

Kv Capacity index of voice-user class L Length of packet in bits Lower bound of capacity index L Matrix of blocking thresholds for shaping traffic LCR Level-crossing-rate parameter ljk Load threshold for blocking traffic of class-(j, k) m Integer index Number of time slots per frequency channel M Number of different bit rates supported Number of different call services M Set {1, 2, ..., M} for indexing different bit rates Set {1, 2, ..., Nm} for indexing different transmission modes mBG Rate of bad-to-good channel state transition mGB Rate of good-to-bad channel state transition mj Mode of user j in M Mu Number of mobiles leaving the reference cell n Integer index of user in U Number of ongoing user transmissions in the system or system state n Occupancy vector in multi-rate multi-class system N Capacity in total number of channels Number of the user population N0 Noise power Nb Number of base stations Nc Number of channels Number of chips in the code sequence Ne Number of correctable bits Nm Number of modes for packet transmissions Nu Number of users Pb Probability of call blocking Pc Probability of correct packet transmission Pd Probability of call dropping Pe Probability of bit error Pl Probability of loss of communication quality Ploss Probability of loss of communication quality pn Probability of steady state n Poutage Probability of system outage Prx Received signal power Prx,j Received signal power of user j Psf Probability of correct packet transmission Ptx Transmitted signal power Ptx,j Transmitted signal power of user j RTT Round-trip time r Distance between transmitter and receiver Number of stages in Erlang type-r distribution r0 Distance between user to the reference base station Cell radius

rm Distance between user m to its own base station Radius of coverage for a bit rate of Rm R Bit rate R0 Primary or reference bit rate Rj Bit rate of user j Rm Bit rate of class m S Throughput S0 Reference throughput SIRj SIR parameter of user j Ss Throughput sensitivity SIRtarget,j SIR target parameter of user j t Time T Packet duration Threshold of channel occupancy tf Fade duration tf_avrg Average fade duration tif Interfade duration tif_avrg Average interfade duration Tmax Maximum packet duration Tmin Minimum packet duration TP Packet duration TSPD Smart-packet-dispatching time interval TTI Transmission-time-interval parameter u Integer index U Upper bound of channel capacity U Set {1, 2, ..., Nu} for indexing different users uj Load factor of a user j v Integer index Velocity of a user w Effective load factor of user w Vector of wk of different user classes W Chip rate of CDMA system wk Effective load factor of class-k user wl Lower bound of effective load factor of user wl,k Lower bound of effective load factor of class-k user wu Upper bound of effective load factor of user wu,k Upper bound of effective load factor of class-k user θj,n Cross correlation between waveforms of two users j and n α Path-loss exponent Coefficient or weighting factor Probability value Path-loss exponent β Coefficient Lower bound of required packet error rate γ SIR target Mean of access timing delay

γj SIR target of user j γn Forced termination rate of handoff calls at system state n η Noise figure Noise to total received interference ratio ι Other-to-own cell interference ratio κj Coefficient in GoS cost function λ Packet/call arrival rate λn Packet/call arrival rate at system state n μ Service rate seen by the reference cell server μ1 Outgoing handoff rate μ2 Call service rate μh Outgoing handoff rate μl Call service rate π Transmission permission probability πn Transmission permission probability at system state n Admission probability at system state n ρ Density of the spatial user distribution σ Standard deviation σ1 Arrival rate of handoff calls σ2 Arrival rate of new calls σk Standard deviation of load factor of class-k user τ Time υ0 Coefficient in GoS cost function υ1 Coefficient in GoS cost function υ2 Coefficient in GoS cost function ξ Lognormal random variable of shadowing attenuation ξk Coefficient in GoS cost function ψ Binomial random variable 1G 1st Generation 2G 2nd Generation 3G 3rd Generation 3GPP 3G Partnership Projects 4G 4th Generation ABR Available Bit Rate AFD Average Fade Duration AMPS Advanced Mobile Phone Service ARQ Automatic Repeat reQuest ATM Asynchronous Transfer Mode B3G Beyond 3G BCC Blocked Calls Cleared BCD Blocked Calls Delayed BER Bit Error Rate BLER BLock Error Rate BS Base Station C-PRMA Centralized Packet Reservation Multiple Access CA Channel Assignment

CAC Call Admission Control CBR Constant Bit Rate CDMA Code Division Multiple Access CIR Carrier to Interference Ratio CLSP Channel Load Sensing Protocol CSMA Carrier Sense Multiple Access DAMA Demand Assignment Multiple Access DDA Dynamic Demand Assignment DFD Doppler Frequency Distribution DL Down-Link DLL Data Link Layer DQRUMA Distributed Queuing Request Update Multiple Access DS Direct Sequence EDGE Enhanced Data-rates for Global Evolution FAMA Fixed Assignment Multiple Access FCFS First Come First Serve FCSI Feedback Channel State Information FDD Frequency Division Duplexing FDMA Frequency Division Multiple Access FEC Forward Error Correcting FER Frame Error Rate FGC Fractional Guard Channel FH Frequency Hopping FIFO First In First Out FRMA Frame Reservation Multiple Access FSMC Finite State Markov Chain GC Guard Channels GoS Grade of Service GPRS General Packet Radio Service GSM Global System for Mobile Communications HARQ Hybrid ARQ HMM Hidden Markov Model HSDPA High-Speed Downlink Packet Access HSPA High-Speed Packet Access IETF Internet Engineering Task Force IMT-2000 International Mobile Telecommunications 2000 ID Identifier IP Internet Protocols IS Idle/Inhibit Signal ISA IS with Acknowledgment ISMA Idle/Inhibit Sense Multiple Access ISO International Organization for Standardization LCR Level Crossing Rate LLC Logical Link Control LSPA Low-Speed Packet Access MAC Medium Access Control

MAI Multiple Access Interference MANET Mobile Ad-hoc NETworks MC-CDMA Multi-Carrier CDMA MIMO Multiple Inputs Multiple Outputs MM Mobility Management MT Mobile Terminal MTIP Multimedia Traffic Intensity Profile NMT Nordic Mobile Telephones OFDM Orthogonal Frequency Division Multiplexing OSI Open Systems Interconnection PC Power Control PDF Probability Distribution Function PDU Protocol Data Unit PER Packet Error Rate PN Pseudo-Noise PRMA Packet Reservation Multiple Access PRN Packet Radio Networks QH Queuing Handoffs QoS Quality of Service R-ALOHA Reservation ALOHA RAMA Resource Auction Multiple Access RF Radio Frequency RL Radio Link RLC RL Control RM Resource Management RP Reservation-request Packet RR Random Reservation RRC Radio Resource Control RTT Round Trip Time RV Random Variable S-ALOHA Slotted ALOHA SCA Smart Call Admission SCAC Soft-decision CAC SDMA Space Division Multiple Access SINR Signal to Interference-plus-Noise Ratio SIR Signal to Interference Ratio SNR Signal to Noise Ratio SPA Smart Packet Access SPD Smart Packet Dispatching SPL Smart Packet Length SPP Smart Packet Persistence SPR Smart Packet Rate SS Spread Spectrum SUD Spatial User Distribution TCP Transmission Control Protocol TDMA Time Division Multiple Access

TF Transport Format TH Time Hopping TM Traffic Management TPP Transmission Permission Probability TTI Transmission Time Interval UBR Unspecified Bit Rate UL Up-Link UMTS Universal Mobile Telecommunications System VBR Variable Bit Rate UWB Ultra WideBand WCDMA Wideband CDMA WLAN Wireless Local Area Networks WPAN Wireless Personal Area Networks

Contents

Abstract Preface List of original papers List of symbols and abbreviations Contents 1 Introduction ...................................................................................................................25

1.1 Resource management in advanced wireless networks ..........................................26 1.1.1 General problem formulation ..........................................................................27 1.1.2 Channel assignment.........................................................................................30 1.1.3 Power control...................................................................................................31 1.1.4 Mobility management......................................................................................31 1.1.5 Traffic management.........................................................................................32

1.2 Focus areas defined ................................................................................................33 1.2.1 Medium access control ....................................................................................33 1.2.2 Call admission control .....................................................................................35

1.3 Aim and contribution of the thesis..........................................................................36 1.4 Organization of the thesis .......................................................................................41

2 Literature survey............................................................................................................43 2.1 Capacity provisioning of cellular systems ..............................................................43

2.1.1 Radio propagation model.................................................................................44 2.1.2 Capacity of FDMA/TDMA systems................................................................45 2.1.3 Capacity of CDMA systems ............................................................................48

2.1.3.1 Single-cell system with one class of users and perfect PC ..................49 2.1.3.2 Uniform multi-cell system with one class of users and perfect PC .....50 2.1.3.3 Load factor of multi-class users...........................................................52

2.2 MAC schemes in centralized packet radio networks ..............................................54 2.2.1 Properties and performance measures .............................................................54

2.2.1.1 Design properties .................................................................................54 2.2.1.2 Performance measures .........................................................................56

2.2.2 Classification of MAC schemes ......................................................................57

2.2.3 Review of centralized MAC schemes..............................................................59 2.2.3.1 Random access schemes ......................................................................60 2.2.3.2 Scheduled access schemes ...................................................................61 2.2.3.3 Hybrid access schemes ........................................................................62 2.2.3.4 DS-CDMA based schemes...................................................................64

2.2.4 Summary and discussion .................................................................................66 2.3 CAC schemes in mobile cellular networks.............................................................69

2.3.1 Properties and performance measures .............................................................69 2.3.1.1 Design properties .................................................................................69 2.3.1.2 Performance measures .........................................................................71

2.3.2 Classification of CAC schemes .......................................................................72 2.3.3 Review of CAC schemes.................................................................................73

2.3.3.1 CAC schemes in FDMA/TDMA cellular networks.............................73 2.3.3.2 CAC schemes in CDMA cellular networks .........................................75

2.3.4 Summary and discussion .................................................................................77 3 Summary of the main contributions ..............................................................................80

3.1 SPA − Smart packet access for centralized CDMA PRNs ......................................80 3.1.1 Implementation losses in adaptive feedback-control MAC schemes ..............80

3.1.1.1 Main problem ......................................................................................80 3.1.1.2 Claim ...................................................................................................81 3.1.1.3 Approach and result .............................................................................82 3.1.1.4 Conclusion ...........................................................................................87

3.1.2 Efficient MAC scheme with adaptive bit rate..................................................87 3.1.2.1 Main problem ......................................................................................87 3.1.2.2 Claim ...................................................................................................88 3.1.2.3 Approach and result .............................................................................89 3.1.2.4 Conclusion ...........................................................................................91

3.1.3 Efficient MAC scheme with adaptive packet length .......................................91 3.1.3.1 Main problem ......................................................................................91 3.1.3.2 Claim ...................................................................................................92 3.1.3.3 Approach and result .............................................................................93 3.1.3.4 Conclusion ...........................................................................................95

3.1.4 Adaptive MAC schemes for optimal trade-offs between throughput and energy efficiency of TCP applications......................................................96 3.1.4.1 Main problem ......................................................................................96 3.1.4.2 Claim ...................................................................................................96 3.1.4.3 Approach and result .............................................................................97 3.1.4.4 Conclusion .........................................................................................101

3.2 SCA − Smart call admission for advanced mobile cellular networks ...................101 3.2.1 Performance analysis of a class of fixed channel-assignment CAC

with generic guard-channel policy and queuing priority handoffs................101 3.2.1.1 Main problem ....................................................................................101 3.2.1.2 Claim .................................................................................................102 3.2.1.3 Approach and result ...........................................................................102 3.2.1.4 Conclusion .........................................................................................104

3.2.2 Simple and efficient soft-decision CAC for CDMA cellular networks .........105 3.2.2.1 Main problem ....................................................................................105 3.2.2.2 Claim .................................................................................................105 3.2.2.3 Approach and result ...........................................................................106 3.2.2.4 Conclusion .........................................................................................108

3.2.3 GoS management for multimedia CDMA cellular networks.........................109 3.2.3.1 Main problem ....................................................................................109 3.2.3.2 Claim .................................................................................................109 3.2.3.3 Approach and result ........................................................................... 110 3.2.3.4 Conclusion ......................................................................................... 114

4 Concluding remarks..................................................................................................... 115 4.1 Summary............................................................................................................... 115 4.2 Remarks................................................................................................................ 116 4.3 Future research directions..................................................................................... 117

References Appendix 1

1 Introduction

Entering the 21st century, wireless communications has emerged as one of the largest sectors of the telecommunication industry, and also one of the most promising areas for further and fast growth. The user population and broad demand for wireless and mobile communications services have steadily increased and this trend is expected to continue. Future wireless systems are envisioned to provide high capacity and ubiquitous information access in various environments, including indoor office, pedestrian, vehicle, and satellite. These will also include a wide variety of applications, which will result in diverse traffic classes being supported with different QoS (Quality of Service) requirements (1-6).

In contrast to the bandwidth demands for supporting an increasing number of users and a continuous expansion of services, wireless systems are constrained to operate within a fixed RF (Radio Frequency) band (1-22). This RF allocation is strictly regulated worldwide and, therefore, rather limited. Furthermore, in future multimedia wireless systems, the MTs (Mobile Terminal) are expected to run more applications more often and for longer periods when relying on limited battery power to conduct communications. The concern of energy conservation for the MTs becomes even more crucial than it is now (23-28). Hence, essential requirements on the capacity and capabilities of wireless systems are posed in term of securing scarce resources, including radio spectrum and MT battery power, to serve as many users or to transfer as much data as possible. This goal can be achieved with proper designs of radio access networks and, in particular, efficient RM (Resource Management).

Thus, together with sufficient system architectures, multiple access techniques, sophisticated physical designs using signal processing techniques, such as advanced receivers, adaptive channel coding and modulation, adaptive equalizers, and diversity combining, efficient RM schemes(*) must be utilized and adopted. RM schemes may include: channel assignment, power control, handoff control, call admission control, medium access control, packet transmission scheduling, congestion control, and other traffic-and-mobility-management schemes. Most often, these are mutually supportive,

(*) The term scheme, instead of algorithm or protocol, is used throughout this thesis to keep it flexible, general, or conceptual.

26

originated from and driven by basic nature, networking paradigms, technologies, and diverse requirements of wireless systems and services. RM schemes systematically define rules, guidelines, and mechanisms on how the network makes use of scarce radio resources while providing the required QoS to simultaneous mobile users. Thus, RM schemes determine the actual serving capacity, coverage, and performance of wireless systems, and are the keys to efficient network operation. These may take on a much more dynamic nature considering various technical, ergonomic, and economic factors, such as highly erratic radio channels, transmit-power restrictions, mobile multimedia supports with different user and service profile characteristics, simplicity and cost-efficiency in network operation. Therefore, efficient RM schemes have to be devised. In fact, efficient RM in advanced wireless networks has been one of the most challenging and widespread research problems in the recent years.

This thesis addresses analysis and synthesis of RM schemes, and will focus on two main areas. The first is MAC (Medium Access Control) for centralized PRN (Packet Radio Networks). In general, MAC schemes represent how the available system bandwidth can be shared among users in accessing and transmitting on the common radio channel in forms of a data frame or radio packet. The second focus area is CAC (Call Admission Control) for connection-oriented service domains of mobile multimedia cellular networks. In general, CAC schemes represent how the network decides to accept or reject a new call request for a connection-oriented service with certain QoS requirements. The decision and initial resource allocation for the user upon accepting the call request is made depending on the available system capacity, user and service profile characteristics, and effects imposed on the existing users. MAC and CAC schemes are commonly used as important and effective tools for managing scarce radio resources, system capacity, and performance in advanced wireless systems.

Chapter 1 is continued as follows. Section 1.1 provides an overview of RM in advanced wireless networks confined to mobile cellular systems. It starts with a formulation of the general RM problem of interest, and then reviews essential RM schemes. Section 1.2 defines the focus areas of the thesis, which are MAC and CAC schemes subject to efficient RM in advanced wireless networks. Section 1.3 presents the aim of the thesis and outlines the author’s contributions. Section 1.4 describes the chosen presentation style and organization of the thesis.

1.1 Resource management in advanced wireless networks

It is believed that evolutions of today’s cellular networks including 3G (3rd Generation) CDMA (Code Division Multiple Access) systems are the most significant segments of the future mobile Internet for a reliable radio access and network operation (1-6). Thus, as for an appropriate high-level modeling abstraction of advanced wireless networks used to address RM problems in this thesis, the author mainly considers the multiple-access channel model in UL (Up-Link) of a reference cell in either single-cell or multi-cell system scenarios, depicted in Figure 1 below. FDD (Frequency Division Duplexing) DS-CDMA (Direct Sequence CDMA) is assumed if not stated otherwise. The results,

27

discussions and, in particular, analytical methods, however, may well be applicable for the DL (Down-Link) direction and wireless networks in general.

Fig. 1. High-level modeling abstraction of the system in question.

1.1.1 General problem formulation

In cellular systems, an optimum usage of frequency-reuse patterns and radio propagation characteristics to minimize co-channel interference and, hence, increase the utility of the radio spectrum is the most restraining factor on the system capacity and performance. This has two major aspects: (i) design and planning of cell sites and network

U1

U2

U3

U4

U5

U6

U7

U8

B1

B2

B3

B4B6

B7

U9Reference Cell

cj, gb,j

Σcn, gb,n

N0,b

BbUj

Un

Own (Ref. Cell)

OtherNetworksNetworks

Ptx,j Prx,j

Ptx,n Prx,n

28

infrastructure, and (ii) radio resource allocation and management in a given service area with the infrastructure in place. The latter is the general RM problem of interest for this thesis, formulated as follows based on (19, 48, 49) and Paper I.

To keep it simple, yet general and insightful, the system is assumed to consist of Nb BSs (Base Station); each serving a single cell. The system has Nc channels or, more generally, signature waveforms for setting up radio links for transmissions between users and BSs. The system provides Nm possible modes for radio transmissions and QoS supports. Each mode corresponds to a specific channel selection, access method, and transport format including user bit rate, packet size, transmission time interval, etc. There is a Nu of active users in the system. Let us define the following sets: B = {1, 2, ..., Nb}; C = {1, 2, ..., Nc}; M = {1, 2, ..., Nm}; and U = {1, 2, ..., Nu}, for numbering the BSs, channels, modes, and active users, respectively. To establish an RL for communications over it, each active user j∈U needs to be assigned to a BS b∈B, with a suitable waveform cj∈C, a mode mj∈M, and a transmit-power, Ptx,j. Then, providing that the required QoS is met, the received SIR (Signal to Interference Ratio) that drives the RL quality and performance model has to be kept above its specified target at all times for each active user during its transmission. The following inequality must hold for the RL between the user j and the serving BS b:

jgetart

jnbnbnjntx

jbjtxj SIR

NgPgP

SIR ,,0,,,

,, ≥+

=∑

≠

θ. (1.1)

In the above expression (1.1), SIRj and SIRtarget,j denote the actual and target SIR values of the user j, respectively. The parameter gb,j is the path gain(1), representing the propagation losses between the j transmitter and the b receiver for all j∈U. Thus, it takes a value in between zero and one. The parameter θj,n denotes the normalized cross-correlation between cj and cn waveforms for all active users. The parameter N0,b is the background noise power at the BS b. The product Ptx,jgb,j gives the value of the received power of the user j signal at the BS, denoted by Prx,j. The product Prx,nθj,n that is Ptx,ngb,nθj,n under the sum in (1.1) represents the effective portion of the received power of the user n signal contributed to the total received interference at the BS against the user j. If the waveforms are designed to be orthogonal as in the FDMA (Frequency Division Multiple Access) and TDMA (Time Division Multiple Access) systems, then the value of the correlations θj,n is either zero or one depending on whether or not the users are assigned the same frequency channel or time slot. Otherwise, the θj,n value is in between zero and one as in a CDMA system with non-orthogonal codes (13-19).

The aforementioned assignment is performed and controlled by a generic resource allocation scheme. This is restricted by limitations and constraints of the available channels as subsets of C for each cell of the BSs in B, transmit power for users in U and BSs in B, and allowable transmission modes as subsets of M. This is, at the same time, conditioned to (1.1) including propagation conditions with path gains, QoS requirements

(1) The path gain may incorporate gains produced by smart solutions of the transmitter and the receiver as well, in addition to radio propagation attenuation due to path loss and fading impacts.

29

with SIR targets, and tolerable received interference levels at the BSs. The users, meanwhile, may be moving around the service area from one cell to another. Therefore, seamless handoffs for active users between cells or BSs, having the required resource allocation ready in time and place, are highly desirable. Furthermore, supporting multimedia applications and services may require a large and dynamic amount of bandwidth that in turn may ask for multiple channels and transmission modes to be assigned to a user time-to-time. This is not to mention the need for power conservation of the MT battery in operating such applications and services. Finding a generic, optimal resource allocation scheme to fulfill all the issues mentioned above and, at the same time, the stringent requirement of network operations, i.e., to serve as many users as possible, is a formidable problem. In practice, a mutual set of simple, heuristic schemes is often adopted; each considering specific areas and aspects of RM in wireless networks. The main tasks of the RM schemes, overall, are to coordinate the users and, once they are admitted into the system, allocate and maintain adequate resources for each active user so that certain criteria and performance requirements can be met and optimized for an efficient network operation. The design of efficient RM schemes can then be targeted to an optimal trade-off between QoS and GoS (Grade of Service) with respect to spectral efficiency, energy efficiency, or other performance measures of interest. The RM schemes, as the results, can be considered as static, dynamic, or somewhat hybrid or flexible.

Note that the term QoS is often used to designate a set of parameters which are intended to represent measurable aspects of the subjective "perceived quality" but are not the causes of this perception. The criteria taken into account by the user to judge a service vary with the nature of the service. The criteria involve simple concepts, such as service availability, transmission characteristics, and subjective estimates. GoS integrates a number of traffic engineering variables used to provide a measure of adequacy of a group of resources under specified conditions. These GoS variables may be probabilities of traffic losses, for examples, a call being blocked, dropped, or delayed for more than a specified time interval. ITU (International Telecommunications Union) E.600 recommendation, “terms and definitions for traffic engineering,” defines the GoS standards as the parameter values assigned as objectives for GoS variables and the GoS results as the values of GoS parameters achieved under actual load conditions. GoS influences and somewhat represents QoS on the application level of a network system, but GoS is less user-oriented than QoS and rather a network performance parameter. Both GoS and QoS, however, are related to network resources and are, via the network resources, linked to the revenue of the network operations. Hence, the key to efficient QoS control and GoS optimization is smart RM.

Based on the above formulation and discussions, four major areas of RM in advanced wireless networks have emerged: CA (Channel Assignment), PC (Power Control), MM (Mobility Management), and TM (Traffic Management). These are interrelated, QoS sensitive, and can be considered jointly.

30

1.1.2 Channel assignment

CA encompasses frequency planning and channel reuse methods which have a major impact on system capacity and performance (14-19, 21, 22, 48-52). In systems using non-spreading multiple access techniques, such as FDMA and TDMA, frequency channels used in a given cell are typically not used in immediately adjacent cells to avoid prohibitive co-channel interference. The frequency planning and channel reuse must assure a sufficient spatial isolation so that the cells using the same frequency channels will not cause excessive co-channel interference to each other. The main idea behind CA schemes is to make use of radio propagation path loss characteristics in order to minimize the CIR (Carrier to Interference Ratio) and hence increase spectral efficiency. In the meantime, certain QoS aspects, such as the signal quality required for each channel, must be respected. CA schemes can be classified as fixed, dynamic, or hybrid (16, 19, 50). Fixed CA schemes permanently assign a specific set of channels to each cell according to some frequency-reuse patterns, then let the users borrow or compete for one or several channels from that set for requested services. Dynamic CA schemes allow the system to fully use the pool of available channels, adapted to different system conditions. The channels are dynamically assigned to the cells, on demand basis, to satisfy the service requests. Hybrid CA schemes combine fixed and dynamic approaches in an efficient way, for instance, by using a fixed assignment up to a certain level of system resources and then dynamic assignment for the rest.

In systems using DS-CDMA techniques, the carrier RF band is universally shared among users/cells. The frequency-reuse factor is one (or one-cell), and therefore the frequency planning can be at ease. If the spreading codes are orthogonal, for example, using Walsh and Hadamard codes (51, 52), the interference of other signals to the desired signal is theoretically zero after de-spreading. However, this can be achieved only if the local de-spreading code is accurately aligned with the arriving code of the desired signal. This condition, in turn, is a costly synchronization problem for a practical system. It is, not to mention, that even in this case perfect code orthogonality is usually broken by multi-path propagation. For those reasons and because the number of orthogonal codes is rather limited, also to support statistical multiplexing on physical level, non-orthogonal code sets are often adopted. The design of a good non-orthogonal code set, however, must keep the cross-correlation between any pair of codes in the code set at any time shift as small as possible to minimize other-channel interference. Examples are Gold codes and Kasami codes (52-54). The population of these code sets can be considered as virtually unlimited, providing that the required minimum length of the code sequences is large enough. Thus, each user can be assigned a unique code sequence for the UL transmission. 3G WCDMA (Wideband CDMA) systems, such as UMTS (Universal Mobile Telecommunications System) and IMT-2000 (International Mobile Telecommunications 2000), adopt multi-stage spreading in the following way. Because all the channels originating from the same transmitter can be readily synchronized, orthogonal codes are used to separate these channels. These channels are then linearly combined and multiplied by a transmitter-specific, non-orthogonal code. In general, whenever synchronization between channels can be easily maintained, orthogonal codes are used; and when asynchronous transmissions are more practical, non-orthogonal codes

31

are used. Considering that simultaneous users should not use the same code channel in the same cell and that orthogonal code sets are rather limited, the CA schemes presented above may be applied for code channel assignment as well.

1.1.3 Power control

PC is another essential area of RM, responsible for setting and maintaining a suitable transmit power for each user at all times (13-22, 55-64). PC is used directly to satisfy the (1.1) condition under its constraints while not inducing excessive interference to other users and systems operating in the same environment. In addition, it is important to minimize the power consumption and prolong the battery life of MTs with all possible energy conservation tools in the system design (19, 23-28). Essential to PC is on-line monitoring of RL performance and adaptation of transmit power to changes induced by user mobility and channel impairments. PC can be done in either centralized or distributed fashion. The centralized fashion requires a central controller that has a complete knowledge of all the radio links and their power levels in the system. In the distributed approach, each wireless terminal adjusts its transmit power based on local measurements. The PC schemes can be based on a closed-loop feedback control or open-loop setting, fast or slow. The resource allocation, as a whole, and the performance of wireless networks depend heavily on PC and its optimization. Through PC, that is, by regularly adjusting the transmit power, the users and the network can perform and interact in an efficient manner while making a direct impact on each other physically in term of interference. Thus, PC can be used as a vehicle for implementing RM schemes. For this reason, PC is considered as fundamental for power-sensitive wireless networks.

In a CDMA system, due to the fact that all users and cells can use the same universal system frequency and that system capacity is interference-limited, sophisticated PC is required to combat the so-called near-far effect, resulting from the path loss, and the inter-cell interference (13-22, 55-64). It is desired that all user signals, regardless of the location of the users in the serving-cell coverage, near to or far from the BS, can be received with the same SIR level to maximize the multiple-access capacity. The interference from other cells is dealt with in the same manner as the interference from the cell itself.

Recent research has paid a lot of attention and effort to energy-efficient designs for advanced wireless networks, developing high-energy-efficiency, low-power-consumption radio-access techniques, architectures, and protocol stacks (22-28). This is of increasing importance as MTs rely on limited battery energy for their operation, especially when running multimedia applications for services that demand extra processing power.

1.1.4 Mobility management

MM, including location management and handoff control, enables the network to track the location of mobile users to deliver incoming calls, and to handle handoffs to ensure continuity of mobile users (16, 19, 21, 65-77). Location management is responsible for

32

locating mobile users and serving cells for incoming calls. Techniques like cell piloting, MT paging, location register-and-updating are adopted in mobile cellular systems (16, 19, 21, 65-67). In advanced wireless networks, the location of mobiles can be monitored on-line and, based on that, different positioning and location information services can be offered. The requirement for inter-system mobility and roaming, i.e., the ability of a mobile user to make and receive calls in systems and places other than the home system, makes mobility management more complex. Roaming may not be possible unless many critical issues, such as location awareness, inter-system similarity or multi-system support, detailed service recording, authorization and authentication, are solved.

Handoff refers to the process of serving-cell change, i.e., shifting the cell and resources allocated for a mobile user connection during cell or system crossing. This is due to the fact that the decreasing gain of the RL connected to the current cell becomes inadequate to maintain the required QoS for the user and, at the same time, the increasing gain toward the other cell becomes more favorable. It happens either naturally at some point when a user is moving from one cell to another along the road or forcedly by the network when there are overlay cells or systems. Handoff control takes care of the handoff process and ensures continuity of the service for mobile users desirably with a seamless experience. The handoff process can be either user-triggered or network-triggered, often carried out in four successive phases: measurement, initiation/control, CA, and connection shifting. It interacts with and is related to other resource allocation schemes, such as PC, CAC, and CA. Thus, a reliable strategy for resource reservation and allocation in the event of handoffs is needed in order to provide seamless communications under often unpredictable radio resource conditions. For this matter, serving priority is usually given to a handoff request against a new request. The handoff requests may also be allowed to queue up temporarily if there are not enough resources available right away at the destination but eventually in a short time before the call is forcedly terminated. The resulting effect of the handoff, in regard to user experiences, can be characterized as seamless or hard. The latter involves some performance losses, such as packet losses, momentary destruction or interruption, or even forced termination of the service.

The handoff is increasingly important and complex in advanced wireless networks as smaller cell sizes, different overlay structures of macro/micro/pico cells and systems can be expected in addition to supporting multimedia services and an increasing number of users. The underlying problem is that handoff control schemes need to deal with a higher handoff frequency and more distributed nature of mobile multimedia connections with stricter requirements on QoS and operational costs. One must consider balancing the expected number of handoffs and the shifting load, as well as minimizing the handoff delay. If the handoff is not carried out quickly enough, the QoS may degenerate to below an acceptable level.

1.1.5 Traffic management

Together with CA, PC, and MM described above, TM handles access of users and transfer of traffic classes over the radio access system to provide an efficient QoS and

33

network performance. TM may apply a range of techniques, both basic and advanced, to implement the functional requirements as well as to boost up the system capacity and performance. CAC, MAC, packet-transmission scheduling, and congestion control are essential to TM (19-22, 31-43, 78-79).

Historically, the demand for wireless communications has consistently exceeded the capacity of wireless networks and the underlying technologies. If there are too many users in the system, the resource allocation cannot find enough resources to serve all users simultaneously, that is, to achieve their target SIR at all times as indicated in the condition (1.1). The choices of which, what and how users can be served are the matters of network policies with certain constraints and conditions. CAC, MAC, packet scheduling, and congestion control schemes jointly provide means of utilizing the available radio resources of each cell in the system to achieve the highest spectral and energy efficiency. In interference-limited CDMA systems, these schemes, together with PC, are used to keep the MAI (Multiple Access Interference) level in control for maximizing system capacity and performance. This is also referred to as interference management or load control problem (15-19, 21-22, 78).

The congestion control, dealing with traffic redundancy or overflow situations, and packet scheduling, providing an in-order transfer of data packets among users, are somewhat familiar concepts studied also in fixed networks (37-39). However, in wireless networks, congestion control and packet scheduling may have to cope with additional problems due to excessive random packet losses, delays, and delay variations over radio links, and also a shifted load from prioritized handoffs. Furthermore, the diverse QoS requirements in supporting multimedia applications add new challenges to the design of efficient congestion control and scheduling schemes as well.

In advanced wireless networks, CAC makes the system environment more stable for other schemes to operate in and prevents congestion situations, in which congestion control actions can result in hard traffic losses or even system resets. Intelligent scheduling schemes are often embedded in MAC for efficient QoS support. MAC and CAC problems are presented next as for the focus areas of this thesis.

1.2 Focus areas defined

1.2.1 Medium access control

In wireless networks, while multiple access techniques such as FDMA, TDMA, and CDMA provide means of allowing a number of users to share the transmission medium simultaneously, MAC schemes govern the users in accessing and transmitting on the common medium on a radio packet basis. MAC is needed in order to avoid or resolve possible contention or collision whenever more than one independent user attempt to use the same resource or channel at the same time. Since the first MAC scheme, called ALOHA, invented by the University of Hawaii in 1970 for data and satellite communications (80), many MAC schemes have been proposed and developed (81-86). MAC schemes can be classified in numerous ways. For example, based on access

34

methods or operation modes, MAC schemes are classified as random, scheduled, or hybrid access. In random access, there is no coordination among the users. In this case, MAC is basically simple. It allows a virtually unlimited number of users to compete for the common channel which may result in high collision probability, low spectral efficiency and induce excessive delays. The ALOHA schemes are typical examples of random access, also referred to as repeated random access. In scheduled access, users are scheduled to transmit within an allotted bandwidth or time interval, which in turn can be fixed or demand-assigned. This case requires a coordination of the users, often serving with a hard limit in the number of users. FDMA/TDMA fixed-assigned schemes are good examples of scheduled access. Hybrid access is a combination of both random access and scheduled access, or somewhat in between. Examples are random reservation schemes. CDMA-based schemes are often considered as hybrid access (17). In principle, a large number of users, each being assigned a unique code channel, can transmit simultaneously in an uncoordinated fashion without contention. However, if the number of simultaneous users rises above a certain threshold, contention will occur.

MAC is of fundamental importance to PRN which is responsible for an efficient and fair sharing of scarce radio resources while providing packet access services to wireless data users. The random and bursty nature of data users makes random or hybrid MAC schemes more attractive to PRN, not to mention their simplicity or flexibility. Those schemes based on DS-CDMA techniques have drawn a lot of attention lately as DS-CDMA has emerged as a dominant technology of 3G cellular systems. The recent convergence of fixed and wireless services has raised new requirements and challenges, also leading to more complex MAC schemes for an efficient QoS support in an integrated multimedia wireless system. The performance of MAC schemes is traditionally evaluated by the throughput, defined as the average number of radio packets successfully transmitted per unit time, and the average packet delay experienced by a radio packet. However, additional measures can be introduced to investigate the performance against a range of concerns, such as spectral efficiency, energy efficiency, and performance sensitivity to different factors in the system design and QoS. The performance characteristics, however, may exhibit significant losses due to contention, burst errors, or fading impacts of the hostile radio channel, imperfection of control or support mechanisms coexisting in the system such as PC, carrier or channel sensing, etc. It is interesting and inspiring to learn that many works, which attempt to investigate and solve MAC issues for various wireless systems, have been reported in the literature so far (13-38, 80-159). However, many problems and challenges still remain, especially in the integrated context of efficient RM for advanced wireless networks. This context can be specified with the following goals: to mitigate interference and increase spectral efficiency; to minimize power consumption and prolong battery life of MTs; and to maintain QoS over wireless environments by adapting to user movements and channel impairments. Thus, one of the focus areas of this thesis is the analysis and synthesis of adaptive MAC schemes for centralized DS-CDMA PRN, in regard to optimal trade-offs of the goals listed above.

35

1.2.2 Call admission control

In conjunction with the other essential RM schemes presented above, CAC is needed in order to prevent congestion and ensure stability and adequate QoS in advanced wireless networks, as the number of users and demands for services increase while the system capacity is constrained. CAC denotes the decision process of a wireless network to accept or reject a new call request. The decision is made depending on the available system capacity to cope with the required QoS and possible impacts imposed by the new user on the existing ones. Thus, through knowledge of the current system conditions and profile characteristics of the users, requested services and network policies, CAC acts in the same way as the passenger check-in process of an airport system. The new call request can be considered for a mobile user either initially arriving or being handed off, in short, a new call or a handoff call. This necessitates a continuous control in CAC due to the user mobility and dynamic nature of the resource availability at each serving cell of the system. The RF bandwidth is the most precious and scarce resource in cellular networks. Thus, CAC may base its decision on and for an optimum utilization of radio resources, in line with the resource allocation strategy of the system (15-19, 21-22, 48-50, 78-79, 160-193). From an operational point of view, CAC schemes can be divided into static or dynamic classes. In static CAC, a call request is accepted only when sufficient dedicated resources are available to meet its required QoS. Examples are CAC schemes of fixed-assignment cellular systems. In dynamic CAC, on the one hand, a new call may be admitted into the system even if instantaneous QoS is violated but, when averaging out over system states in the long run, the required performance can be met. On the other hand, a call request can be rejected even if the system at that time is able to satisfy its required QoS. Thus, dynamic CAC allows implementing various traffic regulating and operation support policies. One may also argue in favor of a hybrid class of CAC schemes because of the following two reasons at least. First, CA schemes underneath CAC can be hybrid as described in Section 1.1.2. Second, wireless systems like those based on CDMA have virtually no hard capacity limit.

However, CAC must be simple and flexible to guarantee a fast response to multiple and, in multimedia wireless networks, diverse QoS requests of mobile users during a connection setup of new or handoff calls. CAC directly affects the chances of calls being blocked or dropped and so also the serving capacity of wireless systems in terms of the number of users. The performance of the CAC schemes can therefore be evaluated by the blocking probability, also referred to as the service denial probability (19), and the dropping or handoff-failure probability, also referred to as the service interruption probability (19). These determine the GoS and Erlang capacity of the system. The simplicity of CAC has been taken for granted in legacy FDMA/TDMA systems. It often relies on a static sharing of a limited number of available channels to accommodate the users; each is allocated a channel from the separate channel pools of UL and DL on a per call basis. In advanced wireless systems supporting multimedia services, different users come with different bandwidth demands and QoS requirements. QoS provisioning is ever more complex as the available system bandwidth is inherently insufficient and unpredictable, considering rather limited radio resources, user mobility, interference and fading impacts, etc. This is not to mention the need of the network operator to configure

36

and manage each serving system scenario where traffic is regulated and QoS is differentiated as desired for maximizing the revenue of network operations. Issues such as traffic characterization and prioritization, resource partitioning, allocation and conservation, and, in particular, CAC are therefore posing many new challenges. The second focus area of this thesis is on the analysis and synthesis of CAC schemes for advanced mobile cellular networks. The author addresses some problems and practical solutions of CAC for an efficient GoS management, and also provides accurate and traceable analytical methods for evaluating, dimensioning, and optimizing the teletraffic performance of multimedia wireless systems.

1.3 Aim and contribution of the thesis

“To raise new questions, new possibilities, to regard old problems from a new angle requires creative imagination and marks real advance in science” (Einstein, 1938). In this regard, the ultimate aim of this thesis is to present the author’s own contributions to the knowledge of the RM problem in advanced wireless networks. In particular, the contributions are made for the two focus areas, MAC and CAC, defined above.

In spite of the substantial amount of research efforts invested in studying the RM problem in wireless networks and the enlightening results reported on MAC and CAC so far, there are still many open issues and problems in the areas of interest. The main objectives of this thesis include:

− To address some open issues and problems on MAC and CAC in the integrated context of RM in advanced wireless networks;

− To introduce new concepts, new MAC and CAC schemes for enhancing and optimizing wireless network performance;

− To provide effective analytical methods for evaluating MAC and CAC schemes as well as wireless network performance;

− To demonstrate the author’s ability to perform research, e.g., his understanding of the subject and his ability to conduct and present the research results as a “contribution to knowledge.”

This thesis presents seven separate contributions: four on adaptive MAC schemes for centralized PRN, shortly referred to as SPA (Smart Packet Access) 1-4; and three on CAC schemes for advanced mobile cellular networks, shortly referred to as SCA (Smart Call Admission) 1-3. These have been published in the eighteen original publications, Papers I-XVIII. The relationship between the contributions and the publications is summarized in Table 1.

37

Table 1. Relationship between contributions and publications.

MAC CAC Papers SPA1 SPA2 SPA3 SPA4

SCA1 SCA2 SCA3

I B × II* J × III* J × IV C × V C × VI C × VII* C × VIII* C × IX* C × X C × XI C × XII* C × XIII C × XIV C × XV C × XVI C × XVII C × XVIII* C × B: book chapter; J: journal papers; C: conference papers; those marked with * are selected for reprint in the appendices.

The following provides a short overview of each contribution.

1. SPA1: Implementation losses in adaptive feedback-control MAC schemes

This contribution provides a study of implementation losses in FCSI (Feedback Channel State Information) based MAC schemes for centralized quasi-asynchronous DS-CDMA PRN. The motivation behind these schemes is to control the packet access in the UL by statically or dynamically changing TPP (Transmission Permission Probability) depending on FCSI. This keeps the system load, that is, the MAI of the UL, under a certain threshold or outage level for maximizing the system throughput. Different probabilistic functions of FCSI for controlling TPP are introduced and investigated.

The contribution herein is twofold. First, it provides a performance analysis of the system taking into account the effects of feedback delay, access delay, and imperfect channel sensing of FCSI simultaneously, which has not been considered before. These effects are referred to as system imperfections. The system sensitivity functions are introduced to characterize the relative performance losses due to system imperfections, which are also referred to as implementation losses. The trade-offs between the average system throughput, packet delay, and outage probability in the presence of system imperfections are discussed based on a number of queuing system models and numerical results. Second, it shows that using a smart probabilistic function of FCSI for dynamically controlling TPP in a MAC operation

38

allows the system not only to outperform the ALOHA and static counterparts, but also to tolerate high degrees of system imperfections. This method is referred to as SPP (Smart Packet Persistence). The SPP thus significantly enhances the robustness of PRN over system imperfections, for example, with up to 15% of the false alarm rate in sensing FCSI and an access timing delay of 20% of the mean packet service time. This contribution is detailed in Papers II and IV.

2. SPA2: Efficient MAC scheme with adaptive bit rate

In centralized DS-CDMA PRN, a high uncertainty of radio channel and PC inaccuracy can cause severe performance degradations. The dependences between the path loss characteristics, user location diversity, transmit power and processing gain of packet transmissions with different bit rates can be exploited. This is in order to compensate for the PC problem of DS-CDMA PRN and mitigate interference; and thus enhance system capacity, coverage, and performance including the energy efficiency of mobile users.

Thus, a location-dependent bit rate adaptation, referred to as SPR (Smart Packet Rate), is proposed. The SPR adapts the user bit rate to the distance between the MT and the serving BS while allowing the transmit power of all users to be kept within a small range of a preset target throughout the cell. That is, the higher the bit rate, the nearer the users are to the BS according to a specified power location rate distribution. By using both theoretical analysis and simulation, we show that the proposed SPR system gains a significant improvement in throughput-delay trade-offs compared to the non-adaptive counterpart having the same coverage, offering traffic and QoS requirements. The SPR substantially reduces the headroom of PC and stabilizes the transmit power of MTs. This, in turn, reduces the peak transmit power and power consumption of MTs, as well as the inter-cell interference. The SPR therefore significantly enhances spectral and energy efficiency. The SPR is gradually developed and evaluated in Papers XI, XII, XV and XVI, considering various system scenarios, models, parameters and their impact on the system performance. This has led to the development of new concepts, such as “green wireless networks” reported in Paper XV, and analytical methods to deal with the complex dynamics of multi-user multi-rate wireless systems including stochastic fluctuations of radio channel and multiple-access capacity, traffic load, user spatial distributions, user mobility, and PC inaccuracy. These can be claimed as additional contributions as well.

3. SPA3: Efficient MAC scheme with adaptive packet length

The impact of highly erratic radio channels (i.e., fading impacts) often causes serious performance losses to PRN. In this regard, the length of the radio packets transmitted over the air, in bits or octets, matters. That is, the smaller the length, the better the chance of a successful packet transmission over a fading radio channel and, at the same time, the heavier the protocol overhead.

This contribution, based on both theoretical analysis and simulation, shows that in burst-error correlated fading environments adapting the length of radio packets to channel conditions significantly improves system performance in terms of robustness, spectral and energy efficiency. This adaptation is referred to as SPL (Smart Packet Length). DS-CDMA PRN is also used as the core system for the

39

investigation approach, similarly as in the previous contribution. The channel model is adapted to low user mobility in PRN applications. First, based on a study of the interrelations between the burst-error or second-order fade statistics, packet length, packet-header size, and their impact on the effective throughput, alternative cost functions are developed and optimal packet lengths are derived. Then, a performance analysis of the adaptive SPL system is carried out, in comparison with the non-adaptive counterpart. The SPL is gradually developed and evaluated in Papers III, XIII and XIV. The SPL is also found in Paper XVIII for an optimal performance of end-to-end applications. The SPL is adopted to further enhance the robustness of packet transmissions over correlated fading channels and, at the same time, the effective throughput over the protocol overhead. Furthermore, similar claims for providing advanced analytical methods as described in the previous contribution are also valid for this contribution.

4. SPA4: Adaptive MAC schemes for optimal trade-offs between throughput and energy efficiency of TCP applications

TCP (Transmission Control Protocol) is a reliable transport layer protocol for end-to-end flow and congestion control. TCP currently carries the largest amount of traffic on the Internet. The convergence of fixed and wireless services, especially the increasing popularity of wireless Internet access, expects that TCP applications also work well in wireless environments. TCP is optimized for wired networks. In wireless networks, TCP exhibits performance degradation because of random packet losses, unpredictable delay, and delay variation over RL. Furthermore, the battery energy conservation for MTs, operating such applications more often and for a longer period, is an important factor to ensure service success.

This contribution develops adaptive MAC schemes to achieve optimal trade-offs between TCP throughput and energy efficiency over burst-error correlated fading channels in CDMA systems. Both HSPA (High-Speed Packet Access) and LSPA (Low-Speed Packet Access) with RL retransmission-and-error-control protocols are studied. By using both theoretical analysis and simulation, this contribution shows that in HSPA adapting the retransmission period of erroneous radio packets to the channel conditions significantly reduces the number of retransmissions while maintaining an adequate TCP performance. This method is referred to as SPD (Smart Packet Dispatching). The SPD schemes are derived based on the development and optimization of alternative cost functions, which represent the balance of a high throughput and low waste of energy in RL error recovery. In LSPA, the SPL, based on the previous contribution, is adopted to significantly improve the system performance. The SPL scheme herein is derived from a new cost function developed for an optimal trade-off between TCP throughput and power consumption. The SPD and SPL are capable of overriding the burst-error states of correlated fading channels and therefore enhance the robustness of radio transmissions and allow the system to operate with a much lower fade margin and transmit power. Furthermore, the SPD and SPL improve the energy efficiency and system performance without a need of excessive signal processing. This contribution is reported in Papers XVII and XVIII.

40

5. SCA1: Performance analysis of a class of fixed channel assignment CAC with generic guard channel policy and queuing priority handoffs

This contribution provides a simple and accurate analytical method for performance evaluation of a class of fixed-assignment CAC schemes with generic guard channel policy and queuing priority handoffs in mobile cellular networks. The handoff requests are allowed to queue up if upon their arrival no idle channel is found. The handoff dwell time, i.e., the time period needed for the mobile to pass through the handoff area or maximum handoff waiting time, is modeled as a general random variable and not just exponential as in earlier works. The new call requests are accepted according to a generic guard channel policy giving priority to handoff calls. Thus, a virtual number of guard channels are reserved for handoff calls in an effective and generic fashion, in which new calls are rejected with a certain probability depending on the number of occupied channels. The effect of cell sizes and random user movements on the handoff rate is taken into account. The analytical model is developed based on a birth-death process with state-dependent Poisson arrivals, state-dependent decomposed service time, multiple servers, finite or infinite handoff buffer and generic handoff impatience until the beginning of service. The measures of interest include the probabilities of a forced termination of a handoff call, handoff failure, and new call blocking. The closed-form solutions are obtained. The results show that there is a significant difference in the probability of a forced termination of a handoff call regarding different PDFs (Probability Distribution Function) used for modeling the handoff dwell time. The deterministic one provides a lower bound for the loss probability. The results also show that it may not necessarily require a large buffer size for queuing handoffs. The suitable queue size, however, still depends on the offered load. This contribution is presented in Papers V and VII with results for micro and macro cellular systems, respectively.

6. SCA2: Simple and efficient soft-decision CAC for CDMA cellular networks

This contribution proposes a SCAC (Soft-decision CAC) scheme for CDMA cellular networks. This exploits the soft-capacity feature of CDMA systems by using a probabilistic decision for admitting calls, adapted to interference fluctuations to compensate for QoS, while expanding the admissible region to accommodate more users. This method not only offers a notable enhancement of the GoS and Erlang capacity, but also avoids complex measurements and mutual exchanges of information between adjacent cells about the states of the network. This is therefore a simple, scalable, signaling, and cost-effective method.

This contribution is twofold. First, alternative probabilistic decision functions are developed for SCAC, i.e., using Gaussian and Incomplete Gamma PDF adapted to the inter-cell interference distribution. Second, a simple and reliable method for evaluating SCAC against fixed-assignment and measurement-based CAC schemes is provided. The analytical model is developed based on a stochastic loss network model with an introduction of the so-called scaled effective bandwidth. This represents the fluctuating load factors of both the connection and cell basis. The priority handoff and user mobility are also considered, based on the previous

41

contribution. Numerical and simulation results are provided. This contribution is presented in Papers VI, IX and X with supplementing results.

7. SCA3: GoS management for multimedia CDMA cellular networks

For the deployment of multimedia cellular networks, such as 3G CDMA systems serving with various user profiles and system scenarios, there is a need for a simple and flexible mechanism that allows the operator to manage rather complex network operations and performances in an efficient way. That is to optimize the GoS and revenue while providing adequate QoS to mobile multimedia users. This contribution defines the GoS for multi-user multi-service cellular systems as a cost function built upon a linear combination of the probabilities of a new call blocking, handoff failure, and loss of communication quality. The weighting factor of each argument in the GoS cost function depends on service and user profile characteristics. This contribution, through incorporating alternative QoS differentiation paradigms and resource partitioning into CAC, provides an in-hand tool for efficient capacity and teletraffic management. It introduces different user prioritization and load-based traffic-shaping schemes for GoS provisioning and optimization. The CAC schemes and analytical methods of the previous contribution are further developed and expanded for an integrated multi-class multi-service system. It is shown that the SCAC exhibits further advantages in maintaining a good and stable GoS performance over a large range of offered traffic with various multimedia traffic patterns. Furthermore, an asymptotic quasi-stationary analysis of the remaining free capacity over a sustained period of time, left for best-effort packet services for instance, is provided. This is important for the design of effective MAC schemes capable of making use of the available system capacity effectively for the background packet access users while not affecting the QoS of the ongoing connection-oriented users. Examples of such adaptive MAC schemes are also presented. The analytical model is based on a stochastic, multi-priority, multi-rate loss network system. Extensive numerical and simulation results are provided. This contribution is reported in Papers I and VIII.

The applications of the above contributions can be recognized, for example, in most recently enhanced concepts of the radio access systems in UMTS and IMT-2000 standards, and tools for cellular network planning, dimensioning, and management (22). These can also be used for future wireless networks in both military and industry and towards building the mobile Internet (6).

1.4 Organization of the thesis

This thesis is presented as a collection of articles based on the eighteen original, peer-reviewed papers of the author, Papers I-XVIII. Seven of them, the most representative papers, are selected for reprint in the appendices. The presentation style is carefully thought of to support the thesis intents, i.e., to highlight the author’s contributions and make it accessible to a broad readership as a book. This introductory chapter, so far, has

42

provided the motivations, defined the focus areas, and outlined the contributions of the thesis. The rest of this thesis is organized as follows.

Chapter 2 provides a literature survey of MAC and CAC schemes confined to advanced cellular systems. It discusses the most common design properties and performance measures, classifies and surveys the most relevant, up-to-date schemes that have been reported in the literature for MAC and CAC, respectively.

Chapter 3 summarizes the main contributions of the thesis. It provides the insights of the main problems, claims, approaches, results, and conclusions making up each of the contributions, as well as for the reading of the original publications, Papers I-XVIII. Chapter 3 is further divided into two sections, one for SPA and one for SCA.

Chapter 4 concludes the thesis by providing commentary remarks on its main results and contributions. Some directions for future research are also provided.

2 Literature survey

This chapter reviews the literature on MAC and CAC schemes confined to centralized PRN and advanced cellular networks. It starts with a capacity provisioning of basic FDMA/TDMA and DS-CDMA cellular systems, as the capacity of multiple-access systems is of great importance to the design of effective RM schemes and the corresponding parameter setting. It then discusses the most common design properties and performance measures, classifies and surveys the most relevant, up-to-date schemes for MAC and CAC, respectively. This chapter, however, is not intended to perform an extensive coverage of the subjects. To remind the readers, the working assumption, as stated in Chapter 1, is mainly focusing on the multiple-access channel in the UL of FDD DS-CDMA cellular networks.

2.1 Capacity provisioning of cellular systems

The capacity of a communications system can be defined in a number of ways. Provided that there is only one class of users, the most straightforward definition of the capacity is the maximum number of users that can be served simultaneously in the system. However, the dynamic nature of the system conditions and offered load has a significant impact on the actual capacity of the system. In cellular networks, the channel capacity, defined as the maximum number of channels or users that can be provided in a fixed RF band (16, 19, 21), is one of the most essential system parameters. The spectral efficiency of a wireless system is determined by the required CIR or SIR and the channel bandwidth. Depending on the statistics of the offered load and the characteristics of the users, for example, how calls or packets arrive at the system for service and how the system handles them, the Erlang capacity, the system throughput, and other performance measures can be derived.

44

2.1.1 Radio propagation model

The propagation of radio signals through a physical channel is subjected to path loss, long-term shadowing, and short-term fast fading. Provided perfect knowledge of the surrounding environment, the channel attenuation can, in theory, be calculated (12, 16, 46). This is impractical and prohibitively complex in most cases. Thus, one has to rely on a propagation model, which provides a simplified, statistically close description of the channel attenuation, i.e., the path gain of the RL in the system under investigation.

The received signal power Prx can be expressed as the transmitted power Ptx multiplied by the path gain g:

gPP txrx = . (2.1)

The path gain g is often given as the product of three separate components: gp for path loss; gs for long-term shadow fading; and gf for short-term fast fading:

fsp gggg = . (2.2)

However, the path gain model can also incorporate additional gains produced by smart solutions of the transmitter and the receiver as noted in Chapter 1 Section 1.1.1. The free-space path loss is usually the dominating factor, especially in satellite communications. It also causes the near-far effect in CDMA systems. It is modeled as

α−= rCg pp , (2.3)

where Cp is a constant, which depends on the gain at the receiving antenna and the wavelength of the radio signals; r is the distance between the transmitter and the receiver; and α is the environment-dependent propagation path-loss exponent, ranging from 2 to 5 (12-19). For cellular systems, Gilhousen (160) suggests the value of α around 4.

The long-term shadow fading is most often modeled as a lognormal distributed RV (Random Variable) (12-19):

10/10ξ=sg , (2.4)

where ξ is the attenuation in dB as a zero-mean Gaussian RV having a mean of zero and a standard deviation of 8 dB as suggested by the Okumura-Hata model (194-195). This model assumes that two consecutive samples of the long-term shadow fading are independent. There are more sophisticated models taking into account, for example, the speed of user movements and the correlation distance representing the size of obstacles (196-197).

The short-term fast fading, and as a result of that the superposition of multi-path propagation of the transmitted signal causes rapid, stochastic fluctuations in the amplitude of the received signal, is often modeled as a Rayleigh, Rician, or Nakagami process (12, 16, 44-46, 198-205). The Rayleigh distribution is most extensively used in

45

the literature, assuming that the received signal has no strong light-of-sight component. In this regard, Jakes’ original work (12) is one of the most cited references. The fast fading can cause local deep fades in the allocated frequency spectrum. This can be devastating, especially in narrowband systems. In wideband systems, sometimes the gain can be larger than 1 (44-46) or, that is, 0 dB when multi-paths constructively interfere.

The multipath propagation may result in shifting or spreading of the signal in three different dimensions: delay spread in time, Doppler spread in frequency, and angular spread in space (12, 16, 44-46). The delay spread is due to a frequency-selective channel when the bandwidth of the transmitted signal is greater than the coherence bandwidth of the channel. The coherence bandwidth is defined as the maximum frequency separation for that the channel impulse responses at two separate frequencies remain strongly correlated. The delay spread causes ISI (Inter-Symbol Interference), which is considered as a distortion of the adjacent symbols at the receiver. The Doppler spread is due to a relative motion between the transmitter and the receiver, that is, the Doppler effect. The Doppler spread is inversely proportional to the coherence time, defined as the maximum time separation for that the channel impulse responses at two time instances remain strongly correlated. The angular spread refers to the distribution of the angle-of-arrival of multipath components at the receiver and the angle-of-departure at the transmitter in [0, 2π]. This is caused by a spatially selective channel and is inversely proportional to the coherence distance. The coherence distance, normalized by the carrier wavelength, determines the maximum spatial separation for that the channel impulse responses at two antennas remain strongly correlated. One can find profound details on the radio propagation and modeling issues of multi-dimensional fading channels in (12, 16, 44-46).

The spreading of signals in multipath fading channels have been exploited in different diversity techniques, in which the user signal is sent/received with multiple copies to increase the robustness of radio transmissions. Diversity and PC are considered as direct methods to combat the impact of fading, especially deep fades. In CDMA systems, a RAKE receiver, capable of receiving and combining multipath signals over frequency-selective fading channels, provides an effective means of implementing the reception diversity and stabilizing the path gain of an RL (22).

2.1.2 Capacity of FDMA/TDMA systems

The total number of simultaneous channels that can be supported in an FDMA/TDMA system is given by

c

guardt

BBBm

N)2( −

= , (2.5)

where Bt is the total bandwidth allocated to the system, Bguard is the guard bandwidth at each size of the allocated RF band, Bc is the channel bandwidth, and m is the number of time slots per channel. The capacity of FDMA/TDMA systems is bandwidth-limited and has a hard constraint.

46

The cellular structure is assumed to consist of a number of uniform hexagonal cells as depicted in Figure 2. The frequency reuse pattern allocates a subset of the total number of frequency channels to a particular cell and different subsets to neighboring (and nearby) cells in such a way that all the available channels are assigned to a relatively small number of neighboring cells forming a cluster. This is repeated with clusters placed side-by-side to cover the service area, meaning that any of two co-channel cells are spatially separated by a minimum distance to ensure a tolerable co-channel interference as a function of path loss. This is characterized by the co-channel reuse ratio, defined as the ratio of the minimum distance between two co-channel cells to the cell radius. The channel capacity of an FDMA/TDMA cellular system, channels/cell, is defined as

kNC = , (2.6)

where k is the number of cells in a cluster of the frequency reuse pattern. For a numerical example, let us consider a GSM (Global System for Mobile Communications) system, which has a system bandwidth of 1.2 MHz divided into channels of 200 kHz and then 8 slots per channel for 8 users. In this case, k = 7 as shown in Figure 2 and the channel capacity is obtained with about 1200/200×8/7 ≅ 7 users per cell.

Fig. 2. Illustration of cellular structure and frequency reuse pattern. The Erlang capacity of the system can be estimated based on the trunk system concept, which allows a large number of users to share the limited number of channels in a cell by providing a channel to each user upon their arrival from the pool of available channels in FCFS (First Come First Serve) fashion (16, 19, 37, 211-213). The statistics of the user behavior such as the user arrival and channel –holding time processes determine the offered traffic intensity of the system. There is a trade-off between the number of available channels and the likelihood of a particular user finding that no channel is available upon their arrival during a peak traffic period. This can be exploited and utilized for the design of efficient CAC.

A

F

D

C

G

B

C

D

E

A

F

G

B

C

D

E

47

There are two basic types of trunk systems: BCC (Blocked Calls Cleared) and BCD (Blocked Calls Delayed) (16, 19). The BCC systems offer no queuing for call requests. The BCD systems, on the other hand, allow the call request to queue up if there is no channel available. Providing that the call inter-arrival time and channel holding time, i.e., the service time, are exponentially distributed, the BCC and BCD systems can be modeled as M/M/C/C and M/M/C/∞ queuing systems leading to the Erlang-B and Erlang-C formulas, respectively (16, 19, 37, 211-213). Let us use the following notation:

− The call arrival is according to the Poisson process with total rate λ; − The channel holding time is exponentially distributed with mean 1/μ; − The offered traffic intensity is A = λ/μ; − The channel capacity of the system is C.

The Erlang-B formula determines the probability that a call is blocked, and is given as

∑=

= C

j

j

C

jA

CA

blocking

0 !

!}Pr{ . (2.7)

The Erlang-C formula determines the probability of a call not having immediate access to a channel, and is given as

∑−

=⎟⎠⎞

⎜⎝⎛ −+

= 1

0 !1

!

!}Pr{ C

j

jC

C

jA

CA

CA

CA

delaying . (2.8)

The formula (2.7) defines the GoS of the BCC systems. In the BCD systems, the GoS is often measured by the probability that the call is dropped after waiting for a specific amount of time, t, in the queue.

))(exp(}Pr{}Pr{}Pr{ tACdelayingtdelaydropping μ−−=>= . (2.9)

The average delay D for all calls in the BCD system is given by:

μ)(1}Pr{AC

delayingD−

= . (2.10)

In an FDMA/TDMA cellular system, a mobile user can communicate with only one BS at any time and different radio channels need to be assigned during a hard handoff. From the user perspective, handoff failure and forced termination of a call, as the result, is more annoying than a new call attempt being blocked occasionally. This leads to different CA strategies in the event of handoffs which often prioritize handoff requests over new call requests when allocating channels in a cell site. For example, the following two policies

48

are often adopted to give priority to handoff calls: GC (Guard Channels) and QH (Queuing Handoffs).

In GC, a fraction of the channel capacity in a cell is reserved exclusively for handoff calls. This reservation, however, may reduce the total carried traffic of the system and lower the utilization of the scarce radio resources.

In QH, queuing of handoff requests is utilized to make time for the handoff process in case there is no free channel available momentarily at the destination cell. However, the mobile may pass through the handoff area, meaning that the signal level drops below a minimum threshold, before getting a new channel. Thus, the waiting time for a handoff request can run out and the call is then forced to terminate. The finite time interval from the instant the handoff is initiated to the instant the call must be handed off or dropped is referred to as the handoff dwell time. The queuing of handoff requests decreases the chances but does not guarantee zero probability of handoff failures.

In the above analysis, the offered traffic intensity is composed of new and handoff calls. The cell size structure and user movement have an impact on the handoff statistics. The Erlang capacity of FDMA/TDMA cellular systems is given as the maximum number of users that can be offered a minimum GoS with a particular configuration of the fixed channels.

2.1.3 Capacity of CDMA systems

Unlike FDMA/TDMA systems, CDMA systems are interference-limited and have no absolute limit on the number of users. The system performance gradually degrades for all users as the number of users and MAI is increased. To ensure quality of radio transmissions in CDMA systems, a lower bound of SIR, also referred to as the SIR target, is required to restrict a maximum number of users in the system. This follows the condition (1.1). The capacity in CDMA cellular systems varies with environmental factors, such as cell coverage, path gain under PC, and interference caused by the load in neighboring cells. This is due to the unique characteristics of CDMA cellular systems, including one-cell frequency reuse, PC, soft handoff, and multidimensional diversity with exploiting multipath propagation, sectorization, and data-source activity of the user nature (10, 15-19, 21-22).

There is a substantial amount of literature studying the capacity of CDMA cellular systems (10, 15-19, 21, 22, 29, 30, 160-168). One of the most commonly accepted models for deriving channel capacity in CDMA cellular systems is originally demonstrated in Gilhousen (160). The system is assumed to consist of uniform hexagonal cells with one-class voice users and perfect PC. The maximum number of users per cell is calculated for both single-cell and multi-cell system scenarios, considering intra-cell or own-cell and inter-cell or other-cell interference. It shows that other-cell interference can be approximated by own-cell interference multiplied by a ratio factor. The sectorization and the on-off activity of voice users are considered as capacity improvement factors. The outage probability is defined as the probability that the required BER (Bit Error Rate) performance is not met, which is equivalently mapped on the condition of the RL quality similar to (1.1) not being satisfied. The Erlang capacity is also studied in (6, 161). It

49

considers that a call arrival is blocked when the total interference of the given cell, normalized by the background noise level, exceeds some predefined threshold (i.e., 10 dB). This “soft blocking” condition depends on the soft capacity of the CDMA system, in contrast to “hard blocking” of the FDMA/TDMA system when there is no free channel.

The capacity of multirate CDMA systems for supporting multimedia services is investigated in (21-22). That whether the UL or the DL capacity is the bottleneck depends on the symmetry and balance of the user applications and traffic distributed between them. For example, emerging data services with applications like interactive web browsing or real-time multimedia downloading can cause UL-DL traffic imbalance and, in this case, the DL capacity becomes the bottleneck. The recent HSDPA (High-Speed Downlink Packet Access) concepts in UMTS and CDMA2000 have been developed to cope with such applications, tailored for TCP/IP based and broadband mobile Internet access (130-136). The impact of imperfect PC, soft handoff, different diversity techniques, QoS requirements, and RM schemes on the capacity of CDMA systems have been the subjects of many recent studies (19, 21, 22, 162-168). However, the common sense is that user transmission with a higher bit rate must be provided with higher resource capability in terms of code channel power and transmit power. The resource allocation for best-effort data users has to adapt to the remaining capacity left by guaranteed users, such as real-time voice/video calls.

Based on the above discussions and references, the following provides a simple and insightful capacity provisioning of DS-CDMA cellular systems, focusing on the UL. The following notation is used in addition to what has been defined so far:

− W denotes the chip rate or the spreading bandwidth of the system; − Rj denotes the bit rate of user j among N active users in the same cell, j∈{1, 2, ..., N}

The ratio of received energy per bit, Eb, to interference-plus-noise density, I0, for user j is expressed as the product of processing gain W/Rj and received SIRj:

jrxtotal

jrx

jjb PI

PRWIE

,

,0 )/(

−= , (2.11)

where Prx,j is the received signal power of user j and Itotal is the total received wideband power at the BS. Itotal includes own-cell interference Iown generated by N active users in the same cell, other-cell interference Iother generated by active users in neighboring cells and background noise N0.

2.1.3.1 Single-cell system with one class of users and perfect PC

For a simple start, let us revisit the Gilhousen model (160) of a single-cell system with one class of users: Rj = R, and perfect PC: Prx,j = Prx. Thus, Equation (2.11) becomes

00 )1(

/NPN

PRWIE

rx

rxb +−

= . (2.12)

50

For a given value of Eb/I0 required for a certain QoS performance, the number of users is:

rxb PN

IERWN 0

0/11 −+= . (2.13)

Equation (2.13) shows that the capacity of the system is inversely proportional to the required Eb/I0. This explains why the design of CDMA networks tends to apply many sophisticated signal-processing techniques, such as advanced receivers and powerful channel coding, in order to keep the required Eb/I0 as low as possible. The required Eb/I0 of voice users is suggested as 5 or 7 dB with R = 9.6 kb/s W = 1.2288 Mc/s in IS-95 and 5 dB with R = 12.2 kb/s W = 3.84 Mc/s in UMTS. The expected noise figure, defined as η =E[N0/(Itotal−N0)] that is equivalent to η =N0/NPrx in this case, is assumed as 10% or –10 dB. Equation (2.13) can be rewritten for a capacity estimation as:

⎟⎟⎠

⎞⎜⎜⎝

⎛+

+=

0/11

11

IERWN

bη. (2.14)

The application of antenna sectorization, which uses directional antennas at the cell site for both receiving in UL and transmitting in DL, offers an increase in cell capacity by as many folds as the number of sectors approximately. Furthermore, monitoring the activity of the user data sources and switching the users on/off accordingly also effectively reduces interference and increases capacity. Equation (2.14) can be rewritten as:

⎟⎟⎠

⎞⎜⎜⎝

⎛+

+=

0/11

11

IERaWN

bvη, (2.15)

where av is the voice activity factor. If av = 3/8 and three sectors per cell site are applied, then about an 8 fold increase in the number of users in a cell can be expected.

2.1.3.2 Uniform multi-cell system with one class of users and perfect PC

Because of the perfect PC assumption and (2.1), a user with index m has to transmit with a power of Ptx,m = Prx/gm, where gm is the path gain of the link towards its own BS. In (15, 160), the path gain of the propagation model is given as g = 10ξ/10r-4. The signal of user m at a distance rm from its own BS will produce interference to the reference site at a distance r0 away from it as

10/)(4

0,00

010)/(),( m

rrPggPrrI m

rxmmrxmξξ −

⎟⎟⎠

⎞⎜⎜⎝

⎛== , (2.16)

51

ξ0 and ξm, defined in Section 2.1.1 at (2.4), are independent so that the difference (ξ0–ξm) has zero mean and variance 2σ2.

The total other-cell interference normalized by the received user signal is given by

∫∫ −⎟⎟⎠

⎞⎜⎜⎝

⎛= − dArr

rr

PI

mmm

rx

other m ρξξφψ ξξ )/,(10 0010/)(

4

0

0 , (2.17)

where m also represents the cell-site index of all the neighboring cells for which

⎪⎩

⎪⎨

⎧≤⎟⎟

⎠

⎞⎜⎜⎝

⎛=−

−

otherwise0

110if1)/,(10/)(

4

000

0 m

rr

rrm

mm

ξξ

ξξφ , (2.18)

ψ is a binomial RV representing the voice activity taking value in {0, 1} for which Pr{ ψ = 1} = av and Pr{ ψ = 0} = 1–av; and ρ is the density of the SUD (Spatial User Distribution) given in number of users per unit area. The computational results in (15, 160) for the system with uniform, 3-sector hexagonal cell sites show that the mean and variance of Iother/Prx are bounded as E[Iother/Prx]≤0.247N and Var[Iother/Prx]≤0.078N. One can notice that the product NPrx represents own-cell interference, Iown. Thus, the expected other-cell interference to own-cell interference ratio ι = E[Iother/Iown] can be defined and used for modeling the impact of other-cell interference on system performance. Based on the above analysis, Equation (2.11) is now applied for this multi-cell system scenario and expressed by

rxrxotherN

i ib

PNPIRWIE

//1/

01

1

0++

=∑ −

=ψ

. (2.19)

The outage probability is defined in (5, 160) as:

})/(/Pr{ 00 voicebboutage IEIEP <= , (2.20)

where (Eb/I0)voice is the minimum or target value of Eb/I0 required to meet QoS requirements, set to 5 or 7 dB as mentioned above for the IS-95 system. That is

}/Pr{ 1

1∑ −

=>+=

N

i rxotherioutage PIP δψ , (2.21)

where the value of rxvoiceb P

NIER

W 0

0 )/(−=δ approaches the single-cell system capacity

given by (2.13), assuming that the processing gain W/R is much larger than 1.

52

∑ ∑∑−

=

==−>=1

0

}Pr{}|/Pr{N

k ii

iirxotheroutage kkkPIP ψψδ , (2.22)

[ ][ ]∑

−

=

−−

⎟⎟⎠

⎞⎜⎜⎝

⎛ −−−⎟⎟

⎠

⎞⎜⎜⎝

⎛ −=

1

0

1

//)1(

1N

k rxother

rxotherkNv

kvoutage PIVar

PIEkQaak

NP δ , (2.23)

where Q(x) is the standard Gaussian distribution resembling the distribution of interference.

The results of (5, 160) show that in the IS-95 DS-CDMA system with a bandwidth of 1.25 MHz, 3 regular sectors per cell, W = 1.2288 Mc/s, R = 9.6 kb/s and (Eb/I0)voice = 5 dB, the capacity per sector for an outage probability of 1% is obtained for about 37 users. This is about 5 times higher than that of the GSM FDMA/TDMA example mentioned above, which has a comparable bandwidth of 1.2 MHz. UMTS, using WCDMA with a bandwidth of 5 MHz and W = 3.84 Mc/s, is expected to have a much higher capacity than IS-95. This is not only due to the fact that the chip rate is now three times larger, but also because the multipath resolution and diversity combined in the wideband receiver operation are more effective and result in lower Eb/I0 required.

2.1.3.3 Load factor of multi-class users

Now, solving Equation (2.11) for Prx,j gives:

totaljjrx IuP =, , 1)/(

111

0

<<⎥⎥⎦

⎤

⎢⎢⎣

⎡+=

−

jbjj IER

Wu . (2.24)

The variable uj represents the load factor of user j contributed to the overall UL interference in the cell. It is inversely proportional to the processing gain W/Rj and proportional to (Eb/I0)j required for a certain QoS with bit rate Rj. The total UL load factor is

1//, <== ∑∑ totalowntotalj

jrxj

j IIIPu . (2.25)

In the equilibrium condition of the capacity limit in terms of the maximum tolerable interference, Imax, we have

[ ]totalownNIINjjNII

IIEuEtotaltotal

/limlimmaxmax →→

=⎥⎥⎦

⎤

⎢⎢⎣

⎡∑ . (2.26)

Providing that Itotal = Iown+Iother+N0, E[N0/(Itotal−N0)⏐Itotal → Imax] =η, E[Iother/Iown] =ι, and the link quality requirement is maintained, the average upper limit of the total UL load

53

factor, representing the expected cell capacity of a multi-cell multi-rate system, can be approximated with

1)1)(1(

1limmax

<++

=⎥⎥⎦

⎤

⎢⎢⎣

⎡∑→ ιηNj

jNIIuE

total

. (2.27)

Note that the ratio Iother/Iown is often denoted by f as well in the literature. On a connection basis, taking into account the data source activity factor aj of user j and given that the received Eb/I0 of user j is kept at its target level γj for an adequate QoS, the mean of the effective load factor of a connection can be defined as

111

−

⎥⎥⎦

⎤

⎢⎢⎣

⎡+=

jjjj Ra

Wwγ

. (2.28)

On a packet transmission basis, aj = 1 is held for all j. The channel capacity Cj for a class of user j, also referred to as the capacity index of a

user class (126, 129, 178), is defined as the maximum number of tolerable simultaneous users of class j in a cell and approximated by

⎥⎥⎦

⎥

⎢⎢⎣

⎢

++=

jj w

C)1)(1(

1ιη

, (2.29)

where ⎣x⎦ denotes the largest integer that does not exceed the argument x as a positive real number. Equation (2.29) is equivalent to (2.15) by setting ι = 0 for a single-cell scenario.

The above analysis has shown that the capacity of an interference-limited DS-CDMA cellular system is bound and varies in time. The mean of the effective capacity in terms of the average upper limit of the total UL load factor, the effective load factor of a user is either a packet or a connection, and the capacity index for a particular class of users have been introduced. There is a trade-off between the system capacity and the communication quality.

In advanced wireless networks, MAC and CAC schemes play crucial roles in controlling the allocation of system capacity to diverse users and, therefore, have a major impact on the system performance. MAC and CAC schemes of interest will be surveyed next.

54

2.2 MAC schemes in centralized packet radio networks

2.2.1 Properties and performance measures

In a wireless network with users transmitting traffic of any type on a shared medium, the procedures guided by MAC schemes must be invoked to distribute packet transmissions among users. The design of a good MAC scheme, however, has to consider many issues. The common properties that MAC schemes possess fall into two main categories. The first is associated with the unique properties and consequences of a wireless medium and system environment. The second is concerned with system performance issues, upon which the essential performance measures of the MAC schemes are defined.

2.2.1.1 Design properties

In the first category, the main properties include:

− Duplex Operation: FDD or TDD (Time Division Multiplexing) must be used to avoid self-interference, as it appears to be difficult, if not impossible, for a user to transmit and receive data at the same time on the same radio frequency. FDD allows full-duplex and more symmetric operations between UL and DL. TDD supports half-duplex operations and is more beneficial for serving with unbalanced traffic between UL and DL by adjusting the size of the UL and DL time frames accordingly.

− Multiple Access Techniques: The applied techniques, for example, FDMA/TDMA or CDMA, establish the basic method of dividing resources into accessible mediums or channels for the system that the MAC schemes must follow and be based upon. The evolution of multiple access techniques as well as wireless networks is associated with and facilitated by the advent of technologies. Therefore, FDMA/TDMA, CDMA, and so forth, also represent the technology platform of wireless systems.

− Impact of Time-Varying Fading Channel: The nature of the wireless medium coupled with the user mobility results in a time-varying fading channel. The user is said to be fading when the received signal strength drops below a certain threshold, also referred to as a fading margin. Thus, errors are likely in wireless transmissions. BER can be as high as 0.1% or even higher in a wireless environment resulting in a high probability of packet errors. Furthermore, errors often occur in a long burst during fading depending on the correlation properties of the fading process and the error-correcting capability of the system on the physical level. The countermeasures against fading impacts that have been proposed at the radio physical layer in wireless networks apply a number of techniques, ranging from direct PC to advanced receivers and powerful FEC (Forward Error Correcting) channel coding methods. These, most often, cannot get rid of the adverse effects of fading entirely. In this regard, it is of interest in the network design to examine the nature of the residual errors that are passed up from the physical layer to the upper layers. In particular, the MAC layer, right on top of the physical layer, is the first one facing the inherent

55

errors. Thus, the impact of the time-varying fading channel has to be dealt with quite directly at the MAC layer. Packet losses due to burst errors can be minimized by using, for example, the following techniques:

− Smaller size packets − FEC coding − Intelligent scheduling and adaptive transmissions − Retransmission methods

The latter, using a proper ARQ protocol over the RL, is widely used in advanced wireless networks. However, most MAC schemes have an immediate acknowledgment to detect packet errors. If an acknowledgment is not received at the end of a transmission, the packet is retransmitted. It is also suggested in (22, 45-46, 147, 228) that, if properly managed, fading can be exploited for contention resolution in the MAC operation as well.

− Location-Dependent Effects: Because of the path loss, the signal strength decays according to a power law of distance between the transmitter and receiver as shown in (2.3). The location of a transmitter relative to a receiver has an important impact on carrier sensing or user detection in multiple-access operations. The location-dependent effects include: hidden or exposed terminals, capture, and near-far effect. The hidden-terminal problem refers to when a terminal is within the range of the intended receiver but out of range of the sender, that is, hidden from the sender (17, 20, 88). The exposed-terminal problem is complementary to the hidden-terminal problem when a terminal is exposed to the sender (17, 20). In distributed CSMA (Carrier Sense Multiple Access) systems (16-20, 87-88), hidden terminals can cause collision on data transmission whereas exposed terminals unnecessarily delay the transmission attempts of the reference sender and waste the channel bandwidth. These problems can be avoided in centralized wireless networks. The capture effect refers to when the receiver can correctly receive (or capture) the user(s) that comes with the strongest signal from simultaneous transmissions within its range. The capture effect improves the throughput-delay performance, but results in an unfair share of the channel bandwidth with preference given to users which have a better path gain, for example, those closer to the BS (86). The near-far effect occurs when the received signal at the BS from a user close to it is much stronger than that of users far from it, for example, at the cell boundary. In centralized CDMA systems, the near-far effect can dominate the received MAI and have a major impact on the system capacity and performance (13-22). To achieve a considerable capacity, all signals, regardless of their distance in the cell coverage, should arrive at the BS with the same mean power. To solve this problem, PC is required. Therefore, the performance of PC is one of the deciding factors on the capacity of a CDMA system.

In general, the design of a MAC scheme is accomplished with a specific goal in mind which targets certain system environments and performance requirements. The main performance properties of the MAC schemes include: efficiency, fairness, stability, flexibility, and robustness. These are briefly discussed in the following.

56

− Efficiency: MAC schemes perform their tasks in such a way that the transmission medium is used efficiently. The efficiency is usually measured in terms of system throughput and average packet delay. This is the most important property in wireless systems with scarce radio resources. This also represents the spectral and energy efficiency of wireless systems.

− Fairness: This is in regard to a fair share of channel capacity toward individual users. Each user of the same class should, on the average, receive the same amount of allocated channel capacity. An unfair share of the channel capacity can cause system unbalance and unpredictable behavior. The capture and the near-far effect are good examples of causing such an unfair share.

− Stability: In general, a MAC scheme is stable if it is capable of moving the system from a current equilibrium state to a new equilibrium state, as the offered load is increased. On the other hand, with an unstable MAC scheme, an increase in the offered load can force the system to fall apart or continue to be jammed with even a higher accumulated load. This results in a drop of system throughput and intolerable prolongation of packet delay. Thus, to enhance stability, traffic-shaping mechanisms such as flow control, congestion control, and CAC are adopted.

− Flexibility: In advanced wireless networks, it is important to have flexible MAC schemes, which allow supporting multimedia traffic with diverse QoS requirements. The packet transmissions of different users need to be treated according to their QoS attributes, such as the required bit rate and delay constraints. Two commonly adopted mechanisms are packet transmission prioritization and scheduling.

− Robustness: This is with respects to changing system conditions and equipment failures. In wireless networks, the robustness of MAC schemes is most often considered against fading and power consumption of the MT battery. The first is due to the fact that a wireless channel is time-varying and error-prone. The impact of fading can occasionally make the RL unusable for a short period of time. MAC schemes should be able to recover from such link failures. The latter is due to the fact that the battery power of a wireless device is rather limited. The power conservation and saving feature of a smart MAC design is of great importance to the success of multimedia services.

Furthermore, the simplicity aspect must be considered and incorporated in the design of MAC schemes and wireless systems in general. This condition ensures efficiency and enables rapid expansion of wireless systems and services.

2.2.1.2 Performance measures

The most essential performance measures of the MAC schemes are:

− Throughput: The throughput can be understood as the portion of the offered load that is successfully delivered to the intended destination. Hence, the maximum throughput is equivalent to the system capacity. Ideally, the throughput is the same as the offered load, which is the amount of traffic or packets being transmitted. This is true in the case that the channel is error-free and the offered traffic intensity is under

57

a certain limit. Providing that all packet transmissions have the same format, the throughput is often defined as the average number of packets successfully transmitted per unit time, which is set equal to TTI (Transmission Time Interval) of a packet (14-19, 37-38). The throughput can therefore be given in bit/s, packet/s, Erlang, or even by a normalized value in [0, 1] if it is calculated as the ratio of the throughput to the offered load.

− Delay: This is often defined as the average time experienced by a successful packet transmission from its initiation to its completion. The delay depends on characteristics of the serving system and traffic load. It is given in seconds, milliseconds or a value of the delay-to-TTI ratio.

To this end, a rule of thumb in the design of MAC schemes is to maximize the throughput and minimize the delay. However, the objective of a well-planned MAC scheme would be to be optimum for a certain QoS profile or group of QoS profiles. Thus, in addition to the throughput delay characteristics, quantitative metrics, such as the fairness factor based on the fair-queuing theory (108, 212-213), goodput as the amount of throughput excluding the protocol overhead (146-147), energy consumption (138-140) and sensitivity characteristics to different system parameters, can be defined and used for the performance evaluation of MAC schemes and systems.

2.2.2 Classification of MAC schemes

The MAC schemes can be classified according to networking paradigms and topologies of wireless networks as distributed (ad-hoc), centralized (cellular), and hybrid (satellite). Each can be further classified based on access methods such as random, scheduled, and hybrid access. The MAC schemes can also be classified according to the multiple access techniques underneath. Figure 3 shows a classification of the MAC schemes for wireless networks.

MAC schemes for ad-hoc networks and satellite communications systems are not in the scopes of this thesis. However, these segments are investigated in (20), for example. One can notice that, with an exception of the ALOHA, all distributed MAC schemes are based on the principles of the CSMA with/without an explicit mechanism for collision avoidance.

The ALOHA is a typical example of random access, which can be used in any type of networks: distributed, centralized, or satellite. In the ALOHA system, users are allowed to transmit whenever they have data to send. If a collision occurs, which is detected by acknowledgement from the network side, the user waits a random period of time and then retransmits the unsuccessful packet. Thus, such schemes are also referred to as repeated random access.

The CSMA refers to MAC schemes in which users listen to the common radio channel to detect any ongoing transmissions before a random access. If the channel is busy, the user waits a random period of time and tries to transmit again. If the channel is idle, the user makes an attempt to acquire the channel. Then the transmission of a packet follows. The reception of each packet is acknowledged. If the acquisition attempt results in a collision, the colliding users try to resolve the collision in an orderly manner. Let us recall

58

the hidden- and exposed-terminal problems that may cause degradation to the CSMA performance. To resolve such a problem, a mechanism for collision avoidance and resolution is adopted. This can be based upon either out-of-band feedback control from the intended receiver or a handshaking approach with an exchange of short request-response messages to test the connection beforehand.

MAC schemes for centralized wireless networks are of our interest. The use of a central BS allows more sophisticated network-controlled functions and optimized MAC schemes. It is reasonable to assume that all the terminals in the cell coverage can talk to and hear from the BS. Hence, this significantly removes the hidden- or exposed-terminal problem.

Fig. 3. Classification of wireless MAC schemes. The random access class of centralized MAC schemes tries to obtain a high efficiency while emphasizing simplicity. It allows all users to contend for the random access channel by transmitting as soon as they have data to send. This class is suitable for bursty data traffic, but not desirable for guaranteed, delay-sensitive traffic. The applications of this class can be found in various PRNs. The ALOHA is still considered as a fundamental scheme of this class. However, in a pure ALOHA channel, providing that all packets have the same format, the maximum throughput is about 18 percent (14, 16). If the channel is slotted into equal time intervals of the packet size or TTI and the transmission attempt is made at the beginning of a slot, the throughput is about doubled. This slotted version of ALOHA is called S-ALOHA (14, 16). The performance of ALOHA schemes is quite poor in general. To improve it, many MAC schemes have been developed ever since.

Wireless MAC schemes

Distributed MAC schemes

Centralized MAC schemes

TDD

Random Access

TDD/FDD

ALOHA-based

CSMA-based

Hybrid AccessRandom Access(contention)

Scheduled Access(non-contention)

ALOHA-based

ISMA-based

RAMA-based

Fixed Assignment

Demand Assignment

Polling

TDMA RR

TDMA DDA

PRMA-based

PRMA-based + DDA

CDMA

Random (ALOHA, ISMA, CLSP, etc.)

Scheduled (Polling, etc.)

Hybrid (RR, DDA)

59

The scheduled access is considered as non-contention, and is applied extensively in FDMA/TDMA cellular networks. This class is based on either fixed- or demand-assignment scheduling. The fixed-assignment scheduling provides each user with a pre-determined, fixed allocation of resources, that is, a fixed channel for the entirety of a session, regardless of the need of the user to transmit during that session. This type of MAC schemes is appropriate for continuous and real-time traffic, but can be wasteful in the case of bursty traffic. The demand-assignment scheduling assigns resources according to requests submitted by users and by polling users in a round-robin fashion. Once the requests are received and processed, users are assigned resources according to the results. In general, a fair queuing scheduling mechanism is embedded in the MAC operation. This type of MAC schemes is useful for VBR or ABR (Available Bit Rate) traffic. However, additional signaling overhead and delay caused by polling or reservation processes can degrade the performance. The scheduled access has to rely on flow control, congestion control, CAC and so forth to ensure stability for its operation.

The hybrid access class combines or inherits the basic features of both the previous two classes. Most hybrid schemes employ request-grant mechanisms for access control. The request packet is usually sent using a random access channel. Each active user sends a request to the BS indicating, for example, how much time and bandwidth is required to send the amount of data that the user currently has in their buffer. There can be other alternatives as well, depending on the system intelligence. The BS, upon the reception of the request, allocates an affordable combination of resources for the user transmission and sends a response to the user indicating the granted channel format. This type of hybrid schemes can be divided into random reservation and dynamic demand-assignment schemes. In a random reservation scheme, the BS has implicit rules for reserving a channel. For example, one rule is that a granted request will result in a periodic reservation of a particular slot for the user transmission. In a dynamic demand-assignment scheme, the BS controls the user transmissions according to their QoS requirements. It collects all the requests from the users and uses scheduling algorithms to make an efficient resource allocation. The resource allocation, as a result, is often dynamic and adaptive. The hybrid access allows more flexible and efficient MAC schemes which can cope with the diverse conditions of multimedia traffic.

Chapter 1 Section 1.2.1 suggests that all MAC schemes for interference-limited DS-CDMA systems can be considered as hybrid because they carry features of both the random access and the scheduled access. That is, regardless of the access methods, contention can happen depending on the MAI level.

2.2.3 Review of centralized MAC schemes

The wireless MAC has received a lot of attention and research works ever since the pioneer ALOHA was proposed in 1970. The research directions can be categorized into three main groups. The first is on the synthesis of new MAC schemes. The second is on the fundamental analysis and performance evaluation of MAC schemes. The third is on the selection and optimization of MAC schemes considering different system environments and technologies. There is a substantial amount of literature on the subject.

60

The results also include a number of MAC schemes ranging from simple ALOHA to complex schemes for supporting diverse multimedia traffic and QoS requirements. MAC protocols have been surveyed in (82) for wireless ATM (Asynchronous Transfer Mode), in (86) for wireless networks considering semantic formats and procedures, and in (85) for multimedia cellular systems. In this section, some centralized MAC schemes for advanced wireless networks are reviewed. Our interest is to keep it somewhat more conceptual and analytical in regard to RM aspects.

2.2.3.1 Random access schemes

In addition to ALOHA schemes, which include pure ALOHA, S-ALOHA, and S-ALOHA with reservation denoted as R-ALOHA (16-17, 80-81, 86), the following are often found in centralized PRNs: ISMA (Idle/Inhibit Sense Multiple Access) and RAMA (Resource Auction Multiple Access) schemes.

ISMA (17, 95) is considered as the centralized version of the distributed CSMA, in which the BS, instead of the user terminals, performs carrier sensing and collision detection. The basic operation of ISMA is as follows. The BS periodically broadcasts an IS (Idle/Inhibit Signal) when the channel is free/busy. FDD is assumed allowing the possibility of almost immediate feedback from the BS. However, in the case of TDD, only an idle signal is issued. The terminals that have packets to send will listen to the broadcast channel and decide whether to transmit or defer according to the received IS. The ISMA can be non-persistent or persistent. If a packet is received correctly, the BS sends an ISA (IS with Acknowledgment). This acknowledges the successful packet transmission and, at the same time, indicates that the channel is free again. If two or more terminals transmit at the same time, a collision occurs. The BS resolves the collision and becomes idle again after it. The terminals that are involved in a collision detect the collision after they receive an IS instead of an ISA. The terminals then back off for a random period of time before the next retransmission attempt.

If the size of the data packets and their TTI is large enough compared to a feedback-control latency, the loss of complete packets due to a collision can result in an inefficient performance. To improve that, a reservation ISMA scheme with polling (95), denoted as R-ISMA, can be used. In R-ISMA, the terminal sends a short RP (Reservation-request Packet) in response to the IS signal in the ISMA fashion. Thus, only small RPs may be involved and lost in a collision. If the BS receives an RP, it issues a polling signal to the corresponding terminal to inhibit others from transmitting at the same time. Upon receiving the polling signal, the terminal can transmit its packet immediately.

RAMA (96-97) uses a deterministic mechanism for random access. Two phases exist in RAMA: the auction phase and the data phase. The terminals that have data to send contend for the data phase, that is, the channel in the auction phase. In RAMA, each terminal is assigned a unique ID (Identifier) in a binary string with a fixed length. In the contention phase, each terminal transmits its ID to the BS symbol-by-symbol. The BS broadcasts what it heard after each symbol to all the contenders. If it does not match the symbol that the terminal transmitted, it drops out of contention right away and tries again in the next auction phase. The collision resolution at the BS is as follows. In case more

61

than one terminal transmit the same symbol simultaneously, there is no collision and the BS broadcasts the symbol it heard. Otherwise, the BS broadcasts a symbol with higher priority, “1”. Thus, at the end of an auction phase, there is always one winner who has the highest ID. This results in an unfair share of the channel.

The channel in RAMA is never wasted. However, the protocol overhead can be quite high. This overhead increases as the data rate increases because of the per-symbol switching involved in collision resolution. RAMA is usually more efficient than R-ALOHA or eventually R-ISMA in only heavily loaded situations. However, if the capture is effective, R-ALOHA or R-ISMA can at least perform as well as RAMA. The fairness problem is addressed in (97). The BS can resolve collision in the auction phase by broadcasting the “1” or “0” symbol randomly, instead of “1”.

2.2.3.2 Scheduled access schemes

This class of MAC schemes avoids the situation in which multiple users try to access the same channel at the same time by scheduling the transmissions of all users in the system. This class is therefore non-contention. The scheduling takes two basic forms: fixed assignment and demand assignment (17). Providing that the channel is error-free, packet transmission is guaranteed as being successful.

In FAMA (Fixed Assignment Multiple Access), each user, if admitted into the system, is assigned a fixed, non-overlapping portion of the available channel capacity, that is, a fixed channel. The channel division is done in FDMA/TDMA fashion. Thus, the channel is a frequency channel or a certain slot in subsequent time frames of the time axis on a frequency channel. The user can transmit its packets on the assigned channel whenever needed, independently from the other users. There is a hard capacity limit on the number of users that can be admitted into the system. The efficiency of FAMA depends on the user activity and traffic. FAMA is suitable for CBR guaranteed real-time services such as voice calls, but not efficient for bursty data services.

In DAMA (Demand Assignment Multiple Access), only the users that have data to send are allocated channel resources. The transmissions of these users are scheduled in an orderly fashion so that contention does not occur. Thus, no capacity is wasted for non-active users. However, the protocol overhead can be quite high. This is because of the signaling and processing needs in order to determine which users require channel access and to schedule the user transmissions, especially when the number of users in the system is large.

The roll-call polling scheme (17, 37-38) is probably the most conventional DAMA. The BS polls the users one after another in a predetermined sequence. Each user is polled once in a sequence. Upon the reception of a polling message, the user transmits either a control message stating it has nothing to send or stating all the packets that have accumulated in its buffer since the last poll. In the latter case, after the last packet is transmitted, the user sends a “ready” message to the BS. The BS then continues polling the next user in the sequence.

The overhead in the roll-call polling grows with the number of users that have nothing to send. The roll-call polling may also suffer from redundancy if a certain user has too

62

much data to send within a protocol operation sequence. More advanced polling schemes have been developed to partially reduce the overhead or increase the flexibility and robustness. For example, (225-227) propose a more clear coherence time frame for the protocol operation with phases for polling the requests, sending the requests, and scheduling the user transmissions. DAMA is most suited for VBR or ABR data users. However, admission control and congestion control are required for a stable operation.

2.2.3.3 Hybrid access schemes

This class allows more flexible and efficient MAC schemes. Most of the recent research works on MAC schemes for advanced wireless networks have addressed and targeted this class. Hybrid MAC schemes can be further divided into three groups: TDMA RR (Random Reservation), TDMA DDA (Dynamic Demand Assignment), and CDMA based as described in Section 2.2.2. This section reviews the essential schemes of the first two groups. The review of DS-CDMA based schemes follows in a separate section, as they are directly related to the focus areas of this thesis.

TDMA RR schemes have been studied in the context of supporting voice and data services in TDMA mobile cellular networks. The UL of such TDMA RR systems is considered as a slotted multiple-access channel, which is organized into frames of an equal length. The frame length is chosen such that a single voice packet is generated per frame. The slot is large enough to carry one voice packet. The length of the data packets is one slot. The TDMA RR operation in DL relies on a broadcast channel. The RR schemes contain two stages: random access and reservation. In the first stage, the terminals that have data to send use a random access channel to make their first transmission. S-ALOHA is usually the first choice because of its simplicity. The reservation stage describes the policy enforced by the network to reserve UL slots for the terminals that have successfully contended for the random access channel. The policy ranges from a simple reservation of a fixed subsequent slot for upstream transmissions of voice packets to a complete scheduling of UL transmissions by the network. The RR schemes with scheduling can be listed under the DDA group as well. TDMA RR schemes are based on the principles of PRMA (Packet Reservation Multiple Access).

PRMA (98) is initially designed for the packet transmissions of voice users in TDMA FDD cellular networks. The main issue is to put an upper bound of 40 ms on the delay of voice packets to maintain intelligible speech, while allowing occasional packet losses. Each active voice user during a talkspurt generates voice packets which are stored in a FIFO (First In First Out) buffer. The user tries to transmit the first packet from the buffer using the S-ALOHA. If this packet is not transmitted successfully within 40 ms, the packet is discarded and the user tries again with the next packet in line in the same fashion. This continues until either all packets are dropped or a successful transmission happens. Upon the first successful transmission, the user gets the reservation of the corresponding slot in future frames for transmitting all the packets in its buffer until the end of the talkspurt.

Many extensions of PRMA have been developed for efficient support of integrated voice/data and multimedia services. In (100), the system allows the data users to contend

63

for a transmission of a data packet using the same S-ALOHA, but repeatedly on packet-by-packet basis. That is, no reservation is made for a data user. Furthermore, to give preference to voice traffic, different persistence probabilities for random access can be used regarding the voice or data user. In (100), it also allows the data users to reserve a limited number of slots to improve efficiency and decrease contention. However, the introduction of data users decreases the performance of PRMA system with respects to voice users. To enhance this aspect, FRMA (Frame Reservation Multiple Access) is introduced to separate the subsystems serving voice and data users (101). In FRMA, a frame contains voice slots and data slots. The distribution of the voice and data slots is dynamically adjusted on a frame-to-frame basis so that the dropping rate of voice packets is kept less than 1%, for instance. PRMA enables more users to be supported than the number of time slots by using voice activity detection and, during the silent period of the voice sources, the free channel capacity is used to serve intermittent data traffic. Thus, PRMA is more flexible and efficient than FAMA.

In more advanced RR schemes (102-107), which expand the concepts of PRMA to accommodate multimedia traffic, the user sends a request to reserve a channel bandwidth for its service before sending the actual data. These are therefore considered as DDA schemes. In DDA schemes, the system tries to allocate bandwidth to the users according to their QoS requirements. In advanced cellular networks, end-to-end QoS is an important design goal and, at the same time, utilization of the network capacity must be maximized. This goal can be achieved by having the central network controller collect QoS requests and schedule transmissions for each user so as to match the requirements against the available system bandwidth. There are three main stages in a DDA scheme: request, scheduling, and data transmission. The request channel, which is typically a random access channel, is used for the users to send QoS requests to the BS. The user can send additional requests or indications during the service session using out-band signaling or piggybacked with data transmissions. This reduces contention on the request channel. Based on the user requests, the network schedules UL transmissions using a scheduling algorithm. The users then transmit their data, contention-free, on the scheduled channel.

In PRMA++, the request and data channels are separated (102). In a UL frame, a few random access slots are assigned for sending requests and the rest are information slots. The request slot can be further divided into mini-slots to increase efficiency, as the size of the request packets is often much smaller than the data packets. If a collision occurs, a collision resolution cycle is initiated and continues until all of the collided mini-packets are successfully transmitted. C-PRMA (Centralized PRMA) is initially designed for micro-cellular environments (104). The goal is to grant the next transmission opportunity at each slot to the user that has the most urgent need or the closest deadline. Thus, C-PRMA attempts a prompt retransmission of corrupted packets. The BS acknowledges a successful reservation. Then, when it issues a polling command, the corresponding user can transmit data packets on the assigned slots. C-PRMA accommodates multimedia traffic via the polling process. The polling sequence for the reserved users is generated by a scheduling algorithm, which is designed to provide QoS to all users in an efficient manner taking into account priority, loss rate, bit rate, and delay constraints. In C-PRMA, slots are dynamically assigned and the scheduling algorithm provides a more accurate statistical combination of the traffic.

64

DQRUMA (Distributed Queuing Request Update Multiple Access) is another flexible DDA scheme for wireless multimedia support (109-111). It was initially designed for FDD TDMA systems with a small and fixed frame size, such as wireless ATM. The UL frame fits a mini-slot of a reservation request and a data slot of a single packet, which includes a piggybacking field for additional requests during the data-transmission stage. In the DL frame, the request mini-slot is for acknowledgement and the piggybacking field is for indicating transmit permission. The data slot in the UL and DL frames, whenever idle, is converted into a number of mini-slots for sending reservation requests and acknowledgements. The data users are treated as distributed queues of packets having different QoS attributes. In DQRMA, when a user has data to transmit, it first attempts to transmit a request on the mini-slot in an UL frame using S-ALOHA. It then listens to the acknowledgement channel in the DL frame immediately for the result. This is repeated until the request is successfully acknowledged. Then, the user listens to the transmit permission field for its ID and more instructions and when the user hears its ID in a DL frame, it transmits its packet in the next UL frame. If the user has more packets to transmit, it sends an additional request in the piggybacking field to reserve more data slots. The BS collects all the requests and adds them to a list of outstanding requests. Thus, based on the list of outstanding requests, the system can serve according to the requirements of the individual users.

2.2.3.4 DS-CDMA based schemes

CDMA techniques, DS-CDMA in particular, have been selected as the technology platform for numerous wireless communications systems including third –generation –and beyond mobile cellular systems (21-22). The features of CDMA, such as its high yet interference-limited capacity and its natural flexibility in supporting multi-rate transmissions, present new possibilities as well as challenges to the design of efficient MAC schemes. CDMA based MAC schemes not only have to share code resources, but also control the MAI level to ensure efficiency and stability of the systems. The requirement embedded in the latter phase is inherently problematic because of the stochastic nature of MAI, which fluctuates within a large dynamic range. This is not to mention the various environmental issues and the user diversity, which must be considered to achieve an efficient network performance.

In a centralized DS-CDMA PRN, regardless of whether or not random access is applied, each user still can be assigned a unique code sequence to spread and transmit its packets in the UL. The contention occurs whenever MAI becomes large enough to make the received SIR inadequate to extract the desired signal without any error. That is, provided it has an appropriate PC and as long as the number of simultaneous user transmissions is kept under a certain limit, the system behaves as non-contention. The performance of ALOHA/DS-CDMA schemes has been investigated extensively for PRNs with different physical layer designs and channels (13-19, 112-120). To prevent the contention and improve the throughput of ALOHA/DS-CDMA in high-load situations, CLSP/DS-CDMA is developed (17, 113-114, 116-120). CLSP is in fact an extension of ISMA used for asynchronous or quasi-asynchronous DS-CDMA systems, in which

65

packets have the length of multiple sensing and accessing-attempt slots. The BS senses the channel load which is the MAI level in general or the number of ongoing transmissions in the case that the PC is reliable enough. It then uses feedback control as in ISMA to allow the users to transmit if the channel load is less than a certain threshold or force them to refrain from transmission otherwise. Thus, the throughput characteristic of CLSP/DS-CDMA is optimized with perfect control.

In (127), a simple MAC scheme is used in multi-rate DS-CDMA cellular systems. The user packets are queued up for different classes of traffic according to their bit rates and priorities forming distributed queues at the user terminals. Each packet is assumed to comprise a synchronization header, a small data unit like an ATM cell, and an error control parity trailer. It is also assumed that an ideal feedback channel exists for the transmission acknowledgment. Based on the status of the buffer and the success of the previous transmission, each user is modeled with three states: “idle”, “active”, and “blocked.” The user is idle if there is no packet in the buffer and active otherwise. The active user is in “blocked” state or remains in it if the previous transmission or retransmission failed. Each active user of a class δ attempts to transmit the head-of-line packet in the next slot with a probability of pδ from its “active” state or qδ from its “blocked” state independently. This scheme is therefore of the ALOHA type, capable of handling a mixture of traffic with different priorities and bit rates. The simplicity and flexibility is emphasized in (127). However, its efficiency and stability depend on how to set pδ and qδ for different traffic classes which in turn depend on the system capacity, offered traffic, and delay attributes of each queue. This aspect is not specified and discussed therein.

Feedback control MAC schemes for integrated voice/data DS-CDMA cellular networks are investigated in (22, 122-126, 129). The voice users have higher priority than the data users. The voice users will be either blocked during the call–set-up phase according to CAC or allowed to transmit until the end of the call. The data users, in the meantime, contend for the remaining resources following a feedback control MAC scheme. The system has a capacity index of Kv with respect to voice users only and Kd with respect to data users only. The condition: Nd(t)/Kd + Nv(t)/Kv <1 must hold for voice/data multiplexing in order to ensure adequate communications quality and that the data transmissions are successful. Nd(t) is the number of data users and Nv(t) is the number of voice users in the system at time slot t. The FCSI has a form of Ψ(Nd(t), Nv(t)), in which Nd(t) represents the number of the backlogged data users due to unsuccessful transmissions at slot t. The data users then independently decide to transmit in the next slot with a probability, which is predicted based on FCSI so as to avoid contention, and, at the same time, to maximize resource utilization and throughput. Thus, depending on how to define and make use of FCSI at the user terminal, there can be numerous variations of such feedback control MAC schemes. However, the optimum solutions for this matter are still open. In (125, 129), the authors present a simple scheme using a power function of π j to define the TPP for data users in the next slot (t +1), where π is a constant and j is a non-negative integer representing the value of the so-called persistence state T(t) at t for FSCI. T(t) is a function of the system load: T(t +1) = T(t) + K, K is a positive integer often set to 1 if the system load at t exceeds a certain threshold or T(t +1) = T(t) −1 otherwise.

66

In more advanced systems like UMTS, on top of a rather complex, flexible physical layer, which is capable of simultaneously transmitting data for different multimedia applications of MTs, the MAC layer also becomes more complex and diverse. Reference (21), for example, provides a good coverage of UMTS. The MAC layer consists of different segments to handle different types of traffic, for which different physical channels, access methods, TFs (Transport Formats), and QoS classes are applied. To keep it simple and general, TF can be understood as the format used for transferring MAC packets over the physical radio channel, which include the bit rate in kb/s, the packet length in bits, TTI in ms, etc. TF has been mentioned in Chapter 1 Section 1.1.1 when discussing the mode of user transmissions. The transmissions of a VBR user can apply a set of TFs. For an efficient handling of user transmissions in such complex systems, MAC schemes are mostly based on dynamic demand assignment scheduling. This includes adaptive control of TFs depending on the system load, that is, MAI, transmit power constraints and QoS requirements. In random access channels, ALOHA schemes are often adopted. The operation of MAC schemes relies on a complex RM of wireless networks. In fact, MAC schemes are a part of RM. The stability and efficiency of MAC schemes depend on functionality of such CAC, congestion control, PC and packet transmission scheduling. The resource allocation behind the MAC schemes in advanced CDMA systems is not hard wired, that is, it is left somewhat unspecified. This offers a great degree of flexibility for the design of efficient MAC schemes. The ultimate design challenge of MAC schemes for such complex systems is therefore related to simplicity/efficiency trade-offs.

2.2.4 Summary and discussion

MAC schemes for wireless networks have been extensively studied since the 1970s. The design and development of a particular MAC scheme depends on the network architecture, service type, traffic demand, constraints of the physical layer, and so forth. This must take wireless properties and performance issues into consideration that makes the optimal design a challenging problem. MAC schemes can be classified based on the distributed/centralized basis of their network architecture; then further classified into random, scheduled, and hybrid categories according to the access methods. The work above has surveyed MAC schemes for centralized PRNs. The random access schemes are simple and efficient for serving a large number of discrete data users while bearing vulnerable chances for successful user transmissions in highly loaded situations. The scheduled access schemes, FDMA/TDMA fixed assignment or polling-based, guarantee successful user transmissions but may waste channel bandwidth for low utility or high overhead. The hybrid schemes are most flexible for supporting diverse multimedia applications in complex cellular networks. TDMA random reservation schemes, i.e., basic PRMA schemes, are hard wired and they explicitly specify how to allocate slots. In this regard, dynamic demand assignment schemes have more room for scheduling alternatives and optimized resource allocation. In particular, CDMA based schemes offer high capacity and natural flexibility to accommodate VBR transmissions and various techniques, at all layers of the protocol stacks, to further enhance network performance.

67

MAC schemes are often designed and evaluated assuming perfect control systems and error-free channels. It is of research interest to study the performance of different MAC schemes in burst error environments, resulting from fading and mobility, and with system imperfections. The future perspective of wireless communications towards building the mobile Internet (1-6) has fostered the development of diverse PRNs to support a broad range of packet-based services and viable solutions for mobile Internet access. MAC schemes are at the core of PRNs. It is therefore motivating to develop more optimal and robust MAC schemes for advanced PRNs.

The residual burst error process of packet transmissions over a physical radio channel is most often characterized through a fading channel model, which has been investigated extensively in the literature. However, most works, dealing with other issues than the physical layer, assume an uncorrelated channel with a certain BER or PER (Packet Error Rate) (80-90, 95-105, 112-121). In more precise works, the most commonly adopted channel model is based on a FSMC (Finite State Markov Chain) (203-210), in which the channel states are defined as a finite sequence of possible SIR levels and the sampling period determines how fast the channel varies. The steady-state probabilities and state-transition probabilities are defined based on distributions and second-order statistics of the fading process (16, 198-202). One of the pioneers is the Gilbert-Elliot model, a two-state, “good” and “bad,” Markov channel (229). More sophisticated channel models are based on an HMM (Hidden Markov Model), for instance, in which each channel state is described by an embedded Markov chain (205). The error process therefore can be studied within a general framework of a renewal-reward process, of which a semi-Markov decision process is a popular example for modeling practice (224). The works in (203-207), investigating the channel model and block-error process with/without FEC coding, are inspiring. The authors also assess the accuracy of FSMC channel models and confirm the conditions under which the error process can be considered as uncorrelated, correlated, or highly correlated. In the latter case, the result found that effective FEC coding requires an impractical interleaving depth with excessive memory and a long delay. The characterization of the residual burst-error process has a number of applications in the optimized design of multi-layer wireless networks, in particular MAC schemes.

In (22, 45-46, 147, 228), it is suggested that the burst error conditions of a fading channel, if properly managed, can be exploited to disable a subset of the contending users. In (230), it is pointed out that delay-tolerant data traffic, which takes up a major segment in advanced wireless networks, can provide some degree of flexibility, in which, instead of transmitting with more power to fight bad channel conditions, it rather defers transmission to better times by using channel-state dependent scheduling. In (1, 22-28, 147), it generalizes that adaptation can be performed at all layers of the protocol stack to accommodate the dynamics of wireless channels. In regard to efficient RM in power-sensitive advanced wireless networks, especially DS-CDMA cellular systems, the above ideas form one of the most effective strategies to achieve the design goals, which target the highest spectral and energy efficiency while providing adequate QoS to mobile users. In particular, these goals include: to mitigate interference and increase spectral efficiency; to minimize the power consumption and prolong the battery life of MTs; and to maintain QoS over wireless environments by adapting to user movements and channel impairments, as mentioned in Chapter 1 Section 1.2.1.

68

In (147), the authors study a framework of an adaptive radio for energy-efficient multimedia wireless links. The degradations of wireless channels are categorized based on interference, flat fading, and frequency-selective fading. It is suggested to use adaptive error control, that is, FEC/ARQ, or an adaptive frame length to combat low levels of interference and maintain an acceptable BER; an adaptive processing gain against high levels of interference; an adaptive frame length against bursty degradations due to flat fading and interference; and finally an adaptive equalizer for frequency-selective fading. The results based on extensive simulation campaigns indicate that the adaptive radio can achieve a 30-80% reduction in mobile battery power consumption compared to the fixed counterpart in the same system environments. Similar ideas are also found in (22, 142-146, 148, 153).

3G radio enhancements have brought retransmission closer to the radio interface. This allows a tighter coupling of ARQ schemes with the physical layer, enabling variable levels of FEC coding, faster operation, and potentially smaller packet sizes. These schemes, denoted as HARQ (Hybrid ARQ), typically incorporate combining retransmissions. This means that retransmitted data is combined with earlier transmissions of the same packet (130-136).

There are also many recent works, for example (151-159), investigating opportunistic scheduling schemes, which exploit system dynamics with different criteria and in various contexts and environments of the integrated services systems. One direction focuses on providing QoS adapted to bursty characteristics of data sources with different scheduling disciplines assuming error-free or error-uncorrelated channels. These result in various schemes based upon round robin, earliest-deadline-first, and fair queuing. The other direction studies the dynamics of time-varying channel conditions and opportunities in time and user diversity to improve efficiency of channel utilization and mobile battery power. In (230), a channel-state-dependent-packet-scheduling scheme is implicitly proposed to avoid transmissions in bad channel states over a WLAN link. The simulation results indicate a 10% to 15% increase in efficiency of resource utilization, assuming perfect knowledge of the good-bad channel states. In (152), highest-SIR-first scheduling schemes select the user with the best SIR or channel condition to maximize the throughput. The fairness of highest-SIR-first is improved with proportional-fair scheduling schemes (151-159) taking into account the priority and delay requirements of different users.

The contributions of this thesis on the MAC area include a study of performance losses of FCSI-based MAC schemes in centralized PRNs in the presence of system imperfections, and an analysis and synthesis of several efficient adaptive MAC schemes for DS-CDMA PRNs. These are summarized in Chapter 3 based on Papers II-IV, XI-XVIII.

69

2.3 CAC schemes in mobile cellular networks

2.3.1 Properties and performance measures

The objective of the CAC schemes is to regulate the operation of a network so as to ensure uninterrupted service provision to existing calls and, at the same time, efficiently accommodate new calls. This is done via managing the available network resources and allocating them among the users according to particular policies. The design properties and performance measures of the CAC schemes for advanced mobile cellular networks are presented in the following.

2.3.1.1 Design properties

The properties associated with technical issues of advanced mobile cellular networks include:

− Multiple Access Capacity: In FDMA/TDMA systems, the capacity is bandwidth-limited. The calls are blocked when there are inadequate frequency channels or time slots available in the system. In CDMA systems, all the users share a universal frequency band and the capacity is interference-limited. The admission of a new user will add interference to the other users in the system and have an impact on the overall transmission quality. Distinguished CAC policies are needed to meet different properties of the applied multiple access techniques.

− User Characteristics: The essential user characteristics in mobile multimedia systems include: randomness of the user access in time and space; QoS requirements; and user mobility. The impacts of different user characteristics are of great concerns for efficient RM. The randomness of the user access influences the offered load and its distribution among the cells, which in turn affect the system performance. The QoS requirements in terms of bandwidth, maximum bit rate, BER, delay tolerance, etc., provide input to calculate the necessary resource allocation for the user and the impact of this allocation on the existing users. This is used for CAC to make a decision. The user mobility contributes to the fading impact and causes handoffs.

− Resource Allocation Strategies: Chapter 1 mentions that the RM schemes are mutual supportive and can be considered jointly. CAC schemes must be in line with and based upon the resource allocation strategies. In FDMA/TDMA systems, CAC schemes particularly reflect the underlying CA schemes including how to handle handoff requests. In CDMA systems, the CAC schemes contribute to MAI management and efficient QoS support, together with code assignment, PC, handoff control, scheduling, etc. In regard to efficient QoS support in multimedia cellular networks, different traffic classes have different characteristics and requirements. Thus, the resource allocation strategies also need to consider how to prioritize and adapt the user transmissions to the available network resources. The real-time users

70

are often given higher priority over non-real-time users; and so are handoff calls over new calls.

− Cell Size Structures: The cell sizes (macro/micro/pico) have a major impact on the system capacity and coverage, as well as the handoff frequency. A smaller cell size increases the channel reuse frequency in FDMA/TDMA systems, and reduces the transmit power in CDMA systems. This improves the overall capacity of cellular systems and, in particular, the coverage for high data-rate users in CDMA systems. However, with a smaller cell size, it is obvious that the number of handoffs will increase. This adds more QoS provisioning requirements and control overhead to the CAC schemes. Furthermore, more precise considerations are required on CAC to handle the impact of other-cell interference in smaller-cell-size CDMA systems.

− Unbalanced Traffic between UL and DL, and among Cells: Because of the traffic asymmetry in the duplex operation of some multimedia applications, the resource requirement can be greatly unbalanced between UL and DL in a particular cell. For example, web browsing, real-time multimedia downloading, etc., require a much larger capacity in the DL direction. CAC schemes need to adjust their decision process to meet this kind of unbalanced load. The traffic imbalance exists among the cells in a system as well due to the randomness and spatial distribution of the users. The resource allocation strategies, including the CAC schemes, need to consider the possibility and ability to adapt to changing traffic patterns and hot spots.

To this point, CAC, based on knowledge of both network and user characteristics, keeps track of the available system capacity and accommodates new call requests while ensuring QoS for all existing users. CAC directly controls the number of users in a cell and therefore plays an important role in determining the system performance. CAC schemes can be evaluated based on the following properties, which are associated with performance issues of wireless networks:

− Efficiency: CAC schemes are derived to achieve optimum trade-offs between GoS and QoS, with respect to efficient RM. The efficiency of CAC schemes is usually measured in terms of call blocking/dropping probabilities. Their linear combination functions define GoS. The Erlang capacity for a certain threshold of GoS is also used to measure the efficiency of the CAC schemes. In FDMA/TDMA systems, due to their hard-capacity feature, the probability of losing communication quality is often ignored once a user is admitted into the system and gets a channel. In interference-limited CDMA systems, the probability of the system being in outage states, in which users experience a loss of the communication quality, is of interest. This is used to assess the trade-offs between GoS and QoS, i.e., the efficiency of the CAC schemes. Furthermore, the delay of the decision process in CAC must be minimized along with the call set-up phase.

− Fairness: The fairness of admission decisions toward individual users upon their arrival depends on the user priority and profile characteristics (e.g., handoff or new user, QoS attributes of the request, importance of the users). In advanced cellular networks, there can be different subscriber classes; each has a different service level agreement and requires different service regulations. CAC schemes provide means of differentiating QoS for an efficient network operation.

71

− Stability: CAC schemes are directly targeted to prevent congestion and outage and ensure stability for cellular network operations. This aspect is intertwined with the efficiency. This is because the CAC schemes try to accommodate as many users into the system as possible to maximize the utilization of radio resources while the reliability of the provided QoS must be continuously maintained for each user.

− Flexibility: It is important to have flexible CAC schemes to cope with various traffic patterns in advanced cellular networks, including traffic imbalances. Traffic patterns are distributed in time (busy hours) and space (hot spots) with diverse QoS requirements (multimedia support). The ability to be reconfigured and extended to support new traffic regulations and services is also of great importance.

− Robustness: The CAC schemes need to be able to adapt to changing system conditions (system load, MAI fluctuations in CDMA systems) and maintain stable network operation. The robustness of the CAC schemes is also considered against the need of redesign due to changes of particular system parameters or RM schemes.

− Simplicity: The doctrine “keep it simple” must be followed by CAC while fulfilling all the aforementioned properties. Only this condition can ensure system efficiency and guarantee a fast response to multiple, increasingly complicated and diverse QoS requests of mobile users.

2.3.1.2 Performance measures

The most essential performance measure of the CAC schemes is the GoS.

− GoS: In trunk systems, GoS is defined as a measure of congestion (16). This is given by the probability of a call being blocked or the probability of a call being delayed beyond a certain amount of time as (2.7) or (2.9), for example. In traditional FDMA/TDMA cellular telephony networks, GoS is defined as, for example, GoS = Pb + 10Pd, where Pb is the probability of a new call being blocked and Pd is the probability of a handoff call being dropped or a handoff failure (19, 170-175). The call dropping probability is weighted by a factor of 10, as it is commonly accepted that dropping a call in progress has more negative impact on the customer perception on the provided service quality than rejecting a new call occasionally. In advanced mobile cellular networks supporting multi-class users and services, Papers I and VIII define GoS as a generic cost function of the probabilities of new call blocking, handoff failure, and loss of communication quality due to system outage. The weighting factor of each argument in the cost function is set depending on the profile characteristics of the corresponding user and service class.

In addition to GoS and its components, the Erlang capacity, the percentage of satisfied users, etc., can be introduced as performance measures of the CAC schemes as well. The satisfied user is defined in (78) as the one who is admitted into the system for service and provided with specified QoS for at least 95% of the time.

72

2.3.2 Classification of CAC schemes

CAC schemes can be classified in numerous ways depending on the basis of comparison. For example, from the operation point of view, CAC schemes can be classified as static, dynamic, and hybrid as mentioned in Chapter 1, Section 1.2.2. Figure 4 shows a classification of CAC schemes for mobile cellular networks. This basically combines the classification aspects reported in (19, 50) for FDMA/TDMA systems, (19, 78) for CDMA systems, and (19, 191, 193) for modeling-based and measurement-based systems.

In static CAC systems, the admission decision is based on a priori knowledge of the system capacity and which resource combinations of user requests can be accepted in a local cell. This is often specified during the network planning. The design of static CAC schemes is therefore simple and decentralized (i.e., distributed on local-cell basis). This, however, has to take into account worst-case scenarios of system conditions and resource consumption in order to guarantee QoS. In FDMA/TDMA cellular systems, fixed CA based CAC policies (16, 19, 48-50, 170) are often adopted because of their simplicity and capacity maximization in line with a fixed channel capacity assigned to a cell. In CDMA cellular systems, such a fixed channel capacity is hard to determine efficiently because qualitative and quantitative features of CDMA systems appear to be heavily intertwined and mutually supportive.

Fig. 4. Classification of CAC schemes. In dynamic CAC systems, it is intended to make full use of the system capacity and to adapt to changing conditions of system load, user distribution, etc. The planning of frequency reuse can be at ease even in FDMA/TDMA cellular systems if the underlying CA strategies are also dynamic. In this case, all the available resources (channels) are placed in a pool and assigned to new calls as needed such that their CIR/SIR targets are meet. The reinforcement learning or measurement-based adaptive CAC is an effective

CAC schemes

HybridStatic Dynamic

Fixed CA-based FDMA/TDMA

Flexible

CDMA

Number-based

Interference-based

SIR-based

GC

QH

GC+QH

Fixed CA-based

Hybrid CA-based

Traffic Adaptive

Dynamic CA-based

Traffic Adaptive

Interference Adaptive

FlexibleModeling-based Measurement-based

Flexible

73

solution for the dynamic class. This, in most cases, relies on online measurements and excessive signaling to keep track of the changes of system conditions among cells. Therefore, the price of better traffic adaptability and performance is the increasing complexity of implementations in both hardware and software. In high-load situations, and when the offered traffic is evenly distributed among the cells, static CAC and resource allocation can be more efficient than the dynamic counterpart.

In advanced cellular networks, the CAC schemes are often based on hybrid resource allocation strategies combining beneficial aspects of both the static and dynamic classes. These form a hybrid or flexible CAC class. For example, the available radio resources can be allocated in a static way up to a certain level, and the rest is utilized in a dynamic fashion. CAC schemes in CDMA cellular networks can be listed under this class regardless of whether they are static or dynamic due to the interference-limited nature of the systems (17, 19).

From the implementation point of view, the CAC schemes can be further divided into, for example, modeling-based or measurement-based, centralized or decentralized, local or global classes. This division basically specifies how networks define, monitor, and get the system parameters that are involved in the decision process of CAC (e.g., system capacity, traffic load, cell interference, user characteristics, resource consumption). However, a practical implementation can also be hybrid or flexible. The decision of CAC can be implemented as hard (i.e., a call request is admitted/rejected with an absolute certainty) or soft (i.e., a call request is admitted/rejected with a specified probability).

In the following section, the CAC schemes are surveyed for FDMA/TDMA and CDMA cellular networks.

2.3.3 Review of CAC schemes

2.3.3.1 CAC schemes in FDMA/TDMA cellular networks

In FDMA/TDMA systems, the channel capacity is bandwidth-limited and bears a hard limit. Thus, a call request can be admitted when there is a spare channel. In (50), an extensive and comprehensive survey of CA schemes for FDMA/TDMA cellular networks is provided. CAC schemes must reflect the underlying CA schemes for efficient RM. The readers are therefore referred to (50). However, there have been some new works investigating CAC schemes for FDMA/TDMA cellular systems since (50) was published in 1996.

In (170), the simplicity and efficiency of fixed CA based CAC with GC policies for handling priority handoffs is reconfirmed. The authors also make one step further with a proposal of a FGC (Fractional GC) scheme, in which new calls are accepted with a certain probability (soft decision) depending on the current channel occupancy state. In a traditional GC scheme, for a cell with a channel capacity of C (i.e., utmost C calls can be served simultaneously), a channel occupancy threshold T is defined as to reserve C−T channels for handoff calls exclusively. Thus, a new call is admitted into the cell only when the number of ongoing calls is less than T; and a handoff call is dropped if there is

74

no free channel available in the cell upon its arrival. In a limited FGP scheme, a new call can be admitted into the cell with a probability of α in between 0 and 1 if the current channel occupancy exceeds T. This is designed to enhance the flexibility and efficiency of static CAC with GC while maintaining the simplicity.

In (171), a distributed CAC scheme is proposed. This takes into account the number of active calls present in both the current cell and its adjacent cells for making a decision on admitting a new call. The rationale for this is that a newly admitted call not only affects the ongoing calls in the cell where the new call is originated, but also affects the calls in the adjacent cells due to possible handoffs. The control is done at each cell in a distributed fashion. The BSs are required to exchange state information of the adjacent cells periodically.

In (172), a predictive CAC scheme based on an estimation of future resource requirements with an introduction of a shadow cluster concept is proposed. The coverage of a shadow cluster for an active mobile user consists of the cell where the mobile is currently located (i.e., the center of the shadow cluster) and all its adjacent cells along the travel path of the mobile. The BS shares the future probabilistic information of the active mobile users currently under its control with the neighboring cells. Then, using this information, the BSs that are involved in the shadow cluster of an active mobile user will predict future demands and reserve resources. The cluster area changes when the mobile is handed off to other cells, thus a tentative shadow cluster needs to be implemented for each new call as well as handoff call. There are three steps in the implementation of such a predictive CAC scheme: to determine the projected active mobile probabilities; to calculate the resource demand of the new calls; and to execute the CAC in a distributed fashion. The simulation results show that the shadow cluster scheme is able to reduce the percentage of dropped calls. The performance of this scheme depends on the prediction accuracy of the future mobile movements. This also requires high storage and computation, and excessive signaling overhead for exchanging real-time information about dynamic movements, traffic patterns, and bandwidth utilization of individual mobile users in the network.

In (177), a dynamic CAC scheme using stochastic control is proposed to provide a precision QoS guarantee against a minimum required target of the handoff dropping probability. This scheme performs a periodical control. The acceptance ratio ai, defined as the maximum fraction of new calls to be admitted into cell i in the following control period, is calculated at the beginning of each control period. This is based on solving a non-linear equation of the predicted average dropping probability set equal to the minimum target. The prediction of the average dropping probability takes into account the available capacity, traffic distribution, and effect of multiple handoffs in the nearest as well as in the second and third nearest neighboring cells. The proposed CAC scheme then admits new calls into cell i with a probability of ai over each control period. Thus, the control is dynamically adjusted to the system and traffic conditions, and avoids a sudden overload of the network. This leads to more effective and stable system operation. The results show that the scheme is insensitive to the network load and the dropping probability is maintained at a stable level over a wide range of the offered traffic. This, however, compromises the practicality of the predictive derivation and expression of the acceptance ratio, that is, the prohibitive computational complexity and signaling load to

75

keep track of the global network states an otherwise large delay and reduced effectiveness if the periodical control interval is large.

The above CAC schemes and, in general, most of the CAC schemes in FDMA/TDMA cellular networks, whether static or dynamic with hard-decision or soft-decision, are actually based on a fixed CA. In this regard, the objective of the CAC schemes is to utilize a fixed number of channels assigned to each cell efficiently, giving priority to handoff calls. The admission decision and call set-up can be made fast. In the survey of CA schemes reported in (50), a number of dynamic and hybrid CA schemes are also included. These are considered as the basis of dynamic and hybrid CAC schemes. The computational complexity and signaling overhead involved in the decision process of such CAC schemes can be prohibitively high, as online and complete knowledge of global system states is required.

2.3.3.2 CAC schemes in CDMA cellular networks

CAC in CDMA cellular networks has been specifically addressed as a means of managing interference (load control) to ensure efficient QoS support, as the capacity of CDMA systems is interference-limited (78, 160-161). CAC directly controls the number of users in a cell in order to keep MAI under a tolerable limit so that an adequate RL performance and required QoS for each user can be maintained. This clearly indicates that there is a trade-off between the system capacity and the overall communications quality, that is, GoS and QoS in effect. In general, due to the time-varying soft capacity of CDMA systems, the decision of CAC exhibits an error in both directions: to accept and to reject. This is with respect to the predictive condition that the system outage state is avoided and that the resource utilization is optimized. Thus, CAC schemes for DS-CDMA cellular networks are considered as hybrid.

In (178-180, 184, 186), number-based CAC schemes are considered for single-service or integrated voice/data CDMA systems. The priority is given to voice users and the packet transmissions of data users have to adapt to the remaining system capacity. Perfect PC is assumed. The capacity index of the voice/data users, i.e., the maximum number of voice/data users that can be accommodated in the system simultaneously, is specified, i.e., based on (2.29). Thus, the operation of number-based CAC schemes is quite similar to the fixed-assignment FDMA/TDMA systems. This provides a great simplicity for implementation and analysis. The efficiency and robustness of number-based CAC schemes, however, depends on the capacity provisioning of CDMA systems. This is mapped on the system characteristics (other/own cell interference ratio, noise figure, etc.) and QoS requirements (bit rate, SIR target, data source on/off factor, etc.). This, however, cannot fully adapt to the actual soft capacity, which varies with fluctuating interference levels from own cell and neighboring cells. Thus, an underestimate of the capacity index will result in a waste of precious network resources and, on the other hand, an overestimate of the capacity index will result in a degradation of communication quality and congestion.

In (181), effective-bandwidth-based CAC schemes are introduced for multi-class CDMA cellular networks. This details the number of mobiles of each class in each cell

76

that the system can admit so as to maintain an acceptable QoS using an effective-bandwidth concept, which is applied in the analysis of statistical multiplexing of ATM networks. The result is an admissible region bound by a finite number of hyper-planes corresponding to multi-class capacity indexes. Thus, this can be considered as a multi-class extension of the number-based CAC schemes for the single-class or dual-class systems presented above.

In (182), SIR-based CAC schemes are introduced for a single-service CDMA cellular system. The effects of the radio propagation, call-arrival process, and traffic distribution among cells are taken into account. The residual capacity, defined as the maximum number of additional calls that the BS can handle while keeping the system-wide outage probability below a certain threshold, is introduced. Perfect PC is assumed. The effects of handoffs and background noise are ignored. The residual capacity is calculated periodically according to the received SIR measurement at the BS. Then, two distributed SIR-based CAC algorithms are proposed. The first is based on the SIR measurement of the local cell, and the second takes the SIR measurements of the adjacent cells into consideration at the same time. In the local CAC scheme, upon the reception of a call request, the BS checks the residual capacity of the local cell and accepts the call if the current value of the residual capacity is non-zero then it reduces it by one; otherwise, it rejects the call. The global CAC scheme follows the same steps. However, the residual capacity is now set to the minimum of the residual capacity of the local cell and the relative residual capacity of all the adjacent cells in effect. This ensures that admitting the new call will not affect the required QoS of the adjacent cells. The simulation results show that, when the spatial distribution of the traffic load is uniform, the local scheme outperforms the global one and the opposite in the hot-spot case. Furthermore, the global scheme tends to balance out the blocking probabilities between the hot-spot cell and its adjacent cells.

In (187), a SIR-based CAC scheme is investigated for CDMA cellular systems supporting multimedia services. This takes into account the load imbalance between UL and DL in a multimedia environment by considering both the UL and DL criteria. In the UL direction, the BS measures the received Eb/N0 for each active call and calculates the average Eb/N0 for each traffic class periodically. The BS estimates the resulting Eb/N0 for class k when a call of class i is accepted. The UL criterion is that a call of class i can be accepted if only the estimate of the resulting Eb/N0 for all class k is above its threshold. The same steps are carried out in the DL direction assuming that the MT of each active user measures the received Eb/N0 and reports to the BS periodically. Thus, a final decision of accepting a new call is made upon the satisfaction of both the UL and DL criteria. The proposed scheme handles a handoff request in a similar fashion. The priority is given to handoff calls over new calls by setting Eb/N0 thresholds for handoff admission lower than that for new calls. The results demonstrate that the proposed scheme can operate well in practical CDMA mobile multimedia systems, such as IMT-2000. The performance properties, such as simplicity/efficiency trade-offs, are not discussed.

In (179, 188), interference-based CAC schemes are introduced for single-service CDMA cellular systems. In (179), the interference-based CAC scheme is based on monitoring MAI generated in a cell instead of the number of ongoing users and is considered as a counterpart of the number-based CAC. The MAI threshold, corresponding to the channel capacity, is set on cell basis. In (188), the interference-based

77

CAC scheme is investigated via a dynamic CA scheme, which is also based on MAI measurements. The idea is that a new channel is assigned if the predictive interference after assigning the channel is less than a certain threshold. To give priority to handoff calls and enhance GoS, appropriate MAI thresholds are proposed to reserve resources (similar to the GC policy). In (78, 185, 190), the previous interference-based CAC schemes are extended to accommodate multi-class users in CDMA mobile multimedia systems. Thus, before accepting a new call, the current level of MAI is measured. If it is larger than a certain threshold, MAIMAX, then the call will be blocked; otherwise, the call will be accepted. The threshold MAIMAX is determined by the outage threshold set for each user class. Two CAC schemes are presented: local and global. The local scheme bases its decision on MAI measurements of the local cell without considering the impact of a new call on the adjacent cells. This can cause quality degradation to the adjacent cells, especially in the case of non-uniform traffic distribution where the traffic load of the local cell is light and that of the adjacent cells is high. The global scheme overcomes such problems by considering the impact of a new call on the residual capacity of the adjacent cells. In this case, the CAC scheme is capable of adapting to the traffic conditions and distributing radio resources among the cells in an efficient manner. This, however, comes with more complexity and overhead due to a larger amount of measurement, processing, and signaling required for keeping track of the global system states.

2.3.4 Summary and discussion

CAC limits and regulates the number of users admitted into the network under the constraints of network capacity and resource allocation strategies. Thus, CAC is an effective tool of RM in cellular networks for providing QoS to mobile users while efficiently sharing scarce radio resources through statistical multiplexing. CAC in FDMA/TDMA systems has been considered in the mutual context of CA, whereas CAC in CDMA systems has been viewed as an interference management problem. The design of CAC schemes, however, must take the properties that are associated with technical and performance concerns of wireless systems into account. CAC schemes are classified as static, dynamic, and hybrid according to the operational characteristics of the underlying decision-making process. From the implementation point of view, these can be further divided into modeling-based or measurement-based, centralized or distributed, local or global, hard-decision or soft-decision, etc. The static schemes based upon hard constraints of local-cell channel capacity are simple and efficient for serving with spatially uniform or highly loaded traffic patterns, providing that capacity and QoS provisioning is sufficient. This condition, in turn, presents a fundamental challenge for the design of mobile multimedia cellular networks, especially CDMA based systems. If the channel capacity is specified for the worst-case behavior of traffic, a potential statistical multiplexing gain can be lost; otherwise, the required QoS cannot be guaranteed. The dynamic schemes adapt to traffic and system conditions, and offer a more optimal performance which results in an increasing complexity and overhead in both hardware and software implementation. This is because most of the dynamic

78

schemes require additional functions performed by the network, such as measurements, predictions and signaling procedures for monitoring the current load states of a global system, i.e., multiple cells in effect. In some schemes, status information and motion detection of every active user are also required. Thus, complex computation, high processing, and a large memory to store the call status and load status are required. The error and delay of the measurements and status information can dramatically degrade the performance of the dynamic schemes. For these reasons, overdone dynamic schemes appear to be impractical. The hybrid schemes offer more flexibility to achieve the optimal performance trade-offs between different desired properties, such as simplicity and efficiency.

The design and evaluation of the CAC schemes often lead to the problem of overall RM, teletraffic modeling and performance evaluation of the entire mobile cellular systems. This is complex and mathematically intractable in real-world situations. Therefore, effective methods for developing and evaluating CAC schemes are of research interest as much as efficient CAC schemes. The related works (16, 19, 21, 22, 68-79, 169-193) have adopted various methods, ranging from using a simple birth-death queuing system model (e.g., Erlang-B formula) to an extensive, sophisticated network simulation campaign (e.g., network planning tool). These methods are categorized into analytical and simulation approaches.

The analytical approach can provide a preliminary idea and solid reasoning on the performance of CAC schemes in a quick and elegant manner. The validity of analytical results, however, is limited by the assumptions made for modeling the systems in question. The system models for studying CAC schemes in mobile cellular networks include basic components such as a cellular structure model, a propagation model, an interference model, a traffic model, a mobility model. The system parameters have to be specified a priori, including those resulting from the provisioning of the system capacity, QoS, and performance of the RM schemes. The analytical approach most often ends up using a simplified, statistical system model, which consists of several basic components in a stationary condition and ignores the rest.

In (16, 19, 69-71, 73, 77, 170, 175, 179, 186), static single-class CAC systems are described using a simple, one-dimensional Markov chain. Then, the probabilities of call blocking and dropping are derived based on the Erlang-B formula. In (178-180), a multidimensional Markov chain is used to extend the model of static single-class systems to multi-class systems. In (181, 190), a product-form traffic model for CDMA cellular networks with multiple classes of mobile users is developed. The effective-bandwidth concepts from the analysis of statistical multiplexing in integrated ATM networks are used to specify the number of admissible mobile users of each class in each cell. This allows applying the existing traffic modeling techniques and management strategies for general loss networks to CDMA cellular networks. Thus, with standard assumptions on the call arrival processes and the channel holding times, the stationary state distribution has a product form on the truncated state space defined by the CAC scheme. The impacts of mobility, propagation, interference, non-uniform traffic, etc., are not addressed explicitly in all the listed references.

In (189, 193), dynamic CAC schemes, based on dynamic programming or reinforcement learning techniques, are described using a semi-Markov decision process.

79

In (161, 191-192), investigating measurement-based CAC schemes, the probability of call blocking is calculated by using Gaussian approximation or Chernoff bound.

The simulation approach is the most common method used for the performance evaluation of the CAC schemes, especially dynamic and hybrid schemes. This approach allows incorporating many features of mobile cellular systems and environments into the evaluation framework. However, extensive modeling is still required on a system component basis. Hence, an appropriate combination of the analytical and simulation approaches can result in powerful tools for reliable and cost-effective evaluations of the CAC schemes and teletraffic performance.

In (189), the work on adaptive interference-based CAC for CDMA cellular networks also includes implementation considerations of CAC schemes in real environments. The estimation algorithm for the average received interference at the BS employs a linear Kalman filter supplied with a measurement of the interference from existing connections and a prediction of the interference generated by a new connection.

CAC is an integral component of RM in mobile cellular networks. Efficient CAC schemes can enhance the system capacity and service quality cost-effectively. In advanced multimedia cellular networks, the success of service deployment also depends on network capabilities in providing differentiated QoS to meet a wide range of customer demands. CAC schemes are vital for supporting the two-fold objective of cellular networks operation: (i) to deliver the required QoS; (ii) to provide operators freedom in controlling the payload and group behavior of traffic classes regarding the required services as well as subscriber classes for optimizing network utility and revenues. The latter is somewhat neglected in the recent research.

The contributions of this thesis on the CAC area include simple and effective analytical methods for a performance evaluation of the CAC schemes in FDMA/TDMA and CDMA cellular networks, an analysis and synthesis of an efficient soft-decision CAC scheme for CDMA cellular networks, and CAC-based GoS management and optimization for efficient operational support of multimedia CDMA cellular networks. These are summarized in Chapter 3 based on Papers I, V-X.

3 Summary of the main contributions

This chapter summarizes the main contributions of the thesis. It provides the insights of the main problems, claims, approaches, results, and conclusions making up each contribution, and it also gives guidelines for reading the original publications, Papers I-XVIII. Chapter 3 is divided into two sections: SPA for the contributions on the MAC area and SCA for the contributions on the CAC area.

3.1 SPA − Smart packet access for centralized CDMA PRNs

3.1.1 Implementation losses in adaptive feedback-control MAC schemes

3.1.1.1 Main problem

In the MAC survey presented in Chapter 2, a number of MAC schemes for the UL of DS-CDMA PRN use feedback control to adapt to the channel load conditions so as to avoid contention and optimize the throughput. In such systems, the BS senses the channel load, i.e., MAI generated by the number of ongoing transmissions, and broadcasts it as FCSI periodically in a DL control channel. The user that has data to send uses the latest received FCSI to adjust its TPP, also referred to as persistence probability. In hard-decision MAC schemes, if the current FCSI is larger than a certain threshold, the user will refrain from transmission; otherwise, it transmits the packet immediately with probability equal to 1. In soft-decision schemes, the user independently attempts to transmit its packet with probability π and refrain from transmission with probability (1−π), where π is a probabilistic function of the current FCSI and also referred to as a decision function. The FCSI can be defined as the number of ongoing transmissions in case the system supports a single class of users. The decision functions in FCSI based MAC schemes are illustrated in Figure 5.

81

The performance of such schemes depends on the availability and correctness of FCSI relative to the frequency at which the channel state varies and the capability to utilize such FCSI in the adaptive operation of MAC by the decision function. Thus, for practical implementations of the FCSI based MAC schemes, some additional issues should be addressed. These include the effects of feedback delay, access delay and imperfect sensing of FCSI, which are usually set aside or ignored in most of the related papers (113-114, 116-121, 125-129). The access delay is defined as a sum of the processing time of the decision-making process in a user terminal and the UL propagation delay, excluding the random back-off delay. The feedback delay is a sum of the processing time of the channel-sensing process in the BS and the DL propagation delay. Furthermore, it is of interest to investigate the optimal decision functions to enhance the efficiency and robustness of corresponding MAC schemes against the aforementioned effects.

Fig. 5. The motivation behind the FCSI based MAC protocols.

3.1.1.2 Claim

This contribution presents a performance analysis of FCSI based MAC schemes in the presence of feedback delay, access delay, and imperfect channel sensing simultaneously. The effects of such issues are referred to as system imperfections. The system sensitivity functions are introduced to characterize the relative performance losses due to system imperfections. These losses are also referred to as implementation losses. The trade-offs between average system throughput, packet delay, and outage probability in the presence of system imperfections are discussed based on a number of queuing system models and numerical results.

This contribution shows that using a smart decision function for the adaptive MAC allows the system not only to outperform the pure ALOHA and static counterparts, but also to tolerate high degrees of system imperfections. This method is also referred to as

1

0

Soft-decision: two constraints, lower- and upper bounds L and U, of channel threshold

TPP

FCSIK UL

Hard-decision: unique hard constraint K of channel threshold

ALOHA: no constraint of channel threshold

82

SPP. The SPP thus significantly enhances the robustness over system imperfections, for example, with up to 15% of the false-alarm rate in sensing FCSI and the access timing delay of 20% of the mean packet service time. The access timing delay is defined as a sum of the feedback delay and the access delay (119).

The basic work was first published in Paper IV. It was then refined and extended with new decision functions and results in Paper II, which is selected to reprint in the appendices.

3.1.1.3 Approach and result

The system model is basically as follows. The UL of a single-cell DS-CDMA PRN is considered, in which N data users of the same class attempt to transmit their packets independently. Each user is assigned a unique code sequence of NC chips. The packets have the same format with TTI of TP milliseconds. It is assumed that PC and transmission/reception in the DL direction are perfect. However, the effects of system uncertainty and errors due to time-varying radio channels are investigated in the context of imperfect channel sensing. The channel threshold effect of the interference-limited DS-CDMA system with perfect PC can be characterized by a capacity constraint K as the maximum number of tolerable simultaneous packet transmissions. This can be specified based on (2.14) or (2.29). The channel state is defined as the number of ongoing packet transmissions, denoted by n; and a packet is considered as successfully transmitted if during its TTI the channel state n never exceeds K. The system outage probability is defined as the probability that the channel state n exceeds K.

In FCSI based MAC systems, if a terminal has a packet to send, it uses the latest received FCSI, identical to the current channel state n in the ideal case, to calculate πn as TPP. It then transmits with probability πn and refrains from transmission with probability (1−πn).

For the hard-decision system, πn is defined by:

⎩⎨⎧ <

= otherwise ,0

if , 1 K nnπ . (3.1)

For the soft-decision system, i.e., the SPP, different πn functions are introduced and investigated as follows.

⎪⎩

⎪⎨

⎧

<≤

<≤

=+

−

otherwise , if ,

0 if , 1

UBn

LAnn U nL

Ln

π

ππ , (3.2)

where L ,U, A and B are positive parameters, L ≤ K ≤ U and 0<π <1. The alternative is to use a tangent hyperbolic function, tgh(n), as:

83

( )[ ]bantghn −−= 121π , (3.3)

where a and b are also positive parameters. Equation (3.3) can be further modified as:

( )[ ] 2,1,,121

=−−= kbntghπ kn βα , (3.4)

where ( ) 0,0 2,1,,, >>= βαβα kntghk is defined as

nn

nn

nn

nn

eeeentgh

eeeentgh βα

βα

βαβαβα

βαβαβα −

−

−

−

+−

=+−

= ),,(;),,( 21 .

The sets of the free parameters in the above functions, i.e., {L, U, A, B} {a, b}, and {α, β}, are introduced to be tuned for optimization purposes.

In order to quantify the performance losses in the presence of imperfect sensing, we introduce two parameters: D ≡ Pr{the BS correctly detects the channel state under the threshold K}; and F ≡ Pr{the BS falsely detects the channel state over the threshold K}. In general, these two parameters are mutually state-dependent. In our system model, these are treated as constants taking different values. In the presence of imperfect sensing, the channel state can be up to the number of users in the system N instead of K, even in the hard-decision system.

The traffic model is based on the following assumption. The system offered load of data packets is generated by the finite population of N users according to a state-dependent Poisson process with a rate λn. The packet arrival (including retransmissions) at each user terminal also follows the Poisson process with a rate λ. The state model of user terminals in the MAC operation is detailed in Papers II and IV.

The throughput improvement offered by using FCSI based MAC schemes depends on the feedback delay and access delay relative to the rate at which the channel state varies. If the channel state is changing fast compared to the controlling delay, then the users can no longer adapt to the channel variations and hence the access control may not be effective. To make it work, proper choices and trade-offs of the model and design parameters must be made. In our system model, to trace the changes of the channel states during the service time of a packet, an Erlang type r distribution function with mean 1/μ ≡ TP is chosen to represent the service process. This service process can be seen as consisting of r identical stages; each follows an exponential distribution with mean 1/rμ (213). It is assumed that the channel state may vary from stage to stage, but remains unchanged within a stage during the packet transmission; and that the BS senses the channel state and broadcasts its FSCI once per stage, i.e., every TP/r interval. For the validity of this assumption, r should take such a value that the probability of having more than one event (arrival/departure) in TP/r interval is negligible and, at the same time, TP/r must be feasible for the updating process of FCSI. For instance, if TP = 20 ms and r = 30, then TP/r = 2/3 ms that is identical to the basic time slot (also used as a contention slot of the random access channel) defined in the 3GPP standards for UMTS (21). The earlier

84

works (125, 129) use the assumptions that utmost one event can occur in a time slot and that feedback delay and access delay are zero to justify the validity of the system model.

To simplify the analysis, the access timing delay, defined as a sum of the access delay and the feedback delay, is modeled with an exponentially distributed RV having a mean of 1/γ and being independent of the service time. If the access timing delay is smaller than the feedback-control period of TP/r ≡ 1/rμ, which is considered to be very small compared to 1/λ and 1/μ, the effects of the aforementioned delays can be ignored. In case the access timing delay is not smaller than TP/r, the effects of the delays may cause an undesired expansion of the channel states. Thus, to incorporate the delay effects in the analytical method, the service time distribution as well as TPP for a transmission attempt, at a given channel state, need to be modified. In (118), a hypothetical packet concept has been developed to take care of the access timing delay, also resulting in an expansion of the channel states. This state expansion is assumed as a Poisson RV. For this reason, in some cases an average system throughput that even exceeds the channel threshold has been obtained. In our model, based on the validity condition presented above, the number of channel states can increase by a maximum factor of ⎣ ⎦γμ /r . TPP for a transmission attempt is now to be modified as:

} 1/in arrivals maximumgiven 1/in arrivals Pr{0

∑=

−=m

iinn mi γγππ ,

where m = min(n, ⎣ ⎦γμ /r ). That is:

∑ ∑=

−

=

−−− ⎟⎟

⎠

⎞⎜⎜⎝

⎛=

m

i

m

j

Gj

aGi

ainn

aa ej

Gei

G0

1

0

//

!)/(

!)/( γμγμ γμγμππ , (3.5)

where Ga is the average offered load of the system and πn is defined above. Meanwhile, each packet generated with rate λ at a terminal will experience a mean

delay of 1/γ before entering service, excluding a random back-off delay in the case of a retransmission. If the wasted time, defined as a minimum of the access timing delay and the time until the next attempt due to a random back-off, is treated as a part of the service time, the system can be seen with no delays. The average value of the wasted time is equal to )/(1 γλ + . The distribution of the modified service time is the convolution of the service time distribution and the wasted time distribution. The mean of the modified service time, as 1/γ is very small compared to 1/μ, is approximated as

γλμμ ++=

11'

1 . (3.6)

The performance analysis is step-by-step expanded in Paper II and IV to cover various scenarios of the ALOHA, hard-decision and soft-decision systems. In order to make it systematic and comparative, the systems with ideal conditions, referred to as perfect systems, are discussed first and then the effects of imperfections are gradually analyzed. The applied generic queuing system model resembles M/Er/K/N, that is, with exponential

85

inter-arrival, r-stage Erlang service, K servers, and N system-state space. The essential here is that the generic queuing system model eventually ends up having a blocking probability given by the Erlang formula (213), e.g., when N is set to K as in a perfect hard-decision system. This special case, referred to as ideal CLSP/DS-CDMA PRN, is used as a core system to investigate the adaptive possibilities in some of the later contributions. The detailed analysis can be found in Papers II and IV.

In the case of imperfect systems, the sensitivity functions representing the relative losses of throughput-delay performance characteristics as functions of the imperfection parameters, that is, the imperfect sensing probabilities and the normalized access timing delay, are introduced. The throughput sensitivity, Ss, is defined as

),max( 0

0

SSSS

Ss

−= , (3.7)

where S is the system throughput in the presence of the corresponding imperfections and S0 is that of the perfect counterpart.

The average-packet-delay sensitivity, Ds, is defined as

),max( 0

0

DDDD

DP

Ps

−= , (3.8)

where DP is the average packet delay in the presence of the corresponding imperfections and D0 is that of the perfect counterpart.

86

Fig. 6. Throughput delay trade-offs with/without system imperfections. For numerical examples, the following set of system parameters is used in Paper II, which is selected for reprint in the appendices: K = 7, N = 30, TP = 20 ms, r = 30. This set is chosen to be compatible and comparable with that used in (112, 115), while the validity condition is kept. The offered load per user is characterized by the product λTP that takes a normalized value in {0.1, 0.2, ..., 1}. The results show that the soft-decision system with a suitable decision function outperforms the hard-decision counterpart in terms of throughput delay trade-offs, especially when the offered load is high (i.e., λTP ≥ 0.5). The throughput improvement can be more than 100% as in the case where the decision function is given by (3.3) with a = 0.5 and b = 2, while the packet delay is kept moderate and stable over the entire range of the offered load. The hard-decision system has a better delay characteristic when the offered load is low. Both the soft-decision and hard-decision systems show more or less the same throughput sensitivity to the access timing delay. In highly loaded situations and if the access timing delay normalized to TP is larger than 20%, more than 90% performance losses can be expected. The hard-decision system is more sensitive to an increase of F, as it causes system outage, than a reduction of D, as it prolongs packet delay. The opposite is the case for the soft-decision system. However, the soft-decision system has much more stable characteristics over the effects of system imperfections as well as a high offered load. The flexibility in the design of soft-decision functions can also be taken for granted to optimize the system performance. Figure 6, which is reproduced resembling Figure 8 in Paper IV but with the parameter set of Paper II, captures some aspects discussed above for an illustration.

0 1 2 3 4 5 6 70

5

10

15A

vera

ge P

acke

t Del

ay in

Tp

System Throughput S

N=30, K=7, r=30, Tp=20ms pi(n)=0.5(1-tgh(0.5n-2)) MeanAccTDelay/MeanServiceTime=5%

ForHardDecision: CircleLine: D=1, F=0 D=0.95, F varies: SquareLine: F=0.05 DiamondLine: F=0.10 HexagramLine: F=0.15

ForSoftDecision: SolidLine: D=1, F=0 D=0.95, F varies: DottedLine: F=0.05 DashedLine: F=0.10 DashDotLine: F=0.15

+Line: PerfectHardDecision*Line: PerfectSoftDecisionxLine: U-ALOHA

87

3.1.1.4 Conclusion

The performance of FCSI-based MAC schemes for centralized DS-CDMA PRNs in the presence of various system imperfections is discussed and evaluated. The results support the claims stated above with a simple and effective analytical method.

There are some possible extensions of the analytical results for future works as follows. In the case of imperfect PC, a continuum of the received signal power level is possible and the overall MAI level should be used as FCSI for access control rather than the number of equivalent users. Thereby, the decision function can be defined in such a way that adapts to the UL interference distribution, e.g., Gaussian, to compensate for the error of the local SIR that dominates the QoS and to benefit the soft-capacity feature of the CDMA systems. The model can be extended to include multimedia users with multiple bit rates and priorities. In this case, the system model should be modified to allow for bulk arrivals. The impact of radio propagation and physical-layer characteristics can also be incorporated more explicitly.

3.1.2 Efficient MAC scheme with adaptive bit rate


In centralized DS-CDMA PRN, the high uncertainty of the radio channel and the inaccuracy of PC may cause severe performance degradations. The PC problem is essential to the capacity, coverage, and performance of interference-limited DS-CDMA systems, in particular in the UL (15-19, 21-22). It is desired to have all signals received at the BS with exact targets of SIR for the required radio performance. To solve the PC problem properly, it often requires a fast closed-loop PC along with an open-loop PC for initial power setting (15-19). In a PRN, due to the connectionless nature of the datagram packet transmissions, the closed-loop PC over a short TTI of a radio packet appears to be impractical. The accuracy of the open-loop PC, on the other hand, is inversely proportional to the dynamic range of the signal variation. This variation is often large and stochastic in nature. The near-far effect, dominating the propagation path loss in DS-CDMA PRN, represents a major factor affecting the design and performance of the open-loop PC as well as the system capacity and coverage. The design of the non-adaptive PRN often ends up having to use worst-case scenarios of power consumption and resource utilization in order to ensure the required radio performance.

The recent research results have demonstrated that adaptation can be used at all layers of the protocol stacks to accommodate the dynamics of a wireless channel (22, 147). This contribution investigates the possibility of user-bit-rate adaptation in the MAC operation to ease the PC problem and to enhance the performance of DS-CDMA PRN in terms of spectral and energy efficiency.

88

3.1.2.2 Claim

This contribution proposes a location-dependent bit rate adaptation, referred to as SPR. This exploits the dependences between the path loss characteristics, user location diversity, transmit power, and processing gain for packet transmissions with different bit rates. The SPR adapts the user bit rate to the distance between the terminal and the serving BS while allowing the transmit power of all users to be kept within a small range of a preset target throughout the cell. That is, the higher the bit rate, the nearer the users are to the BS according to a specified power-location-rate distribution. This is illustrated in Figure 7. The proposed SPR thus compensates for the near-far effect and PC problem in DS-CDMA PRN.

Fig. 7. SPR rate-location-power resolution. This contribution, by using both theoretical analysis and simulation, shows that the proposed SPR system gains a significant improvement in throughput delay trade-offs compared to the non-adaptive counterpart having the same coverage, offered traffic, and QoS requirements. The SPR substantially reduces the headroom of PC and stabilizes the transmit power of MTs. This, in turn, reduces the peak transmit power and power consumption of MTs, as well as the inter-cell interference. The SPR therefore significantly enhances spectral and energy efficiency.

The SPR is gradually developed and evaluated in Papers XI, XII, XV, and XVI considering various system scenarios, models, parameters, and their impact on the system performance. This has led to the development of new concepts, such as “green wireless networks” reported in Paper XV, and analytical methods to deal with complex dynamics of multi-user multi-rate wireless systems including stochastic fluctuations of the radio channel and multiple-access capacity, traffic load, user spatial distributions, user mobility,

rm

r0

d

log(d)

log(ΔPtx)

R0 R1 R2 R3

adaptive

non-adaptive

Note: ΔPtx is the normalized average of transmit-power due to the near-far effect

89

and PC inaccuracy. These can be claimed as additional contributions as well. Paper XII is selected to reprint in the appendices.


The system model is based on a CLSP/DS-CDMA PRN with an infinite user population. CLSP is described in Chapter 2 Section 2.2.3 and considered as a static variation of FCSI-based MAC schemes investigated above. The sensing of the number of ongoing packet transmissions is assumed to be perfect and the controlling delay in the MAC operation to be zero. The radio packets, i.e., MAC frames formed after data segmentations and coding, have the same length of L bits.

The traffic model is assumed as follows. The scheduling of packet transmissions, including retransmissions, at the user terminals is randomized to such an extent that the offered traffic of each user is the same and the overall number of packets is generated according to the Poisson process with rate λ.

The propagation path-loss model is given in Chapter 2 Section 2.1.1. The imperfection of PC is characterized by a lognormal error of the received SIR against its target with a mean of zero and a standard deviation of σ.

Two fading channel models are considered. In Papers XI, XII, and XV, a less correlated fading channel is considered, in which each channel state is described by an HMM. The channel state is defined as the number of ongoing packet transmissions. The HMM consists of two states, which are characterized by a different value of the bit error probability relative to the targeted BER. In Paper XVI, a burst-error correlated fading channel is considered, taking into account the impact of low user mobility. The channel model is developed based on fading duration statistics: LCR (Level Crossing Rate) and AFD (Average Fade Duration) (16, 198-202). The LCR and AFD denote the rate of occurrence and the average length of burst errors, respectively. These depend on the fading margin and the maximal Doppler frequency.

In the case of adaptive bit rate transmissions, the channel state is also referred to as the occupancy vector. The channel load state, defined as the product of the occupancy vector, and the vector of the load factors correspond to different bit rates of the transmissions. The load factor is given by (2.24) and (2.28). The following notation is defined:

R0 − the primary bit rate specified for the cell coverage and the transmit power restriction of users; T0 − the packet duration, i.e., TTI corresponding to the primary rate R0, T0 = L/R0; Rm− the bit rate supported by the system, Rm = 2mR0, m ∈ M = {1, 2, ..., M} assuming that the system supports such (M +1) different bit rates; Tm− the packet duration corresponding to the bit rate Rm, Tm = L/Rm.

The SPR adaptation is specified as follows:

r0 − the cell radius, referred to as the normalized unit of distance taking the value of 1, corresponding to the primary bit rate R0; rm − the radius for the coverage of bit rate Rm in the SPR, defined as rm

−α/Rm = r0−α/R0 = 1/R0, where α is the path-loss exponent. Thus, rm = r02−m/α = 2−m/α,

90

and the cell area are divided into (M +1) rings centered to the BS as for a spatial resolution of the user location. This is illustrated in Figure 7, providing that rM+1 = 0. The SPR then adapts the bit rate R of a user transmission to the distance d between the user and the base station as:

If rm+1 < d ≤ rm then R = Rm for all m∈M, and rM+1 = 0. (3.9)

The factor d−α/R is thus kept moderate throughout the cell, and so it allows approximately a constant transmit power Ptx to be used for all terminals at any time according to (2.1) and (2.3), assuming no fading. In the equilibrium condition, the SUD is generally modeled with a function of f(d,θ), where θ is the angle coordinate taking value in [0, 2π].

The equilibrium probability of a successful packet transmission is defined as the product of three factors: the probability that the packet is not blocked due to CLSP; the probability that the packet is not corrupted due to system outage; and the probability that the packet is not faded.

Fig. 8. Throughput-delay improvement of SPR. The throughput delay performance characteristics, including goodput, are derived. This is based on a stochastic multi-rate loss network model, coupling with a Gaussian approximation for the interference distribution and the outage probability. The details are presented in Papers XI, XII, XV, and XVI. In the SPR system, a smaller processing gain due to a higher bit rate transmission is compensated for by a shorter TTI and a smaller SIR target required for maintaining the QoS requirement. Furthermore, the SPR

0 5 10 15 20 25 30 35 400

0.5

1

1.5

2

2.5

System Throughput

Ave

rage

Pac

ket D

elay

fixed systemadaptive system with 1-dim uniform SUDadaptive system with 2-dim uniform SUD

91

significantly reduces MAI, especially inter-cell interference, as the number of simultaneous transmissions in the adaptive system is smaller than that in the non-adaptive counterpart for the same offered load and the transmit power of the users is stabilized at a preset level. The results, overall, show that the SPR significantly enhances the throughput delay performance. The performance of the SPR system, however, is sensitive to SUD and the path-loss law exponent. This can be desirable because as more users are put to use higher bit rates this boosts up the rate and reduces the time of communications. Figure 8, which is the same as Figure 4 in Paper XII, presents the throughput delay improvement offered by the SPR, for example.

3.1.2.4 Conclusion

The performance of DS-CDMA PRN using SPR is evaluated against the non-adaptive counterpart based on both analysis and simulation. The SPR is a simple and effective scheme to improve the efficiency and robustness of the centralized PRN. The SPR allows the users to keep approximately the same transmit power and consume the same energy per packet at any time. This not only reduces interference to adjacent cells so as to improve cell capacity comparable to that of an isolated cell, but also enhances cell coverage and radiated power towards the users. Thus, the SPR can be considered as environment-friendly and this aspect can be further developed into a “green wireless networks” concept caring for human wellbeing as reported in Paper XV.

3.1.3 Efficient MAC scheme with adaptive packet length


The impacts of highly erratic radio channels (i.e., fading impacts) often cause serious performance losses to PRN, which non-adaptive MAC schemes of PRN are not able to compensate for. The length of radio packets, which is controlled by the MAC layer, matters. The smaller the packet length, the better the likelihood of the packet being successfully transmitted over a fading radio channel and, at the same time, the heavier the protocol overhead. This is shown in the following formula of the goodput G, defined as the average data rate successfully transmitted excluding the protocol overhead or, that is, the effective throughput (146, 147):

G/R0 = Pc(L−H)/L, (3.10)

where:

R0 − the bit rate of packet transmissions in kb/s; L − the packet length in bits; H − the length of the packet header plus trailer in bits; Pc − the probability of correct packet transmission.

92

For perfect codes and channels with independent bit-errors (16, 207)

∑=

−−⎟⎟⎠

⎞⎜⎜⎝

⎛=

eN

i

iLe

iec PP

iL

P0

)1( , (3.11)

where:

Pe − the average bit error probability; Ne − the maximum number of correctable error bits that represents the error-correcting capability of the applied FEC coding.

Equations (3.10) and (3.11) can be used for the derivation of an optimum packet length as in earlier works (137, 142, 146-147). To obtain applicable results for a specific system, the dynamics of the channel conditions and the effects of the system parameters on the performance have to be studied. This contribution investigates the possibility of packet length adaptation in MAC operation for enhancing the performance of CLSP/DS-CDMA PRN over burst-error correlated fading channels. In certain conditions, such adaptive MAC schemes improve the efficiency and robustness of wireless systems without the need of excessive, power-consuming signal processing as required for using advanced FEC coding and diversity techniques.

3.1.3.2 Claim

This contribution develops two alternative schemes for packet length adaptation over burst-error correlated fading channels in DS-CDMA PRN, referred to as SPL. The adaptation criteria eliminate the impact of fading with respect to an optimal trade-off between the system throughput, average packet delay, and goodput. The first SPL alternative keeps the packet length as large as possible to avoid degradation of the goodput while fulfilling a specified lower limit of PER. The second SPL alternative optimizes the goodput of individual users. The channel model is adapted to low user mobility in PRN. The correlation between the fade/interfade duration, packet length, packet header size and their impact on system performance characteristics is studied in order to develop alternative cost functions for packet length optimization and adaptation.

This contribution, by using both theoretical analysis and simulation, shows that the proposed SPL significantly improves the system performance in terms of robustness, spectral and energy efficiency. This is carried out in comparison with the non-adaptive counterpart. This contribution also provides a comprehensive analytical method for modeling the system dynamics such as user mobility and offered traffic in highly correlated fading environments, investigating the effects of different system parameters, and deriving the performance measures.

The SPL is gradually developed and evaluated in Papers III, XIII, and XIV. The SPL is also found in Paper XVIII for an optimal performance of end-to-end TCP applications presented in the next contribution. Paper III is selected to reprint in the appendices.

93


The core of the system model is also based on a CLSP/DS-CDMA PRN with an infinite user population as in the previous contribution. It is assumed that the system operates at a 2.4-GHz carrier frequency with omni-directional antennas, a user bit rate of 64 kb/s, and a processing gain of 64. The user velocity is in the range of 0-4 m/s. Thus, it takes at least 32 ms for a mobile user to travel a distance of one wavelength and the maximal Doppler frequency is up to 32 Hz. The following modeling extensions are added.

It is well-known that at a selected carrier frequency the physical characterization of the fading channels can be expressed in terms of the fading margin and the normalized Doppler bandwidth. For low user mobility, the Doppler bandwidth is small and the fading process is highly correlated. In this case, the interdependence of channel errors cannot be neglected and (3.11) is not applicable. The fading channel model is based on fade/interfade duration distribution, i.e., LCR and AFD statistics of a multi-path Rayleigh fading channel taking into account diversity-combining gains at the RAKE receiver of the BS.

The probability of a correct packet transmission over a fading channel depends on the number of fades occurring during the transmission period, the burst-error duration under fading, and the error-correcting capability of the applied FEC coding. The simulation results in Paper III as well as the earlier works (138-140, 144, 147, 206-210) demonstrate that FEC coding is not efficient in such highly correlated fading channels. This is because the burst-error duration can be expected in the range of multiple milliseconds, and thus an error-block under fading can contain hundred(s) of bits depending on the bit rate. In order to correct them with FEC coding, the coding rate has to be impracticably small or the interleaving depth has to be impracticably large. Thus, in highly correlated fading channels, the probability of a correct packet transmission, as a function of packet duration T = L/R0 and maximal Doppler frequency fd, can be approximated as follows

)/()/()/exp(),( _1__ avrgifavrgifavrgifdsf tTEtTtTfTP −−= , (3.12)

where tif_avrg = 1/LCR−AFD denotes the average interfade duration; T takes value in [Tmin, Tmax] as 10 ms or multiples thereof; and

∫∞ −−=y

dtttyE 11 )exp()( .

The detailed derivation of (3.12) is given in Paper III.

(A1): First SPL Alternative. The following system parameter is introduced: β − the required rate of successful packet transmission, i.e., 1−PER, depending on the nature of systems and services. To satisfy such a requirement in the system with a fixed packet length, the packet duration may need to be set to Tmin resulting in a significant degradation of goodput. This SPL alternative adapts the packet duration T (and so L = R0T) to the time-varying channel conditions so that Psf(T, fd) is kept above β while keeping T as large as possible for maximizing the goodput. This can be formally described by

94

T = arg max{T: Psf(T, fd) ≥ β, Tmin ≤ T = 10x ≤ Tmax}, (3.13)

where x is a positive integer.

(A2): Second SPL Alternative. This SPL alternative adapts packet duration T to the time-varying channel conditions so as to maximize the normalized goodput of individual users, which is similar to that of (146, 147)

T = arg max{Gsf(T, fd) = Psf(T, fd)(R0T−H)/R0T; Tmin ≤ T = 10x ≤ Tmax}. (3.14)

Equations (3.13) and (3.14) are solved by numerical methods. Depending on the maximal Doppler frequency of the mobile, there is a constraint of T that satisfies (3.13) or (3.14). Thus, to implement (A1) and (A2) alternatives, only an online estimation of the maximal Doppler frequency is required. This is available and can be extracted from the parameter estimations of the RAKE receiver.

For modeling the dynamics of user mobility and the distributions of the conditions that drive the SPL adaptation, a DFD (Doppler Frequency Distribution) function is introduced. Let the DFD function be v(fd), where fd is limited between fd_0 = 0 and fd_max due to constraints of the system model. In a simple and dynamic case of user mobility, v(fd) can be modeled with a uniform PDF; and in a more general case, v(fd) can modeled with a general Gamma PDF

bfadad

defab

fv /1

)(1)( −−=

Γ, (3.15)

where a is the shape parameter, b is the scale parameter, and Γ(x) is the Gamma function. Thus, by adjusting the a and b parameters, a suitable PDF can be obtained to fit numerous scenarios of user mobility and expansion of the Doppler frequency range. This PDF family also includes deterministic PDF and exponential PDF as its special members, in which the shape parameter a is infinite or 1 respectively. For example, in case that the majority of mobile users are moving very slowly, the exponential or the Gamma PDF with a = 2, b = 3, denoted with Gamma PDF (a = 2, b = 3), can be used for modeling. The uniform or the Gamma PDF (a = 2, b = 6) can be used for more dynamic systems.

The details of the system modeling and the derivation of the performance characteristics are presented in Papers III, XIII, and XIV. The results clearly show the following. The non-adaptive system with a small packet size can offer better throughput and delay performance, but suffers from worse goodput performance; and vice versa with a large packet size. There is an optimal packet size for the overall system performance depending on the traffic load, MAI, and DFD. This can be used to obtain a global optimum solution for the SPL, as the proposed SPL alternatives, however, are sub-optimum. For instance, setting Tmax according to the system load condition can further enhance the system performance against the effect of MAI without major changes in the SPL. That Tmax is reduced in high-load situations will reduce the number of simultaneous transmissions and MAI, and thus the chances against system outage. This is especially effective in less dynamic systems. The (A1) SPL alternative offers the best overall system

95

performance: best goodput and stable throughput delay characteristics over a large range of system dynamics. The (A2) SPL alternative provides a fair gain in light-load situations. Figure 9, which is the same as Figure 8 in Paper III, illustrates the goodput improvement offered by the SPL with (A1) and (A2) alternatives in comparison with that of non-adaptive systems with different fixed packet lengths of 10, 20, 40, and 80 ms, referred to as F10, F20, F40, and F80.

Fig. 9. Comparison of normalized system goodput for SPL.

3.1.3.4 Conclusion

The performance of DS-CDMA PRN using SPL is evaluated against the non-adaptive counterpart based on both analysis and simulation. The SPL is a simple and energy efficient method to improve system performance in highly correlated fading channels. The claims stated in the above are justified. For further studies, one may consider other fading and interference scenarios, impacts of adaptive FEC coding and different physical layer transport formats, effects of and imposed on higher-layer protocols.

5 10 15 20 25 30 35 40 45 500.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Normalized Packet Arrival Rate

Nor

mal

ized

Sys

tem

Goo

dput

(A1)

(A2)

(F80)

(F10)

(F40)

(F20)

GammaPDF(a=2,b=6) Line: analytical Non-line: simulation

96

3.1.4 Adaptive MAC schemes for optimal trade-offs between throughput and energy efficiency of TCP applications


TCP currently carries the largest amount of traffic in the Internet (6, 39). The convergence of fixed and wireless services, especially the increasing popularity of wireless Internet access, expects that existing TCP applications also work well in wireless environments. TCP is optimized for wired networks. In wireless networks, TCP exhibits performance degradation due to random packet losses, unpredictable delay and delay variation over the RL. Furthermore, the battery energy conservation for MTs, operating such applications often and for a long period of time, is an important factor to the service success.

Existing attempts to solve these problems fall into three categories: split-connection, end-to-end, and link-layer schemes (143). The latter for loss recovery on the RL fits naturally into the protocol-layering structure of mobile cellular networks. These are the most promising solutions to improve TCP performance independently from which TCP version is used. The commonly used link-layer schemes include FEC, ARQ, and packet-scheduling schemes. The energy conservation aspect is often set aside.

The objective hereby is to develop adaptive MAC schemes for packet access over correlated fading channels in CDMA cellular systems to achieve optimal trade-offs between throughput and energy efficiency with TCP applications. Both HSPA and LSPA modes using ARQ protocols are considered. In typical HSPA, different users share a common transport channel with a peak bit rate, for example, over 2 Mb/s, separated from each other in time/code domains. The packet transmissions are synchronized at a fixed TTI of, e.g., 2 ms. In typical LSPA, each user transmits on a dedicated transport channel, for example, with a peak bit rate of 64 kb/s and a TTI of 10 ms or multiples thereof. The performance of such HSPA and LSPA has been investigated in many papers (21-22, 132-136). The earlier works, based on extensive simulation campaigns, have observed:

− The retransmission rate over the RL in HSPA is typically in the order of 10%-40%; − The retransmission attempts often fail when the mobile user moves slowly; − The link-layer protocol overhead, especially in LSPA, can easily take up more than

50% of the overall RL throughput; − Furthermore, depending on the applied scheduling mechanisms, prioritized

retransmissions may cause an unfair share of the channel bandwidth.

These observations also give us motivation for further enhancements with the design of adaptive MAC schemes, along with the aforementioned objective.

3.1.4.2 Claim

This contribution develops alternative SPD schemes adapting the scheduling period between two successive retransmissions to the channel conditions for HSPA, and an SPL

97

scheme based on the previous contribution for LSPA. The adaptation criteria optimize the efficiency of resource utilization for TCP applications over correlated fading channels in low user-mobility environments. The adaptive schemes are derived based on an optimization of alternative cost functions, which are developed to express performance trade-offs between the TCP throughput and energy consumption depending on the corresponding system parameters as arguments of interest.

This contribution, by using both analysis and network simulation, shows that the SPD in HSPA significantly reduces the number of retransmissions for saving energy while maintaining adequate TCP performance. The SPL significantly reduces the protocol overhead and increases the TCP throughput in LSPA without a major increase in energy consumption. The comparison is made against non-adaptive counterparts. The SPD and SPL are capable of overriding the burst-error states of correlated fading channels and therefore enhance the robustness of radio transmissions and allow the system to operate with a much lower fading margin and transmit power. Furthermore, the SPD and SPL improve the energy efficiency and system performance without a need of excessive signal processing.

This contribution is gradually reported in Papers XVII and XVIII. Paper XVIII is selected to reprint in the appendices.


The system model is based on single-user scenarios. The user data (e.g., TCP/IP packets) are coded and segmented into transport blocks. Then, a header that contains, e.g., system-specific addresses, transmission-control and error-control information fields, is added to each transport block to form a radio packet which is transmitted over the air. The stop-and-wait ARQ protocol over the RL is applied. It is assumed that ACK/NACK is perfectly received and the radio propagation delay is zero. RTT (Round Trip Time) of the ARQ process is defined as the duration from the instant a packet is transmitted to the instant that ACK/NACK for that packet is received. The user velocity is assumed as being less than 3 m/s.

In HSPA, a shared transport channel with a peak bit rate of 2.56 Mb/s, a TTI of 2 ms, and a fixed packet length of 5120 bits for radio packets, including 320 bits of the protocol overhead, is used. In LSPA, a dedicated transport channel of 64 kb/s is used with a TTI of 10 ms or multiples thereof. The radio packet includes 160 bits of the protocol overhead.

The RL is considered as the bottleneck of the end-to-end TCP connections in both HSPA and LSPA. The TCP data packets have to pass through three other wired links, each with a capacity of 10 Mb/s and a delay of 50 ms. All TCP buffers are Drop-Tail with the buffer length of 50 packets. TCP Reno is chosen because of its popular use in the current Internet. The energy efficiency herein means the average amount of energy consumed to transfer a data unit of TCP applications, e.g., in Joules/Octet, using either HSPA or LSPA.

The fading channel model is based on that of the previous contribution, but simplified with a two-state Markov channel. The states, good (G) and bad (B), correspond to the channel conditions during the interfade duration and during the fade duration,

98

respectively. The faded packet transmission is considered as erroneous and needs to be retransmitted. It is desired to use a fading margin as small as possible to minimize the transmit power while fulfilling the radio performance requirements. The state-transition rates for the channel model are:

mGB =1/tif_avrg and mBG =1/tf_avrg, (3.16)

where tif_avrg and tf_avrg are the average interfade duration and the average fade duration, respectively. It is required that tf_avrg/tif_avrg<<1 for reliable data communications over burst-error correlated fading channels and also to justify the Markov channel model.

SPD for HSPA. In HSPA, a bad channel state can last for tens of RTT. The motivation behind the SPD is shown in Figure 10. The SPD does not retransmit the packet immediately after receiving a NACK. Instead, it waits a dispatching time interval of TSPD before the retransmission. TSPD is adapted to tf_avrg, which in turn is determined by the fading margin and the estimated maximal Doppler frequency, subject to an optimal trade-off between the TCP throughput and the energy consumption performance. Thus, to calculate the optimal TSPD for the SPD, one needs to define a reasonable cost function Φ(TSPD) which represents the desired performance trade-off. Let TSPD,opt be the value of the optimal TSPD that minimizes the cost function Φ(TSPD)

)(minarg, SPDoptSPD TT Φ= . (3.17)

Fig. 10. SPD concept over burst-error correlated fading channels. Depending on the QoS constraints, such as the maximum RL delay or RL timeout, different cost functions can be developed and adopted. Because the average energy consumption is proportional to the average number of transmission attempts for each

Fade Margin F = 5 dB

Fade Margin F = 2 dB

B G

B G

TimeNACK Retransmit

TSPD

99

successful packet E[k] and the TCP throughput is a decreasing function of the average packet delay D, the following cost function is proposed:

Φ (TSPD) = E[k]D, (3.18)

where E[k] and D are given as functions of TSPD. The detailed derivation is presented in the appendices, as it is missing from Papers XVII and XVIII. The optimization of the above cost function can be solved by numerical methods.

For a more simple and more flexible implementation of the SPD, we adopt an alternative cost function based on the following considerations. On the one hand, if we underestimate TSPD, the retransmission fails and thus the energy is wasted, which can be represented by a cost function as

fSPDSPD tTTTIcTC <= ,)( 11 ,

where c1 is the cost coefficient of the energy consumption and tf is the fade duration. On the other hand, if we overestimate TSPD, the packet delay is increased resulting in a waste of the channel bandwidth, represented by the following cost function

fSPDfSPDSPD tTtTcTC ≤≤−= 0),()( 22 ,

where c2 is the cost coefficient of the extra delay and wasted bandwidth. Thus, the average of the overall wasted cost, as an alternative cost function to be minimized in (3.17), is

∫∫ −∝

− +=SPD

fBG

SPD

fBG

T

ftm

BGSPDT

ftm

BGSPDSPD dtemTCdtemTCT0

21 )()()(Φ (3.19)

and TSPD,opt is then obtained in closed form as:

BG

BG

optSPD mc

TTImc

T)1ln(

2

1

,

+= . (3.20)

By choosing the ratio c1mBGTTI/c2 in the above formula, we can have a suitable TSPD,opt for the various system scenarios with different QoS constraints. For example, if the RL delay requirement can be relaxed, one can set c1mBGTTI/c2 = 1 then TSPD,opt = 0.7/mBG; otherwise c1mBGTTI/c2 = 0.5 then TSPD,opt = 0.4/mBG. In our system model, 1/mBG is less than 200 ms.

100

Table 2. Numerical illustration of SPD benefit.

Retransmission schemes TCP throughput in kb/s Number of retransmissions over RL Non-SPD 218.85 34485 SPD1 209.92 10264 SPD2 206.22 7743 SPD1 is based on (3.18) and SPD2 is based on (3.20) with TSPD = 0.7*AFD. The channel has a fading margin of 2 dB and a Doppler frequency of 7 Hz, that is, a user velocity of about 1m/s.

The simulation results confirm the claims with a number of retransmissions reduced by many times, while the TCP throughput does not change much (i.e., less than 5%). The SPD is expected to be even more effective in beyond-3G wireless networks, which have much higher data rates and shorter TTI. Table 2 gives a numerical example, based on the simulation results provided in Paper XVIII, to show the benefit of the SPD.

SPL for LSPA. Similar to the SPD derivation, the optimal packet length Topt in milliseconds is the value minimizing the cost function Ψ(T)

)(minarg TTopt Ψ= . (3.21)

Let L0 denote the expected length of the TCP packets and T0 denote the expected effective radio transmission time of a TCP packet, which is the total transmission-plus-retransmission time of the radio packets resulting from the segmentation of that TCP packet. Thus, T0 is a function of the radio packet duration T and proportional to the energy consumption for transferring a TCP packet.

⎡ ⎤ [ ]TkEHRTLTT )/()( 00 −= , (3.22)

where ⎡x⎤ denotes the minimum integer exceeding the argument, E[k] is the average number of transmission attempts for each successful packet as a function of T, R is the bit rate, and H is the size of the radio packet header.

The expected delay of a TCP packet D0 as a function of T is calculated as

⎡ ⎤ [ ][ ]TRTTkEHRTLTD +−−= )1()/()( 00 . (3.23)

The following cost function is adopted for packet length optimization and adaptation in (3.21):

)()()( 00 TDTTT =Ψ . (3.24)

The simulation results show that the SPL increases the TCP throughput by about 30% while the energy consumption is also slightly increased by less than 6%. Table 3 gives a numerical example, based on the simulation results provided in Paper XVIII, to show the benefit of the SPL.

101

Table 3. Numerical illustration of SPL benefit.

Retransmission schemes TCP throughput in kb/s Retransmission overhead over RL Non-SPL 32.32 16% SPL 41.26 18% Non-SPL has a fixed TTI of 10 ms, whereas SPL has a TTI of 40 ms, as the Doppler frequency is set to 1 Hz. The channel has a fading margin of 2 dB.

The SPD and SPL schemes require neither excessive signal processing nor sophisticated algorithms. The only crucial parameter that has to be estimated is the maximal Doppler frequency which is available and can be extracted from parameter estimations of the advanced receiver, such as the RAKE receiver used in CDMA systems.

3.1.4.4 Conclusion

The SPD and SPL schemes have been developed and demonstrated as simple and effective methods to improve the energy efficiency and throughput performance of HSPA and LSPA with TCP applications in CDMA systems. These schemes are flexible in order to be implemented and integrated into the functionality of the MAC layer in wireless systems. These schemes, at the same time, give room for further enhancements with advanced techniques, such as adaptive modulation and coding, multi-user detection, interference cancellation, smart antennas. Future works may further investigate the benefits of the SPD combined with packet scheduling for multi-user systems. For example, during the SPD period of a given user, the system schedules for other users to utilize the channel bandwidth and improve fairness. The fairness aspect is important to TCP networks and can be extended. One also may consider possible applications of such adaptive schemes in wireless LANs, ad-hoc, and beyond-3G wireless networks. Further incorporation of cross-layer optimization in order to develop more efficient adaptive schemes is also an attractive research direction.

3.2 SCA − Smart call admission for advanced mobile cellular networks

3.2.1 Performance analysis of a class of fixed channel-assignment CAC with generic guard-channel policy and queuing

priority handoffs


The related topics have been investigated in many papers (16, 19, 69-71, 73, 77, 170, 175). Even so, there are several reasons motivating us to consider the performance

102

analysis of fixed CA based CAC schemes with generic GC and QH in mobile cellular networks:

− The reported results were exclusively based on computer simulations in case that the handoff dwell time, i.e., the time period needed for a mobile to pass through the handoff area, is modeled with something else than exponentially distributed RV;

− The classical queuing theory has addressed those queuing systems with general customer impatience rather moderately;

− There was no unified performance analysis of different GC and QH policies; − It is valuable for both industrial and academic purposes to have an effective method,

with low computational complexity and high accuracy, to evaluate the teletraffic performance of mobile cellular networks.

3.2.1.2 Claim

This contribution provides a simple and accurate analytical method for performance evaluation of a class of fixed CA based CAC schemes with generic GC and QH in mobile cellular networks. The handoff dwell time is considered as a general RV. The impact of cell sizes and random user movements on the handoff rate is taken into account. The closed form solutions are obtained for the performance measures of interest, i.e., the probabilities of a forced termination of a handoff call, handoff failure, and new-call blocking.

This contribution is detailed in Papers V and VII. Paper V presents the results for micro cellular systems, whereas Paper VII presents the results for both micro and macro cellular systems. Paper VII is selected to reprint in the appendices.


The system model considers a reference cell of a uniform multi-cell system with a channel capacity of c. The handoff requests are allowed to queue up in a finite or infinite buffer upon their arrival if no idle channel is found. It is assumed that there are two cases causing handoff failures. The first is that a handoff call is blocked if the queue is filled up in case of a finite buffer. The second is that a handoff call is forced to terminate if the mobile passes through the handoff area before getting a new channel. The handoff dwell time is denoted by ν, which is an independent RV having a general PDF of Fν(t), where Fν(0) = 0.

The call arrivals are according to Poisson processes with rate σ1 for handoff and σ2 for new calls. The service discipline is non-preemptive and FCFS. The service time is exponentially distributed with rate μ, which is decomposed into two parts (69-73, 170, 175):

μ = μ1 + μ2, (3.25)

103

where 1/μ2 is the mean call-holding time given that there are no handoff failures; and μ1 is the outgoing handoff rate. In the condition of mobility equilibrium, i.e., the mean number of incoming mobiles is equal to the mean number of outgoing ones per unit time in regard to the reference cell, we have (68):

μ1 = E[Mu]/E[Nu], (3.26) σ1 = E[n]μ1, (3.27)

where E[Mu] is the mean number of users leaving the cell per unit time, E[Nu] is the mean number of users in the cell area, and E[n] is the mean number of calls being served in the cell. These depend on cell sizes and user movements. The system state n is defined as the number of ongoing calls found in the cell, including handoff requests.

The call request is accepted into the system according to a generic GC policy giving priority to handoff calls along with QH. Thus, a virtual number of guard channels are reserved for handoff calls by rejecting a new call with a certain probability that depends on the number of occupied channels. Let πn be the acceptance probability of a new call from the channel occupancy state n–1; πn is a probabilistic function of n and πn = 0 if n > c. The freedom of choosing the function for πn gives flexibility to tune/optimize system performance.

The queuing system model is developed based on a birth-death process as follows. The birth rate at state n is defined as

⎪⎩

⎪⎨

⎧+<≤

<≤+=

+

otherwise0,

0,

1

211

Ncnccnn

n σσπσ

λ , (3.28)

where N is the queuing buffer size to allocate handoff requests. The death rate at state n is defined as

⎩⎨⎧

+≤<+≤≤

=Ncncc

cnn

nn ,

1,γμ

μμ , (3.29)

where γn is the forced termination rate of handoff calls at state n > c and found in the following formula, which is derived in Papers V and VII based on (212, 213, 223):

μμΦμΦγ c

cccn

cn

cnn −−=

−

−−

)()()( 1 , (3.30)

where Φn(s) is the Laplace transformation of ∫ −=t

nn dxxFtg

0

)))(1(()( ν .

104

Fig. 11. Loss probabilities for different PDF of the dwell time. The steady-state characterization and the derivation of the closed form performance measures are detailed in Papers V and VII. The numerical results, for example, in Figure 11 which is the same as Figure 2 in Paper V for a micro cell and similar to Figure 2 in Paper VII for a macro cell, show that there is a notable difference in the probabilities of a handoff forced termination and failure regarding different PDFs used for modeling the handoff dwell time. The deterministic one provides a lower bound for the loss probabilities. The results also show that it may not necessarily require a large buffer size for QH. The suitable queue size, however, still depends on the offered load.

3.2.1.4 Conclusion

The provided analytical method can be used as an effective tool for quick network dimensioning and planning, and also to investigate different models and system parameters for performance optimization. The application of the contribution can also be recognized in many QoS related problems with real-time data services or in HSPA, where individual packets have some delay constraint.

5 10 15 20 25 3010-20

10-15

10-10

10-5

100P

b an

d P

f Pro

babi

litie

s

Normalized System Offered Traffic sig1+sig2

c= 30, r= 27, N= 10 with CGCP micro-cell, radius= 100m, E[V]= 1m/s mean call-holding time mu2= 120s, normalized mean dwell time= 0.25 Exp. Distr. Dwell Time +Line: Pb xLine: Pf Det. Distr. Dwell Time squareLine: Pb circleLine: Pf

Pb

Pf

105

3.2.2 Simple and efficient soft-decision CAC for CDMA cellular networks


On the one hand, number-based CAC schemes for CDMA cellular networks appear to be simple. However, it is hard to define an effective, fixed value of the channel capacity for such schemes because the qualitative and quantitative features of CDMA systems are heavily intertwined and mutually supportive. The QoS guaranteed scenario would end up using the worst case of the channel capacity and results in the worst GoS.

On the other hand, SIR-based or interference-measurement based CAC schemes are robust, but require much more complex hardware and software for implementation. In high load and uniform traffic distribution among adjacent cells, such measurement-based CAC schemes may not offer any capacity gain over number-based or modeling-based counterparts and may even suffer from performance degradations due to measurement and estimation errors.

The capacity provisioning of the CDMA systems presented in Chapter 2 Section 2.1.3 indicates that, as the traffic load of the own cell increases beyond an average capacity, the users may experience a degradation of the communication quality depending on the other-cell interference level. Hence, the objective is to develop simple and robust CAC schemes capable of utilizing the soft capacity of CDMA systems for an optimal trade-off between QoS and GoS.

3.2.2.2 Claim

This contribution proposes a SCAC scheme for CDMA cellular networks. The idea behind SCAC is illustrated in Figure 12. This uses a probabilistic decision for admitting calls, adapted to the interference distribution to compensate for QoS while expanding the admissible region to accommodate more users. This method not only offers a notable enhancement in GoS and Erlang capacity, but also avoids complex measurements and mutual exchanges of information between adjacent cells about states of the network. This is therefore simple, scalable, signaling and cost-effective.

106

Fig. 12. Illustration of SCAC. This contribution is twofold. First, alternative probabilistic decision functions are developed for SCAC, i.e., using Gaussian and Incomplete Gamma PDF adapted to the inter-cell interference distribution. Second, a simple and reliable method for evaluating SCAC against fixed-assignment and measurement-based CAC schemes, referred to as FCAC and MCAC, is provided. The analytical model is developed based on a stochastic loss-network model and Gaussian approximation.

This contribution is presented in Papers VI, IX, and X with supplementing results. Paper IX is selected to reprint in the appendices.


The system model considers a uniform multi-cell single-class system scenario, similar to that used for capacity provisioning in Chapter 2. PC is assumed as being sufficient enough so that the error of the received SIR is kept in the allowed order and the load factor of each user is modeled with a bound Gaussian RV having a mean of w given by (2.28) and a variance of σ2 and limited in [wl, wu]. The total UL load factor of a reference cell is also viewed as a bound Gaussian RV limited in [Gl, Gu]. The lower bound Gl is determined by the maximum of other-to-own cell interference factor up to 1 and the upper bound Gu can be extended to that of an isolated cell. Thus, based on the knowledge of maximum and minimum guaranteed resource consumption, i.e., load factors of user connections and cell basis, lower-limit and upper-limit constraints of the cell capacity can be determined. These are denoted by cl and cu, respectively.

The proposed SCAC scheme utilizes both cl and cu, instead of an average capacity cavrg as in traditional number-based CAC schemes surveyed in Chapter 2. In the SCAC system, a new call will be admitted immediately if less than cl users are occupying the cell or with a probability defined below if the number of users occupying the cell is at least cl and less than cu; otherwise a new call will be rejected immediately. The priority is given to handoff calls meaning that a handoff failure occurs when cu users are occupying the cell.

1

0clow er cavrg cupper

Hard-decision

Soft-decision

Decision function

Admissible region

clower, cav rg, cupper: lower-bound, average, upper-bound of system capacity

107

The system state n is defined as the number of calls occupying the cell. The permission probability π(n) to accept a call at state n is adapted to the other-cell interference distribution, which is assumed resembling Gaussian PDF (15, 17-19):

[ ][ ]

⎪⎪

⎩

⎪⎪

⎨

⎧

<≤⎟⎟⎠

⎞⎜⎜⎝

⎛ −−−−

<

=

otherwise , 0

if , 11

if , 1

)( ulother

other

l

cncGVar

GEnwQ

cn

n ηπ , (3.31)

where η is the ratio of the noise power to the maximum total received interference (i.e., equivalent to η/(1+η) if η is defined as the noise figure as in Chapter 2) and Gother is the load factor caused by other-cell interference. Thus, the error of SIR on average, which dominates the QoS in the long run, is compensated. However, an overestimate of cu may still affect the overall communication quality. To prevent that, other than the Q-function can be used to define the decision function such as the following Incomplete Gamma PDF:

( ) ul cncnwn <≤−−−= for ,)1(,1)( ηανΓπ , (3.32)

where Γ(ν, x) is the Incomplete Gamma function with α and ν parameters having the following relations: E[Gother] = ν/α and Var[Gother] = ν/α2. E[n], E[Gother] and Var[Gother] can be calculated by using the steady-state solution and inversion technique detailed in Papers VI, IX, and X.

The SCAC scheme harmonizes the advantage of the traditional number-based and measurement-based CAC schemes, that is, simplicity and robustness, respectively. This also gives more flexibility for the implementation.

The traffic model is based on similar assumptions as in the previous contribution. The loss probability of the communication quality on average is defined by the ratio of the average number of users hit by the system outage and the average number of users being served in the system:

[ ][ ]

1

00

1−

==⎟⎟⎠

⎞⎜⎜⎝

⎛⎟⎟⎠

⎞⎜⎜⎝

⎛ −−−= ∑∑

uu c

nn

c

n other

othernloss np

GVarGEnwQnpP η , (3.33)

where pn is the steady-state probability. The performance of measurement-based CAC schemes is analyzed based on the

central limit theorem and Gaussian approximation. Then, the performance measures of the measurement-based CAC schemes are obtained by using simulations.

The comparative performance evaluation of the SCAC scheme and the traditional number-based and measurement-based counterparts is detailed in XVII-XIX. The results show that the proposed SCAC scheme, on the one hand, causes a slight degradation in the communication quality and, on the other hand, provides a significant improvement in the

108

GoS. The SCAC scheme overall offers much better Erlang capacity. This is, for example, illustrated in Figure 13, which is the same as Figure 1 in Paper X.

Fig. 13. Performance comparison of CAC policies.

3.2.2.4 Conclusion

The SCAC scheme provides a simple and robust method to benefit soft capacity and optimize radio resource utilization in CDMA cellular networks. The simplicity is hereby emphasized as one of the most important requirements because only this can ensure a fast response to a high intensity of call requests with increasingly complex and diverse QoS requirements in advanced cellular networks. However, this contribution not only proposes the SCAC scheme, but also gains an understanding of different CAC schemes for CDMA cellular networks and provides a comprehensive method for teletraffic performance evaluation.

4 6 8 10 12 14 1610-12

10-10

10-8

10-6

10-4

10-2

100

Offered Traffic

Gra

de o

f Ser

vice

GoS of MCACGoS of FCACGoS of SCACQoS Loss Pr of SCAC

109

3.2.3 GoS management for multimedia CDMA cellular networks


For the deployment of multimedia cellular networks, such as 3G CDMA systems, there is a need for an in-hand tool that allows the operator to manage rather complex network operations and performances in an efficient way. That is, to optimize GoS while providing adequate QoS to mobile multimedia users.

The GoS management and teletraffic engineering framework must be simple and flexible for supporting a variety of services with QoS differentiation to satisfy a wide range of customer demands and to maximize revenues.

CAC is an integral component of RM in mobile cellular networks. Efficient CAC schemes can enhance the system capacity and service quality in a cost-effective way. CAC schemes can therefore play an important role in supporting the above needs and requirements.

The operational aspects of multimedia cellular networks have been somewhat neglected in the recent research works despite the recognition of their practical value.

3.2.3.2 Claim

This contribution incorporates alternative QoS differentiation paradigms and resource partitioning into CAC and provides an in-hand tool for efficient capacity and teletraffic management. It introduces different user prioritization and load-based traffic-shaping schemes for GoS provisioning and optimization. The GoS of multi-user multi-service cellular systems is hereby defined as a cost function built upon the probabilities of new-call blocking, handoff failure and loss of communication quality as arguments. The weighting factor of each argument depends on the service and user profile characteristics.

This contribution includes an expansion and further evaluation of the SCAC, number-based CAC and measurement-based CAC schemes that were investigated in the previous contribution for an integrated multi-class multi-service system. SCAC exhibits further advantages in serving multiple traffic classes. It gives more chances for calls of high resource-consuming classes to access the system when the traffic load or the offered traffic intensity of low resource-consuming classes is high. This aspect cannot be improved with the other CAC schemes. Furthermore, this contribution provides an asymptotic quasi-stationary analysis of the remaining free capacity over a sustained period of time. This is essential to the challenge of the design of efficient MAC schemes for controlling background packet access, which has to adapt to the free capacity left by the ongoing calls. The performance analysis is carried out with a simple and effective analytical method based on a stochastic, multi-priority, multi-rate loss-network model. Extensive numerical and simulation results are provided to investigate the system performance with various traffic patterns.

This contribution is detailed in Papers I and VIII. Paper I, as a chapter of (6), is based on Paper VIII. However, an explicit GoS cost function for a multi-user multi-service

110

mobile cellular system with QoS differentiation is first defined in Paper I. Paper VIII is selected to reprint in the appendices.


The system model used for the previous contribution is now extended to support M different connection-oriented services. For all k∈ K, K = {0, 1, ..., Μ}, the load factor of each class-k call user is modeled with a bound Gaussian RV having a mean of wk given by (2.28) and a variance of σk

2 and limited in [wl,k, wu,k]. Let us define n = (nk, k∈ K) is the occupancy vector, where nk is the number of class-k calls in progress; w = (wk, k∈ K) is the load vector, where wk is the average load factor of a class-k call; c = nw is the system load state that is identical to the effective load factor of the reference cell, also referred to as the own cell.

In the SCAC, the permission probability to accept a class-k call at system load state c is adapted to the other-cell interference distribution. Equations (3.31) and (3.32) are modified as

[ ][ ]

⎪⎪

⎩

⎪⎪

⎨

⎧

≤+<⎟⎟⎠

⎞⎜⎜⎝

⎛ −−−−−

≤+

=

otherwise , 0

if , 11

if , 1

)( uklk

lk

k CwcCcfVar

cfEwcQ

Cwc

c ηπ (3.34)

or alternatively

( ) uklkk CwcCwcc ≤+<−−−−= for ,)1(,1)( ηανΓπ . (3.35)

where η is the ratio of the noise power to the maximum total received interference (i.e., equivalent to η/(1+η) if η is defined as the noise figure as in Chapter 2); Cl and Cu are the lower bound and upper bound of the cell capacity normalized in the load factor; f is the other-to-own cell interference ratio (i.e., equivalent to ι used in Chapter 2). The calculation of E[c] and Var[c] are explained in more detail in the appendices, as it has not been included in the published papers of the author.

To regulate the operation of multimedia cellular systems in GoS management, CAC schemes not only have to know about the characteristics of the requested services, but also the subscription profiles of the users. This ensures a fair decision in resource allocation for the service-level agreement upon accommodating a new call request. QoS differentiation paradigms specify the serving traffic patterns depending on the offered traffic intensity of each traffic class during different busy hours of the day, subject to optimal GoS and revenues. These have to be simple, that is, fast responding to call requests, easy to configure and extend for supporting a variety of user and network profile characteristics. The QoS differentiation of call service classes can be viewed as the traffic prioritization in CAC schemes. This can be realized by, for example, defining

111

multiple load-based thresholds or fracturing factors for hard or soft blocking of less desirable call requests. Thus, the network resources are now shared among users in a non-complete fashion.

Let us assume there are J classes of subscribers sorted in the decreasing order of the priority, for example, 1 is “gold”, 2 is “silver”, etc. Let J = {1, 2, ..., J}. Together with M different service classes sorted in the increasing order of the resource consumption, it can be considered as there are J×M classes of prioritized user traffic, which form different group behaviors according to the user or service classes. Thus, we introduce a J×M matrix of the prioritized admission probability as follows

A0 = [ajk(c)] with j∈J, k∈K, 0 ≤ ajk(c) ≤ 1, (3.36)

where ajk(c) is the prioritized admission probability of a new call request from a class-j user for a class-k service at a given system load state c.

In the case of threshold-based hard blocking, it is straightforward that J×M thresholds can be defined; each corresponds to a traffic class:

L = [ljk] for all j∈J, k∈K, ljk ≥ luv if j ≥ u and k ≥ v (3.37)

and ajk(c) in (3.36) can be given by

⎩⎨⎧ ≤+

=otherwise 0

if 1)( jkk

jk

lwcca . (3.38)

The number of necessary thresholds can be reduced significantly if only group behaviors are of interest. For example, the upper bound Cu can be used as a threshold for the “gold” class, whereas the lower bound Cl can be used to limit the access of the “bronze” class regardless of the requested services.

In the case of threshold-based soft blocking, ajk(c) can take any value in [0, 1] depending on the system load state and priority of the requested traffic class. This results in load-based fracturing factors for each traffic class or group behavior. The fractional paradigm, on the one hand, gives the operator better flexibility to tune the GoS performance and, on the other hand, allows unifying the analysis of a class of guard resource (similar to GC) schemes as in Section 3.2.1.

Taking into account QoS differentiation, the admission probability of class-(j, k) calls for all j∈J and k∈K in the SCAC system at the system load state c is given by

As = [πk(c)ajk(c)]. (3.39)

The handoff calls are assumed to have the highest priority regardless of their associated traffic classes. QoS differentiation of the background non-real-time packet services can be done not only in a blocking fashion, but also in granting different throughput delays.

The SCAC scheme is specified to optimize the GoS, defined as the following cost function:

112

∑∑∈ ∈

+−=J Kj k

jkjkkjl FBP )()1(GoS 210 υυξκυ , (3.40)

where Bjk is the probability of class-(j, k) new call blocking; Fjk is the probability of class-(j, k) handoff failure; and Pl is the probability of loss of communication quality. The weighting factors in (3.40) can be set according to the QoS differentiation paradigms. For example, υ0 = 0.95 is for an overall 5% of the allowable communication quality loss, υ1 = 0.33 and υ2 = 1−υ1 as a handoff has higher priority than a new call. The parameter κj for the “gold” user class is set to 0.45, the “silver” to 0.33, and the “bronze” to 0.22 meaning that the “gold” users provide more revenues than others and

∑∈

=Jj

j 1κ .

The parameter ξk represents the relative resource consumption of each service class, which is proportional to the cost of the service, given that all calls have the same duration:

∑∈

=Kk

kkk ww /ξ .

The GoS cost function defined by (3.40) thus takes value in [0,1] and needs to be minimized, depending on the offered traffic patterns, also referred to as MTIP (Multimedia Traffic Intensity Profile). It is also obvious that (3.40) also represents a trade-off between QoS and GoS.

113

Fig. 14. Example of GoS management with different CAC schemes (MdCAC refers to modeling-based fixed-assignment CAC, MsCAC to measurement-based CAC, ThresholdBlkQoSDiffSCAC to SCAC with threshold-based hard-blocking QoS differentiation, and FractionalBlkQoSDiffSCAC to SCAC with fractional soft-blocking QoS differentiation. The state characterization and the derivation of the performance measures are detailed in Papers I and VIII. The system performance with measurement-based CAC schemes is analyzed based on the central limit theorem and multidimensional Gaussian approximation. Then, a network simulation is carried out to obtain the performance measures of the measurement-based CAC system. The results of a comparative study between the SCAC system and the traditional number-based and measurement-based counterparts show that the SCAC scheme can maintain a good and stable GoS performance over a large range of offered traffic with various MTIPs. The results also show that the SCAC with QoS differentiation provides a flexible tool to tune the group behavior of different traffic classes and to optimize the GoS. This significantly enhances the Erlang capacity for a cost-effective network operation. This is illustrated in Figure 14, which is the same as Figure 12.1(a) in Paper I.

In the meantime, the background non-real-time packet access needs to be controlled so that GoS of the real-time traffic is not affected and optimal throughput delay trade-offs can be achieved. Thus, there is a need for a precise dimensioning of the dynamics of the free capacity or the available resources left for packet access in the design of such adaptive MAC schemes. The quasi-stationary probability of being in load state c over a sustained period of time τ is defined by

0.5 1 1.5 2 2.5 3 3.5 40.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

Common Divisor of the Offered Traffic

1-G

oS

MdCACMemorylessMsCACMemoryMsCACSCACThresholdBlkQoSDiffSCACFractionalBlkQoSDiffSCAC

Offered RT traffic in 4:3:3 MTIP

114

p(c,τ) = lim t→∞ Pr{c(t +τ) = c⎟ c(t) = c}. (3.41)

This is estimated using the Pascal approximation, detailed in Papers I and VIII. Based on the results, optimal transport formats for packet transmissions (i.e., bit rate, TTI, and persistent probability) with FCSI MAC schemes are discussed.

3.2.3.4 Conclusion

The contribution provides a comprehensive method for GoS management in multimedia CDMA cellular systems. It demonstrates that CAC with QoS differentiation is a simple and effective means of optimizing GoS and Erlang capacity for network operation.

The results, especially the SCAC scheme, are highly applicable for CDMA networks as segments of future wireless IP networks where the simplicity and scalability requirements favor a simple and distributed resource management.

4 Concluding remarks

“To raise new questions, new possibilities, to regard old problems from a new angle requires creative imagination and marks real advance in science” (Einstein, 1938). To this end, the present thesis has analyzed and synthesized novel MAC and CAC schemes for efficient RM in advanced wireless networks. This chapter summarizes the work presented so far in this thesis, emphasizing the new results with further commentaries and remarks. It also provides some directions for further development and future research on the related topics.

4.1 Summary

Chapter 1 provided the reasoning, solid background, and context for the research topic and focus areas of interest: the RM problem in advanced wireless networks and MAC and CAC schemes in advanced wireless networks. It formulated and presented a comprehensive overview of the integrated RM problem. Essential RM schemes were addressed. It defined the focus areas of interest on MAC and CAC, identified problems, and shortly stated the aim, objectives, and contributions of the thesis.

Chapter 2 presented a literature review of MAC and CAC schemes. It addressed the fundamentals as well relevant up-to-date developments on the subjects. It provided a systematic understanding of MAC and CAC schemes in centralized PRN and advanced cellular networks: basic properties, classifications, existing solutions and their performance, open issues and problems, and trends for future developments. These were considered in the integrated context of RM with respect to optimal spectral and energy efficiency for advanced wireless networks.

Chapter 3 summarized the research results of the author as his contribution to knowledge. It presented seven separate contributions: four on the SPA with adaptive MAC schemes for centralized PRN and three on the SCA with CAC schemes for advanced mobile cellular networks. The main problem, claim, approach, result, and conclusion of each contribution were clearly outlined. The supporting original publications are enclosed in the appendices for further details.

116

On the SPA, the first contribution studied the implementation losses of FCSI based MAC schemes in DS-CDMA PRN due to feedback delay, access delay, and imperfect channel sensing. It also proposed optimal control functions in MAC operation so as to enhance throughput delay performance and robustness against system imperfections for PRN. The second contribution proposed the SPR to boost up the performance of DS-CDMA PRN in terms of spectral and energy efficiency. The third contribution proposed the SPL to enhance the efficiency and robustness of DS-CDMA PRN over burst-error correlated fading channels. The fourth contribution developed the SPD and SPL for HSPA and LSPA in CDMA systems, respectively, in order to achieve optimal trade-offs between the throughput and energy consumption of TCP applications.

On the SCA, the first contribution provided a simple and accurate analytical method to unify the performance evaluation of a class of fixed-assignment CAC schemes with generic GC policies and priority handoffs in mobile cellular networks. The second contribution proposed the SCAC as a simple and effective scheme to benefit the soft capacity of CDMA cellular networks for an optimal trade-off between GoS and QoS. The third contribution provided a simple and effective method for GoS management and optimization in multimedia CDMA cellular networks with an integrated CAC and QoS differentiation.

4.2 Remarks

The RM problem that has been addressed in this thesis is one of the most challenging and widespread research problems in the field of wireless communications. The fundamental goal of the RM is to maximize resource utilization of wireless networks, especially the limited radio spectrum and MT battery power, so as to serve as many users as possible while respecting the QoS required for each user. Given an anticipated rapid growth of user population and diverse demands for wireless communications services, efficient RM becomes ever more crucial and has to cope with many sophisticated requirements. Therefore, it is essential to design and use adaptive RM schemes capable of guaranteeing QoS requirements and, at the same time, achieving efficient utilization of system resources.

The recent research works have demonstrated that adaptation mechanisms can be performed at all layers of the protocol stacks to accommodate the dynamics of a wireless system resulting from, e.g., the randomness of the traffic load, the user diversity, and the hostile nature of wireless channels. This thesis has addressed various issues and approaches on MAC and CAC schemes for efficient RM in advanced wireless networks. The MAC layer, directly interfacing with the radio physical layer, is responsible for coordinating and controlling the user transmissions of radio packets, i.e., the amount of capacity allocated to each user. It also has to deal with the residual errors or losses of the physical transmissions first-hand. The CAC of the network layer is considered as the user check-in process. It reflects the overall RM strategy: handling QoS requests of handoffs and new calls, resource partitioning, resource allocation, etc. It ensures stability for efficient QoS support and GoS optimization.

117

This thesis has introduced new concepts, new adaptive MAC and CAC schemes for enhancing and optimizing wireless network performance with respect to efficient RM. The SPP significantly enhances the robustness of FSCI based MAC schemes over feedback delay, access delay and imperfect channel sensing for DS-CDMA PRN. The SPR is capable of compensating for the PC problem, stabilizing the transmit power of user terminals, reducing inter-cell interference, and boosting up throughput delay performance for DS-CDMA PRN. The SPL and SPD enhances the robustness of radio transmissions in burst-error correlated fading environments while significantly increasing the energy efficiency for DS-CDMA PRN and the performance of TCP applications. The SCAC, together with QoS differentiation paradigms, is capable of providing stable, optimal GoS and Erlang capacity for multimedia CDMA cellular networks. These schemes are simple and cost-effective, improving the performance of wireless networks without a need of excessive signal processing and signaling. The simplicity aspect must be emphasized in the design of RM schemes in general, as perhaps only this can ensure practicality of the RM schemes in coping with an increasing complexity and diversity of wireless networks and operation environments.

This thesis has provided simple and effective analytical methods for evaluating MAC and CAC schemes as well as wireless network performance. The impact of different system parameters and various issues concerning the dynamics of wireless systems, being stochastic in nature and often complex, are investigated and taken into account. The essential performance characteristics of wireless networks, such as throughput, packet delay, outage probability, call blocking probability, handoff failure or call dropping probability, and GoS, are derived and quantified with both numerical and simulation results. The analytical results can be used extensively for network provisioning and optimization.

There have of course been some limitations in conducting the research results. The modeling has had to use certain assumptions to simplify the system in question, without loss of the insight. The work has focused on the UL direction to present the results. The traffic asymmetry between UL and DL has not been taken into consideration. It has been assumed that the real-time services are of CBR, whereas the non-real-time packet services are of best-effort classes. The traditional Poisson arrivals have been adopted for the offered traffic load. An explicit RF system, channel coding and modulation have not been modeled. These limitations, however, are kept under a commonly acceptable range of working assumptions, which have been justified in related literature or often experienced in related works. The research results of this thesis have provided enhancing concepts and solid groundwork for further investigations of more realistic system integration scenarios and advanced applications for future wireless networks.

4.3 Future research directions

There is a substantial amount of open issues remaining in the subjects that have been touch upon in this thesis. For example, considering the limitations listed above and then attempting to take a bolder approach to network modeling and end-to-end QoS provisioning, which are capable of capturing the interactions of protocol stacks, will open

118

up many interesting questions and problems for performance analysis as well as optimization across system layers. The breadth of cross-layer optimization techniques will remain challenging and attractive for research.

The motivations behind the adaptive MAC and CAC schemes in this thesis can be applied to develop simple and effective RM schemes in different system environments, with multiple antennas and carriers. Thus, by exploiting diversity in time, frequency, and space as well as diversity with regard to users and services, efficient RM schemes can be designed for both UL and DL. For the DL, there has been considerable work recently on opportunistic scheduling and other RM schemes (130-136, 151-153, 155). For the UL, there are still many interesting issues to be resolved. For instance, there is a need for optimally exploiting the burstiness of data sources in a decentralized MAC scheme. This also leads to issues related to the fairness of resource allocation amongst competing data users. The adaptation in RM can also be performed based on mobile context awareness (e.g., application, location, mobility, environment).

Investigating adaptive RM schemes for efficient QoS support in multimedia wireless IP networks is another promising area. Future mobile IP networks will consist of multiple segments evolved from today’s wireless networks. The simplicity and scalability requirements may favor distributed, smart RM schemes. In addition, efficient RM schemes have to deal with the protocol overhead in supporting different QoS models and mobility management over IP.

Looking at the 4G (4th Generation) system concepts (231-232), e.g., supporting multimedia wireless IP with an air interface capacity of 100 Mb/s to 1 Gb/s and a good coverage by using MC-CDMA (Multi-Carrier CDMA) or OFDM (Orthogonal Frequency Division Multiplexing) with MIMO (Multiple Inputs Multiple Outputs) techniques at a 5GHz carrier frequency, the MAC layer of a high data rate system optimized for the transport of IP packets, even with enhanced mobility, is likely to be closer to today’s WLAN access methods than that of 2G or 3G cellular systems. The challenge is to find a hybrid that will correctly combine the best of both worlds: low access delay comparable with that of WLAN; stability in loaded conditions, guaranteed QoS and low terminal power consumption as that of cellular systems.

There are also many challenges ahead for efficient RM in WLAN. The recent introduction of UWB (Ultra Wide-Band) technology for IEEE 802.15 WPAN (Wireless Personal Area Networks) raises interesting possibilities for adaptive RM. The current research on UWB mostly concentrates on physical channel characterization, antenna and transceiver design (233). However, a comprehensive study on RM schemes, MAC in particular, is highly necessary, as the existing narrowband or wideband MAC schemes may not be suitable for UWB systems (234). For instance, since UWB is capable of recovering positional information with high precision, smart position of data users through RM can lead to better organization and performance of communication networks (235). Hence, designing MAC protocols for UWB systems will be an interesting topic.

Thus, the debates on simplicity/efficiency trade-offs of RM schemes for existing wireless networks and the needs of new RM schemes for future wireless networks continue. Furthermore, research and development for such concepts like our “green wireless networks” that care for living environments and human wellbeing will also be of great contribution.

References

1. Rappaport TS, Annamalai A, Buehrer RM & Tranter WH (2002) Wireless communications: past events and a future perspective. IEEE Communications Magazine, 40(5): 148-161.

2. Macker JP, Park VD & Corson MS (2001) Mobile and wireless Internet services: putting the pieces together. IEEE Communications Magazine, 39(6): 148-155.

3. European Union, Information Society Technologies - Wireless Strategic Initiative (IST-WSI) (2000) The book of visions 2000: visions of the wireless world [online]. Available from: http://www.ist-wsi.org/.

4. Wireless World Research Forum (WWRF) (2001) The book of visions 2001: visions of the wireless world [online]. Available from: http://www.wireless-world-research.org/.

5. McCann PJ & Hiller T (2000) An Internet infrastructure for cellular CDMA networks using mobile IP. IEEE Personal Communications, 7(4): 26-32.

6. Dixit S & Prasad R (2002) Wireless IP and Building the Mobile Internet. Artech House. 7. Noble D (1962) The history of land-mobile radio communications. IEEE Transactions on

Vehicular Technology, 11: 1406-1416. 8. McDonald VH (1979) The cellular concept. Bell System Technical Journal, 58: 15-41. 9. Lee WCY (1987) New cellular schemes for spectral efficiency. IEEE Transactions on

Vehicular Technology, 36: 188-192. 10. Lee WCY (1991) Overview of cellular CDMA. IEEE Transactions on Vehicular

Technology, 40: 291-302. 11. Pickholtz RL, Milstein LB & Schilling DL (1991) Spread spectrum for mobile

communications. IEEE Transactions on Vehicular Technology, 40: 313-322. 12. Jakes WC (1974) Microwave Mobile Communications. John Wiley and Sons, New York. 13. Steel R (1992) Mobile Radio Communications. John Wiley & Sons. 14. Pahlavan K & Levesque AH (1995) Wireless Information Networks. John Wiley & Sons. 15. Viterbi AJ (1995) Principle of Spread Spectrum Communication. Addison-Wesley. 16. Rappaport TS (1996) Wireless Communications: Principles and Practice. Prentice-Hall, New

Jersey. 17. Prasad R (1996) CDMA for Wireless Personal Communications. Artech House. 18. Glisic SG (1997) Spread Spectrum CDMA Systems for Wireless Communications. Artech

House. 19. Zander J & Kim SL (2001) Radio Resource Management for Wireless Networks. Artech

House. 20. Holma H & Toskala A (2002) WCDMA for UMTS Radio Access for Third Generation

Mobile Communications. John Wiley & Sons. 21. Glisic SG (2003) Adaptive WCDMA: Theory and Practice. John Wiley & Sons. 22. Perkins C (2001) Ad Hoc Networking. Addison-Wesley.

12023. Zorzi M & Rao RR (1997) Energy constrained error control for wireless channels. IEEE

Personal Communications, 4(6): 27-33. 24. Biglieri E, Caire G & Taricco G (1998) Coding and modulation under power constraints.

IEEE Personal Communications, 5(3): 32-39. 25. Bambos N (1998) Toward power-sensitive network architectures in wireless

communications: concepts, issues, and design aspects. IEEE Personal Communications, 5(3): 50-59.

26. Lorch J & Smith A (1998) Software strategies for portable computer energy management. IEEE Personal Communications, 5(3): 60-73.

27. Agrawal P (1998) Energy efficient protocols for wireless systems. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’98), 2:564-569.

28. Ephremides A (2002) Energy concerns in wireless networks. IEEE Wireless Communications, 9(4): 48-59.

29. Jung P, Baier PW & Sreil A (1993) Advantages of CDMA and spread spectrum techniques over FDMA and TDMA in cellular mobile applications. IEEE Transactions on Vehicular Technology, 42: 357-363.

30. Frullone M, Riva G, Grazioso P & Missiroli M (1994) Comparisons of multiple access schemes for personal communication systems in a mixed cellular environment. IEEE Transactions on Vehicular Technology, 43: 99-109.

31. Pahlavan K & Levesque AH (1994) Wireless data communications. Proceedings of the IEEE, 82: 1398-1430.

32. Steele R, Whitehead J & Wong WC (1995) System aspects of cellular radio. IEEE Communications Magazine, 33(1): 80-87.

33. Pandya R, Grillo D, Lycksell E, Mieybegue P, Okinaka H & Yabusaki M (1997) IMT-2000 standards: network aspects. IEEE Personal Communications, 4(4): 20-29.

34. Samukic A (1998) UMTS universal mobile telecommunications system: development of standard for the third generation. IEEE Transactions on Vehicular Technology, 47: 1099-1104.

35. Ojanperä T & Prasad R (1998) An overview of air-interface multiple access for IMT-2000/UMTS. IEEE Communications Magazine, 36(9): 82-86, 91-95.

36. Cimini LJ, Chuang JC-I & Sollenberger NR (1998) Advanced cellular Internet service (ACIS). IEEE Communications Magazine, 36(10): 150-159.

37. Schwartz M (1987) Telecommunication Networks: Protocols, Modeling and Analysis. Addison-Wesley.

38. Bertsekas D & Gallager R (1992) Data Networks. Prentice Hall. 39. Stevens R (1994) TCP/IP Illustrated, Volume 1: The Protocols. Addison-Wesley. 40. 3rd Generation Partnership Project (3GPP) (2005) TS 23.107 UMTS: Quality of Service

(QoS) concept and architecture [online]. Available from: http://www.3gpp.org/. 41. 3rd Generation Partnership Project (3GPP) (2003) TS 22.105 UMTS: Service aspects,

services and service capabilities [online]. Available from: http://www.3gpp.org/. 42. Internet Engineering Task Force (IETF) (1994) RFC 1633 Integrated services in the Internet

architecture: an overview [online]. Available from: http://www.ietf.org/. 43. Internet Engineering Task Force (IETF) (1998) RFC 2475 An architecture for differentiated

services [online]. Available from: http://www.ietf.org/. 44. Sklar B (1997) Rayleigh fading channels in mobile digital communication systems part i:

characterization. IEEE Commununications Magazine 35(7): 90-100. 45. Sklar B (1997) Rayleigh fading channels in mobile digital communication systems part ii:

mitigation. IEEE Commununications Magazine 35(7): 102-109. 46. Biglieri E, Proakis J & Shamai S (1998) Fading channels: information theoretic and

communication aspects. IEEE Transactions on Information Theory, 44: 2619-2692. 47. Guerin R, Ahmadi H & Naghshineh M (1991) Equivalent capacity and its application to

bandwidth allocation in high-speed networks. IEEE Journal on Selected Areas in Communications, 9: 968-981.

12148. Papavassiliou S & Tassiulas L (1995) The joint resource allocation problem in wireless

networks. Proc. IEEE International Conference on Communications (ICC’95), 1: 395-399. 49. Zander J (1997) Radio resource management in future wireless networks: requirements and

limitations. IEEE Commununications Magazine 35(8): 30-36. 50. Katzela I & Naghshineh M (1996) Channel assignment schemes for cellular mobile

telecommunication systems: a comprehensive survey. IEEE Personal Communications, 3(6): 10-31.

51. Adachi F, Sawahashi M & Okawa K (1997) Tree-structured generation of orthogonal spreading codes with different lengths for forward link DS-CDMA mobile radio. IEEE Electronic Letters, 33: 27-28.

52. Dinan EH & Jabbari B (1998) Spreading codes for direct sequence CDMA and wideband CDMA cellular networks. IEEE Communications Magazine, 36(9): 48–54.

53. Gold R (1967) Optimal binary sequences for spread spectrum multiplexing. IEEE Transactions on Information Theory, 13: 619-621.

54. Sarwate DV & Pursley MB (1980) Crosscorrelation properties of pseudorandom and related sequences. Proceedings of the IEEE, 68: 593-619.

55. Zander J (1992) Performance of optimum transmitter power control in cellular radio systems. IEEE Transactions on Vehicular Technology, 41: 57-62.

56. Zander J (1992) Distributed cochannel control in cellular radio systems. IEEE Transactions on Vehicular Technology, 41: 305-311.

57. Grandhi SA, Vijayan R, Goodman DJ & Zander J (1993) Centralized power control in cellular radio systems. IEEE Transactions on Vehicular Technology, 42: 466-468.

58. Zander J & Frodigh M (1994) Comment on “Performance of optimum transmitter power control in cellular radio systems.” IEEE Transactions on Vehicular Technology, 48: 636.

59. Yates RD (1995) A framework for uplink power control in cellular radio systems. IEEE Journal on Selected Areas in Communications, 13: 1341-1347.

60. Bambos N, Chen SC & Pottie GJ (1995) Radio link admission algorithm for wireless networks with power control and active link quality protection. Proc. IEEE INFOCOM (INFOCOM’95), 1: 97-104.

61. Wu Q (1999) Performance of optimum transmitter power control in CDMA cellular mobile systems. IEEE Transactions on Vehicular Technology, 48: 571-575.

62. Caire G, Taricco G & Biglieri E (1999) Optimum power control over fading channels. IEEE Transactions on Information Theory, 45: 1468-1489.

63. Viswanath P, Anantharam V & Tse D (1999) Optimal sequences, power control and capacity of synchronous CDMA systems with linear MMSE multiuser receivers. IEEE Transactions on Information Theory, 45: 1968-1983.

64. Goodman DJ & Mandayam NB (2000) Power control for wireless data. IEEE Personal Communications, 7(2): 48-54.

65. Bar-Noy A & Kessler I (1993) Tracking mobile users in wireless communications networks. IEEE Transactions on Information Theory, 39: 1877-1886.

66. Akyildiz IF & Ho JSM (1995) Dynamic mobile user location update for wireless PCS networks. ACM Journal of Wireless Networks, 1: 187-196.

67. Akyildiz IF, Ho JSM & Lin YB (1996) Movement-based location update and selective paging for PCS networks. IEEE/ACM Transactions on Networking, 4: 629-638.

68. Thomas R, Gilbert H & Mazziotto G (1988) Influence of the movement of the mobile station on the performance of a radio cellular network. Proc. 3rd Nordic Seminar on Digital Land Mobile Radio Communication, Copenhagen, Denmark, Paper 9.4.

69. Hong D & Rappaport SS (1989) Priority oriented channel access for cellular systems serving verhicular and portable radio telephones. IEE Proceedings, 136: 339-346.

70. Rappaport SS (1991) Modeling the hand-off problem in personal communications networks. Proc. IEEE Vehicular Technology Conference (VTC’91), 2: 517-523.

71. Xie H & Kuek S (1993) Priority handoff analysis. Proc. IEEE Vehicular Technology Conference (VTC’93), 2: 855-858.

72. Chuang JC (1993) Performance issues and algorithms for dynamic channel assignment. IEEE Journal on Selected Areas in Communications, 11: 955-963.

12273. Lin YB, Mohan S & Noerpel A (1994) Queueing priority channel assignment strategies for

PCS handoff and initial access. IEEE Transactions on Vehicular Technology, 43: 704-712. 74. Qing-An Z, Mukumoto K & Fukuda A (1994) Performance analysis of mobile cellular radio

system with priority reservation handoff procedures. Proc. IEEE Vehicular Technology Conference (VTC’94), 3: 1829-1833.

75. Tripathi ND, Reed JH & VanLandinoham HF (1998) Handoff in cellular systems. IEEE Personal Communications, 5(6): 26-37.

76. Ortigoza-Guerrero L & Aghvami AH (1999) A prioritized handoff dynamic channel allocation strategy for PCS. IEEE Transactions on Vehicular Technology, 48: 1203-1215.

77. Re ED., Fantacci R & Giambene G (1999) Handover queuing strategies with dynamic and fixed channel allocation techniques in low earth orbit mobile satellite systems. IEEE Transactions on Communications, 47: 89-102.

78. Dimitriou N, Tafazolli R & Sfikas G (2000) Quality of service for multimedia CDMA. IEEE Communications Magazine, 38(7): 88-94.

79. Dixit S, Gou Y & Antoniou Z (2001) Resource management and quality of service in third generation wireless networks. IEEE Communications Magazine, 39(2): 125-133.

80. Abramson N (1970) The ALOHA system - another alternative for computer communications. Proc. AFIPS Fall Conference Joint Computer Conference, 37: 281-285.

81. Abramson N (1994) Multiple access in wireless digital networks. Proceedings of the IEEE, 82: 1360-1370.

82. Sanchez J, Martinez R & Marcellin M (1997) A survey of MAC protocols proposed for wireless ATM. IEEE Network Magazine, 11(6): 52-62.

83. Prasad R, Dasilva JS & Arroyo-Fernandez B (1999) Air interface access schemes for wireless communications. IEEE Communications Magazine, 37(9): 104-105.

84. Progler M, Evci C & Umehira M (1999) Air interface access schemes for broadband mobile systems. IEEE Communications Magazine, 37(9): 106-115.

85. Akyildiz IF, McNair J, Carrasco L & Puigjaner R (1999) Medium access control protocols for multimedia traffic in wireless networks. IEEE Network Magazine, 13(4): 39-47.

86. Chandra A, Gummalla V & Limb JO (2000) Wireless medium access control protocols. IEEE Communications Surveys: 2-15.

87. Kleinrock L & Tobagi F (1975) Packet switching in radio channels: part I - Carrier Sense Multiple Access modes and their throughput-delay characteristics. IEEE Transactions on Communications, 23: 1400-1416.

88. Tobagi F & Kleinrock L (1975) Packet switching in radio channels: part II - the hidden-terminal problem in Carrier Sense Multiple Access and the busy-tone solution. IEEE Transactions on Communications, 23: 1417-1433.

89. Tobagi F & Kleinrock L (1976) Packet switching in radio channels: part III - polling and (dynamic) split-channel reservation multiple access. IEEE Transactions on Communications, 24: 832-845.

90. Abramson N (1985) Development of the ALOHANET. IEEE Transactions on Information Theory, 31: 119-123.

91. Wu C & Li VOK (1987) Receiver-initiated busy-tone multiple access in packet radio networks. Proc. ACM Workshop on Frontiers in Computer Communications Technology: 336-342.

92. Glisic S, Rao R & Milstein LB (1990) The effect of imperfect carrier sensing on nonpersistent carrier sense multiple access. Proc. IEEE SUPERCOMM International Conference on Communications (SUPERCOMM ICC'90), 3: 1266 -1269.

93. Glisic SG (1991) 1-persistent carrier sense multiple access in radio channels with imperfect carrier sensing. IEEE Transactions on Communications, 39: 458-464.

94. Huang JH & Kleinrock L (1992) Throughput analysis and protocol design for CSMA and BTMA protocols under noisy environments. IEE Proceedings, 139: 289-296.

95. Wu G, Mukumoto K & Fukuda A (1994) Performance evaluation of reserved idle signal multiple-access scheme for wireless communication networks. IEEE Transactions on Vehicular Technology, 43: 653-658.

12396. Amitay N (1992) Resource auction multiple access (RAMA): efficient method for fast

resource assignment in decentralised wireless PCS. IEEE Electronics Letters, 28: 799-801. 97. Pinheiro ALA & de Marca JRB (1996) Fair deterministic packet access protocol: F-RAMA

(fair resource assignment multiple access). IEEE Electronics Letters, 32: 2310-2311. 98. Goodman DJ, Valenzuela RA, Gayliard KT & Ramamurthi B (1989) Packet reservation

multiple access for local wireless communications. IEEE Transactions on Communications, 37: 885-890.

99. Mitrou NM, Orinos TD & Protonotarios EN (1990) A reservation multiple access protocol for microcellular mobile communication systems. IEEE Transactions on Vehicular Technology, 39: 340-351.

100. Wong WC & Goodman DJ (1993) Integrated data and speech transmission using packet reservation multiple access. Proc. IEEE International Conference on Communications (ICC'93), 1: 172-176.

101. Narasimhan P & Yates RD (1995) A new protocol for the integration of voice and data over PRMA. IEEE Journal on Selected Areas in Communications, 14: 623-631.

102. DeVile JM (1993) A reservation-based multiple access scheme for future Universal Mobile Telecommunications System. Proc. IEE European Conference on Mobile and Personal Communications (ECMPC’93): 210-215.

103. Qiu X & Li VOK (1996) Dynamic reservation multiple access (DRMA): a new multiple access scheme for personal communication systems (PCS), ACM Journal of Wireless Networks, 2: 117-128.

104. Bianchi G, Borgonovo F, Fratta L, Musumeci L & Zorzi M (1997) C-PRMA: a centralized packet reservation multiple access for local wireless communications. IEEE Transactions on Vehicular Technology, 46: 422-436.

105. Qiu X & Li VOK (1998) A unified performance model for reservation-type multiple-access schemes. IEEE Transactions on Vehicular Technology, 47: 173-189.

106. Brand BE & Aghvami AH (1998) Multidimensional PRMA with prioritized bayesian broadcast - a MAC strategy for multiservice traffic over UMTS. IEEE Transactions on Vehicular Technology, 47: 1148- 1161.

107. Qiu X & Chuang JC-I (1999) Data throughput enhancement in wireless packet systems by improved link adaptation with application to the EDGE system. Proc. IEEE Semiannual Vehicular Technology Conference (VTC’99-Spring), 1: 456-460.

108. Lu S, Bharghavan V & Srikant R (1999) Fair scheduling in wireless packet networks. IEEE/ACM Transactions on Networking, 7: 473-489.

109. Karol MJ, Eng KY & Liu Z (1995) An efficient demand-assignment multiple access protocol for wireless packet (ATM) networks, ACM Journal of Wireless Networks, 1: 269-279.

110. Karol MJ, Liu Z & Eng KY (1995) Distributed-queueing request update multiple access (DQRUMA) for wireless packet (ATM) networks. Proc. IEEE International Conference on Communications (ICC’95), 2: 1224-1231.

111. Passas N, Paskalis S, Vali D & Merakos L (1997) Quality of service oriented medium access control for wireless ATM networks. IEEE Communications Magazine, 35(11): 42-50.

112. Morrow RK & Lehnert JS (1989) Bit-to-bit error dependence in slotted DS/SSMA packet system with random signature sequences. IEEE Transactions on Communications, 37: 1052-1061.

113. Abdelmonem AH & Saadawi TN (1989) Performance analysis of spread spectrum packet radio network with channel load sensing. IEEE Journal on Selected Areas in Communications, 7: 161-166.

114. Yin M & Li VOK (1990) Unslotted CDMA with fixed packet lengths. IEEE Journal on Selected Areas in Communications, 8: 529-541.

115. Holtzman JM (1992) A simple, accurate method to calculate spread-spectrum multiple-access error probabilities. IEEE Transactions on Communications, 40: 461-464.

116. Toshimitsu K, Yamazato T, Katayama M & Ogawa A (1994) A novel spread slotted ALOHA system with CLSP. IEEE Journal on Selected Areas in Communications, 12: 665-672.

124117. Makrakis D & Murthy KMS (1992) Spread slotted ALOHA techniques for mobile and

personal satellite communication systems. IEEE Journal on Selected Areas in Communications, 10: 985-1002.

118. Sato T, Okada H, Yamazato T, Katayama M & Ogawa A (1996) Throughput analysis of DS/SSMA unslotted ALOHA system with fixed packet length. IEEE Journal on Selected Areas in Communications, 14: 750-756.

119. Sato T, Okada H, Yamazato T, Katayama M & Ogawa A (1996) Performance degradation due to access timing delay on CDMA unslotted ALOHA with channel load sensing. Proc. IEEE International Conference on Universal Personal Communications (ICUPC’96), 1: 111-114.

120. Judge G. & Takawira F (1998) Performance analysis of a channel load sensing protocol for CDMA packet radio networks with finite populations and variable length messages. Proc. IEEE International Symposium on Spread Spectrum Techniques and Applications (ISSSTA’98), 1: 282-286.

121. Glisic S & Vikstedt J (1998) Effect of wireless link characteristics on packet-level QoS in CDMA/CSMA networks. IEEE Journal on Selected Areas in Communications, 16: 875-889.

122. Wilson ND, Ganesh R, Joseph K & Raychaudhuri D (1993) Packet CDMA versus dynamic TDMA for multiple access in an integrated voice/data PCN. IEEE Journal on Selected Areas in Communications, 11: 870-884.

123. Zhang Z & Liu YJ (1993) Performance analysis of multiple access protocols for CDMA cellular and personal communication services. Proc. IEEE INFOCOM (INFOCOM'93), 3: 1214-1221.

124. Tan L & Zhang Q (1996) A reservation random-access protocol for voice/data integrated spread-spectrum multiple-access systems. IEEE Journal on Selected Areas in Communications, 14: 1717-1727.

125. Viterbi AJ (1994) Capacity of a simple and stable protocol for short message service over CDMA network. In: Blahut R (ed) Communication and Cryptography: 423-429.

126. Soroushnejad M & Geraniotis E (1995) Multi-access strategies for an integrated voice/data CDMA packet radio network. IEEE Transactions on Communications, 43: 934-945.

127. Pichna R & Wang Q (1996) A medium access control protocol for a cellular packet CDMA carrying multirate traffic. IEEE Journal on Selected Areas in Communications, 14: 1728-1736.

128. Akyildiz IF, Levine DA & Joe I (1999) A slotted CDMA protocol with BER scheduling for wireless multimedia networks. IEEE/ACM Transactions on Networking, 7: 146-158.

129. Sampath A & Holtzman JM (1997) Access control of data in integrated voice/data CDMA systems: benefits and tradeoffs. IEEE Journal on Selected Areas in Communications, 15: 1511-1526.

130. Parkvall S, Dahlman E, Frenger P & Persson M (2001) The high speed data evolution of WCDMA. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’01), 2: 27-31.

131. 3rd Generation Partnership Project (3GPP) (2004) TS 25.308 UMTS: HSDPA overall description [online]. Available from: http://www.3gpp.org/.

132. Nakamura M, Awad Y & Vadgama S (2002) Adaptive control of link adaptation for high speed downlink packet access (HSDPA). Proc. IEEE International Symposium on Wireless Personal Multimedia Communications (WPMC’02), 2: 382-386.

133. Kolding TE, Frederiksen F & Mogensen PE (2002) Performance aspects of WCDMA systems with high speed downlink packet access (HSDPA). Proc. IEEE Semiannual Vehicular Technology Conference (VTC’02-Fall), 1: 477-481.

134. Frederiksen F & Kolding TE (2002) Performance and modeling of WCDMA HSDPA transmission/H-ARQ schemes. Proc. IEEE Semiannual Vehicular Technology Conference (VTC’02-Fall), 1: 472-476.

135. Zhang J, Li L & Poon T (2002) Turbo coded HSDPA systems with transmit diversity over frequency selective fading channels. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’02), 1: 90-94.

125136. Das A, Gopalakrishnan N, Hu T, Khan F, Rudrapatna A, Sampath A, Su H, Tatesh S &

Zhang W (2003) Evolution of UMTS towards high-speed downlink packet access. Bell Labs Technical Journal, 7(3): 47-68.

137. Siew CK & Goodman DJ (1989) Packet data transmission over mobile radio channels. IEEE Transactions on Vehicular Technology, 38: 95-101.

138. Zorzi M, Rao RR & Milstein LB (1998) Error statistics in data transmission over fading channel. IEEE Transactions on Communications, 46: 1468-1477.

139. Chockalingam A & Zorzi M (1998) Energy efficiency of media access protocols for mobile data networks. IEEE Transactions on Communications, 46: 1418-1421.

140. Chockalingam A & Zorzi M (1998) Energy consumption performance of a class of access protocols for mobile data networks. Proc. IEEE Vehicular Technology Conference (VTC’98), 2: 820-824.

141. Chang LF (1991) Throughput estimation of ARQ protocols for a Rayleigh fading channel using fade- and interfade-duration statistics. IEEE Transactions on Vehicular Technology, 40: 223-229.

142. Hara S, Ogino A, Araki M, Okada M & Morinaga N (1996) Throughput performance of SAW-ARQ protocol with adaptive packet length in mobile packet data transmission. IEEE Transactions on Vehicular Technology, 45: 561-569.

143. Balakrishnan H, Padmanabhan VN, Seshan S & Katz RH (1997) A comparison of mechanisms for improving TCP performance over wireless links. IEEE/ACM Transactions on Networking, 5: 756-769.

144. Zorzi M (1998) Performance of FEC and ARQ error control in bursty channel under delay constraints. Proc. IEEE Vehicular Technology Conference (VTC’98), 3: 1390-1394.

145. Annamalai A & Bhargava VK (1998) Efficient ARQ error control strategies with adaptive packet length for mobile radio networks. Proc. IEEE International Conference on Universal Personal Communications (ICUPC'98), 2: 1247-1251.

146. Lettieri P & Srivastava MB (1998) Adaptive frame length control for improving wireless link range and energy efficiency. Proc. IEEE INFOCOM (INFOCOM’98), 2: 564-571.

147. Chien C, Srivastava MB, Jain R, Lettieri P, Aggarwal V & Sternowski R (1999) Adaptive radio for multimedia wireless links. IEEE Journal on Selected Areas in Communications, 17: 793-813.

148. Oh SJ & Wasserman KM (1999) Dynamic spreading gain control in multiservice CDMA networks. IEEE Journal on Selected Areas in Communications, 17: 918-927.

149. Lu Y & Brodersen RW (1999) Integrating power control, error correction coding, and scheduling for a CDMA downlink system. IEEE Journal on Selected Areas in Communications, 17: 978-989.

150. Chockalingam A & Bao G (2000) Performance of TCP/RLP protocol stack on correlated fading DS-CDMA wireless links. IEEE Transactions on Vehicular Technology, 49: 28-33.

151. Liu X, Chong EKP & Shroff NB (2001) Opportunistic transmission scheduling with resource-sharing constraints in wireless networks. IEEE Journal on Selected Areas in Communications, 19: 2053-2064

152. Holtzman JM (2001) Asymptotic analysis of proportional fair algorithm. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’01), 2: F33-F37.

153. Rajan D, Sabharwal A & Aazhang B (2001) Delay and rate constrained transmission policies over wireless channels. Proc. IEEE Global Telecommunications Conference (GLOBECOM'01), 2: 806-810.

154. Bambos N & Kandukuri S (2002) Power-controlled multiple access schemes for next-generation wireless packet networks. IEEE Wireless Communications, 9(3): 58-64.

155. Barriac G & Holtzman J (2002) Introducing delay sensitivity into the proportional fair algorithm for CDMA downlink scheduling. Proc. IEEE International Symposium on Spread Spectrum Techniques and Applications (ISSSTA’02), 3: 652-656.

156. Piazza D & Milstein LB (2002) Multiuser diversity-mobility tradeoff: modeling and performance analysis of a proportional fair scheduling. Proc. IEEE Global Telecommunications Conference (GLOBECOM'02), 1: 906-910.

126157. Kumaran K & Qian L (2003) Uplink scheduling in CDMA packet-data systems. Proc. IEEE

INFOCOM (INFOCOM’03), 1: 292-300. 158. Kulkarni SS & Rosenberg C (2003) Opportunistic scheduling policies for wireless systems

with short term fairness constraints. Proc. IEEE Global Telecommunications Conference (GLOBECOM'03), 1: 533-537.

159. Liu Y, Gruhl S & Knightly EW (2003) WCFQ: an opportunistic wireless scheduler with statistical fairness bounds. IEEE Transactions on Wireless Communications, 2: 1017-1028.

160. Gilhousen KS, Jacobs IM, Padovani R, Viterbi AJ, Weaver Jr. LA & Wheatley CE (1991) On the capacity of a cellular CDMA system. IEEE Transactions on Vehicular Technology, 40: 303-312.

161. Viterbi AM & Viterbi AJ (1993) Erlang capacity of a power controlled CDMA system. IEEE Journal on Selected Areas in Communications, 11: 892-900.

162. Viterbi AJ, Viterbi AM, Gilhousen KS & Zehavi E (1994) Soft handoff extends CDMA cell coverage and increases reverse link capacity. IEEE Journal on Selected Areas in Communications, 12: 1281-1288.

163. Abu-Dayya AA & Beaulieu NC (1994) Outage probabilities in the presence of correlated lognormal interferers. IEEE Transactions on Vehicular Technology, 43: 164-173.

164. Ariyavisitakul S (1994) Signal and interference statistics of a CDMA system with feedback power control II. IEEE Transactions on Communications, 42: 597-605.

165. Hanly SV (1996) Capacity and power control in spread spectrum macrodiversity radio networks. IEEE Transactions on Communications, 44: 247-256.

166. Goldsmith AJ & Varaiya PP (1997) Capacity of fading channels with channel side information. IEEE Transactions on Information Theory, 43: 1986-1992.

167. Hanly SV & Tse DNC (1998) Multiaccess fading channels II, delay-limited capacities. IEEE Transactions on Information Theory, 44: 2816-2831.

168. Lee CC & Steele R (1998) Effect of soft and softer handoffs on CDMA system capacity. IEEE Transactions on Vehicular Technology, 47: 830-841.

169. Everitt D (1994) Traffic engineering of the radio interface for cellular mobile networks. Proceedings of IEEE, 82: 1371-1382.

170. Ramjee R, Nagarajan R & Towsley D (1996) On optimal call admission control in cellular networks. Proc. IEEE INFOCOM (INFOCOM’96), 1: 43 –50.

171. Naghshineh M & Schwartz M (1996) Distributed call admission control in mobile/wireless networks. IEEE Journal on Selected Areas in Communications, 14: 711-717.

172. Levine DA, Akyildiz IF & Naghshineh M (1997). A resource estimation and call admission algorithm for wireless multimedia networks using the shadow cluster concept. IEEE/ACM Transactions on Networking, 5: 1-12.

173. Andersin M, Rosberg Z & Zander J (1997) Soft and safe admission control in cellular networks. IEEE/ACM Transactions on Networking, 5: 255-265.

174. Orlik PV & Rappaport SS (1998) A model for teletraffic performance and channel holding time characterization in wireless cellular communication with general session and dwell time distributions. IEEE Journal on Selected Areas in Communications, 16: 788-803.

175. Kulavaratharasah MD & Aghvami AH (1999) Teletraffic performance evaluation of microcellular PCNs with prioritized handoff procedures. IEEE Transactions on Vehicular Technology, 48: 137-152.

176. Lozano A & Cox DC (1999) Integrated dynamic channel assignment and power control in TDMA mobile wireless communication systems. IEEE Journal on Selected Areas in Communications, 17: 2031-2040.

177. Wu S, Wong KY & Li B (2002) A dynamic call admission policy with precision QoS guarantee using stochastic control for mobile wireless networks. IEEE/ACM Transactions on Networking, 10: 257-271.

178. Yang WB & Geraniotis E (1994) Admission policies for integrated voice and data traffic in CDMA packet radio networks. IEEE Journal on Selected Areas in Communications, 12: 654-664.

127179. Ishikawa Y & Umeda N (1997) Capacity design and performance of call admission control

in cellular CDMA systems. IEEE Journal on Selected Areas in Communications, 15: 1627-1635.

180. Wu C, Tsai Y & Chang J (1999) A quality-based birth-and-death queueing model for evaluating the performance of an integrated voice/data CDMA cellular system. IEEE Transactions on Vehicular Technology, 48: 83-89.

181. Evans JS & Everitt D (1999) Effective bandwidth-based admission control for multiservice CDMA cellular networks. IEEE Transactions on Vehicular Technology, 48: 36-46.

182. Liu Z & Zarki ME (1994) SIR-based call admission control for DS-CDMA cellular systems. IEEE Journal on Selected Areas in Communications, 12: 638-644.

183. Huang CY & Yates RD (1996) Call admission in power controlled CDMA systems. Proc. IEEE Vehicular Technology Conference (VTC’96), 3: 1665-1669.

184. Liu T & Silvester JA (1998) Joint admission/congestion control for wireless CDMA systems supporting integrated services. IEEE Journal on Selected Areas in Communications, 16: 845-857.

185. Kim D (2000) Efficient interactive call admission control in power-controlled mobile systems. IEEE Transactions on Vehicular Technology, 49: 1017-1028.

186. Ho CJ, Copeland JA, Lea CT & Stuber GL (2001) On call admission control in DS/CDMA cellular networks. IEEE Transactions on Vehicular Technology, 50: 1328-1343.

187. Jeon WS & Jeong DG (2002) Call admission control for CDMA mobile communications systems supporting multimedia services. IEEE Transactions on Wireless Communications, 1: 649-659.

188. Shin SM, Cho CH & Sung DK (1999) Interference-based channel assignment for DS-CDMA cellular systems. IEEE Transactions on Vehicular Technology, 48: 233-239.

189. Dziong Z, Ming J & Mermelstein P (1996) Adaptive traffic admission for integrated services in CDMA wireless access networks. IEEE Journal on Selected Areas in Communications, 14: 1737-1747.

190. Mitra D, Reiman MI & Wang J (1998) Robust dynamic admission control for unified cell and call QoS in statistical multiplexer. IEEE Journal on Selected Areas in Communications, 16: 692-707.

191. Grossglauser M & Tse D (1999) A framework for robust measurement-based admission control. IEEE/ACM Transactions on Networking, 7: 293-309.

192. Breslau L, Jamin S & Shenker S (2000) Comments on the performance of measurement-based admission control algorithms. Proc. IEEE INFOCOM (INFOCOM’00), 3: 1233-1242.

193. Tong H & Brown TX (2000) Adaptive call admission control under quality of service constraints: a reinforcement learning solution. IEEE Journal on Selected Areas in Communications, 18: 209-221.

194. Okumura Y, Ohmori E, Kawano T & Fukuda K (1968) Field strength and its variability in VHF and UHF land-mobile radio service. Review of the Electrical Communication Laboratory, 16: 9-10, 1968.

195. Hata M (1980) Empirical formula for propagation loss in land mobile radio services. IEEE Transactions on Vehicular Technology, 29: 317-325.

196. Gudmundson M (1991) Correlation model for shadow fading in mobile radio systems. IEE Electronics Letters, 27: 2145-2146.

197. Giancristofaro D (1996) Correlation model for shadow fading in mobile radio channels. IEE Electronics Letters, 32: 958-959.

198. Rice SO (1958) Distribution of the duration of fades in radio transmissions: Gaussian noise model. Bell System Technical Journal, 37: 581-635.

199. Bodtmann WF & Arnold HW (1982) Fade-duration statistics of Rayleigh distributed waves. IEEE Transactions on Communications, 30: 549-553.

200. Adachi F, Feeney MT & Parsons JD (1988) Effects of correlated fading on level crossing rates and average fade durations with predetection diversity reception. IEE Proceedings, 135: 11-17.

128201. Durgin GD & Rappaport TS (1999) Level-crossing rates and average fade duration for

wireless channels with spatially complicated multipath. Proc. IEEE Global Telecommunications Conference (GLOBECOM’99), 1: 427-431.

202. Dong X & Beaulieu NC (2002) Average level crossing rate and fade duration of maximal ratio diversity in unbalanced and correlated channels. Proc. IEEE Wireless Communication and Networking Conference (WCNC’02), 2: 762-767.

203. Zorzi M, Rao RR & Milstein LB (1995) On the accuracy of a first-order Markov model for data transmission on fading channels. Proc. IEEE International Conference on Universal Personal Communications (ICUPC'95): 211-215.

204. Wang HS & Chang PC (1996) On verifying the first-order markovian assumption for a Rayleigh fading channel model. IEEE Transactions on Vehicular Technology, 45: 353-357.

205. Turin W & Nobelen RV (1998) Hidden Markov modeling of flat fading channels. IEEE Journal on Selected Areas in Communications, 16: 1809-1817.

206. Zorzi M, Rao RR & Milstein LB (1996) A Markov model for block errors on fading channels. Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’96), 3: 1074-1078.

207. Tan CC & Beaulieu NC (2000) On first-order Markov modelling for the Rayleigh fading channel. IEEE Transactions on Communications, 48: 2032-2040.

208. Lai J & Mandayam NB (1998) Packet error rate for burst-error-correcting codes in Rayleigh fading channels. Proc. IEEE Vehicular Technology Conference (VTC’98), 2: 1568-1572.

209. Rahman MG, Rajatheva RMAP & Ahmed KM (1999) Performance of serial concatenated convolutional code in Rayleigh multipath fading channel. Proc. IEEE Semiannual Vehicular Technology Conference (VTC'99-Fall), 5: 2510-2514.

210. Damen MO, Abdi A & Kaveh M (2001) On the effect of correlated fading on several space-time coding and detection schemes. Proc. IEEE Semiannual Vehicular Technology Conference (VTC'01-Fall), 1: 13-16.

211. Papoulis A (1991) Probability, Random Variables and Stochastic Processes. McGraw Hill. 212. Kleinrock L (1975) Queueing Systems, I: Theory. John Wiley & Sons. 213. Gross D & Harris CM (1998) Fundamentals of Queuing Theory. John Wiley & Sons. 214. Kaufman JS (1981) Blocking in a shared resource environment. IEEE Transactions on

Communications, 29: 1474-1481. 215. Roberts JW (1981) A service system with heterogeneous user requirements. Performance of

Data Communications Systems and Their Applications, North-Holland: 423-431. 216. Kelly FP (1986) Blocking probabilities in larger circuit-switched networks. Advances in

Applied Probability, 18: 473-505. 217. Ross KW & Tsang DHK (1989) The stochastic knapsack problem. IEEE Transactions on

Communications, 37: 740-747. 218. Kelly FP (1991) Effective bandwidths at multi-class queues. Queueing Systems, 9: 5-16. 219. Elwalid AI & Mitra D (1993) Effective bandwidth of general Markovian traffic sources and

admission control of high-speed networks. IEEE/ACM Transactions on Networking, 1: 329-343.

220. Chung SP & Ross KW (1993) Reduced load approximations for multirate loss networks. IEEE Transactions on Communications, 41: 1222-1231.

221. Mitra D & Morrison JA (1994) Erlang capacity and uniform approximations for shared unbuffered resources. IEEE/ACM Transactions on Networking, 2: 558-570.

222. Bean NG, Gibbens RJ & Zachary S (1995) Asymptotic analysis of single resource loss systems in heavy traffic, with applications to integrated networks. Advances in Applied Probability, 27: 273-292.

223. Movaghar A (2000) On queueing with customer impatience until the end of service. Proc. IEEE International Computer Performance and Dependability Symposium (ICPDS’00): 167-174.

224. Puterman ML (1994) Markov Decision Processes: Discrete Stochastic Dynamic Programming. John Wiley & Sons.

129225. Zhang Z & Acampora AS (1991) Performance of a modified polling strategy for broadband

wireless LANs in a Harsh fading environment. Proc. IEEE Global Telecommunications Conference (GLOBECOM'91), 2: 1141-1146.

226. Haines RJ & Aghvami AH (1994) Indoor radio environment considerations in selecting a media access control protocol for wideband radio data communications. Proc. IEEE International Conference on Communications (ICC'94), 3: 1306-1311.

227. Acampora AS & Krishnamurthy SV (1997) A new adaptive MAC layer protocol for wireless ATM networks in harsh fading and interference environments," Proc. of IEEE International Conference on Universal Personal Communications (ICUPC'97), 2: 410-415.

228. Rao RR (1997) Perspectives on the impact of fading on protocols for wireless networks. Proc. IEEE International Conference on Personal Wireless Communications (ICPWC’97): 489-493.

229. Elliott EO (1963) Estimates of error rates for codes on burst-noise channels. Bell System Technical Journal, 42: 1977-1997.

230. Bhagwat P, Bhattacharya PK, Krishna A & Tripathi SK (1996) Enhancing throughput over wireless LANs using channel state dependent packet scheduling. Proc. IEEE INFOCOM (INFOCOM’96), 3: 1133-1140.

231. Evans BG & Baughan K (2000) Visions of 4G. IEEE Electronics & Communication Engineering Journal, 12(6): 293-303.

232. Hwang SK, Lyu DS & Chang K (2003) 4G vision and technology development in Korea. Proc. International Conference on Communication Technology (ICCT’03), 1: 26-27.

233. Hussain M (1998) Ultra-wideband impulse radar - an overview of the principles. IEEE Aerospace and Electronics Systems Magazine, 13(9): 9-14.

234. Cuomo F & Martello C (2001) MAC principles for an ultra-wideband wireless access," Proc. IEEE Global Telecommunications Conference (GLOBECOM'01), 6: 3548-3552.

235. DeNardis L, Baldi P & DiBenedetto M (2002) UWB ad-hoc networks. Proc. IEEE Conference on Ultra Wideband Systems and Technologies: 219-223.

236. The ns-2 Network Simulator (2005) [online]. Available from: http://www.isi.edu/nsnam/ns/. 237. OMNeT++ Discrete Event Simulation System (2005) [online]. Available from:

http://www.omnetpp.org/.

Appendix 1 Derivation of some formula

This section provides some detailed derivations of some formulas, which have not been included in the original papers of the author.

Expected number of retransmissions

This is related to Papers XVII and XVIII supporting SPA4 contribution, in which the formula of the expected number of retransmissions per successful radio packet, E[k], is given without detailed derivation.

The following provides a detailed derivation of the equations (9) and (21) in Paper XVIII, which help to calculate (3.18) and (3.20) above. Let us deal with (21) first for a more general case, and then, using either the same method or the results of (21) with approximations made for HSPA due to its short TTI, we obtain (9).

From (19) we have:

[ ] TmGGBG

kBBGB

TmG

TmGB

GBGBGB eTRTTRTTTePePPkP −−−− +−+= )()()()()1()( 2 ππππ ,

where the equilibrium state and state-transition probabilities are given by (1)-(4). The equation (20) becomes

∑∞

=

−=2

)(1)1(k

kPP

[ ]∑∞

=

−−−− +−+−=2

2 )()()()()1(1)1(k

TmGGBG

kBBGB

TmG

TmGB

GBGBGB eTRTTRTTTePePPP ππππ

[ ] ∑∞

=

−−−− +−+−=2

2 )()()()()1(1)1(k

kBB

TmGGBGGB

TmG

TmGB RTTeTRTTTePePPP GBGBGB ππππ

[ ])(1

1)()())(1(1)1(RTT

eTRTTTePPPPBB

TmGGBGGB

TmGGB

GBGB

ππππ

−−−+−= −−

[ ] TmGG

TmGGG

GBGB eTeTPP −−−−= )()(11)1( ππ .

For the equation (21) we have

132

[ ] ∑∞

=

=1

)(k

kkPkE

[ ] ∑∞

=

+=2

)()1(k

kkPPkE

[ ]

[ ]∑∞

=

−−−− +−+

+=

2

2 )()()()()1(

)1(

k

TmGGBG

kBBGB

TmG

TmGB

GBGBGB eTRTTRTTTePePPk

PkE

ππππ

[ ]

[ ] ∑∞

=

−−−− +−+

+=

2

2 )()()()()1(

)1(

k

kBB

TmGGBGGB

TmG

TmGB RTTkeTRTTTePePP

PkE

GBGBGB ππππ.

The sum at the end of the above equation, denoted by S, can be calculated as follows.

∑∞

=

−=2

2 )(k

kBB RTTkS π

∑∞

=

−=2

1 )()(

1k

kBB

BB

RTTkRTT

S ππ

⎥⎦

⎤⎢⎣

⎡ −= ∑∞

=

− 1)()(

11

1

k

kBB

BB

RTTkRTT

S ππ

⇔ ⎥⎦

⎤⎢⎣

⎡ −′= ∑∞

=

1))(()(

11k

kBB

BB

RTTRTT

S ππ

⎥⎥

⎦

⎤

⎢⎢

⎣

⎡−

′

⎟⎟⎠

⎞⎜⎜⎝

⎛−

= 1)(1

1)(

1RTTRTT

SBBBB ππ

⇔ ⎥⎦

⎤⎢⎣

⎡−

−= 1

))(1(1

)(1

2RTTRTTS

BBBB ππ

2))(1()(2

RTTRTTS

BB

BB

ππ

−−

= .

Now we have [ ]

[ ] [ ])(1)(2)()(1)()(1

1

RTTRTTeTeTPeTeTP

kE

BB

BBTmGG

TmGGG

TmGG

TmGGG

GBGBGBGB

ππππππ

−−

−+−

−=

−−−−

[ ] [ ])(1

1)()(11RTT

eTeTPkEBB

TmGG

TmGGG

GBGB

πππ

−−+= −− .

Using the formula of πBB(RTT) given by (4) for the above equation, we obtain the final formula of E[k] as given by (21):

[ ] [ ]RTTmm

G

TmGG

TmGGG

BGGB

GBGB

ePeTeTPkE )(1

1)()(11 +−

−−

−−

+=ππ .

Based on this result, we can easily obtain (9), either by carrying out a similar derivation or by applying the approximation πGG(T)exp(-mGBT) ≅ 1, resulting from that TTI ≡ T of 2

133

ms in HSPA is so small compared to the average duration of the good channel state ranging from about 100 ms to about 2000 ms.

Documents

Smart packet access and call admission control for ...jultika.oulu.fi/files/isbn9514276779.pdf · Phan, Vinh V., Smart packet access and call admission control for efficient resource