REFINED SYSTEM DESIGN PRINCIPLES FOR CONTEMPORARY DATA CELLULAR COMMUNICATION SYSTEMS

Transcription

1 REFINED SYSTEM DESIGN PRINCIPLES FOR CONTEMPORARY DATA CELLULAR COMMUNICATION SYSTEMS by Ahmed Alsohaly A thess submtted n conformty wth the requrements for the degree of Doctor of Phlosophy Department of Electrcal and Computer Engneerng Unversty of Toronto Copyrght 2015 by Ahmed Alsohaly

2 Abstract Refned System Desgn Prncples for Contemporary Data Cellular Communcaton Systems Ahmed Alsohaly Doctor of Phlosophy Department of Electrcal and Computer Engneerng Unversty of Toronto 2015 Cellular communcaton systems have transformed from systems deployed by a sngle moble network operator whle utlzng a sngle Rado Access Technology (RAT) to mult RAT, mult operator systems. owever, the nadequate applcaton of the fundamental desgn prncples of classcal cellular communcaton systems when deployng contemporary cellular communcaton systems leads to large system deployment costs, hgh system complexty, lmted system scalablty and suboptmal utlzaton of system resources. Refned system desgn prncples for contemporary cellular communcaton systems are developed and presented n ths thess to crcumvent the neffcences ntroduced by currently employed system desgn prncples, wth emphass on practcal system confguratons and deployment scenaros. The proposed enhanced system desgn prncples substantally mprove the performance of contemporary data cellular communcaton systems by explotng system multuser dversty and maxmzng the utlzaton of radofrequency resources.

3 After establshng the foundaton for contemporary cellular communcaton systems, attenton s shfted towards mult operator systems, n whch multple moble network operators share network nfrastructure elements to reduce system deployment costs. The feasblty of spectrum poolng n mult operator systems s thoroughly studed to determne the gans and lmtatons of ont resource allocaton under spectrum poolng, wth spectrum poolng verfed to maxmze the overall performance of mult operator systems. Mult RAT systems, n whch multple RATs are co deployed to accommodate varyng user equpment capabltes, are then consdered, wth emphass on enhancng neffcent system structurng and operaton, user access and spectrum management. Smlar to mult operator systems, ont resource allocaton wth autonomous spectrum assgnment at system access ponts s verfed to maxmze the performance of mult RAT systems. Fnally, ths thess addresses rregular system deployment scenaros subect to lmted multuser dversty. Specfcally, small cell access pont performance s enhanced usng a system aded dynamc spectrum access framework desgned to enable full spectrum utlzaton at all system cells.

4 Dedcaton To my father and my late uncle Mohammad. v

5 Acknowledgements Ths work was sponsored by Telus and the Mnstry of Educaton n Saud Araba. Words fal me n expressng my grattude to my famly for provdng me wth boundless love and support, and for bearng wth the selfsh decsons I have made over the course of the past fve years. I would lke to express my utmost grattude to my supervsor, Professor Elvno Sousa. The completon of ths work would not have been possble wthout hs contnuous support and gudance. I wll forever be ndebted to hs kndness, as I wll forever be nspred by hs dedcaton to hs students and passon for engneerng. I would also lke to express my grattude to members of my supervsory commttee: Professor Dmtrs atznakos and Professor Shahrokh Valaee, for ther support and feedback throughout my PhD. I am also grateful for my examnng commttee members: Professor Marcelo Alencar, Professor Ravra Adve, Professor Ashsh hst, Professor Deepa undur, Prof. Xaodong Ln and Professor Josh Taylor, for commttng the tme to serve on my commttee and for ther valuable feedback. Grattude s also due to Dr. Ivo Malevc and Dr. Adam Tenenbaum for ther techncal and personal support throughout my PhD. Specal thanks to Ms. Dane Slva, Mr. Vlad Crllo and the ECE Graduate Offce members: Ms. Darlene Gorzo, Ms. Judth Levene, Mr Shawn Mtchell and Ms. Gaa Sanmugaratnam, for ther contnuous support throughout my graduate studes. Specal thanks also for my dear frends and companons n my ourney at the Unversty of Toronto: Professor Waleed Alasmary, Engneer Fars Alfarhan, Dr. Sherf nawy and Dr. James Soldatc. I was truly blessed to have ther company! I would lke to extend my thanks as well to all my colleagues and frends at Chestnut Resdence for hostng and supportng me throughout my graduate studes. v

6 Table of Contents Dedcaton... v Acknowledgements... v Table of Contents... v Lst of Tables... x Lst of Fgures... x Lst of Appendces... xv Lst of Abbrevatons... xv Lst of Symbols... xx Chapter 1 Introducton and Background Classcal Cellular Communcaton Systems Contemporary Cellular Communcaton Systems Thess Organzaton and Contrbutons Adaptve Spectrum Poolng n Mult Operator Systems Mult Mode Access for Mult Mode UE n Mult Rado Access Technology Systems Autonomous spectrum assgnment n mult Rado Access Technology systems Unfed RAN structurng and operaton n mult Rado Access Technology systems System Aded Dynamc Spectrum Access for Small Cell Access Ponts Scope Rado Access Modelng n Data Cellular Communcaton Systems General System Setup Lnk Adaptaton v

7 1.5.3 Proportonal Far Schedulng Performance of Data Cellular Communcaton Systems under Proportonal Far Schedulng Proportonal Far Schedulng n Mult Carrer Data Cellular Communcaton Systems Rado Access Smulaton n Data Cellular Communcaton Systems Methodology System Layout and Antenna Model Propagaton Model User SINR and Throughput Rate Chapter 2 Mult Operator Systems: Spectrum Poolng Gans and Lmtatons Rado Resource Allocaton n RAN Sharng Mult Operator Systems Practcal Consderatons: Preventng Operator Performance Degradaton under Spectrum Poolng Spectrum Poolng Performance Evaluaton Spectrum Poolng Performance Adaptve Spectrum Poolng and Excess Loadng Tolerance Excess Loadng Tolerance Conclusons Chapter 3 Mult Rado Access Technology Systems: Enhanced User Access User Access n Mult Rado Access Technology Systems Practcal Consderatons: Contanng Sngle Mode User Equpment Performance Degradaton under Mult Mode Access Mult Mode Access Performance Evaluaton User Performance Cell Throughput Performance Overall System Performance v

8 3.4 Summary and Conclusons Chapter 4 Mult Rado Access Technology Systems: Enhanced Spectrum Management Spectrum Management n Mult Rado Access Technology Systems Practcal Consderatons: Enablng the Realzaton of Autonomous Spectrum Assgnment Spectrum Parttonng Spectrum Assgnment Adaptaton Rate Pooled Spectrum Blocks Assgnment Implementaton of the Autonomous Spectrum Management Functon Effect of Inter Cell, Inter Rado Access Technology Interference Autonomous Spectrum Assgnment Performance Evaluaton User Performance Cell Throughput Performance Overall System Performance Summary and Conclusons Chapter 5 Mult Rado Access Technology Systems: Enhanced Rado Access Network Structurng and Operaton Rado Access Network Structurng and Operaton n Mult Rado Access Technology Systems Practcal Consderatons: Enablng the Realzaton of Jont Resource Allocaton n Mult RAT Systems Jont Resource Allocaton Performance Evaluaton User Performance Cell Throughput Performance Overall System Performance Summary and Conclusons v

9 Chapter 6 Explotng Lmted Multuser Dversty: Enhanced Small Cell Spectrum Access Small Cell Access Pont Densfcaton Small Cell Spectrum Access Small Cell Deployment Strateges System Aded Small Cell Dynamc Spectrum Access Mult Rado Access Technology System Aded Dynamc Spectrum Access Mult Operator System Aded Dynamc Spectrum Access System Aded Dynamc Spectrum Access Performance Evaluaton Effect of User Offloadng on the Performance of Macro Cell Access Ponts Effect of Small Cell Access Pont Densfcaton on the Macro Cell Area Performance System Performance under SADSA Summary and Conclusons Chapter 7 Summary, Conclusons and Future Work Summary and Conclusons Future Work Refned Rado Access Network Topologes Refned Rado Access Technologes Optmzaton of Mult Ter Cellular Systems Integraton of Wreless Ad oc Access wth Cellular Access Bblography Appendx A: 3GPP Cellular Bands Appendx B: Canadan Assgnment of LTE Band Appendx C: LTE UE Classes x

10 Lst of Tables Table 1.1: Man Rado Access Network Functons... 4 Table 1.2: Transmsson parameters for 3GPP Rado Access Technologes... 9 Table 1.3: Alpha Shannon formula downlnk parameters for SPA and LTE Table 1.4: Mult path sgnal profle under the ITU VA channel model Table 6.1: Mult path sgnal profle under the ITU PA channel model Table A1: 3GPP cellular bands Table B1: Canadan Parttonng of LTE Band Table B2: Canadan Assgnment of LTE Band Table C3: LTE UE Classes x

11 Lst of Fgures Fgure 1.1: Illustraton of the cellular concept Fgure 1.2: Archtectural structure of cellular communcaton systems Fgure 1.3: Illustraton of the two level dstrbuted RAN topology Fgure 1.4: Illustraton of the flat dstrbuted RAN topology Fgure 1.5: Illustraton of the centralzed RAN topology Fgure 1.6: Archtectural structure of a Sngle Mode User Equpment Fgure 1.7: Functonal structure of a contemporary cellular communcaton system deployed by M moble network operators and employng I RATs Fgure 1.8: Spectral Effcency of Modulaton and Codng Schemes defned by the LTE Standard Fgure 1.9: LTE channel capacty models Fgure 1.10: System smulaton flowchart Fgure 1.11: System smulaton layout Fgure 1.12: AP antenna radaton pattern Fgure 1.13: A statc channel frequency response sample as defned by the ITU VA channel Fgure 1.14: A sample of wreless channel gan fluctuatons at 2 Gz, assumng a user velocty of 30 km/h, under the ITU VA channel model x

12 Fgure 2.1: Illustraton of network nfrastructure sharng confguratons n a mult operator system shared by M operators Fgure 2.2: Rado resource allocaton structurng n RAN sharng mult operator systems Fgure 2.3: Effect of system traffc varatons on the spectrum poolng gans for operator 1 when W = 2W Fgure 2.4: Effect of system traffc and total transmsson bandwdth varatons on the spectrum poolng gans for operator 1 when = Fgure 2.5: Effect of system traffc and total transmsson bandwdth varatons on the spectrum poolng gans for operator 1 when 1 = Fgure 2.6: Effect of system traffc and total transmsson bandwdth varatons on the overall spectrum poolng gans when 1 = 10 and M = Fgure 2.7: Effect of total transmsson bandwdth and system traffc varatons on the maxmum system loadng that mantans G1 > Fgure 3.1: Archtectural structure of a Mult Mode User Equpment capable of utlzng I Rado Access Technologes Fgure 3.2: Illustraton of Sngle Mode Access n a mult RAT system Fgure 3.3: Illustraton of Mult Mode Access n a mult RAT system Fgure 3.4: Illustraton of Restrcted Mult Mode Access n a mult RAT system.. 53 Fgure 3.5: Effect of system traffc varatons on the average SPA user throughput performance gans under SMA, MMA and RMMA Fgure 3.6: Effect of system traffc varatons on the average LTE user throughput performance gans under SMA, MMA and RMMA x

13 Fgure 3.7: Effect of system traffc varatons on the average SPA cell throughput performance gans under SMA, MMA and RMMA Fgure 3.8: Effect of system traffc varatons on the average LTE cell throughput performance gans under SMA, MMA and RMMA Fgure 3.9: Effect of system traffc varatons on the overall average cell throughput performance gans under SMA, MMA and RMMA Fgure 4.1: Illustraton of system level Fxed Spectrum Assgnment n mult RAT systems Fgure 4.2: Examples of traffc varatons and fluctuatons at dfferent cells and locatons n cellular communcaton systems Fgure 4.3: Illustraton of spectrum refarmng n mult RAT systems Fgure 4.4: Illustraton of Autonomous Spectrum Assgnment Fgure 4.5: Illustraton of the autonomous spectrum management functon Fgure 4.6: Illustraton of spectrum parttonng and assgnment under ASA Fgure 4.7: Illustraton of a centralzed mplementaton of the spectrum management functon n a mult RAT system co deployng four RATs Fgure 4.8: Illustraton of an alternatve centralzed mplementaton of the spectrum management functon n a mult RAT system co deployng four RATs.. 75 Fgure 4.9: Illustraton of the dstrbuted mplementaton of the spectrum management functon n a mult RAT system co deployng four RATs Fgure 4.10: Consdered Inter Cell Inter Rado Access Technology nterference scenaros Fgure 4.11: SPA system performance under IRI from LTE x

14 Fgure 4.12: LTE system performance under IRI from SPA Fgure 4.13: SPA system performance under IRI from GSM Fgure 4.14: LTE system performance under IRI from GSM Fgure 4.15: Effect of system traffc varatons on the average SPA user throughput performance gans under ASA and FSA Fgure 4.16: Effect of system traffc varatons on the average LTE user throughput performance gans under ASA and FSA Fgure 4.17: Effect of system traffc varatons on the average SPA cell throughput performance gans under ASA and FSA Fgure 4.18: Effect of system traffc varatons on the average LTE cell throughput performance gans under ASA and FSA Fgure 4.19: Effect of system traffc varatons on the overall average cell throughput performance gans under ASA and FSA Fgure 5.1: Dsont RAN structurng for a mult RAT system employng I RATs.. 88 Fgure 5.2: Unfed RAN structurng n a mult RAT system employng I RATs Fgure 5.3: Independent and ont rado resource allocaton structurng n mult RAT systems Fgure 5.4: Effect of system traffc varatons on the average SPA user throughput performance gans under JRA, IRA wth MMA and IRA wth SMA Fgure 5.5: Effect of system traffc varatons on the average LTE user throughput performance gans under JRA, IRA wth MMA and IRA wth SMA Fgure 5.6: Effect of system traffc varatons on the average SPA cell throughput performance gans under JRA, IRA wth MMA and IRA wth SMA xv

15 Fgure 5.7: Effect of system traffc varatons on the average LTE cell throughput performance gans under JRA, IRA wth MMA and IRA wth SMA Fgure 5.8: Effect of system traffc varatons on the overall average cell throughput performance gans under JRA, IRA wth MMA and IRA wth SMA Fgure 6.1: Illustraton of cell splttng n classcal cellular communcaton systems Fgure 6.2: Illustraton of Planned Small Cell Deployment AP densfcaton and Autonomous Small Cell Deployment AP densfcaton Fgure 6.3: An example of spectrum parttonng and assgnment under FFR n a cell cluster of three macro cells and 15 small cells Fgure 6.4: Flowchart for System Aded Dynamc Spectrum Access Fgure 6.5: Flowchart for Mult RAT System Aded Dynamc Spectrum Access Fgure 6.6: Flowchart for System Aded Mult Operator Dynamc Spectrum Access Fgure 6.7: A statc channel frequency response sample as defned by the ITU PA channel Fgure 6.8: A sample of wreless channel gan fluctuatons at 2 Gz, assumng a user velocty of 3 km/h, under the ITU PA channel model Fgure 6.9 Effect of user offloadng on the average LTE macro cell user performance for operator Fgure 6.10: Effect of user offloadng on the average LTE macro cell AP performance for operator Fgure 6.11: Effect of small cell AP densfcaton on the LTE macro cell area throughput for operator 1 under PSCD xv

16 Fgure 6.12: Effect of small cell AP densfcaton on the LTE macro cell area throughput for operator 1 under ASCD Fgure 6.13: Effect of ncreasng the small cell deployment densty on the offloadng capabltes of PSCD Fgure 6.14: Effect of ncreasng the small cell deployment densty on the offloadng capabltes of ASCD Fgure 6.15: Effect of ncreasng the SPA small cell deployment densty on MRDSA gans under PSCD Fgure 6.16: Effect of ncreasng the SPA small cell deployment densty on MRDSA gans under ASCD Fgure 6.17: Effect of ncreasng the small cell deployment densty of operator 1 on MODSA gans under PSCD Fgure 6.18: Effect of ncreasng the small cell deployment densty of operator 1 on MODSA gans under ASCD xv

17 Lst of Appendces Appendx A: 3GPP Cellular Bands Appendx B: Canadan Assgnment of LTE Band Appendx C: LTE UE Classes xv

18 Lst of Abbrevatons 3GPP AMC AP APC ASA ASCD ASP AWGN BER CDF CDMA CN CoMP CSI DRP DU EAP 3 rd Generaton Partnershp Proect Adaptve Modulaton and Codng Access Pont Access Pont Controller Autonomous Spectrum Assgnment Autonomous Small Cell Deployment Adaptve Spectrum Poolng Addtve Whte Gaussan Nose Bt Error Rate Cumulatve Dstrbuton Functon Code Dvson Multple Access Core Network Coordnated Mult Pont Channel State Informaton Dstrbuted Resource Parttonng Dgtal Unt Enhanced Access Pont xv

19 FD FDD FFR FR FSA GPS SPA d IRA IRI JRA LTE MCS MIMO MMA MMUE MODSA MRDSA NGMN NI Full Duplex Frequency Dvson Duplex Fractonal Frequency Reuse Fractonal Reuse Factor Fxed Spectrum Assgnment Global Postonng System gh Speed Packet Access ndependent and dentcally dstrbuted Independent Resource Allocaton Inter Cell Inter Rado Access Technology Interference Jont Resource Allocaton Long Term Evoluton Modulaton and Codng Scheme Multple Input Multple Output Mult Mode Access Mult Mode User Equpment Mult Operator Dynamc Spectrum Access Mult RAT Dynamc Spectrum Access Next Generaton Moble Networks Network Infrastructure xx

20 non RTF OTA PDF PFS PSCD RAN RAT RMMA RTF RU SADSA SINR SMA SMUE SOI SP TDD UE non Rado Transmsson Functon Over The Ar Probablty Densty Functon Proportonal Far Schedulng Planned Small Cell Deployment Rado Access Network Rado Access Technology Restrcted Mult Mode Access Rado Transmsson Functon Rado Unt System Aded Dynamc Spectrum Access Sgnal to Interference Plus Nose Rato Sngle Mode Access Sngle Mode User Equpment Spectrum Occupancy Informaton Spectrum Poolng Tme Dvson Duplex User Equpment xx

21 Lst of Symbols (,k,t) Path loss exponent Attenuaton factor of the employed RAT Wavelength correspondng to the transmsson frequency Maxmum spectral effcency of the employed RAT Standard devaton of a Normal random varable SINR of user k connectng to AP at tme t 0 Addtve Whte Gaussan Nose (AWGN) power spectral densty Attenuaton factor for RAT,k Phasor sum of multpath sgnal components receved by user k from AP,k Sgnal to Interference plus Nose Rato of user k at cell max SINR value requred to utlze the hghest performng AMC confguraton supported by the employed RAT mn SINR value requred to utlze the lowest performng AMC confguraton supported by the employed RAT A max Maxmum AP antenna attenuaton B B (TA n ) Number of pooled spectrum blocks Number of pooled blocks assgned to RAT durng pooled spectrum block assgnment nterval TA n xx

22 C Carrer bandwdth of RAT C k (n) schedulng tmeslot n Subset of system carrers allocated by Q(n) to user k (n) at d(,k,t) Dstance between AP and user k at tme t D user connectvty Transmsson bandwdth contnuously assgned to RAT to mantan d,k Dstance between AP and user k F X (x) F Y (y) f Y (y) F Z (z) CDF of user SINR CDF of achevable user throughput rate PDF of achevable user throughput rate CDF of the user SINR averaged over L g(θ,k (t)) AP antenna gan for user k havng an azmuth angle θ,k (t) wth respect to AP at tme t G ASA Gan n the long term average cell throughput rate of RAT due to the employment of ASA over FSA G JRA Gan n the long term average cell throughput rate of RAT due to the employment of JRA over IRA G MMA Gan n the long term average cell throughput rate of RAT due to the employment of MMA over SMA G RMMA Gan n the long term average cell throughput rate of RAT due to the employment of RMMA over SMA g,k Antenna gan of AP at the locaton of user k xx

23 G k ASA Gan n the long term average throughput rate of user k due to the employment of ASA over FSA G k JRA Gan n the long term average throughput rate of user k due to the employment of JRA over IRA G k MMA Gan n the long term average throughput rate of user k due to the employment of MMA over SMA G k RMMA Gan n the long term average throughput rate of user k due to the employment of RMMA over SMA G m ASP Gan n the long term average performance of operator m due to the employment of ASP over IRA G m JRA Gan n the long term average performance of operator m due to the employment of JRA over IRA g max Maxmum AP antenna gan G T ASA Gan n the overall long term average cell throughput rate due to the employment of ASA over FSA G T ASP Gan n the overall long term average cell throughput rate due to the employment of ASP over IRA G T JRA Gan n the overall long term average cell throughput rate due to the employment of JRA over IRA G T MMA Gan n the overall long term average cell throughput rate due to the employment of MMA over SMA G T RMMA Gan n the overall long term average cell throughput rate due to the employment of RMMA over SMA Multuser dversty factor of order xx

24 I' I' k Set of RATs wth ' '' Set of RATs ontly supported by the connectng AP and UE of user k satsfyng the condton ' '' J n Indcator of user satsfyng the PFS selecton crtera at tmeslot n '' Set of system users connected by operators wth G m JRA 1 (n) Total number of system users Set of system users chosen by schedulng Q(n) tmeslot n * Altered total number of system users * Altered number of system users utlzng RAT as the prmary mode of operaton ' '' Number of system users capable of utlzng RAT User support threshold for RAT operaton Number of system users utlzng RAT as the prmary mode of * (TA n ) Number of system users utlzng RAT as the prmary mode of operaton durng pooled spectrum block assgnment nterval TA n m Number of system users connected by operator m Max Excess loadng tolerance n a mult operator system L L F (,k,t) L P (,k,t) PFS averagng wndow sze Log normal shadowng loss between AP and user k at tme t Path loss between AP and user k at tme t xxv

25 M'' Set of operators wth G m JRA 1 M MC m Number of system operators n a mult operator system Set of actve macro cell APs deployed by operator m wthn the coverage area of SC m MC m Subset of MC m utlzng RAT M F (,k,t) tme t Channel gan due to multpath components between AP and user k at N F Nose fgure P,k Power receved by user k connectng to AP Pn Nose power P r (,k,t) power receved by user k from AP at tme t P t AP transmsson power P t (,t) Transmsson power of AP at tme t Q(n) schedulng tmeslot n Feasble schedulng for allocatng system radofrequency resources at r k Achevable throughput rate for user k r k (n) R k (n) Throughput rate achevable by user k at schedulng tmeslot n Runnng average throughput rate of user k at schedulng tmeslot n r k,l (n) Achevable throughput rate by user k at schedulng tmeslot n usng carrer component l C k (n) R m,k (n) Runnng average throughput rate of user k connected by operator m at schedulng tmeslot n xxv

26 S(n) SC m Schedulng allocatng system radofrequency resources at tmeslot n Set of actve small cell APs deployed by operator m wthn the coverage area of SC m SC m Subset of SC m utlzng RAT SC m Small cell AP of operator m utlzng RAT SC' ml Set of actve small cell APs wthn the coverage area of SC m havng secondary access to W ml SC' n Set of actve small cell APs wthn the coverage area of SC m havng secondary access to W n s,k Effect of shadow fadng on the power receved by user k from AP s k (n) T(n) Effectve user throughput rate of user k at schedulng tmeslot n Total cell throughput rate at schedulng tmeslot n TA mn Mnmal pooled spectrum block assgnment nterval T Transmsson frame duraton of RAT T m (n) u k (n) W W W' Cell throughput rate for operator m at schedulng tmeslot n PFS ndcator functon at schedulng tmeslot n Set of system spectrum blocks Total transmsson bandwdth ASA transmsson budget W'' Total transmsson bandwdth of operators wth G m JRA 1 W* Altered transmsson bandwdth assgned to RAT xxv

27 W B Pooled spectrum block bandwdth W W * (TA n ) Transmsson bandwdth assgned to RAT Total transmsson bandwdth assgned to RAT durng pooled spectrum block assgnment nterval TA n W m Set of system spectrum blocks assgned to operator m W m Transmsson bandwdth of operator m W m Subset of W m assgned to RAT W' m Unutlzed subset of W m W' m Subset of W' m accessble to SC m W' n Unutlzed subset of W n W' n Subset of W' n accessble to SC m θ 3dB AP antenna 3dB beamwdth xxv

28 Chapter 1 1 Chapter 1 Introducton and Background Cellular communcaton systems reuse avalable radofrequency resources at suffcently spaced wreless Access Ponts (APs), as Fgure 1.1 shows, to provde the requred system capacty under the constrant of lmted spectrum avalablty [1], [2]. Cells denote the coverage area of system APs and are typcally modelled by hexagons to smplfy the analyss of cellular communcaton systems [2]. The avalable system radofrequency resource are apportoned between groups of cells referred to as reuse clusters, wth the example of Fgure 1.1 llustratng a cellular system employng a reuse cluster consstng of three cells. Reducng the reuse cluster sze ncreases the area capacty of a cellular communcaton system, due to ncreased frequency reuse, at the expense of ncreasng the nterference levels at system cells [2], [8]. Therefore, the maxmum tolerable nterference level at a cellular system sets a lmt on the mnmum reuse cluster sze, wth state of the art cellular communcaton systems reducng the reuse cluster sze to a sngle cell [9], [10]. f 1 f 1 f 1 f 1 f 1 f 2 f 2 f 2 f 2 f 2 f 3 f 3 f 3 f 3 f 3 f 1 f 1 f 1 f 1 f 1 f 2 f 2 f 2 f 2 f 2 f 3 f 3 f 3 f 3 f 3 Fgure 1.1: Illustraton of the cellular concept.

29 Chapter Classcal Cellular Communcaton Systems Specfc radofrequency bands, lsted n Appendx A and referred to as cellular bands, are set asde for wreless cellular communcatons by spectrum regulatory authortes of dfferent ursdctons, such as Industry Canada n Canada and the Federal Communcatons Commsson n the Unted States of Amerca. As Appendx A shows, cellular bands may be contguous or non contguous and are typcally dvded nto sub bands that may also be contguous or non contguous when assgned to moble network operators, as llustrated by Appendx B. The terms spectrum and radofrequency resources are nterchangeably used to refer to the aggregate bands utlzed by a wreless cellular communcaton system. Upon obtanng a spectrum utlzaton lcense, moble network operators deploy the Network Infrastructure (NI) of cellular communcaton systems to wrelessly connect User Equpment (UE) employed by system users. The NI of a cellular communcaton system s dvded nto a Rado Access Network (RAN) and a Core Network (CN) as Fgure 1.2 shows [2], [3]. RANs facltates UE access to system radofrequency resources, usng a wreless rado nterface referred to as the Rado Access Technology (RAT) [3], and carry user traffc to the CN through backhaul lnks. The CN nterconnects cellular systems wth other communcaton systems and prmarly conssts of core lnks and data routers [2], [3]. UE RAT RAN Backhaul Lnks CN Core Lnks Other Systems NI Fgure 1.2: Archtectural structure of cellular communcaton systems.

30 Chapter 1 3 Rado access networks cover specfc geographcal areas by deployng fxed APs to mantan the connectvty of moble system users [1] [6]. Rado access network elements consst of processng unts and rado equpment such as antennas, crcuts and connectng cables that perform both Rado Transmsson Functons (RTFs) and non Rado Transmsson Functons (non RTFs) as specfed by Table 1.1. Common RAN topologes are detaled n Fgures The two level dstrbuted RAN topology, comprsng APs and Access Pont Controllers (APCs), s shown n Fgure 1.3. APs perform rado transmsson functons and connect to APCs, where non rado transmsson functons are performed and user traffc s transported to the CN through backhaul lnks [4], [5]. Access pont controllers connect to the CN and each APC typcally connects a group of APs, wth the number of APs connectng to an APC determned by the processng capabltes of the APC hardware and the performance of lnks between APs and the APC. Furthermore, APCs connectng adacent groups of APs are nterconnected to mplement RAN functons spannng multple APs such as user connecton handover. Access pont controller functonalty s embedded nto APs to form Enhanced APs (EAPs) that perform all RAN functons n the flat RAN topology [6], [12]. EAPs thus drectly connect to the CN and adacent EAPs are nterconnected to mplement RAN functons spannng multple EAPs as Fgure 1.4 shows. On the other hand, the centralzed RAN topology abstracts APs to Rado Unts (RUs) for wreless sgnal transmsson and recepton only whle connectng to centralzed Dgtal Unts (DUs) performng all remanng RAN functons [7], [13], [14]. Smlar to the two level dstrbuted RAN topology, DUs connect to the CN, wth DUs connectng adacent RUs nterconnected to mplement RAN functons spannng multple RUs as Fgure 1.5 shows. owever, unlke the two level dstrbuted RAN topology, the number of RUs connectng to a centralzed DU s typcally sgnfcantly larger than the number of APs connectng to an APC. The choce of RAN topology s entaled by the performance of avalable network nfrastructure hardware and qualty of lnks between network nfrastructure elements [7].

31 Non Rado Transmsson Functons Chapter 1 4 Table 1.1: Man Rado Access Network Functons [5]. Wreless sgnal transmsson and recepton. Rado Transmsson Functons Modulaton and demodulaton of carrer waveforms. Baseband sgnal processng. UE/AP synchronzaton n frequency and tme. System Access Control Rado Resource Management System nformaton broadcast. Admsson of authentcated system users. Facltaton of user access to system radofrequency resources. User Moblty Management User connecton transfer between dfferent rado channels and system APs. Pagng system users to contact the RAN. System user locaton postonng.

32 Chapter 1 5 Backhaul Lnk CN APC APC APC Lnk APC AP APC Lnk AP AP AP AP AP AP Fgure 1.3: Illustraton of the two level dstrbuted RAN topology (Source: [4], [5]). CN Backhaul Lnk EAP EAP EAP EAP EAP EAP Lnk EAP EAP Fgure 1.4: Illustraton of the flat dstrbuted RAN topology (Source: [6]).

33 Chapter 1 6 Backhaul Lnk CN DU DU DU Lnk DU RU DU Lnk RU RU RU RU RU RU Fgure 1.5: Illustraton of the centralzed RAN topology (Source: 13). Fgure 1.6 shows the archtectural structure of UE, employed by system users to wrelessly connect to system APs. The connectvty capabltes of UE, such as the supported RATs, Modulaton and Codng Schemes (MCSs), number of supported bands and maxmum transmsson bandwdth, are determned by the UE transcever module, wth UE cost typcally ncreasng wth the connectvty capabltes of the employed transcever module [8], [99]. The UE transcever module thus determnes the RATs and bands that can be utlzed by UE when connectng to system APs and, as Appendx C specfes, sets an upper bond on the maxmum UE achevable throughput rate.

34 Chapter 1 7 Applcaton Processor Power IC Memory Sensors and/or Perpherals Applcaton Software Communcaton Control Software Baseband Software Battery Communcaton Software Communcaton Processor Radofrequency Processor ardware Analog Front End Transcever Module Fgure 1.6: Archtectural structure of a Sngle Mode User Equpment, capable of utlzng a sngle Rado Access Technology (Source [69]).

35 Chapter 1 8 The amount of user traffc that can be carred by a cellular communcaton system,.e. the capacty of a cellular communcaton system, s determned by the system AP densty,.e. area frequency reuse, transmsson bandwdth and spectral effcency of employed RAT [2]. To crcumvent lmted spectrum avalablty and restrctons on system AP densfcatons, RATs are typcally desgned wth the obectve of achevng maxmal spectral effcency whle enablng full spectrum reuse at all system APs. State of the art RATs thus reuse system radofrequency resources at all system APs and new RATs, wth mproved spectral effcency, are contnuously developed and deployed by contemporary cellular communcaton systems [9] [12]. Rado access technologes are characterzed by the transmsson bandwdth, transmsson frame duraton, frequency reuse factor between system APs, user multple access scheme, modulaton and codng confguratons along wth the sgnalng and transmsson protocols [8] [10]. Table 1.2 specfes the transmsson parameters for the 3 rd Generaton Partnershp Proect (3GPP) RATs: Global System for Moble (GSM), gh Speed Packet Access (SPA) and Long Term Evoluton (LTE).

36 Chapter 1 9 Table 1.2: Transmsson parameters for 3GPP Rado Access Technologes [8] [10]. RAT GSM SPA LTE Frequency Reuse Factor (FR) Carrer Bandwdth Frame Duraton 3, 7, kz 5 Mz 1.4, 3, 5, 10, 15 and 20 Mz ms 10 ms 10 ms Number of Carrers (FR = 3), 7 (FR = 7) and 4 5 (FR = 12) at 10 Mz Up to 8 wth Mult Carrer support Up to 5 wth Carrer Aggregaton Transmsson Power (Downlnk) Up to 43 dbm/carrer Up to 43 dbm/carrer <10 Mz: Up to 43 dbm/carrer 10 Mz: Up to 46 dbm/carrer Transmsson Power (Uplnk) Up to 33 dbm/carrer Up to 21 dbm/carrer Up to 23 dbm/carrer Multple Access (Downlnk) TDMA, tme slot duraton = s WCDMA, chp rate = 3.84 Mz, spreadng factor = 16 OFDMA, subcarrer spacng = 15 kz, FFT sze = Multple Access (Uplnk) TDMA, tme slot duraton = s WCDMA, chp rate = 3.84 Mz, spreadng factor = 2256 SC FDMA, subcarrer spacng = 15 kz, FFT sze = Modulaton (Downlnk) GMS, delay bandwdth product = 0.3 QPS, 16 QAM and 64 QAM QPS, 16 QAM and 64 QAM Modulaton (Uplnk) GMS, delay bandwdth product = 0.3 QPS and 16 QAM QPS, 16 QAM and 64 QAM

37 Chapter Contemporary Cellular Communcaton Systems Whle the desgn and structurng of classcal cellular communcaton systems assumes system deployment by a sngle moble network operator whle utlzng a sngle RAT to connect system users, contemporary cellular systems are ontly deployed by multple moble network operators whle utlzng multple RATs to connect system users [33], [53] [55], [69]. Sharng the NI by multple moble network operators, through the deployment of mult operator systems, s prmarly drven by system deployment cost reducton [53]. On the other hand, the large varaton n UE capabltes compels the co deployment of multple RATs to mantan user connectvty n contemporary cellular communcaton systems [33], [69]. Nevertheless, based on the nadequate applcaton of the fundamental desgn prncples of classcal cellular communcaton systems, contemporary cellular communcaton systems mantan ndependent structurng and operaton of employed RATs for all system operators as Fgure 1.7 shows. The structurng and operaton of contemporary cellular systems as ndependent sngle RAT, sngle operator subsystems ntroduces a multtude of neffcences n user access, spectrum management and resource allocaton that result n the suboptmal utlzaton of system resources as shall be dscussed n the followng chapters. Operator 1, RAT 1 Operator 2, RAT 1 Operator M, RAT 1 Operator 1, RAT 2 Operator 2, RAT 2 Operator M, RAT 2 Operator 1, RAT I Operator 2, RAT I Operator M, RAT I Fgure 1.7: Functonal structure of a contemporary cellular communcaton system deployed by M moble network operators and employng I RATs.

38 Chapter Thess Organzaton and Contrbutons Chapter 1 descrbes the cellular concept, defnes classcal and contemporary cellular communcaton systems, models data cellular communcaton systems under Proportonal Far Schedulng (PFS) and specfes the smulaton framework for evaluatng the performance of data cellular communcaton systems. The followng system desgn prncples are ntroduced n the remanng chapters of ths thess to mprove the performance of contemporary cellular communcaton systems Adaptve Spectrum Poolng n Mult Operator Systems Chapter 2 consders mult operator systems, n whch moble network operators share NI elements to reduce system deployment costs. In partcular, focus n Chapter 2 s centered on mult operator systems deployng a common RAN; as the deployment of a common RAN enables poolng operator spectrum and unfyng the system rado resources allocaton. owever, poolng unequal amounts of spectrum, combned wth msmatched operator traffc condtons, may degrade operator performance under ont resource allocaton wth spectrum poolng. Nevertheless, performance evaluaton studes of spectrum poolng mult operator systems manly assume symmetrc spectrum poolng, n whch operators pool equal amounts of spectrum, under matched operator traffc demand [57] [59], [104]. A thorough analyss of spectrum poolng mult operator systems, for varous spectrum poolng confguratons and traffc varaton scenaros, s performed n Chapter 2 to determne spectrum poolng gans and lmtatons n practcal mult operator system deployment scenaros. Smlar to [32], Adaptve Spectrum Poolng (ASP) s adopted as a system desgn prncple for mult operator systems to prevent operator performance degradaton n spectrum poolng mult operator systems. The feasblty of spectrum poolng n mult operator systems s verfed by the wde

39 Chapter 1 12 bonds on spectrum poolng gans even under asymmetrc spectrum poolng and msmatched operator traffc condtons Mult Mode Access for Mult Mode UE n Mult Rado Access Technology Systems Mult RAT systems, n whch multple RATs are co deployed to accommodate varyng UE capabltes, are consdered n Chapters 3 5, wth Chapter 3 focusng on enhancng user access n mult RAT systems. Independent RAT operaton n mult RAT systems restrcts UE capable of utlzng multple RATs, referred to as Mult Mode UE (MMUE), to utlzng a sngle RAT when connectng to system APs. MMUE thus utlze the hghest performng RAT ontly supported by the connectng AP and the MMUE transcever module under Sngle Mode Access (SMA), leadng to the suboptmal utlzaton of system radofrequency resources and MMUE hardware capabltes. owever, efforts for enhancng MMUE connectvty n mult RAT systems are lmted to enablng the opportunstc shftng of the utlzed RAT when connectng to system APs to mprove the performance of MMUE under SMA, [70], [71], [102], [103], Chapter 3 develops Mult Mode Access (MMA), n whch all RATs ontly supported by UE and connectng APs are smultaneously utlzed to connect UE, as a system desgn prncple to mprove user connectvty and spectrum utlzaton n mult RAT systems. Provdng MMUE wth access to addtonal system radofrequency resources substantally mproves the performance of MMUE under MMA. Furthermore, MMA mproves the overall system performance by enhancng the system multuser dversty of employed RATs. Smlar to ASP, Restrcted Mult Mode Access (RMMA) s ntroduced n Chapter 3 to contan the degradaton n the performance of UE capable of utlzng a sngle RAT only, referred to as Sngle Mode UE (SMUE), when enablng MMA. RMMA s verfed to mantan consstent performance for SMUE at the expense of reducng MMA performance gans.

40 Chapter Autonomous spectrum assgnment n mult Rado Access Technology systems The employment of multple RATs n a cellular communcaton system, under ndependent RAT operaton, entals parttonng the system radofrequency resources between co deployed RATs. Due to ndependent RAT deployment and operaton, spectrum parttonng between co deployed RATs s typcally appled at the system level,.e. the same spectrum parttonng between co deployed RATs s appled at all system APs [33]. owever, traffc fluctuatons and varatons between system APs result n unbalanced system loadng and suboptmal spectrum utlzaton n mult RAT systems under fxed system level spectrum assgnment and constrans the gans of system level spectrum management technques [33], [63] [68]. Chapter 4 presents an Autonomous Spectrum Assgnment (ASA) framework, n whch spectrum assgnment at each system AP s autonomously adapted to AP traffc condtons, as a fundamental system desgn prncple to contan traffc varatons and fluctuatons n mult RAT systems. Whle matchng spectrum assgnment to system traffc condtons mproves the performance of RATs wth ncreasng traffc demand at the expense of degradng the performance of RATs wth declnng traffc demand, ASA s verfed to mprove the overall performance of mult RAT systems as the user densty of hgher performng RATs ncreases. Therefore, ASA enables supportng RATs wth declnng traffc demand wthout underutlzng system radofrequency resources. The requrements and provsons for enablng the realzaton of ASA n mult RAT systems are also detaled n Chapter 4.

41 Chapter Unfed RAN structurng and operaton n mult Rado Access Technology systems Structurng RANs based on ndependent RAT deployment and operaton results n the redundant duplcaton of RAN functons and elements n mult RAT systems. Ineffcent RAN structurng n mult RAT systems thus leads to large system deployment costs and operatonal complexty, lmted system scalablty and non trval mplementaton of ont mult RAT functons. Unfcaton of RAN structurng and operaton s proposed n Chapter 5 as a system desgn prncple to enable the ont optmzaton of non rado transmsson RAN functons whle addressng the shortcomngs of ndependent RAT deployment and operaton n mult RAT systems. The unfcaton of non rado transmsson RAN functons under unfed RAN structurng elmnates the redundant duplcaton of RAN functons and elements, thus sgnfcantly reducng system deployment costs and operatonal complexty. Furthermore, the scalablty of mult RAT systems s substantally mproved by ontly enhancng RAN functons for all RATs and mnmzng the system ntegraton requrements for ntroducng new RATs. ASA s exploted under unfed RAN structurng to enable the ont allocaton of system radofrequency resources for all employed RATs at each system AP. By maxmzng multuser dversty n mult RAT systems, JRA s verfed to enhance the performance of all employed RATs and users whle maxmzng the overall system performance System Aded Dynamc Spectrum Access for Small Cell Access Ponts Whle Chapters 2 5 explot system multuser dversty to enhance the performance of contemporary cellular communcaton systems under conventonal system deployment scenaros, Chapter 6 consders rregular system deployment

42 Chapter 1 15 scenaros subect to lmted multuser dversty, wth focus on enhancng spectrum utlzaton at small cell APs. In addton to specfyng small cell spectrum access, the performance and offloadng capabltes of dfferent small cell AP densfcaton strateges s evaluated and studed. Lmtng small cell spectrum access to a small subset of the avalable system radofrequency resources s dentfed as the prmary cause for suboptmal spectrum utlzaton at small cells. A small cell System Aded Dynamc Spectrum Access (SADSA) framework s ntroduced n Chapter 6 to enable full spectrum utlzaton at small cells n mult RAT, mult operator deployment envronment. In the presented SADSA framework, System APs broadcast Spectrum Occupancy Informaton (SOI) to allow small cell APs wth lmted capabltes to determne accessble system radofrequency resources whle crcumventng the lmtatons of spectrum sensng and spectrum access coordnaton [77] [92]. Rather than lmtng small cell AP spectrum access to a subset of the system radofrequency resources assgned to a sngle RAT of a sngle operator, small cell APs access unutlzed spectrum of all operators and RATs based on AP SOI broadcast. In addton to enablng full spectrum utlzaton at small cell APs wth lmted connectvty and backhaul capabltes, SADSA provdes sgnfcant system performance gans and enables the full realzaton of the potental of small cell AP densfcaton. As detaled n the concludng chapter (Chapter 7), the contrbutons of ths thess mprove the performance of contemporary cellular communcaton systems by enhancng the system structural effcency and spectrum utlzaton wthout alterng the system AP densty, RAT spectral effcency or system transmsson bandwdth. To buld on the fndngs of ths thess, future research drectons are dscussed n Chapter 7. The results, fndngs and contrbutons of ths thess have been publshed n [33] [41] and fled n [45] [52], wth [42] [44] submtted for publcaton.

43 Chapter Scope Analyses and performance evaluatons of data cellular communcaton systems throughout ths thess consders system downlnk transmssons,.e. transmsson lnks from system APs to UE, n Frequency Dvson Duplex (FDD) systems, where a separate frequency band s assgned for system downlnk, as detaled n Appendx A. The results and fndngs are applcable to uplnks n FDD systems along wth Tme Dvson Duplex (TDD) and Full Duplex (FD) systems, n whch the same frequency band s utlzed for system downlnk and uplnk transmssons. Furthermore, sngle antenna and sngle AP transmsson confguratons are consdered n ths work, wth the results and fndngs expandable to antenna beamformng, Multple Input Multple Output (MIMO) and Coordnated Mult Pont (CoMP) transmsson schemes. In addton, UE are assumed to be capable of utlzng all MCSs supported by the employed RAT at the hghest supported transmsson bandwdth and the Full Buffer traffc model s adopted throughout ths thess, wth the obtaned results and fndngs adaptable to dfferent UE classes and traffc types. The utlzaton of a sngle, contnuous transmsson band s assumed throughout ths thess, wth all results and fndngs applcable to non contguous transmsson scenaros. 1.5 Rado Access Modelng n Data Cellular Communcaton Systems Numerous approaches, wth varous degrees of abstracton, have been developed to model cellular communcaton systems [15] [23]. In ths secton, the modelng structure adopted by [15], [16], [20] and [30] s expanded to provde a modelng framework for data cellular communcaton systems utlzng Proportonal Far Schedulng (PFS), wth emphass on quantfyng the effect of enhancng system multuser dversty and spectrum utlzaton.

44 Chapter General System Setup A cellular communcaton system comprsng J hexagonal cells s consdered, wth a sngle system AP connectng unformly dstrbuted statonary users at each cell. System users are assumed to employ UE utlzng a sngle sotropc antenna whle contnuously generatng nfntely backlogged, delay tolerant traffc. The power receved by user k at cell, P,k, under AP transmsson power P t s modeled by, k, k, k, k 2 / d. P P g s (1.1) t, k n whch,k s the phasor sum of multpath sgnal components receved by user k from AP ; g,k s the antenna gan of AP at the locaton of user k; s,k represents the effect of shadow fadng on the power receved by user k from AP and s typcally modelled by a zero mean Log Normal random varable wth standard devaton ; s the wavelength correspondng to the transmsson frequency; d,k s the dstance between AP and user k; s the path loss exponent. The dstrbuton of P r,k s thus prmarly determned by the channel multpath propagaton profle; as s a fxed parameter and g,k, s,k and d,k are statc at any user locaton. Based on the nose power, P n, the Sgnal to Interference plus Nose Rato (SINR) of user k at cell,,k, s equal to P, k P, k, k. (1.2) P P P n y y, k y y, k n whch the effect of the nose power on the SINR s overlooked n (1.2) snce the nose power s typcally sgnfcantly lower than the nterference power [24]. Smlar to the receved user power, the dstrbuton of user SINR s prmarly determned by the multpath propagaton profle. Multpath fadng channels between users and system APs are assumed to be ndependent and dentcally dstrbuted (d), resultng n an d user SINR dstrbuton.

45 Chapter Lnk Adaptaton Rado access technologes typcally defne multple MCSs, as Fgure 1.8 shows, to explot varyng channel condtons at cellular communcaton systems [8] [10]. The transmsson MCS s consstently adapted based on user SINR fluctuatons, usng Adaptve Modulaton and Codng (AMC), to maxmze user throughput rate whle mantanng the requred Bt Error Rate (BER) [8] [10]. Under AMC, the achevable throughput rate for user k, r k, s approxmated by the Alpha Shannon formula, llustrated n Fgure 1.9 and defned as follows [25] 0, k mn r k W log 2 (1 k ), mn k max. (1.3) W, k max where s the attenuaton factor for the employed RAT; s the maxmum spectral effcency of the employed RAT; W s the transmsson bandwdth; mn s the SINR value requred to utlze the lowest performng AMC confguraton supported by the employed RAT; max s the SINR value requred to utlze the hghest performng AMC confguraton supported by the employed RAT. The Alpha Shannon formula parameters for SPA and LTE are provded at Table 1.3. Assumng k fluctuatons range between mn and max, the normalzed user throughput rate can be wrtten as rk k log 2 1 (1.4) Table 1.3: Alpha Shannon formula downlnk parameters for SPA and LTE [25]. RAT (b/s/z) mn (db) max (db) SPA db 20 db LTE db 17 db

46 Throughput, bts per second per z Chapter MCS-1 [QPS,R=1/8] MCS-2 [QPS,R=1/5] MCS-3 [QPS,R=1/4] MCS-4 [QPS,R=1/3] MCS-5 [QPS,R=1/2] MCS-6 [QPS,R=2/3] MCS-7 [QPS,R=4/5] MCS-8 [16 QAM,R=1/2] MCS-9 [16 QAM,R=2/3] MCS-10 [16 QAM,R=4/5] MCS-11 [64 QAM,R=2/3] MCS-12 [64 QAM,R=3/4] MCS-13 [64 QAM,R=4/5] Shannon SNR, db Fgure 1.8: Spectral Effcency of Modulaton and Codng Schemes defned by the LTE Standard (Source: 3GPP [25]). Fgure 1.9: LTE channel capacty models (Source: 3GPP [25]).

47 Chapter Proportonal Far Schedulng Achevable user throughput rates n data cellular systems are subect to sgnfcant varatons and fluctuatons; due to varyng user proxmty to connectng APs and the rapdly changng wreless channel gan [26]. Data cellular systems thus utlze opportunstc schedulng technques for rado resource allocaton to explot the sgnfcant varatons n achevable user throughput rates [20], [22]. Specfcally, Proportonal Far Schedulng (PFS) [27] s adopted n data cellular systems to balance the compromse between maxmzng the overall system performance and attanng farness between system users [15] [17], [28] [30]. Such balance s acheved by allocatng system radofrequency resources to users that maxmze a tme varyng utlty functon of the rato between achevable and averaged user throughput rates [27]. Proportonal far schedulng thus dvdes the tme doman nto tmeslots of equal duraton, wth tmeslot duraton chosen to be shorter than the channel fluctuaton rate to ensure quas statc channel condtons at all tmeslots [28]. Relable estmaton of r k s assumed to be avalable at the connectng AP; as the user Channel State Informaton (CSI) reportng frequency exceeds the channel fluctuaton rate by choce of transmsson tmeslot duraton. Allocaton of system radofrequency resources at tmeslot n s determned by the PFS selecton crtera J n n n rk arg max (1.5) {1,..., } Rk n whch J n {1,, } denotes the user satsfyng the PFS selecton crtera at tmeslot n; r k (n) s the throughput rate achevable by user k at tmeslot n; R k (n) s the runnng average throughput rate of user k at tmeslot n, defned as n 1 R k n rk u k. (1.6) L nl1